Finite Element Method With Applications in Engineering

Finite Element Method with Applications in Engineering
i
PREFACE
The revolution in computer technology has led to the development of numerous
computational techniques. The closed form solutions are either extremely difficult to obtain
or not available for many engineering problems. Hence one has to resort to numerical
methods. The Finite Difference Method (FDM), Finite Element Method (FEM), Boundary
Element Method (BEM) and Mesh Free Method are few examples of important numerical
techniques, which are available in the field of computational mechanics. Each of these
methods has its own merits and demerits. The choice of an appropriate method for the
solution of any particular problem depends on investigator's familiarity of the method and
nature of the problem. The aim of this book is to introduce FEM as a general numerical
technique for the solution of various engineering problems.
The main emphasis in this book is given to important applications in engineering and
science. Since majority of the applications of FEM are in the realm of mechanics including
solids, fluids and structural systems, descriptions are primarily in terms of these fields of
study. This book is written in a simple structured way starting with the basics of various
FEM approaches, a description of finite element interpolation functions, FEM formulations
and applications in one-, two- and three-dimensional domains and further advanced
applications for a few complex problems. Further, some of the commonly used FEM
software and web resources available for the solution of various engineering and science-
related problems are described briefly.
An illustrative and simple computer program, sfeap, with a sophisticated Graphical User
Interface for pre-processing and post-processing has been provided for structural and
solid mechanics related problems. A few simple MATLAB and FORTRAN-based programs
for fluid mechanics related problems in one- and two-dimensional domains are also given.
In addition, source codes of the programs are provided so that a reader can easily
understand the computer implementation of the method and modify the codes to integrate
their own elements and routines.
One may take a physical or intuitive approach to learning FEM. On the other hand, one
may develop a rigorous mathematical interpretation of the method. A middle path has
been adopted in this book with more importance to the physical approach. Efforts have
been made to keep the mathematics simple. We feel that the book will be useful for a
basic understanding of FEM by beginners, further understanding by FEM practitioners,
and to a certain extent for the advanced level understanding by researchers.
We encourage readers to convey feedback and suggestions for the improvement of the
contents of the book, and point out any necessary corrections in it.
Y. M. Desai
T. I. Eldho
A. H. Shah
ii
Contenido
i
PREFACE ........................................................................................................................... ii
1 Introduction ..................................................................................................................... 9
1.1 INTRODUCTORY REMARKS .................................................................................. 9
1.2 MATHEMATICAL MODELLING OF ENGINEERING PROBLEMS ............................ 9
1.3 TYPE OF GOVERNING EQUATIONS .................................................................... 12
1.4 SOLUTION METHODOLOGIES ............................................................................. 14
1.4.1 Analytical Method ............................................................................................. 15
1.4.2 Physical Method ............................................................................................... 15
1.4.3 Computational Method ..................................................................................... 15
1.5 NUMERICAL MODELLING..................................................................................... 15
1.6 PRE-PROCESSING AND POST-PROCESSING ................................................... 16
1.7 SCOPE OF THE BOOK.......................................................................................... 18
1.8 HIGHLIGHTS OF THE BOOK ................................................................................ 19
1.9 HOW TO USE THE BOOK? ................................................................................... 19
1.10 CLOSING REMARKS ........................................................................................... 20
REFERENCES AND FURTHER READING.................................................................. 20
2 Approximate Methods of Analysis ................................................................................. 22
2.1 INTRODUCTION .................................................................................................... 22
2.2 APROXIMATING METHODS ................................................................................. 22
2.3 METHOD OF WEIGHTED RESIDUALS ................................................................. 23
2.4 RAYLEIGH–RITZ METHOD ................................................................................... 27
2.5 FURTHER NUMERICAL EXAMPLES .................................................................... 29
2.6 CLOSING REMARKS ............................................................................................. 35
EXERCISE PROBLEMS .............................................................................................. 35
3 Finite Element Method—An Introduction ....................................................................... 37
3.1 GENERAL .............................................................................................................. 37
3.2 WHAT IS FEM? ...................................................................................................... 37
3.3 HOW DOES FEM WORK? ..................................................................................... 38
3.4 A BRIEF HISTORY OF FEM .................................................................................. 40
3.5 FEM APPLICATIONS ............................................................................................. 41
3.6 MERITS AND DEMERITS OF FEM ........................................................................ 42
Merits of FEM ........................................................................................................... 42
Demerits of FEM ....................................................................................................... 43
iii
3.7 CLOSING REMARKS ............................................................................................. 43

4 Different Approaches in FEM ........................................................................................ 45
4.1 INTRODUCTION .................................................................................................... 45
4.2 GENERAL STEPS OF FEM ................................................................................... 45
4.3 DIFFERENT APPROACHES USED IN FEM .......................................................... 50
4.3.1 Direct Approach ............................................................................................... 50
4.3.2 Variational Approach ........................................................................................ 52
4.3.3 Energy Approach ............................................................................................. 56
4.3.4 Weighted Residual Approach ........................................................................... 57
4.4 CLOSING REMARKS ............................................................................................. 60
5 Finite Elements and Interpolation Functions .................................................................. 63
5.1 INTRODUCTION .................................................................................................... 63
5.2 INTERPOLATION FUNCTIONS ............................................................................. 64
5.2.1 One-Independent Spatial Variable ................................................................... 64
5.2.2 Two-Independent Spatial Variables .................................................................. 65
5.2.3 Three-Independent Spatial Variables ............................................................... 65
5.3 ONE-DIMENSIONAL ELEMENTS .......................................................................... 65
5.3.1 Line Element: Linear Interpolation Function ..................................................... 65
5.3.2 Quadratic Interpolation Function ...................................................................... 69
5.3.3 Cubic Interpolation Function ............................................................................. 71
5.3.4 Lagrangian Form of Interpolation Function ....................................................... 72
5.3.5 Further Higher Order Elements in One-Dimension ........................................... 74
5.4 TWO-DIMENSIONAL ELEMENTS ......................................................................... 79
5.4.1 Triangular Element: Linear Interpolation Function in Cartesian Co-ordinates ... 79
5.4.2 Triangular Element—Area Co-ordinates........................................................... 83
5.4.3 Integration Formula for Triangular Elements .................................................... 83
5.4.4 Triangular Element—Quadratic Function ......................................................... 84
5.4.5 Triangular Element—Cubic Interpolation Function ........................................... 86
5.4.6 Two-Dimensional Rectangular Elements .......................................................... 87
5.4.7 Rectangular elements—Lagrangian Form in natural and Cartesian co-ordinates
................................................................................................................................. 89
5.4.8 Isoparametric elements .................................................................................... 91
5.4.9 Lagrangian Interpolation Functions for two-Dimensional elements ................... 93
iv
5.4.10 Two-Dimensional serendipity elements .......................................................... 96

5.5 THREE-DIMENSIONAL ELEMENTS...................................................................... 99
5.5.1 Tetrahedral Elements ..................................................................................... 100
5.5.2 Tetrahedral elements: Quadratic Interpolation Function ................................. 104
5.5.3 Tetrahedral elements: cubic Interpolation Function ........................................ 104
5.5.4 Three-Dimensional elements—prismatic elements......................................... 104
5.5.5 Three-Dimensional elements in local co-ordinates ......................................... 107
5.5.6 Three-Dimensional serendipity elements ....................................................... 108
5.6 CLOSING REMARKS ........................................................................................... 109
EXERCISE PROBLEMS ............................................................................................ 109
REFERENCES AND FURTHER READING................................................................ 110
6 One Dimensional Finite Element Analysis ................................................................... 112
6.1 INTRODUCTION .................................................................................................. 112
6.2 LINEAR SPRING .................................................................................................. 112
6.2.1 Expressions for Equivalent Spring Constant and Nodal Forces ...................... 123
6.3 TRUSS ELEMENT ............................................................................................... 125
6.3.1 Plane Truss .................................................................................................... 125
6.3.2 Element Equations by Minimizing Potential Energy ........................................ 128
6.3.3 Local and Global Element Equations for a Bar in the X–Y Plane .................... 134
6.4 SPACE TRUSS .................................................................................................... 140
6.5 ONE-DIMENSIONAL TORSION OF A CIRCULAR SHAFT .................................. 147
6.6 ONE-DIMENSIONAL STEADY STATE HEAT CONDUCTION ............................. 150
6.7 ONE-DIMENSIONAL FLOW THROUGH POROUS MEDIA.................................. 153
6.8 ONE-DIMENSIONAL IDEAL FLUID FLOW THROUGH PIPES (INVISCID FLUID
FLOW) ....................................................................................................................... 157
6.9 BEAM ELEMENT ................................................................................................. 157
6.9.1 Review of Beam Theory ................................................................................. 158
6.9.2 Finite Element Formulation of a Beam Element ............................................. 160
6.9.3 Illustrative Examples ...................................................................................... 165
7 Two Dimensional Finite Element Analysis ................................................................... 213
7.1 INTRODUCTION .................................................................................................. 213
7.3 TWO-DIMENSIONAL STRESS ANALYSIS .......................................................... 222
7.3.1 Review of Theory of Elasticity ........................................................................ 222
7.3.2 Application of Three-Dimensional Equations for Two-Dimensional analysis ... 223
7.3.3 CST Element for Plane Stress and Plane Strain Analyses ............................. 225
7.3.4 Triangular Element for Axisymmetric Analysis ................................................ 232
v
7.3.5 Some remarks on Triangular Elements .......................................................... 234

7.3.6 Four-Node Rectangular Element for Plane Problems ..................................... 235
7.4 ISOPARAMETRIC FORMULATION ..................................................................... 238
7.4.1 Two-Node Isoparametric Line Element (Bar Element) .................................... 238
7.4.2 Four-Node Isoparametric Element for Plane Problems (Quadrilateral Element)
............................................................................................................................... 241
7.5 FINITE ELEMENT SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS BY
METHOD OF WEIGHTED RESIDUAL ....................................................................... 249
7.6 FEM FORMULATION BASED ON VARIATIONAL PRINCIPLE ............................ 252
7.7 FINITE ELEMENT SOLUTION OF STOKES FLOW EQUATIONS ....................... 254
7.7.1 Problem Statement ........................................................................................ 254
7.7.2 FEM Solution ................................................................................................. 255
7.8 ILLUSTRATIVE EXAMPLES ................................................................................ 260
8 Three Dimensional Finite Element Analysis ................................................................ 274
9 Computer Implementation of FEM............................................................................... 302
9.3 STATIC CONDENSATION ................................................................................... 304
9.4 COMPUTER IMPLEMENTATION OF FEM–SFEAP ............................................. 308
9.5 STORAGE SCHEMES FOR GLOBAL STRUCTURAL STIFFNESS MATRIX ...... 309
10 Further Applications of Finite Element Method .......................................................... 319
10.3 DYNAMICS WITH FINITE ELEMENT METHOD ................................................ 339
10.3.1 Introduction .................................................................................................. 339
10.3.2 Governing Equations .................................................................................... 339
10.4.3 Illustrative Examples .................................................................................... 349
10.4 NON-LINEAR ANALYSIS ................................................................................... 355
10.4.1 Finite Element Formulation for Non-Linear Analysis ..................................... 356
10.4.2 Solution of Non-Linear Equations ................................................................. 357
10.4.3 Illustrative Examples .................................................................................... 358
10.6 HYDRODYNAMICS SIMULATION OF SHALLOW WATER FLOW .................... 364
10.6.1 Introduction .................................................................................................. 364
10.6.2 Governing Equations and Boundary Conditions ............................................... 365
10.6.3 Finite Element Formulation .......................................................................... 366
10.7 FEM–SOFTWARE AND WEB RESOURCES ..................................................... 372
10.7.1 Introduction .................................................................................................. 372
10.7.2 FEM in Structural Engineering...................................................................... 372
vi
10.7.3 FEM in Geotechnical Engineering ................................................................ 374

10.7.4 FEM in Fluid Mechanics ............................................................................... 375
10.7.5 FEM in Thermal and Automobile Engineering .............................................. 375
10.7.6 FEM in Physics ............................................................................................ 376
10.7.7 Multi-Field FEM Software ............................................................................. 376
10.7.8 FEM in Other Fields ..................................................................................... 377
10.8 CONCLUDING REMARKS ................................................................................. 377
Appendix A .................................................................................................................... 381
A.1 REVIEW OF MATRIX ALGEBRA ......................................................................... 381
A.2 MATRIX CALCULUS............................................................................................ 386
A.3 SOLUTION OF LINEAR SIMULTANIOUS EQUATIONS ...................................... 387
Appendix B .................................................................................................................... 394
B.1 GENERAL ............................................................................................................ 394
B.2 THE EULER–LAGRANGE EQUATION ................................................................ 394
EXAMPLE B.1: ....................................................................................................... 397
B.3 THE VARIATIONAL OPERATOR......................................................................... 398
EXAMPLE B.2 ........................................................................................................ 399
Appendix C .................................................................................................................... 401
C.1 GENERAL............................................................................................................ 401
C.2 EXAMPLE ............................................................................................................ 401
Appendix D .................................................................................................................... 406
D.1 GENERAL............................................................................................................ 406
D.2 FUNDAMENTAL CONCEPT ................................................................................ 406
D.3 GAUSS QUADRATURE FOR TWO DIMENSION ................................................ 411
Appendix E .................................................................................................................... 414
E.1 PREAMBLE ......................................................................................................... 414
E.2 PROGRAM LAYOUT ........................................................................................... 415
E.2.1 Global Variables and Arrays .......................................................................... 416
E.2.2 Notes on Element Subroutines ELMTn .......................................................... 419
E.2.3 Input Instructions ........................................................................................... 421
E.3.3 EXAMPLES ................................................................................................... 431
E.3.4 Listing of SFEAP............................................................................................ 442
Appendix F..................................................................................................................... 503
F.1 PROGRAMS ........................................................................................................ 503
F.1.1 Preamble ....................................................................................................... 503
vii
F.2 EXECUTING PROGRAMS ................................................................................... 504

1. CONTROL INFORMATION................................................................................. 504
2. CO-ORDINATES ................................................................................................ 505
3. ELEMENT DATA ................................................................................................ 507
4. BOUNDARY CONDITIONS AND NODAL DISPLACEMENT DATA .................... 510
5. MATERIAL CHARACTERISTIC DATA ............................................................... 511
6. NODAL LOAD DATA AND DISTRIBUTED LOAD ............................................... 511
Appendix G .................................................................................................................... 515
G.1 COMPUTER PROGRAM FOR THE FLOW NETWORK ANALYSIS .................... 515
G.2 COMPUTER PROGRAM FOR THE ELECTRICAL NETWORK ANALYSIS ........ 518
G.3 COMPUTER PROGRAM FOR THE LAPLACE EQUATION – HOMOGENEOUS
AND NON-HOMOGENEOUS MEDIA PROBLEMS .................................................... 520
G.3.1 Example 1: .................................................................................................... 524
G.3.2 Example 2: .................................................................................................... 526
ACKNOWLEDGEMENTS .............................................................................................. 532
viii
1 Introduction
1.1 INTRODUCTORY REMARKS
Most practical engineering problems are complex in nature. A model is often needed to
solve a system consisting of any domain of interest and various phenomena that take
place in it. A model may be defined from the engineering point of view as the selected
simplified version of an engineering system that approximately simulates the system's
excitation response relations, which are of interest to the user. Information obtained from
the model can be used to understand the system, to provide information for design
purposes, to provide information to comply with regulations and to, finally, take appropriate
management decisions to solve particular problem concerned.
Modelling can be done through physical means with laboratory and field experiments or
through mathematical means. Physical modelling in laboratory or field can be
cumbersome, expensive and time consuming. Due to these difficulties in physical
modelling in laboratories or in the real world, use of mathematics in solving the
engineering problems has become widespread in the recent times. Use of systems
approach, increasing power of digital computers and the development of numerous
computational methodologies have made the mathematical modelling inevitable for all
branches of learning.
1.2 MATHEMATICAL MODELLING OF ENGINEERING PROBLEMS

Mathematical modelling of an engineering problem is a process by which the problem is
represented as it appears in the real world and interpreted in an abstract form. Depending
on the modelling approach, different mathematical models are possible for tackling the
same problem. Also, identical mathematical model may be used to tackle quite different
physical situations.
Generally, basic elements of modelling can be represented by five components: state
variables, external variables, mathematical equations, coefficients or parameters and
universal constants. State variables describe the state of the system that is to be
modelled. External variables, or the forcing functions, on the other hand, are variables
external in nature that can influence the system. Mathematical equations are relations
between the external variables and the state variables. Mathematical representation of
process in any system contains coefficients or parameters and universal constants.
Calibration is often required in modelling to find the best parameters in accordance with
the computed and the observed state variables by variation of number of parameters.
Modelling procedure and methodology depends on the nature of the problem, aims and
objectives of the modelling. Different workable modelling procedures are possible. A
typical modelling procedure is shown in Figure 1.1. Following steps are involved in a
typical mathematical modelling procedure.
Step 1 - Problem Definition: As the first step, it is essential to define the problem clearly
with aims and objectives. Sources of facts and data should be known.
Step 2 - Problem Simplification: As the real world problems are very complex to any
model, various assumptions may be used to make the problem solvable with sufficient
accuracy.
Step 3 - Problem Solution in Space, Time and Subsystems: Once the simplified form
of the problem is available, decision can be made for a three-dimensional, two-
9
dimensional or one-dimensional modelling to obtain the required solution. Need for time-
dependent (dynamic) or time-independent (static) modelling should also be ascertained.
Step 4 - Data Requirement: Data collection and handling may be simple or complex,
depending on the problem to be solved. Availability and quality of the available data
should be checked before proceeding further with modelling. Accuracy of the model
prediction depends on the quality of the available data.
Step 5 - Conceptual Model: Finally, a conceptual model for the problem should be
evolved from the above steps. The conceptual model would tend to reduce the real
problem and the real domain to simplified versions, which are satisfactory in view of the
objectives of the modelling, available resources and management of the problem.
Step 6 - Mathematical Model: Conceptual model can be expressed in the form of a
mathematical model. Mathematical model, generally, consists of the definition of domain,
the governing equations (algebraic, ordinary differential equations, partial differential
equations or integral equations), initial conditions (known state of the system at some
initial time) and boundary conditions. For most engineering problems, mathematical
representations are possible in terms of equations, depending on varying conditions.
Step 7 - Solution Methodology: Solution methodology should be decided once the
mathematical model is ready. Depending on the problem and its complexity, solution
methodology can be analytical, numerical or physical in nature. Analytical solution is
generally possible only for simple problems with regular domain and simple boundary
conditions. Physical solution can be obtained through laboratory modelling or field
simulations. Generally, physical solutions are time consuming and expensive. If an
analytical or easier physical solution is not plausible, numerical process is required.
Step 8 - Model Development: A reasonable model to solve the given problem may
generally be available as a public domain or commercial model. However, if the model is
not available, it should be developed, depending on the objectives of the modelling. If the
solution methodology is physical, laboratory or field level investigation would be required.
On the other hand, analytical or numerical solution can be developed with the help of a
computer.
Step 9 - Model Validation: The model cannot be used effectively for the considered
problem unless the numerical values of all the coefficients appearing in the model are
known. Once a model has been developed or proposed for a particular problem domain, it
must be validated for the given conditions. Through model validation, it is ensured that the
model includes and describes all the relevant processes, which affect the excitation and
model responses of interest to the acceptable degree of accuracy. Further, the solution
obtained may be cross checked for few parameters/conditions for which analytical or other
standard results are available.
Step 10 - Model Sensitivity Analysis: The model sensitivity analysis involves
investigations on model behavior for the important model variables. Generally, sensitivity
analysis is performed by varying each variable to its upper and lower extremes, while
keeping other variables as constant. Sensitivity analysis indicates over all model behavior
for extreme conditions of the problem.
Step 11 - Model Predictions: Once the model is calibrated and validated for the
considered problem, the model is ready for use. Model runs are made to provide required
results for the given problem conditions and data.
10
Figure 1.1 A typical mathematical modelling procedure
11
Step 12 - Results and Interpretations: Results obtained from the model should be
presented in a tabular or graphical form. This includes various post-processing techniques.
Results should be interpreted appropriately by considering sensible behavior of the model
variables.
Step 13 - Summary and Conclusions: At the end, information that the model was
expected to provide, including accuracy, uncertainty, suggested follow-up works or other
activities should be concluded.
Step 14 - Model Reporting: Model reporting should be done in such a way that the
important features of the model are clear and the results and interpretations are presented
appropriately. Suggested contents of a model report include: abstract, introduction,
problem definition, assumptions, simplifications, data, details of the conceptual model,
details of the mathematical model, solution methodology, code (program) selection or
development, details of model sensitivity analysis, verification, calibration and validation,
details of the model runs and predictions, results, discussions, conclusions and
references.
1.3 TYPE OF GOVERNING EQUATIONS

It is possible to derive mathematical equations governing the problem for most physical
problems in nature. Generally, the governing equations can be classified into algebraic
equations, ordinary differential equations (ODE), partial differential equations (PDE) and
integral equations.
Many problems can be represented in the algebraic equation form, which can be generally
expressed as
where g is a function, x and y are state variables and θ is a set of parameters. The
equation is said to be linear if x and y appear linear in g, otherwise the equation is non-
linear. Many engineering problems can be represented by systems of linear or non-linear
algebraic equations and generally a set of values are sought as solution that satisfy the
system. Large systems of algebraic equations are generated, for example, in the analysis
of structures, electric circuits and fluid networks.
Most fundamental laws of mechanics, physics, thermodynamics and electricity are based
on empirical observations, which explain variations in physical properties and states of the
system in terms of spatial and temporal changes. As many physical laws are expressed in
terms of the rate of change of a quantity rather than the magnitude itself, the ODE are of
great importance in engineering practice. By using ODE, generally two types of problems,
i.e., the initial-value and the boundary value problems are addressed. Differential
equations are classified according to their order. An equation is said to be of order n if the
highest derivative of the dependent variable is of order n. For example, the deflection y of
a mass spring system with damping can be expressed with a second order differential
equation as
where m is the mass, c is the damping coefficient, k is the spring coefficient and t is the
time.
Dependent physical quantities vary with two or more independent variables and hence
PDE formulations are used in many practical situations. If the dependent variable and its
12
partial derivatives appear linearly with coefficients, depending only on the independent
variable, the equation is said to be linear, otherwise it is termed as a non-linear equation.
The order of a PDE is that of the highest-order partial derivative that appears in the
equation. For example, consider a second order PDE with two independent variables as
given below.
Here a, b and c are functions of x and y and α is a function of x, y, h, and . Such a

linear second order PDE can be further classified as elliptic, parabolic and hyperbolic,
depending on the values of the coefficients of the second derivative terms a, b and c. An
equation is called elliptic if the quantity (b2 – 4ac) < 0. For example, the Laplace equation
used in many engineering problems, given below is elliptic.
Elliptic equations are generally used to represent steady-state problems. Steady-state heat
conduction problem or groundwater flow in homogeneous isotropic porous media without
any pumping and recharge are typical examples where elliptic equations are used.
An equation is said to be parabolic if the quantity (b2 – 4ac) = 0. For example, the diffusion
equation in one dimension, used in many engineering problems, given below is parabolic.
Parabolic equations typically determine how a dependent quantity varies in space and
time. Time-dependent heat conduction problem is an example where parabolic type
equations can be used.
In the third category, an equation is said to be hyperbolic if the quantity (b2 – 4ac) > 0. For
example, the wave equation in one dimension, given below is hyperbolic.
Hyperbolic equations are used to deal with propagation problems and characterized by a
second order time derivative as in the case of wave propagation.
Sometimes, physical laws are expressed in terms of the summation or integrals of one or
more variables over the entire domain or a range, and hence, integral equations are of
great importance in engineering practice. Generally, the integral equations can be
expressed as follows.
Analytical solutions can be derived only for a small class of formulations involving ODE,
PDE or integral equations. Generally, approximate methods such as numerical methods
are used to obtain solution.
13
1.3.1 Initial and Boundary Conditions

Solutions of ODE, PDE or integral equations are attempted in conjunction with initial and
boundary conditions.
Generally, initial conditions describe values of dependent variable throughout the domain
at the given point in time. For example, temperature distribution at the beginning of the
calculations should be stated to get appropriate solution in the further time steps, in the
transient analysis of heat-conduction problem.
Boundary conditions, on the other hand, are described on the boundary of the domain
throughout the simulation time, based on which the solution is obtained for the concerned
problem. Generally, three types of boundary conditions are used in engineering problems,
which are called as essential, natural and the mixed boundary conditions.
The essential boundary condition, which is also known as fixed, forced, geometric or
Dirichlet boundary condition, describes variation of the state variable (unknown function)
concerned on the boundary of the problem domain. For example, h in Eq. (1.3) is the state
variable, which can be temperature or potential or any other quantity, depending on the
problem.
The natural boundary condition, also known as the Neumann boundary condition,
describes variation of derivative of the state variable concerned on the boundary of the
problem domain. For example, specification of the heat flux rather than the temperature on
the boundary is the natural boundary condition for the heated rod problem.
Both the state variable and its derivative together are described as the boundary condition
in the case of mixed boundary conditions. For example, consider the partial differential Eq.
(1.3) in a two-dimensional domain Ω and satisfy the homogeneous conditions on the
boundary Γ, the above boundary conditions can be described as the following:
Essential or Dirichlet boundary condition
Natural or Neumann boundary condition
Mixed boundary condition
where f1, f2 and f3 are known values on the boundary, s is the arc-length along Γ,
measured from some fixed point on Γ and represents the differentiation along the
outward normal n to Γ.
In the mathematical modelling, accurate and appropriate representation of the initial and
boundary conditions are essential to obtain accurate results.
1.4 SOLUTION METHODOLOGIES

It is necessary to decide analysis and solution methodology for the problem concerned,
once the mathematical formulation is ready. Analysis of mathematical formulation can be
either qualitative or quantitative. Qualitative aspects of the mathematical formulation are
14
explored without explicitly solving it in qualitative analysis. Quantitative analysis, on the

other hand, can be divided into analytical, physical and computational methods. These
methodologies are described briefly in the following sections.
1.4.1 Analytical Method

An analytical expression involving parameters and the independent variables are obtained
in an analytical method using various mathematical procedures. Solution can be exact or
approximate depending on how the solution satisfies the formulation. Generally,
approximate solutions are sought when it is either impossible or extremely difficult to get
an exact solution. Unfortunately, analytical solutions can be obtained only for a small class
of mathematical formulations with simplified governing equations, boundary conditions and
geometry.
1.4.2 Physical Method

A mathematical model represents a real physical system, albeit, on the basis of idealized
assumptions, variables and parameters. Thus, such a model can be considered as having
physical dimensions and can be analyzed sometimes in the laboratory or in the field itself.
Scaling is often required for such models and accuracy depends on scale and simulation
of the actual field conditions. Presently, the physical models are used less frequently since
they are expensive, cumbersome and difficult to employ.
1.4.3 Computational Method

Solution is obtained with help of some approximate methods using a computer in
computational method. Solution, thus, obtained would be approximate. Generally,
numerical methods are used to obtain solution in computational method. Computational
methods of analyses are applicable to a much wider class of mathematical formulations
and hence most engineering problems can be solved. With the advent of fast computers,
computational models have become very widely used valuable tool for solving engineering
problems.
1.5 NUMERICAL MODELLING

Variety of numerical methods such as method of characteristics, Finite Difference Method
(FDM), Finite Volume Method (FVM), Finite Element Method (FEM), Boundary Element
Method (BEM) and so on have been developed by engineers and scientists for solution of
engineering problems. Each of these methods has its own advantages and disadvantages,
and the choice of method largely depends upon the complexity of the problem and
investigator's familiarity with the method. Out of many available numerical methods, FDM,
FEM and BEM are the most popular techniques among engineers and scientists.
Continuous variation of the function concerned is represented by a set of values at points
on a grid of intersecting lines in the FDM. Gradients of the function are then represented
by differences in the values at neighboring points and a finite difference version of the
equation is formed. This equation is used at points in the interior of the grid to form a set of
simultaneous equations, which yield value of the function at a point in terms of values at
nearby points. At the edges of the grid, value of the function is fixed, or a special form of
finite difference equation is used to provide the required gradient of the function.
In FEM, the region of interest is divided in a much more flexible way than what is possible
with the FDM. Nodes at which value of the function is found do not have to lie on a grid
system but can be on a flexible mesh. Boundary conditions are handled in a more
convenient manner so that a standard program can deal very easily with most conditions.
The result is a program, which can be used for problems involving many shapes of domain
15
and types of boundaries. Direct approach, variational principle or weighted residual

methods are used to approximate the governing equations.
Partial differential equation describing the domain is transformed into an integral equation
relating only to boundary values in the BEM. The method is based on Green's integral
theorem. Here, the boundary is discretized instead of the domain. Hence, the
computational dimension of the problem is reduced by one in BEM. For examples, a three-
dimensional problem reduces to a two-dimensional problem and two-dimensional problem
reduces to a one-dimensional problem. Thus, BEM is ideally suited for solution of many
two- and three-dimensional problems in elasticity and potential theory for which other
methods are inefficient. However, the method requires explicit expression of Green's
functions as kernels appearing in the integral equation.
From the above discussion, it is observed that FDM is simple and easy to implement, but
difficult to implement for irregular geometry problems. FEM is more versatile and can be
applied to most problems compared to the FDM and BEM. BEM has some specific
advantages over FEM, such as reduction in computational dimension and ease in handling
of data. FDM is confined to the grid system and is not versatile compared to FEM or BEM.
Hence, the best choice in many cases is the combination of these three methods in one or
other form like FEM-BEM combination or BEM-FDM combination or FEM-FDM
combination to suit any particular problem to the benefit, accuracy, efficiency and
economy. Discretization of an earthen dam problem using FDM, FEM and BEM is shown
in Figure 1.2. Differences in discretization between the FDM, FEM and BEM can be clearly
seen from the figure. As discussed earlier, FDM, FEM and BEM have got their own merits
and demerits. Choice of the method is largely governed by the complexity of the problem,
geometry, investigator's familiarity of the method and other factors such as availability of
particular packages, computational facilities and so on. This book deals with the theory
and practice of the well-established FEM.
1.6 PRE-PROCESSING AND POST-PROCESSING

When numerical methods like FDM, FEM or BEM are used in mathematical modelling, the
ensuing computer model generally consists of pre-processing, processing and post-
processing. In the pre-processing part of the model, input required to solve the problem
are generated or read in. Input data required include domain geometry, initial and
boundary conditions, coefficients and constants for the particular problem, value of
universal constants, the grid information (finite difference grid, finite element mesh or
boundary element mesh) and various options for the concerned problem such as one-,
two- or three-dimensional analysis; steady state or transient analysis. In the processing
part, real simulation of the problem with the concerned numerical model is done. For
example, processing in the context of FEM includes generation of element matrices,
assembly of element equations, and imposition of boundary conditions and solution of
system of equations. In the post-processing part, results obtained are processed in terms
of tables, charts, graphs, contours, bar chart and so on.
Mesh generation is an important task of pre-processing. Mesh or grid generation includes
discretization of the concerned domain into elements or cells and determining grid points
or nodal positions. Subsequently, nodes and elements are numbered and nodal positions
and connectivity between elements are specified such that the processor can identify the
problem domain, discretization and nodal position. Mesh or grid generation depends on
the type of numerical model used and dimension of the model. Mesh generation is a
complex process when numerical methods such as FEM or FVM are used. On the other
hand, such process is easier when BEM is used. Mesh or grid can be generated manually
16
or through computer models. For automatic mesh generation, computer programs can be
written and incorporated in the model itself or separate mesh generating packages can be
utilized.
Figure 1.2 Flow through an earth dam—discretization using FDM, FEM and BEM
Results from a problem analysis consist of numbers pertaining to various unknown
variables such as potential, displacements, stresses and so on. It is thus better to present
17
such details either graphically or in tabular forms. Graphical display may include deformed
shape of mesh, vectors, charts, contours, and so on. The post-processing can be done
manually or through computer models. For post-processing, computer program can be
written such that results are displayed graphically or separate post-processing packages
can be used.
1.7 SCOPE OF THE BOOK

Purpose of the present book is to introduce FEM as a general numerical technique for the
solution of various engineering problems. The main emphasis is given to important
applications in engineering and science. An effort has been made to keep the
mathematics simple. Since majority of applications of FEM are in the realm of mechanics
including solid, fluids, structural and soil, descriptions in this book are primarily in terms of
these fields of study.
To effectively use the book, necessary pre-requisites are relatively moderate such as
understanding of basic mechanics and physics, basic courses in advanced calculus and
linear algebra and basic concepts of various engineering theories. For effective study and
application of FEM, the reader must have working capability in several fundamental areas
such as matrices, algebra, calculus and basic computer skills. The reader is assumed to
have these backgrounds. However, to make the text as self-contained as possible, a brief
description of these topics are given in appendices.
An overview of various approximate methods of analyses has been presented in Chapter
2wherein the basics of FEM have been evolved. An introduction to FEM with history,
applications, merits and demerits are detailed in Chapter 3. Important approaches in FEM
have been described in Chapter 4. Interpolation functions used in FEM are explained in
Chapter 5. Detailed formulation and applications of FEM in one-dimensional for various
engineering problems are presented in Chapter 6. Formulation, implementation and
applications of FEM in two-dimensional are elaborated in Chapter 7. Three-dimensional
formulations of some engineering problems are presented in Chapter 8. Details of the
computer implementation of FEM with details of intuitive, teaching purpose program are
given in Chapter 9. Further advanced level of FEM formulation and applications are
described in Chapter 10. Various FEM software and web resources available are also
discussed in this chapter. Brief review of matrix algebra and calculus, elements of calculus
of variation, illustration on use of Galerkin's method, review of Gauss quadrature
procedure for numerical integration, user's manual for the simplified finite element analysis
program (SFEAP), further illustrative one- and two-dimensional computer programs are
detailed in various appendices.
One may take a physical or intuitive approach to learning of FEM. On the other hand, one
may develop a rigorous mathematical interpretation of the method. A middle path has
been adopted in this book giving more importance to the physical approach, with essential
mathematical basis. The book can be used for basic understanding of FEM by beginners,
further understanding by FEM practitioners and to certain extent for the advanced level
understanding of researchers. For the benefit of beginners, explanation of FEM with basic
theories and fundamental applications in structural, geo- and fluid-mechanics are
presented in one-, two- and three-dimensional domains. Once the basic concepts of FEM
are understood with illustrative examples, one can study advanced topics and apply the
method to complex problems and to other areas.
18
1.8 HIGHLIGHTS OF THE BOOK

This book is written to introduce the FEM as a numerical method that employs the
philosophy of virtual work/equilibrium approach, variational principle and method of
weighted residual for solution of various engineering problems. The book is written in a
simple structured way starting with the basics of various FEM approaches, description of
finite element interpolation functions, FEM formulations and applications in one-, two- and
three-dimensional domains and further advanced applications for more complex problems.
The book is intended as an introduction to FEM as well as oriented towards various
applications in engineering and science. It is envisaged that several practical applications
and examples presented in the book will enable an engineer to assess the potential of
FEM and its applicability to specific engineering problems. Various exercise problems are
also proposed in the book. Further, the important FEM software and Web resources
available for solution of various engineering and science related problems are briefly
discussed.
Emphasis of the book is on development of FEM formulation and implementation of the
method to various problems. A very illustrative and simple computer program SFEAP with
sophisticated graphical user interface for pre-processing and post-processing are given for
structural and solid mechanics related problems. Few simple MATLAB and FORTRAN
based programs for the fluid-mechanics related problems in one- and two-dimensional
domains are also given. These programs can be operated on desktop PCs. Source codes
of these programs are also given so that the reader can easily understand the computer
implementation of the method. These codes not only demonstrate the implementation of
the methods described in the book, but also help readers who are contemplating in
preparing their own computer codes.
1.9 HOW TO USE THE BOOK?

A large number of books are available on FEM. Thus, it may be rather difficult for a user
(student, teacher or practitioner) to effectively follow-up a plan-of-study. Of course, the
plan-of-study may depend on the time available and the objectives of the study such as
1. To study the FEM systematically for the solution of complex problems (students
and teachers).
2. To have an in-depth understanding of FEM so as to further pursue research on
FEM topics (researchers).
3. To study FEM for minimum understanding to use software for practical design and
analysis problems (practitioners).
This book has been written primarily to serve the first objective. However, it addresses the
other two objectives also.
To meet the first objective for students and teachers, the FEM has been presented
systematically from the basics to one-, two- and three-dimensional applications. Users
may study all chapters systematically from Chapters 2 through 10. For researchers or
advanced learners, in addition to the systematic approach, some advanced level topics
have been presented in Chapter 10. For FEM practitioners, in addition to understanding of
basics of FEM, the book is presented with software with pre- and post-processing
graphical user interface, which can be used in practical application of problems and get
training. Depending on the interest, to meet the third objective, users may concentrate on
Chapters 3, 6, 7, 8and 9.
This book has not been written to provide a broad survey of FEM as it may require much
more comprehensive volumes. The book has concentrated on more commonly used and
19
effective FEM techniques that can be applied to engineering problems. As shown in this
book, FEM can be systematically programmed to accommodate complex and difficult
problems such as linear/non-linear stress–strain behavior, fluid-flow problems, non-
homogeneous materials problems and complicated boundary conditions. It is important,
however, that good judgement be developed and exercised concerning the solution of any
complex engineering problem to achieve the desired results using FEM. To achieve this,
the analyst must be sufficiently familiar with possible available finite element procedure
and models. In this direction, the SFEAP software provided with this book will provide
sufficient background of learning and in reinforcing the understanding of FEM.
1.10 CLOSING REMARKS

Various aspects of mathematical modelling of engineering problems have been discussed
in this chapter. Detailed procedures of mathematical modelling are illustrated, which are
directly applicable to most engineering problems. Modelling methodology may vary.
However, procedure more or less remains unchanged. Further, various types of
mathematical equations and boundary conditions used in modelling are discussed.
Accurate results can be obtained only if the mathematical model developed is appropriate.
Various analyses and solution methodologies for modelling are further illustrated in the
subsequent chapters. Since a few analytical solutions are possible for simplified problems
and the physical modelling is quite cumbersome and expensive, most engineering
problems are solved using numerical methods such as FDM, FVM, FEM and BEM.
Various aspects of pre-processing and post-processing are also discussed briefly in the
context of various numerical methods. The scope of the book, highlights of the book and
details of ‘how to use the book’ are also explained. FEM is elaborated in the subsequent
chapters.
REFERENCES AND FURTHER READING

Aris, R. (1978). Mathematical Modeling Techniques, Pitman, London.
Beer, G., and Watson, J. O. (1992). Introduction to Finite and Boundary Element Methods
for Engineers, John Wiley and Sons, Chichestor.
Berry, J. S., et al. (eds.) (1987). Mathematical Modelling Courses, Ellis Howard, New York.
Boyce, W. E. (ed.) (1981). Case Studies in Mathematical Modeling, Pitman, Boston.
Burghis, A. D. Wood (1980). Mathematical Models in the Social, Management and Life
Sciences, Elliz Horwood, Chichster.
Chapra S. C., and Canale, R. P. (2002). Numerical Methods for Engineers, 4th ed.,
McGraw-Hill, New York.
Clements, R. R. (1989). Mathematical Modeling–A Case Study Approach, Cambridge
University Press, Cambridge.
Cook, R. D. (1981). Concepts and Applications of Finite Element Analysis, John Wiley and
Sons, New York.
Dym, C. L, and Ivey, E. S. (1980). Principles of Mathematical Modeling, Academic Press,
Boston.
Haberman R. (1977). Mathematical Models, Prentice Hall, New Jersey.
Kapur, J. N. (1988). Mathematical Modeling, Wiley Eastern, New Delhi.
Krishnamoorthy, C. S. (1994). Finite Element Analysis: Theory and Programming, Tata
McGraw Hill, New Delhi.
20
Levine G., and Burkee, C. J. (1992). Mathematical Model Techniques for Learning
Theories, Academic Press, New York.
Mooney D. D., and Swift R. J. (1999). Course in Mathematical Modeling, Mathematical
Association of America, USA.
Murray S. K. (eds.) (1987). Mathematical Modeling: Class Notes in Applied Mathematics,
Siam, Philadelphia.
Reddy, J. N. (1985). An Introduction to the Finite Element Method, McGraw-Hill
International Edition, New York.
Sanorskii, A. A., and Mikchailov A. P. (2002). Principles of Mathematical Modeling, Taylor
and Francis, London.
Singh, B. et al. (eds.) (2001). International Conference on ‘Mathematical Modelling’, Tata
Smith, T. M. (1977). Mathematical Modeling and Digital Simulation for Engineers and
Scientists, Wiley Inter-sciences, New York.
Zienkiewicz, O. C., and Taylor, R. L. (1989). The Finite Element Methods, 4th ed., Vol. 1,
Basic Formulation and Linear Problems, Vol. 2: Solid and Fluid Mechanics; Dynamics and
Non-linearity, McGraw Hill, UK.
21
2 Approximate Methods of Analysis

2.1 INTRODUCTION
Engineering problems can be classified into two fundamental categories on the basis of
adopted viewpoints. In one view point called Lagrangian concept, all matter consists of
particles that retain their identity and nature as they move through space. The position of
particles in space at any instant is obtained by the coordinate systems, which are functions
of time. In the other view point, known as Eulerian concept, the problem is considered to
be in continuum rather than consisting of particle. All the processes are characterized by
field quantities that are defined at every point in space and time and one point is focused
to observe the phenomena occurring there. Most engineering problems can be considered
as continuum problems and can be approximately represented by differential or partial
differential equations. In mathematical physics, the continuum problems are often referred
as boundary value problems. Generally, the solution is sought in some domain defined by
the known boundary on which the boundary conditions are specified. For the problem
concerned, boundary can be closed or open. Boundary is known as closed if the boundary
conditions are specified everywhere on the boundary. Boundary is called open if part of
the boundary extends to infinity and no boundary conditions can be specified over it.
Concept of the mathematical modeling was illustrated in Chapter 1 through governing
equations and boundary conditions applicable to the problem domain. Governing
equations are required to be solved with the appropriate boundary conditions over the
prescribed domain. Depending on the problem, it may be possible to get an exact or
approximate solution. An exact solution is one that exactly satisfies the governing
differential equation and all the boundary conditions for the problem concerned. On the
other hand, an approximate solution may satisfy the differential equation exactly but may
not satisfy all the boundary conditions or vice versa.
Various approximating methods used in the solution of engineering problems are
discussed here to set the stage for the introduction of finite element method (FEM). After
discussing some popular approximating methods for different classes of problems,
necessary definitions, and terminologies will be established to show FEM as one of the
important approximating method.
2.2 APROXIMATING METHODS

It is possible to obtain exact analytical solutions for few simplified engineering problems.
The exact solution can be obtained by direct integration of the concerned differential
equations. Generally, the direct integration is made possible by: (i) separation of variables;
(ii) similarity solutions; or (iii) Fourier and Laplace transformations. Since most engineering
problems are complex in nature, number of problems having exact solutions is severely
limited. Hence, approximate solutions are generally sought.
Generally used approximating methods are as follows.
1. Perturbation Techniques
2. Power Series Solutions
3. Probability Methods
4. Finite Difference Method (FDM)
5. Method of Weighted Residuals (MWR)
6. Rayleigh–Ritz Method
7. Finite Element Method (FEM)
22
Usefulness of the perturbation method is limited because they are applicable only when
the nonlinear terms in the equation are small in relation to the linear terms. The power
series method, on the other hand, has been employed with some success. However, as
the method requires generation of a coefficient for each term in the series, it is relatively
tedious to apply. Also, it is difficult to prove that the series converges. The probability
methods are used if the desired variables in the problem concerned are statistical
parameters. The probability method is used to obtain statistical estimate of variables by
the process of random sampling. The first three methods briefly mentioned above have
limited use in engineering problems.
Due to wide applicability and simplicity, the most commonly used approximating methods
include FDM, MWR, Ritz Method and FEM. Out of various approximating methods, FDM is
the most popular method due to its simplicity and wide applications. As discussed earlier,
FDM has some merits and demerits compared to FEM. Incorporation of variation of
physical properties of the medium in the solution domain is a difficult task in application of
FDM to engineering problems. A large number of text books are available describing
various applications of FDM such as Forsythe and Wasow (1960), Mitchell and Griffiths
(1980), Davis (1986) and Thomas (1995). The MWR, Ritz method and FEM are related
methodologies. Hence, discussion here is limited only to methods of weighted residual and
Raleigh–Ritz, which are important in obtaining element equations in finite element
analysis.
2.3 METHOD OF WEIGHTED RESIDUALS

The MWR is a very useful tool to approximate solutions to differential equations. Here the
differential equation is considered directly rather than its variational forms. An approximate
solution has some error with respect to the exact solution. Depending on the problem,
error may be in the domain or on the boundary or in both. The objective of the method is to
reduce the error. The procedure through which the error minimization is implemented is
called the weighted residual technique.
When method of weighted residual is used, the concerned problem is represented by
differential equation valid over the domain Ω with the prescribed boundary conditions on
the boundary Г. Initially, a trial function is assumed as solution. The function may not
satisfy the differential equation and the boundary conditions exactly. By substituting the
trial function in the differential equation and boundary condition, an error known as
residual will be produced. The unknown parameters in the trial function are determined in
such a way that the residual is made as small as possible. A more detailed treatment of
weighted residual techniques can be found in Finlayson and Scriven (1966) and Finlayson
(1972).
Some of the most commonly used weighted residual techniques in engineering
applications are Method of Point Collocation, Method of Least Squares, and Galerkin's
Approach. These important weighted residual techniques are elaborated next with
examples.
2.3.1 Method of Point Collocation
A number of discrete points referred to as collocation points are chosen in the collocation
method and the error is forced to become zero at these points. As a result, the differential
equation is satisfied at these chosen points.
Let Lɸ = 0 be the differential equation under consideration. Approximate solution should be
such that the polynomial or any other function chosen should satisfy the boundary
conditions of the problem, i.e., ɸ can be expressed by
23
Here, ɸi are undetermined parameters and Ni are known functions linearly independent in
dimensionless forms. Parameters ɸi are determined by enforcing the error or residual by
ε{=L(ɸ)} = 0 at n points in the domain resulting in n equations to determine n values of ɸi.
The number of collocation points depends on the number of unknowns in the assumed
function.
Depending upon the problem, the point collocation technique can be classified into three
methods. In interior collocation method, the admissible function satisfies all the boundary
conditions but not the differential equation in the domain. Here, errors arising are set to
zero at chosen points in the interior of the domain. In boundary collocation method, the
admissible function satisfies the differential equation exactly in the domain, but does not
satisfy the boundary conditions. Here the collocation points are chosen at the boundary.
Inmixed collocation method, the admissible function does not satisfy the differential
equation and boundary conditions exactly. Thus, collocation points are chosen in the
domain as well as on the boundaries.
2.3.2 Method of Collocation by Sub-Regions
This method is very similar to point collocation method. However, error over the integral of
the function, rather than collation points is made zero over different regions. Hence if R is
the residual function,
Example 2.1
Solve following differential equation using point collocation method.
Use boundary conditions ɸ(x = 0) = 0 and ɸ(x = 1) = 0. Choose x = 0.25 and x = 0.5 as
collocation points.
Solution:
The exact solution for the expression (2.3) is
Let the approximate solution that satisfies the boundary conditions be of the form
Considering only the first two coefficients,
24
are derived.
Substitution of Eqs. (2.6) and (2.8) into Eq. (2.3) yields an error, ε, where
Choosing x = 0.25 and x = 0.5 as collocation points, where the error is made equal to zero,
following expressions are obtained.
Solving Eqs. (2.10) and (2.11) simultaneously, it can be shown that α1 = −0.1459 and α2
=−0.15263.
Hence, the solution is
Here it should be noted that the number of collocation points should be equal to the
number of unknown parameters αi.
2.3.3 Method of Least Squares
In the method of least squares, the sum of squares of the residuals are minimal (or made
zero) after substitution of the approximate solution in the differential equation. That is, the
sum of the squares of the error is minimized as
For F to be minimum that yields n equations to determine n values of

ɸi.
Thus, the final equations are
where are the weights.

Method has been demonstrated in the following example.
Example 2.2
Solve the following differential equation using least square method.
Use boundary conditions ɸ(x = 0) = 0 and ɸ(x = 1) = 0.

Solution:
Here the approximation function is assumed to be the same as given in Eq. (2.6). Then the
error is given by
25
Simplifying Eqs. (2.19) and (2.20), the resulting equations are
2.3.4 Galerkin's Method

In the Galerkin's method, approximating or trial functions are considered to be the
weighting functions. Consider a system with homogeneous boundary conditions
An approximate function that satisfies these conditions can be used such that
Here Ni are the assumed functions and Ψi are either the unknown parameters or unknown
functions of one of the independent variables. Substituting this approximating function in
Eq. (2.24) may produce a residual as
In the Galerkin's method, the error is made orthogonal to the same trial function Ψi i.e.,
This method also yields n linear equations to determine n values of Ψi. Galerkin's method
may produce symmetrical coefficients in many cases since the weighting function and
approximating functions are same.
Many finite element formulations are based on the Galerkin's type approximating method.
In the later chapters, use of Galerkin's method in development of FEM will be described in
details.
Example 2.3
Solve the following differential equation using Galerkin's method.
26

Solution:
The functional form of error is given earlier by Eq. (2.16). Here the error ε is made
orthogonal to the trial functions,
Simplifying equations (2.30) and (2.31), resulting equations are
Solving equations (2.32) and (2.33) yields α1 = −0.14588 and α2 = −0.16278

Hence,
Finally, the results of examples 2.1, 2.2 and 2.3 are compared in the following table.
Table 2.1 Comparison of Results Obtained by Different Methods
The above comparison shows that Galerkin's method yields better results in comparison
with the Point collocation or least square method.
2.4 RAYLEIGH–RITZ METHOD

Form of the unknown solution is assumed in terms of known functions (trial functions) with
unknown adjustable parameters in the Rayleigh–Ritz Method. From the family of trial
functions, the function that renders the functional stationary are selected and substituted
into the functional, which is a function of the functions. Thus, the functional is expressed in
terms of the adjustable parameters. The resulting functional is differentiated with respect
to each parameter and the resulting equation is set equal to zero. If there are n unknown
parameters in the functional, there will be n simultaneous equations to be solved for the
parameters. Thus the best solution is obtained from the family of assumed solutions.
Accuracy depends on the trial functions chosen. Generally, the trial function is constructed
27
from the polynomials of successively increasing degree. It may be noted that the method
requires the knowledge of the functional.
Thus the main aim of Rayleigh–Ritz method is to replace the problem of finding the minima
and maxima of integrals by finding the minima and maxima of functions of several
variables. For example, consider search of a function L(x) that will extremize certain given
functional (L). As mentioned, L(x) can be approximated by a linear combination of suitably
chosen coordinate functions c1(x), c2(x),………cn(x). Then L(x) can be written as
where gi are unknown constants to be found. Since each of ci(x) is an admissible function,
the functional I(L) becomes a function of g. By taking the differential of the function,
unknown g can be determined as follows.
Using Eq. (2.36), n algebraic equations are obtained from which the unknown constants gj
are determined. Method has been explained in the following example.
Example 2.4
Solve the following differential equation using Rayleigh–Ritz Method.

Solution:
Solution is obtained from the corresponding variational functions. Variational function for
the above differential equation can be written as
F satisfies the Euler–Lagrangian equation of variational calculus given as
yielding the differential equation chosen. Further,
substituting all these into Eq. (2.39), , can be obtained,

which is the given differential equation. Substituting Eqs. (2.6) and (2.7) into
28
and integrating separately term by term within the limits x = 0 to 1,
can be derived.
Finally, the variational function can be written as
Minimizing Ψ with respect to undetermined coefficients α1, and α2
Solving equations (2.42) and (2.43) simultaneously, α1 = –0.1460 and α2 = –0.1628.

Hence, the expression is,
This result indicates that whenever a functional exists for any differential equation, both
Galerkin and Raleigh–Ritz methods give almost identical results.
2.4.1 Relation Between FEM and Rayleigh–Ritz Method
From the mathematical point of view, the FEM can be considered as a special case of
Rayleigh–Ritz Method. Both FEM and Rayleigh–Ritz Method use a set of trial functions as
the starting point for obtaining the approximate solution, take linear combinations of trial
functions and try to get combinations of trial functions, which make a functional stationary.
However, the trial function is considered for the entire domain in the Rayleigh–Ritz Method
and hence simple domain only can be modeled. On the other hand, functions are not
defined over the whole solution domain in FEM. The trial function is defined element wise
and they do not have to satisfy boundary conditions.
2.5 FURTHER NUMERICAL EXAMPLES

Example 2.5
Solve the differential equation ɸ″ - ɸ = x2 with boundary conditions ɸ(x = 0) = 0 and ɸ(x = 1)
=0, by Point Collation, Least Square, Galerkin and Rayleigh–Ritz Method and compare
results with the exact solution.
Solution:
1. Point Collocation
2. Solution that satisfies boundary conditions is
29
Limiting it to two terms,
Then
and
Substituting values of in original equation, error ε is found to be
Simplifying,
Choosing x = 0.25 and x = 0.5 as collocation points,
Simplifying and making error = 0 at x = 0.5,
Solution of Eqs. (2.50) and (2.51) yields
Therefore,
Method of Least Squares

Assuming same trial function as above,
Minimize the square of error with respect to α1 and α2 over the domain to get α1 and α2.
This can be achieved by differentiating with respect to α1 and α2. Therefore,
30
Solving Eqs. (2.53) and (2.54) for α1 and α2, α1 = –0.06284 and α2 = –0.17903. Hence the
solution is
Galerkin's Method
In this method, the error is made orthogonal to the trial functions over the domain. That is,
Integrating and substituting the limits, it can be shown that
Raleigh–Ritz Method
Solution is obtained from the corresponding variational functional, which can be written as
When this equation is substituted into the Euler–Lagrangian equation, the original
differential equation should be obtained.
31
Substituting all these into Eq. (2.60), the governing differential equation
or ɸ″ – ɸ = x2 is obtained. Substituting Eqs. (2.46) and (2.47) into
and integrating separately term by term, following equations are obtained.
Solving, α1 = –0.05496 and α2 = –0.16279

Therefore,
Exact Solution
The exact solution of differential equation ɸ″ – ɸ = x2 can be derived as:
Table 2.2 shows a comparison of results by different methods with the exact solution. It
can be seen that Galerkin’s and Rayleigh–Ritz Methods give better results.
Table 2.2 Comparison of Results Obtained by Different Methods
32
Example 2.6
Solve ɸ″ + ɸ' – 2ɸ = 0, 0 ≤ x ≤ 1 with boundary conditions ɸ(x = 0) = 0, ɸ(x = 1) = 1 by
using (i) Point Collocation and (ii) Least Square Method.
Solution:
Let ɸ = α1x + α2x2 + α3 x3
The boundary conditions are: at
Now, ɸ' = α1 + 2α2x + 3α3x2 and ɸ″ = 2α2 + 6α3x
Therefore, ε = 2α2 + 6α3x + α1 + 2α2x + 3α3x2 – 2α1x – 2α2x2 – 2α3x3
Or ε = α1 (1 – 2x) + α2 (2 + 2x – 2x2) + α3(6x + 3x2 – 2x3)
(i) Collocation Method:
Choosing x = 0.25 and 0.5 as collocation points,
On solving, α1 =1.136, α2 = –0.4765, α3 = 0.3404.

Hence the solution is ɸ = 1.136x – 0.4765x2 + 0.3404x3.
(ii) Least Squares Method:
Solving (2.66), (2.67), and (2.68),
Example 2.7
Solve ∇2h= f(x,y) in the region (x, y) є (0, a); x (0, b) and f(x, y) = c, with boundary condition
h(x = 0) = 0, h(x = a) = 0, h(y = 0) = 0, h(y = b) = 0, by Galerkin's method.
Solution:
Let h = α x (x – a) y (y – b)
33
According to Galerkin criterion,

or,
which yields
Example 2.8
Solve the problem of laminar flow through a rectangular duct, wherein the governing
differential equation for flow is of the form
with boundary conditions h = 0 at x = ± a, y = ± b, using the Galerkin's method.

Solution:
Assume the trial function of the form
Substitution of Eq. (2.71) into Eq. (2.70) yields
By using Galerkin's method,
The resulting equations are
34
From which α1, α2, α3 can be calculated. For a = 1, b = 1, s = slope of the pressure
gradient = 0.001, g = 9.8, υ = 0.01 m2/sec (kinematic viscosity),
the equations are as follows:
The solution is: α1 = –0.6521; α2 = –1.0485; α3 = –0.3434.

Finally, h = (a2 – x2)(b2 – y2)(–0.6521 – 1.0485x – 0.3434y).
2.6 CLOSING REMARKS

In this chapter, various approximating methods are introduced. Procedures for applications
of various approximating methods are discussed in detail. As FEM is also one of the
approximating methods, it is essential to understand how various methods can be applied
to solve differential equations. The MWR and the Rayleigh–Ritz Method based on the
variational principles are the most important approximating methods. The basic principles
used in FEM are very similar to that used in weighted residual and Rayleigh–Ritz Methods.
Generally, most FEM formulations are either based on the variational principles (similar to
Rayleigh–Ritz Method) or Galerkin's method based on weighted residuals technique.
Approximate methods described in this chapter are applied to the entire domain of the
problem. On the other hand, it will be shown later that approximation is applied initially
element wise in the FEM. Total system is then assembled for all elements to obtain
solution. Various FEM principles, methodologies and applications are elaborated further in
the subsequent chapters.
EXERCISE PROBLEMS
1. Solve y” – x = 0, 0 ≤ x ≤ 1 with the boundary conditions y(x = 0) = 0 y(x = 1) = 0 by
using (i) Collocation method; (ii) Least square method, and (iii) Galerkin's method
and compare the results.
2. Solve y” + y + x = 0, 1 ≤ x ≤ 2 with boundary condition y(x = 0) = 0, y(x = 1) = 0 by
using (i) Point collocation, (ii) Least square and (iii) Galerkin method.
3. Solve x2y” – 2xy’ + 2y = 0, 1 ≤ x ≤ 4 with following boundary conditions y(x = 1) = 0,
y(x= 4) = 12 by using (i) Point Collocation; (ii) Least square, and (iii) Galerkin's
method and compare the results.
4. Solve x2y” + 6xy’ + 6y = 1/x2, 1 ≤ x ≤ 2 with boundary conditions y(x = 1) = 1, and
y(x = 2) = 2. The exact solution is Use (i) Collocation method,

(ii) Least square, and (iii) Galerkin's method and compare the results.
5. Solve y”’ – 6y” + 12y’ – 8y = 0, 0 ≤ x ≤ 1 with boundary conditions y(x = 0) = 0, y(x
= 1) =e2. The general solution is y = (1 – 2x + 2x2) e2x. Use (i) Collocation method,
(ii) Least square, and (iii) Galerkin's method and compare the results.
6. Solve y” + 4y’+ 4y = cos(x), 0 ≤ x ≤ 1 with boundary conditions y(x = 0) = 0 and y(x
= 1) = 1. The general solution is y = (A + Bx) e–2x + 1/25 (3cos x + 4sin x). Use (i)
Collocation method, and (ii) Least square and compare the results.
7. Solve y” + 4y’+ 3y =10e–2x, 0 ≤ x ≤ 1 with boundary conditions y(x = 0) = 0, y(x = 1)
= 1. Use (i) Collocation method, (ii) Least square, and (iii) Galerkin's method and
compare the results.
35
8. Solve (1–x2)y”–xy' = 0, 0 ≤ x ≤ 1 with boundary conditions y(x = 0) = 0, y(x = 1) =

π/2 by (i) Least square, and (ii) Galerkin's method and compare the results.
9. Solve y”+ y = x, 0 ≤ x ≤ 2 with boundary conditions y(x = 0) = 0, y'(x = 0) = 2, by
Rayleigh–Ritz method.
10. Solve y” + 2y + x2 = 0, 0 ≤ x ≤ 1 with boundary condition y(x = 0) = 0, y(x = 1) = 0
by Rayleigh–Ritz method.
11. Solve x2y”– 2y – 1 = 0, 1 ≤ x ≤ 2 with boundary conditions y(x = 1) = 0 and y(x = 2)
= 0 by using (i) Point collocation, (ii) Least Square method, and (iii) Galerkin's
method.

Bathe, K. J. (2001). Finite Element Procedure, Prentice Hall of India, New Delhi.
Chandrupatla, T. R., and Belegundu, A. D. (2003). Introduction to Finite Elements in
Engineering, Prentice Hall of India, New Delhi.
Davis, J. L. (1986). Finite Difference Methods in Dynamics of Continuous Media,
Macmillan, New York.
Finlayson, B. A. (1972). The Method of Weighted Residuals and Variational Principles,
Academic Press, New York.
Finlayson, B. A., and Scriven, L. E. (1966). The Method of Weighted Residuals: A Review,
Applied Mechanics Review, 19(9).
Forsythe, G. E., and Wasow, W. R. (1960). Finite Difference Methods for Partial
Differential Equations, John Wiley, New York.
Huebner, K. H. (1975). The Finite Element Method for Engineers, John Wiley and Sons,
New York.
Hilderband, F. B. (1965). Methods of Applied Mathematics, Prentice-Hall, Englewood
Cliffs.
Mitchell, A. R., and Griffiths, D. F. (1980). Finite Difference Method in Partial Differential
Equations, John Wiley, Chichester.
Reddy, J. N. (1985). An Introduction to the Finite Element Method, McGraw-Hill
International Edition, New York.
Thomas, J. W. (1995). Numerical Partial Differential Equations: Finite Difference Methods,
Springer-Verlag, New York.
Zienkiewicz, O. C. (1980). Finite Element Method, 3rd ed., McGraw Hill, New York.
36
3 Finite Element Method—An Introduction

3.1 GENERAL
In the last few decades, revolution in the computer technology has led to development of
numerous computational techniques for solving many engineering problems. As
mathematical modelling became an integral part of analysis of engineering problems, a
variety of numerical methods such as finite difference method (FDM), finite element
method (FEM), finite volume method (FVM) and boundary element method (BEM) have
been developed. Each of these methods has its own merits and demerits depending on
the problem to be solved. Out of the available numerical techniques, FEM is one of the
most flexible and versatile method for solving engineering problems.
FEM is used widely for solving engineering problems in solid mechanics, heat transfer,
structural mechanics, aerospace, automobiles, biomechanics, fluid mechanics, etc.
Development of FEM for solution of practical engineering problems began with the advent
of the digital computers. FEM envisions the solution process as built up of many small
interconnected sub-regions or elements and gives a piecewise approximation to the
problem concerned. Hence, most engineering problems, which can be considered to be in
a continuum, can be solved using FEM by the piecewise approximation.
3.2 WHAT IS FEM?

In the most numerical methods, the unknown state variables are solved at a discrete
number of points in the problem domain considered to obtain approximate solutions.
Process of dividing the problem domain into an equivalent system of smaller domains or
units and selecting a discrete number of points is called discretization. Once a problem
domain is discretized, solution can be obtained for each of the smaller domains or units
considered. Finally, such domain wise solutions can be combined together to obtain
solution for the entire domain. Hence, the solution is obtained from the approach known as
‘going from part to whole’. Through this approach, analysis is simplified even though large
amount of data may have to be handled. Basic idea of FEM is developed from the above
principle. Discrete points considered in the domain are called nodes and the smaller
domains or units considered are called elements. Elements and nodes together constitute
the mesh. Fineness of the mesh increases accuracy of the solution but at the cost of
computation time. On the other hand, the number of unknown parameters at each node,
determines the degrees of freedom.
FEM offers a way to solve wide variety of complex continuum problems by sub-dividing
them into a series of simpler interrelated problems. Essentially, it provides a consistent
technique for modelling whole system as assemblages of discrete parts or finite elements.
The whole system may be a body of matter or a region of space in which some
phenomenon of interest is occurring. The degrees of freedom to which the assemblage of
elements represents the whole system usually depend on the number, size and type of
elements chosen for the representation. Sometimes it is possible to choose elements in
such a way that they lead to exact representation. However, this occurs in some special
cases only. Most often, the choice of an element is a matter of engineering judgment
based on accumulated experience.
Fundamental idea of discretization in the finite element method stems from the physical
procedures used in the network analysis or framed structures. An engineering problem
may be solved in one-dimensional, two-dimensional or three-dimensional. Accordingly, the
FEM discretization can also be in one-dimensional, two-dimensional or three-dimensional
37
as shown in Figs. 3.1, 3.2 and 3.3. In mathematical modelling, the problem to be solved
can be generally represented by mathematical equations. As FEM includes discretization
of the domain into elements, the finite element solution gives a piecewise approximation to
the governing mathematical equations.
3.3 HOW DOES FEM WORK?

In engineering problems of continuum in nature, the field variable (such as displacement,
potential, pressure, velocity and temperature) possesses infinitely many values because it
is a continuous function of generic point in the body or solution domain. Hence, the
problem becomes one with an infinite number of unknowns. The discretization procedure
reduces the problem to one of a finite number of unknowns by dividing the solution domain
into elements. Then the unknown field variable is expressed in terms of assumed
approximating functions within each element. Approximating functions (or interpolating
functions) are defined in terms of values of the field variables at specified nodes or nodal
points. Nodes usually lie on the element boundaries (boundary nodes) where adjacent
elements are considered to be connected or inside the element as interior nodes.
Behavior of the concerned field variable is defined by the nodal values of the field variable
and the interpolation functions for the element within the elements. The nodal values of the
field variable become the new unknowns for the finite element representation of the
problem. Once the nodal unknowns are obtained, the interpolation functions define the
field variable throughout the assemblage of elements of the problem. Thus, solutions are
formulated for individual elements before putting them together to represent the entire
problem. Nature of the solution and degree of accuracy depend on the size and number of
elements and the kind of interpolation function used. The interpolation functions are
chosen such that the field variable or its derivatives are continuous across adjoining
element boundaries.
For example, Figure 3.1 shows the FEM discretization for an axially loaded bar member,
which is considered as a one-dimensional problem. The physical system includes the bar
subjected to an axial load g(x) and the problem to be solved is to determine the
displacement u = u(x). The problem can be mathematically represented by the following
equation:
where AE represents the axial rigidity of the bar assumed to be constant. To obtain a
solution using FEM, the bar considered is discretized into elements using connecting
nodes as shown in Figure 3.1.
Figure 3.1 Finite element discretization of a bar in one dimensional domain

Here, the unknown variable is the displacement u. The unknown variable is approximated
by a set of piecewise continuous functions (referred to as interpolation or shape function)
assembled over the elements at the nodes. Then the equations governing the behavior of
the system are derived by assembling the equations formulated for all the elements.
Finally, the response of the system (here the bar deflection) is obtained by solving the
approximated system of equations.
38
Consider another example of a two-dimensional steady flow through a homogeneous

isotropic earth dam resting on an impermeable formation. The problem domain and FEM
discretization is shown in Figure 3.2. The physical problem includes the ‘potential’
distribution in the domain of the dam due to the differential heads at upstream and
downstream and determination of the free surface position in the dam. The upstream head
value is known as h1(=10 m) and the downstream head value is known as h2 (=2 m). The
problem can be mathematically represented by the Laplace equation as
where h(x,y) is the head or ‘potential’. To get a solution, initially a free surface is assumed
and the dam domain including the free surface considered is discretized into triangular
elements using connecting nodes as shown in Figure 3.2. Here, the unknown variable is
the head or ‘potential’ (h) at various nodes. The unknown variable h is approximated by
the two-dimensional interpolation or shape functions assembled over the triangular
elements at the nodes. Then, the equations governing the behavior of the problem are
derived by assembling the equations formulated for all the triangular elements. Finally, the
‘potential’ is obtained by solving the approximated system of equations after introducing
the upstream and downstream head values. As the location of free surface is not known a
priori for this problem, the final solution is obtained through an iterative procedure which is
discussed later in the book.
For demonstration purpose, consider a problem dealing with steady-state three-
dimensional heat conduction in an isotropic homogeneous plate. The problem domain and
three-dimensional FEM discretization are shown in Figure 3.3. The domain is discretized
using prism elements. The physical problem includes the ‘temperature’ distribution in the
plate due to the differential temperature from left to right, with values T1 and T2. The
problem can be represented mathematically by the Poisson equation as
Figure 3.2 Earth dam discretization using two-dimensional finite elements
39
Figure 3.3 Three-dimensional finite element discretization of a plate with prism elements
where T is the temperature, k is the thermal conductivity and Q is the heat source or sink.
The domain is discretized into prism elements using connecting nodes to obtain solution
as shown in Figure 3.3. Here, the unknown variable is the temperature at various nodes.
The unknown variable ‘temperature’ is approximated by the three-dimensional
interpolation or shape functions assembled over the prism elements at the nodes. Then,
the equations governing the behavior of the problem are derived by assembling equations
formulated for all the prism elements of the problem. Finally, the ‘temperature’ distribution
is obtained by solving the approximated system of equations after introducing the
Above mentioned problems indicate a broad perspective of working of FEM in the one-
dimensional, two-dimensional and three-dimensional problems. Further chapters will
explain in detail how FEM is used in the solution of engineering problems.
3.4 A BRIEF HISTORY OF FEM

Basic idea of FEM originated from analysis procedure used in framed structures like
trusses, aircraft structural analysis and flow network analysis. For the origin and
development of FEM, there are different perspectives from the viewpoint of a
mathematician, a physicist and an engineer. From the mathematician's perspective,
solutions to boundary value problems of continuum mechanics were sought by finding
approximate upper and lower bounds for eigenvalues. The physicists were trying to solve
continuum mechanics problems by means of piecewise approximating functions. The
engineers were investigating methodologies for the solution of complex aero-elasticity
problems such as stiffness of shell type structures reinforced with ribs.
In the 1930s, when civil engineers dealt with truss analysis, they identified the solution
procedure by solving the component stresses and deflections as well as the overall
strength of the system. They recognized the truss as an assembly of members of rods
whose force deflection characteristics can be easily obtained. By combining these
individual characteristics using laws of equilibrium and solving the resulting system of
equations, the unknown forces and deflections for the overall truss were obtained. Efforts
of mathematicians, physicists and engineers finally resulted in the development of basic
ideas of finite element method in 1940s.
Hrenikoff (1941) proposed the ‘frame work method’ for the solution of elasticity problems.
On the other hand, Courant (1943) presented an assemblage of piecewise polynomial
interpolation over triangular elements and the principle of minimum potential energy to
solve torsion problems. Some mathematical aspects related to eigenvalues were
40
developed for boundary value problems by Poyla (1954), Hersch (1955) and Weinberger
(1958). Foundation of finite element was laid by Argyris in 1955 through his book on
‘Energy Theorems and Structural Analysis’. Turner et al. (1956) developed the stiffness
matrices for truss, beam and other elements for engineering analysis of structures.
For the first time in 1960, the terminology ‘Finite Element Method’ was used by Clough
(1960) in his paper on plane elasticity. In 1960s, a large number of papers appeared
related to the applications and developments of the finite element method. Engineers
applied FEM for stress analysis, fluid flow problems and heat transfer. A number of
international conferences related to FEM were organized and the method got established.
The first book on FEM was published by Zienkiewicz and Cheung in 1967. With the advent
of digital computers and finding the suitability of FEM in fast computing for many
engineering problems, the method became very popular among scientists, engineers and
mathematicians. By now, a large number of research papers, proceedings of international
conferences and short-term courses and books have been published on the subject of
FEM. Many software packages are also available to deal with various types of engineering
problems. As a result, FEM is the most acceptable and well-established numerical method
in engineering sciences.
3.5 FEM APPLICATIONS

FEM is now applicable to a wide range of engineering problems. Majority of applications of
FEM are in the realm of solid mechanics and fluid mechanics. In the last two decades,
FEM has also been used in electrical and electromagnetic problems as well as in
bioengineering problems. Categories of problems that can be solved using FEM can be
divided into equilibrium, eigenvalue and transient problems. The equilibrium problems are
generally steady-state problems such as determination of stress and displacement in solid
mechanics-related problems, determination of temperature distribution in thermal
problems, estimation of potential, velocity and pressure in fluid mechanics problems. The
eigenvalue problems are also steady state in nature but include estimation of vibration and
natural frequencies in solids and fluids. In the transient problems, FEM is used in
propagation problems of continuum mechanics with respect to time. A brief description of
the applications of FEM in various engineering fields is given below.
Structural mechanics and aerospace engineering: FEM applications in equilibrium
conditions include analysis of beams, plates, shell structures, stress and torsion analysis
of various structures. On the other hand, eigenvalue analyses include stability of
structures, viscoelastic damping, vibrations and natural frequency analysis of structures.
The transient or propagation analysis using FEM includes dynamic response of structures
to periodic loads, viscoelastic and thermo-elastic problems and stress wave propagation.
Geotechnical engineering: FEM applications include stress analysis, slope stability
analysis, soil structure interactions, seepage of fluids in soils and rocks, analysis of dams,
tunnels, bore holes, propagation of stress waves and dynamic soil structure interaction.
Fluid mechanics, hydraulic and water resources engineering: FEM applications
include solutions of potential and viscous flow of fluids, steady and transient seepage in
aquifers and porous media, movement of fluids in containers, external and internal flow
analysis, seiche of lakes, ocean and harbors, salinity and pollution studies in surface and
sub-surface water problems, sediment transport analysis and water distribution networks.
Mechanical engineering: In mechanical engineering, FEM applications include steady
and transient thermal analysis in solids and fluids, stress analysis in solids, automotive
design and analysis and manufacturing process simulation.
41
Nuclear engineering: FEM applications include steady and dynamic analysis of reactor
containment structures, thermo-viscoelastic analysis of reactor components, steady and
transient temperature-distribution analysis of reactors and related structures.
Electrical and electronics engineering: FEM applications include electrical network
analysis, electromagnetics, insulation design analysis in high-voltage equipment’s, thermo-
sonic wire bond analysis, dynamic analysis of motors, molding process analysis in
encapsulation of integrated circuits and heat analysis in electrical and electronic
equipment.
Metallurgical, chemical and environmental engineering: In metallurgical engineering,
FEM is used for the metallurgical process simulation, molding and casting. In chemical
engineering, FEM can be used in the simulation of chemical processes, transport
processes (including advection and diffusion) and chemical reaction simulations. FEM is
used in environmental engineering widely in the areas of surface and sub-surface pollutant
transport modelling, air pollution modelling, land-fill analysis and environmental process
simulation.
Meteorology and bioengineering: In the recent times, FEM is used in climate
predictions, monsoon prediction and wind predictions. FEM is also used in bioengineering
for the simulation of various human organs, blood circulation prediction and even total
synthesis of human body.
3.6 MERITS AND DEMERITS OF FEM

As discussed above, FEM can be applied to almost all branches of engineering. The fact
that FEM can be used to solve a particular problem does not mean that it is the most ideal
solution technique. To solve a given problem, often several attractive numerical
techniques are available. Each method has its own merits and demerits. Depending on the
problem, ‘the best’ method should be chosen by comparing the merits and demerits of the
method. Here, the merits and demerits of FEM are discussed.
Merits of FEM
Compared to other numerical methods some of the merits of FEM are as follows.
Modelling of complex geometries and irregular shapes are easier as varieties of finite
elements are available for discretization of domain.
Boundary conditions can be easily incorporated in FEM.
Different types of material properties can be easily accommodated in modelling from
element to element or even within an element.
Problems with heterogeneity, anisotropy, nonlinearity and time-dependency can be easily
dealt with.
The systematic generality of FEM procedure makes it a powerful and versatile tool for a
wide range of problems.
FEM is simple, compact and result-oriented and hence widely popular among engineering
community.
FEM can be easily coupled with computer-aided design (CAD) programs in various
streams of engineering.
An FEM model can be developed at different levels and it is possible to interpret the
method in physical terms.
42
In FEM, it is relatively easy to control the accuracy by refining the mesh or using higher
order elements.,
Availability of large number of computer software packages and literature makes FEM a
versatile and powerful numerical method.
Demerits of FEM
Some demerits of FEM are as follows.
Closed-form expressions in terms of problem parameters are not available in FEM.
Numerical solution is obtained at one time for a specific problem case only. Hence, unlike
analytical solutions, there is no advantage of flexibility and generalization.
Large amount of data is required as input for the mesh used in terms of nodal connectivity
and other parameters depending on the problem.
Generally, voluminous output data must be analyzed and interpreted.
Experience, good engineering judgment and understanding of the physical problems are
required in FEM modelling. Poor selection of element type or discretization may lead to
faulty results.
3.7 CLOSING REMARKS

In this chapter, the finite element method is introduced. The basic idea in FEM ‘going from
part to whole’ of the problem concerned is explained. Various aspects of FEM
discretization are elaborated. Domain can be discretized in one-, two- or three-dimensions.
Various aspects of working of FEM in one-dimension, two-dimensions and three-
dimensions are explained with example problems. Further, history of FEM is briefly
reviewed by mentioning the important developments. FEM is now used in almost all
branches of engineering. The important applications of FEM in different branches of
engineering are further described briefly. As a numerical method, FEM has its own merits
and demerits, which are briefly outlined. Various FEM principles, methodologies and
applications are elaborated in subsequent chapters.
EXERCISE PROBLEMS
With the help of appropriate domain discretization, governing equations and boundary
conditions, explain the one-dimensional application of FEM in heat conduction through
cylindrical bar of 5 m length and 5 cm diameter. Use linear line elements for discretization.
conditions, explain the two-dimensional application of FEM for deformation and stress
distribution in a cantilever bar of 1 m and 10 cm and 2 cm thickness and a load of 1 KN at
the free end. Use triangular elements for discretization.
conditions, explain the three-dimensional application of FEM for potential flow in a cubic
cavity of size 1 m. Use linear prism elements.

Argyris, J. H., (1955), Energy theorems and structural analysis, Aircraft Engineering, Vol.
27.
Chandrupatla T. R., and Belegundu, A. D. (2003). Introduction to Finite Elements in
43
Clough, R. W. (1960). The finite element method in plane stress analysis, Proceedings of
ASCE, 2nd Conference on Electronic Computation, Pittsburgh, 23: 345–378.
Courant, R. (1943). Variational methods for the solution of problems of equilibrium and
vibrations, Bulletin of the American Mathematical Society, 49: 1–23.
Hersch, J. (1955). Equations Differentielles et Fonctions de Cellules, C. R. Acad. Sci., Vol.
240.
Hrenikoff, A. (1941). Solution of problems in elasticity by the frame work method, Journal
of Applied Mechanics, Transactions of the ASME, 8: 169–175.
New York.
Poyla, G. (1954). Estimates for Eigenvalues: Studies Presented to Richard von Mises,
Reddy J. N. (1985). An Introduction to the Finite Element Method, McGraw-Hill, New York
(Int. ed.).
Turner, M. J., Clough, R. W., Martin, H. C., and Topp, L. J. (1956). Stiffness and deflection
analysis of complex structures, Journal of Aeronautical Science, 23(9): 805–824.
Weinberger, H. F. (1958). Upper and lower bounds for eigenvalues by finite difference
methods, Commun, Pure Appl. Math., Vol. 9.
Zienkiewicz, O. C., and Cheung, Y. K. (1967). The Finite Element Method in Structural and
Continuum Mechanics, Mac-Graw Hill, London.
Zienkiewicz, O. C. (1980). Finite Element Method, 3rd ed., Mac-Graw Hill, New York.
44
4 Different Approaches in FEM

4.1 INTRODUCTION
Most conventional methods are cumbersome, uneconomical and time consuming to apply
to many engineering applications, due to introduction of new materials such as
composites, fiber reinforced materials etc. and complicated geometries to be modelled.
FEM has become one of the most popular numerical methods to deal with complex
geometries and material properties with the availability of high speed digital computers.
Generally, FEM can be considered as a variant of Rayleigh–Ritz method in combination
with the variational principle applied to continuum mechanics. In the Rayleigh–Ritz
method, trial functions are used to represent the entire domain. On the other hand, the trial
functions used in FEM are approximated over a sub-domain, referred to as an element.
Hence, FEM can better handle any material inhomogeneity within the domain and can
model any irregular geometry by using varieties of shapes of elements. General steps
used in FEM are explained next.
4.2 GENERAL STEPS OF FEM

Even though FEM procedure may vary depending upon problem and approach used for
solution, general steps in the FEM analysis remain more or less the same. The first step to
the solution of the considered problem is to define the problem in detail with the available
data. Based on the problem statement and available data, a mathematical model is
developed defining the geometry of the problem, material properties, assumptions and
simplifications used, governing equations, boundary and initial conditions.
In this section, the general steps used in the solution using the finite element method are
described with explanations.
Step 1: Discretize and select element types
This step involves subdividing body into an equivalent system of small bodies, called ‘finite
elements’. Points at which the primary unknowns are required to be evaluated, are called
‘nodes’ or ‘nodal points’, and the interfaces between the elements are called nodal lines
(or nodal planes/surfaces). Number of unknowns at a node is termed as ‘nodal degrees-of-
freedom (DOF)’. Most appropriate element type is chosen for the analysis required. Total
number of elements used and their variation in size and type within a given body are
primarily matters of engineering judgment. One may choose one-dimensional (1D), two-
dimensional (2D), three-dimensional (3D) or an axisymmetric element, depending upon
the physical system under consideration. These elements are shown in Figure 4.1 through
Figure 4.4.
Figure 4.1 One-dimensional (line) elements

Axisymmetric elements are used when the geometry and material properties of the
physical system are axisymmetric. An axisymmetric element is actually a degenerated,
45
three dimensional element and is constructed by rotating a two-dimensional finite element

about the axis of symmetry by 360°.
Discretization is the most important step in the finite element analysis. In a typical problem,
a substantial amount of the time is spent in this phase of the analysis. As a result, many
commercially available finite element software packages have sophisticated pre-
processing part with graphical user interface to expedite data preparation. It may be noted
that accuracy of results depends upon the details of finite element discretization. Hence,
sound engineering judgment and understanding of physics of the problem are essential to
get meaningful results from the finite element analysis.
Figure 4.2 Two-dimensional elements
Figure 4.3 Three-dimensional elements
Step 2: Select approximation functions

This step involves choosing a pattern or shape for the
distribution of the unknown quantity u, within each
element. Unknown quantity can be displacement for Figure 4.4 Axisymmetric element
stress-analysis problems, temperature in heat flow
problems, fluid pressure and/or velocity for fluid flow problem, and both temperature (fluid
pressure) and displacement for coupled problems involving effects of both flow and
deformation. Approximation function is defined within the element using the nodal values
of the element. Linear, quadratic and cubic polynomials are frequently used functions
because they are simple to work with in the finite element formulation. However,
trigonometric series could also be used. For an n-node element, the approximation
function can be expressed as
46
where u1, u2, …, un are the values of the unknowns at the nodal points and N1, N2, …, Nn
are the interpolation functions or shape functions. Accuracy of a solution depends greatly
on the selection of approximation function. Linear polynomials N1 and N2 are shown in
Figure 4.5for a one-dimensional two-node element.
Figure 4.5 One-dimensional approximation of u
Figure 4.6 Constitutive laws and gradients of primary unknowns

Step 3: Define the gradients of the unknown quantity and constitutive relationships
These relationships are necessary for deriving the equations for each finite element. In
stress-analysis problems, gradients and constitutive relationships are simply the strain-
displacement and the stress–strain expressions, respectively, as illustrated in Figure
4.6(a). The examples are given below.
(a) One-dimensional–Stress analysis
Strain/displacement relation:
Stress-strain relation: σx = E εx (Hooke's Law)
47
where, εx = strain in the x-direction, σx = stress in the x-direction and E = modulus of

elasticity (material property)
(b) One-dimensional flow through porous media (Figure 4.6(b))
Fluid gradient =
Constitutive relation:
vx = − kxx gx (Darcy's Law)
where kxx = coefficient of permeability and vx = velocity.
(c): One-dimensional heat flow
temperature gradient =
constitutive relation: (Fourier's law of heat conduction)

where qx is the heat flow and kxx is the thermal conductivity.
Step 4: Derive element equations
In this step, equations governing the behavior of a generic (typical) finite element are
obtained by invoking available laws and principles. These equations describe a
relationship between the nodal DOF and the nodal forcing parameters for the generic
element. This relationship can be written in compact matrix form as
where [ke] = element property matrix (or element stiffness matrix)

e
{q } = element vector of unknown DOF
{fe} = vector of element nodal forcing parameters
Basic approaches to derive the element equations will be explained in later sections of this
chapter.
Step 5: Assemble element equations to obtain the global or total equation and introduce
boundary conditions
Repeated application of generic element equations results in the element equations for
other elements. Then, element equations are added together using a method of
superposition to obtain global or total equations for the entire body. This process of super-
position is called ‘assembling’. Implicit in the assembly process is the concept of
continuity, or compatibility, which requires that the body remains together and that no tears
occur anywhere in the structure. The assembled equation can be written in matrix form as
where [K] = assembled (global) property (stiffness) matrix (assembly of [ke])

{q} = assembled (global) vector of nodal unknowns (assembly of {qe})
{F} = assembled (global) vector of nodal forcing parameters (assembly of {f e}).
48
Figure 4.7 General process of finite element analysis
49
Above equations indicate the capabilities of a body to withstand applied forces. To

evaluate performance of a body, certain boundary conditions need to be introduced.
Boundary conditions are the physical constraints or supports that must exist so that the
structure or body is not mobile. These conditions are commonly specified in terms of
known values of unknowns on a part of the surface or boundary of the body.
At this stage, it is sufficient to note that the global equations are modified when the
boundary conditions are incorporated.
Step 6: Solve for the unknown DOF (primary unknowns)
The assembled equations (after the modification for the boundary conditions) are solved
for the q's by using the Gauss elimination or iterative methods. The q's are called the
‘primary unknowns’ or primary quantities because they are the first quantities determined
using the FEM.
Step 7: Solve for secondary quantities
In most problems, it is necessary to compute secondary quantities from the primary
quantities. In the case of stress-analysis problems such quantities can be strains,
stresses, moments and shear forces; for fluid flow problems they can be velocities and
discharges. Once the primary quantities are known, in most cases, the relationship defined
in Step 3 can be employed to find the secondary quantities.
Step 8: Interpret the results
Steps 7 and 8 are essentially post-processing parts of a finite element analysis. Usually, a
tabulated or graphical presentation of results helps in making the design/analysis
decisions.
General process of finite element analysis is shown in Figure 4.7.
4.3 DIFFERENT APPROACHES USED IN FEM

There are number of ways in which one can formulate the properties of individual elements
of the domain. Most commonly used approaches to formulate element matrices are: (1)
direct approach, (2) variational approach, (3) energy approach and (4) weighted residual
approach.
4.3.1 Direct Approach

As discussed earlier, the basic idea of finite element method was conceived from the
physical procedure used in framed structural analysis and network analysis (pipe network
and electric network). It may be possible to choose elements in a way that leads to an
exact representation of the problem in certain applications. Hence, element properties are
derived from the fundamental physics and nature of the problem in the direct approach.
Analysis based on stiffness method is an example of this approach in structural
mechanics.
Main advantage of this approach is that an easy understanding of techniques and
essential concepts is gained without much mathematical illustrations. In solving a problem
using the direct approach, first, the elements are defined and then their properties are
determined. Once the elements have been selected, direct physical relationships are used
to establish element equations in terms of concerned variables. Finally, element equations
for various elements or members are combined to generate a system of equations which
are solved for the unknowns.
50
Some engineering problems which can be solved using the direct approach include spring
systems, trusses, beams, fluid flow in pipe networks, electric resistance networks etc.
Here, the linear spring system problem is analyzed to demonstrate the development of
finite element technique using the direct approach formulation.
4.3.1.1 Linear Spring System

A system of linear springs is used in many engineering problems. To demonstrate the
finite element procedure using the direct approach, a system of three linear springs as
shown in Figure 4.8 is analyzed. The system is connected in series with springs of
different stiffness coefficients. One end of the spring on left-hand side (LHS) is rigidly
fixed, while the other end is free to move, as shown in Figure 4.8a. All springs can be
subjected to either tension or compression. Physical parameters involved in this system
are forces, displacements and spring stiffnesses. In the finite element procedure, each
spring is considered as an element and the connection as nodes. The system considered
here has three springs and hence consists of three elements and four nodes as shown in
Figure 4.8b.
Figure 4.8a A linear spring system with three springs
Figure 4.8b Finite element discretization of linear spring system
Figure 4.8c Single spring element

Considering only one spring as shown in Figure 4.8c, which is being treated as isolated
from system, force displacement equations can be developed as follows. Forces and
displacements are defined at each node and field variable is the displacement. There is no
51
need for an interpolation function to represent variation of field variable over the element
because an exact representation is available from Hooke's law, which relates the nodal
displacements and the applied nodal forces F given as
where k is the spring stiffness constant, δ is the displacement, F is the force applied and C
is the deflection coefficient. In other words, k can be treated as the force required to
produce unit deflection or C as the deflection caused by a unit force.
Force displacement equations for node 1 and node 2 are
In matrix form, Eqs. (4.2) and (4.3) can be written as
The square matrix [ke] is known as the element stiffness matrix, the column vector {qe} is
the nodal displacement vector, and {f e} is the resultant nodal force vector for the element.
Depending on the problem, the stiffness coefficients of matrix [ke] are determined exactly
or approximately. In the present case, a typical stiffness coefficient kij of [ke] is defined as
the force required at node i to produce a unit deflection at node j. It should be noted that
the left hand side element is constrained to have zero displacement at one node which
does not influence derivation of element matrix. Constraint conditions or boundary
conditions are taken into account only after element equations are assembled to form
system of equations.
4.3.2 Variational Approach

Problems in continuum mechanics can be represented by two different but equivalent
formulations—the variational formulation and the differential formulation. Unknown
function or functions are formulated such that extremize (maximize or minimize) or make
stationary a functional or system of functionals subject to the given boundary conditions in
the variational formulation. On the other hand, a differential equation or a system of
differential equations, are integrated in the differential formulation, subject to the known
boundary conditions. A close observation of both methods indicates that both are
equivalent because the functions that satisfy differential equations and boundary
conditions also extremize or make the functional stationary.
Variational approach relies on the calculus of variations and involves extremizing a
functional. A functional is a quantity whose value depends on entire shape of some
functions rather than a number of discrete variables. Functional involves unknown
functions and the aim of calculus of variation is to find conditions which are imposed when
integrals attain stationary values.
It can be observed in variational approach that function that is extreme of the functional is
also solution of the corresponding operator equations. Also, it is easy to prove the
52
existence of a solution and the conditions involved at the interfaces can be easily tackled.
Variational formulation has also got the advantages that more complex boundary
conditions can be modelled. Moreover, the functional contains lower order derivatives than
the differential operator and the problem may possess reciprocal variational formulation
(Zienkiewicz, 1980).
The main task in using variational principle is to find the variational functional for the
particular problem concerned. It may be possible to find functional for most problems in
engineering. However, mathematical manipulation may be required to get the functional, if
a classical variational functional is not available for the problem concerned. These
mathematical manipulations, whenever required, make the approach more complex.
Depending on the problem, functional could be an integrated quantity which may be
characteristic of the problem. Generally, it can be a physically recognizable quantity such
as energy or a mathematically defined entity without physical interpretation. If an example
from solid mechanics is considered, the functional turns out to be the potential energy or
some derivative of it. In fluid mechanics problems, the functional can be kinetic energy.
If only the differential equation and the boundary conditions are known and no physical
principle giving a variational functional is apparent in a problem, it should be first decided
whether a classical variational principle exists before the problem can be attempted using
the variational approach. Different methodologies are available in literature to decide
existence of the variational principle and determination of the functional (Mikhlin, 1964;
Vainberg, 1964; Finlayson, 1972). Most commonly used methods include (Heubner,
1975): formulation of variational principle based on classical procedure available in
mathematical literature (Mikhilin, 1964), derivation of variational principle by mathematical
manipulations and use of Frechet derivatives (Tonti, 1969, Finlayson, 1972).
Consider a differential equation of the form L(u) = 0. For this differential equation if
functional I exists, a function u(x) is required in the variational approach, such that the
integral within the range a and b
is extreme (maximum or minimum) for given functional ƒ. Function u(x), which maximizes
or minimizes ƒ, is called the stationary function. In the calculus of variational principle, it
may be shown that the stationary function ∫ƒ [x,u,u′(x)] dx must satisfy the Euler–
Lagrangian equation given by (refer to Appendix B)
Similar equations can be obtained for higher order functions (Mikhlin, 1964; Vainberg,
1964; Finlayson, 1972). This will yield the given differential equation.
For example, consider the potential flow in terms of h in a domain Ω with boundary Г = Г1 +
Г2. The governing equation is Laplace equation
with boundary conditions
53
In variational approach, the above problem can be represented by an equivalent problem

of finding function h that minimizes
subject to the same boundary conditions. Eq. (4.10) can be used instead of the original
Laplace equation when the variational approach is employed.
FEM procedure using the variational approach includes following steps. First step is to
discretize the domain into elements and substitute the trial functions into the functional for
each element. The functional is then differentiated with respect to each parameter and the
resulting equation is set equal to zero. Then the systems of equations are assembled for
all elements and a system of equations are formed which are solved for unknowns after
substituting boundary conditions.
In the variational approach, nodal values of u are required to be found so as to make the
functional I(u) stationary such that
where n is the number of discrete solution u assigned to the domain. From Eq. (4.11), it
follows that
Then as per the variational procedure for FEM, for whole domain discretization
where m is the total number of elements, and the superscript (e) denotes an element in the
domain. Adding contribution of every element in the domain
can be obtained. From Eq. (4.14), it can be shown that
where p is number of nodes per element. These p equations give characteristics of the
element. Finally, similar equations are written for all elements and assembled. Solution of
this system of equations after application of boundary conditions gives solution to the
problem.
For structural mechanics problems, variational functional or some kind of relationships
useful in FEM analysis can be derived using the principle of virtual work, principle of
minimum potential energy and Castigliano theorem (Shames and Dym, 1991). These
relations are mainly used in stress analysis problems. In structural mechanics problem, the
minimum potential energy method is most commonly used (Shames and Dym, 1991;
Zienkiewicz, 1980). Principle of minimum energy is based on the idea of finding the
consistent equilibrium states of body or structures associated with stationary values of a
54
scalar quantity assumed by the loaded bodies. This scalar quantity is referred as
‘functional’. This approach is similar to the variational approach discussed in this section.
Following numerical example demonstrates the variational approach in FEM.
Example 4.1
Using FEM based on the variational approach develop the stiffness matrix for the following
differential equation. Use one-dimensional linear elements for discretization.
in the range 0 ≤ x ≤ 1, with the boundary conditions h(x = 0) = 0 and h(x = 1) = 0.

Solution:
Variational functional for the above differential equation that should be minimized can be
written as
Here, the highest order derivative appearing in I(h) is a first order derivative and hence the
simplest interpolation functions satisfying the compatibility and completeness requirements
are linear functions (explained in Chapter 5). As minimum 2 nodes are required for linear
variation to be represented, two node linear elements are considered here.
Let the domain be divided into n elements and hence the solution can be written as
where Ni (discussed in Chapter 5) is the interpolating function Ni = Ni(x) and hi is the

unknown nodal value which is to be determined. For linear interpolation function, at any
node i, Ni = 1 and at other nodes Ni = 0.
Within any element, assuming linear variation of h, the expression for h can be written as
where and L is the length of the element and x is the distance

measured from node i within the element.
For nodes 1 and 2 in the element concerned, variation of h can be written as
and the variation of x can be written as
Then the derivative of h can be obtained as
Since the interpolation functions for approximating h satisfy the compatibility and
completeness requirements, the functional in Eq. (4.17) can be written as
55
where m is the number of elements and (e) is the element concerned. To find I(e)(h)(e), the
equation for h given in Eq. (4.19) is substituted in the expression for I(h) as follows.
Here the prime indicates differentiation with respect to x. Equations for an element are
obtained by performing minimization of I(e)(h)(e) or differentiation of I(e)(h)(e) with respect to
nodal values of h and equating to zero as follows.
Both the equations can be written together as
Using linear interpolation functions,

Now the element property matrix equations can be written as
In a similar way, element matrices can be written for all other elements. Finally, element
property matrices are assembled for all the elements. After introducing the boundary
conditions, system of equations is solved for the unknown nodal values of h.
4.3.3 Energy Approach

Principle of virtual work, the principle of minimum potential energy and Castigliano's
theorem are used frequently to derive equations of elements used in stress analysis
problems. Here, the principle of minimum potential energy, which is applicable only to
conservative system, will be considered since it is probably the most well-known of the
three approaches mentioned above.
Principle of minimum energy is based on the idea of finding the consistent states
(equilibrium states) of the body or structure associated with stationary values of a scalar
quantity assumed by the loaded bodies. This scalar quantity is generally referred to as a
‘functional’ (defined to be a function of another function). For example, let f (x) be a
function of the variable x and π be the functional defined such that π = π[f (x)]. Then, π is
a function of the function f.
Stationary values are given by (see Figure 4.9)
56
Figure 4.9 Stationary values of a functional

In structural analysis problems, π is the potential energy of the body or structure. In this
case, element equations can be obtained by invoking the principle of minimum potential
energy (minimizing the potential function).
FEM can be applied generally to any problem provided this scalar function is available.
As an illustration, consider one spring of Section 4.3.1.1.
The potential energy for this elastic system is
4.3.4 Weighted Residual Approach

In many engineering problems, it is difficult to obtain the functional based on variational
principles. As discussed earlier, method of weighted residual is a technique for obtaining
approximate solutions to linear and non-linear partial differential equations. It offers
another means to formulate finite element equations. In Chapter 2, various methods of
weighted residuals techniques have been discussed.
In FEM using the weighted residuals approach, the domain considered is first discretized
with suitable elements. Then, general functional behavior of dependent field variable is
assumed so as to approximately satisfy given differential equations over elements. Initially,
approximation is applied over each element and subsequently this approximation is
substituted into the original differential equation and assembled for the whole domain.
Approximation results in some error called a ‘residual’, which is required to vanish in some
average sense over the entire solution domain. Resulting system of equations is solved
after substitution of boundary conditions to get approximate solution. This approach is
advantageous because it thereby becomes possible to extend FEM to problems where no
functional is available.
57
Use of weighted residual technique is explained here with a sample problem. The aim is to
find an approximate functional representation for a field variable h governed by the
differential equation
in the domain Ω bounded by surface Г with appropriate boundary conditions. Function g is

a known function of the independent variables. Finite element procedure using the method
of weighted residual is illustrated in the following steps.
In the first step, the unknown exact solution h is approximated by approximating functions
. Here, it should be noted that the functional behavior of is completely specified in
terms of unknown parameters or the functional. The functional dependence on all but one
of the independent variable is specified while the functional dependence on the remaining
independent variable is left unspecified. Accordingly, the dependent variable can be
approximated as
Here, Ni an assumed functions and Di are the unknown parameters or the unknown
functions of one of the independent variables. As is the approximated functions, it may
not satisfy the equation when it is substituted into Eq. (4.23). Hence, it can be written as
Here, R1 is the residual or error that results from approximating h by . In the method of
weighted residuals, the m unknowns Di are determined in such a way that the error R1
over the entire solution domain vanishes or becomes very small.
In the weighted residual approach, the error is minimized by forming a weighted average
of the error and specifying that this weighted average vanish over the solution domain. It is
achieved in the weighted residual technique by using m linearly independent weighting
functions Ψi. Finally, it can be written as
After this approximation, the residual approaches to zero in weighted sense.

Here, the error depends on the choice of the weighting functions. After the weighting
functions are specified, Eq. (4.26) is solved for the Di to get an approximation of the
unknown field variable h given in Eq. (4.24).
As discussed in Chapter 2, there are a variety of weighted residual techniques. FEM
procedure using the weighted residual method can be built on these techniques. One of
the most commonly used weighted residual techniques is Galerkin's method. Details of the
Galerkin method as an approximating method have been given in Chapter 2. As stated,
the weighting functions are chosen to be the same as the approximating functions used to
represent h according to Galerkin's method. In other words, Ψi = Ni for i = 1, 2,….. m.
Thus, Eq. (4.26) can be written as
As the domain is discretized into elements and nodes, initially the weighted residual
technique based on Galerkin's approach can be formulated for an element and then it can
be assembled for the entire domain. Ni are the interpolation functions, Ni (e) defined over
the element and Di are the undetermined parameters, which may be the nodal values of
58
the field variable or its derivatives. For the element considered, equations governing the
behavior of the element can be written as
Here superscript (e) refers to an element, h(e) = [N(e)]{h}(e), g(x,y)(e) is the forcing function
defined over element (e) and r is number of unknown parameters assigned to the element.
In FEM using the Galerkin's technique, a set of equations, like Eq. (4.28), is obtained for
each element of the whole assemblage. The choice of approximating function Ni should
guarantee the inter-element continuity necessary for the assembly process. While many
interpolation functions provide continuity of value, fewer provide continuity of slope. By
applying integration by parts to the integral expression of Eq. (4.28), expressions
containing lower-order derivatives can be obtained (called weak formulation). Hence,
approximating functions with lower order inter-element continuity can be used. Such a
process offers a convenient way to introduce the natural boundary conditions that must be
satisfied on some portion of the boundary. Finally, the system is assembled for all the
elements, boundary conditions are applied and the system equations are solved for the
unknowns.
Example 4.2
By using the Galerkin's weighted residual FEM approach, derive the element equations for
the second order differential equation
where a, b and c are constants.

Solution:
For an element of length L, according to Galerkin's criterion in FEM,
The first term can be written as
The second term can be written by using h = [N]{h} as
The third term can be written as
Within any element, assuming linear variation of h, the expression for h can be written as
59
where and L is the length of the element and x is the distance

measured from node i within the element.
Therefore, noting and expressions
4.4 CLOSING REMARKS

In this chapter, different approaches and detailed analysis procedure in finite element
methods are introduced. Procedures for applications of various FEM approaches are
discussed in detail. After the problem domain is discretized in FEM, variables are
approximated in each element using some relationships or trial functions to form element
property matrix. Most commonly used FEM approaches for derivation of element property
matrix are direct approach, variational approach, energy approach and method of
weighted residual. As discussed in this chapter, each FEM approach has its own
advantages and limitations. Depending upon the problem to be solved, suitable FEM
approach can be chosen for the solution.
Use of direct approach is possible only if a direct relationship based on fundamental
physics and nature of the problem is available for the problem concerned. Applicability of
variational approach depends on the possibility of deriving a variational functional for the
problem concerned. In most structural mechanics problems, either direct approach or
variational approach based on the potential energy can be used to derive element property
matrix. The energy approach has limited applications mainly in thermo-mechanic
problems. It is observed that the FEM based on weighted residual approach has wide
applications in most engineering problems. Further, a detailed description of finite element
analysis procedure is presented in this chapter. Finite element analysis is deliberated
60
essentially in eight steps. In further development of FEM, these steps will be followed for
the systematic approach.
EXERCISE PROBLEMS
Using the direct approach, develop element stiffness matrix for a cantilever beam of length
L subjected to a concentrated load W at the free end and find the expression for the
deflection and slope at the free end.
Using FEM based on the variational approach develop element property matrix for the
following differential equation. Use one-dimensional linear elements for discretization. ɸ″−
ɸ = x2 with boundary conditions ɸ(x = 0) = 0 and ɸ(x = 1) = 0.
Using FEM based on the weighted residual approach (Galerkin's method), develop
element property matrix for the following differential equation. Use one-dimensional linear
elements for discretization. in the range 0 ≤ x ≤ 1, with the boundary conditions

y (x = 0) = 0 and y (x = 1) = 0.
Using FEM based on the weighted residual approach (Galerkin's method), develop
element property matrix for the following differential equation. Use one-dimensional linear
elements for discretization. ɸ″ + ɸ′ − 2ɸ = 0, 0 ≤ x ≤ 1, with boundary conditions ɸ(x = 0) =
0, ɸ(x =1) =1

New York.
Mikhlin, S. G. (1964). Variation Methods in Mathematical Physics, Macmillan, New York.
Oden, J. T. (1969a). A general theory of finite elements I: Topological considerations,
International Journal for Numerical Methods in Engineering, 1(3).
Oden, J. T. (1969b). A general theory of finite elements II: Applications, International
Journal for Numerical Methods in Engineering, 1(3).
Oden, J. T. (1970). Finite element analogue of the Navier stokes equations, Proceedings
of ASCE, Journal of the Engineering Mechanics Division, 96(EM4).
Rao, S. S. (1989). The Finite Element Method in Engineering, Pergamon Press, Oxford.
Reddy J. N. (1985). An Introduction to the Finite Element Method, McGraw-Hill
International, New York.
Segerlind, L. J. (1987). Applied Finite Element Analysis, John Wiley and Sons, New York.
Shames, I. H., and Dym, C. L. (1991). Energy and Finite Element Methods in Structural
Mechanics, Wiley Eastern, New Delhi.
Stasa, F. L. (1985). Applied Finite Element Analysis for Engineers, CBS Publishers Japan,
New York.
Tonti, E., (1969). Variational formulation of nonlinear differential equations, I, II, Bull. Acad.
Roy. Belg. (Classe Sci.) 55(5): 137–165, 262–278.
61
Vainberg, M. M. (1964). Variational Methods for the Study of Nonlinear Differential

Operators, Holden-Day, San Francisco.
White, R. E. (1985). An Introduction to the Finite Element Method with Applications to
Nonlinear Problems, John Wiley and Sons, New York.
62
5 Finite Elements and Interpolation Functions

5.1 INTRODUCTION
As discussed earlier, domain of the problem is discretized into sub-domains or elements
as the first step in finite element modelling. Each sub-domain is referred to as finite
elements. A field variable is approximated in the next step by using trial functions, which
are also known as interpolation or shape functions. These functions are used to form the
element property matrix. A subject of utmost importance in a finite element analysis is the
selection of a specific finite element for a domain discretization and a definition of the
appropriate trial functions within each element. Selection of a particular shape or
configuration of element depends on the problem and its dimension. For example, a line
element is used for one-dimensional problem. On the other hand, triangular, quadrilateral
or rectangular elements are employed for two-dimensional problems. Similarly,
tetrahedron or any shape of a prism can be used to discretize three-dimensional problems.
Interpolation functions can be linear, quadratic, cubic, quartic and quintic depending upon
the desired accuracy and anticipated variation of the primary field variable. For example,
field variable ɸ can be expressed in terms of interpolation function N of any element and
its nodal value ɸ i as: ɸ = ΣNiɸi = [N] {ɸ}, where Ni = Ni(x) or Ni(x, y) or Ni(x, y, z) for one-,
two- and three-dimensional analysis, respectively. Variable ɸi is the nodal value of the field
variable at any specified point i. [N] is the array of interpolation functions and {ɸ} is the
array of nodal variables. A problem is said to have C0 continuity if the field variable ɸ is
continuous at the element interface. Additionally, if the first derivatives of the field variable
are continuous across element interfaces, the problem is said to have C1 continuity.
Similarly, Cr continuity can be defined if the r-th derivatives are continuous at the element
interfaces.
Conditions of compatibility and completeness are required in a finite element analysis. If
functions appearing under integrals in the element equations described earlier contain
derivatives up to the (r + 1)-th order, the compatibility requirement states that Cr continuity
must be satisfied at the element interfaces and Cr + 1 continuity must be satisfied within the
element. These requirements hold regardless of whether element equations (integral
expressions) are derived using variational, Galerkin or any other method.
Following information is required in the finite element discretization: shape of the element,
number of nodes per element, type of nodal variables and type of the interpolation function
used in analysis. As discussed earlier, various shapes of elements can be used depending
on dimension of the problems concerned. In an analysis, number of nodes assigned to a
particular element depends on the type of the nodal variable, type of interpolation function
and the degree of continuity required. For domains with curved boundaries and non-
linearity, the linearly varying elements may not accurately represent variation of different
field variables. It would be better to use higher order finite elements for accurate
representation of field variables in many non-linear problems and irregular boundary
problems with a finite element grid. Any element with a second or higher order
interpolation function is called higher order element. With higher order elements such as
quadratic or cubic, primary variable will vary non-linearly within the element.
Total number of unknowns necessary to completely specify variation of primary unknown
field in an element is referred to as the degrees of freedom of the element. Degrees of
freedom are specifically identified with a single nodal point and represent field variables
having a clear physical interpretation. Degrees of freedom occurring at the external and
63
internal nodes are distinguished by referring to them as nodal degrees of freedom and
internal degrees of freedom. The minimum number of degrees of freedom (or generalized
co-ordinates) necessary for a given element is determined, according to the completeness
requirements for convergence, geometric isotropy requirement and the necessity of an
adequate representation of the field variable in the formulation (See Chapter 10 for
details). Beyond the minimum number of degrees of freedom, additional degrees of
freedom may be included for a finite element by adding secondary external nodes or by
specifying degrees of freedom of higher order derivatives of the field variable at the
primary nodes. Such higher order elements with additional degrees of freedom may
increase the complexity of the individual element properties.
Use of higher order interpolation functions, generally, results in more accurate result in
those application areas where the gradient cannot be properly approximated by a set of
constant values. Fewer higher order elements may be required to obtain the same degree
of accuracy in the final answers. A reduction in number of elements reduces the number of
element data, which in turn decreases possibility of errors in preparing the element data.
Use of higher order elements does not, however, always lead to a reduction in the total
computation time. Numerical integration techniques are needed to obtain element
matrices. Hence, these techniques can involve a large number of calculations by
increasing the total computation time per element. It has been observed in many cases
that the numerical integration itself may constitute about 70% to 80% of the total
computation time.
Characteristics of some commonly used simple linear type of finite elements as well as
higher order finite elements in one-, two- and three-dimensions will be discussed in this
chapter.
5.2 INTERPOLATION FUNCTIONS

Functions used to represent behavior of a field variable within an element are called
interpolation functions. These functions are also called shape functions or approximating
functions. Different types of functions could be used as interpolation functions such as
polynomials and trigonometric functions. As polynomials are easy to manipulate, they are
used most widely as the interpolation functions. In this book, polynomials will be mainly
used as the interpolation functions. Mixed or hybrid formulations will not be discussed
(refer to Zienkiewicz and Taylor, 1989 for details). Depending on the problem dimension,
polynomials in one, two or three independent variables may be used in the interpolation
functions.
5.2.1 One-Independent Spatial Variable

A general complete nth order polynomial may be written for one-dimensional analysis as
where Pn(x) is the nth order polynomial in x, and ai (i = 0, 1, … n) are the coefficients of xi.
Some examples of polynomials in one-independent variable are as follows.
64
5.2.2 Two-Independent Spatial Variables

A general polynomial for two-dimensional analysis may be written as
Some examples of polynomials in two-independent variables are,
Any higher order polynomial can be written in a similar manner.
5.2.3 Three-Independent Spatial Variables

A polynomial for three-dimensional analysis may be written as
Some examples of polynomials in three-independent variables are
Any higher order polynomials for three-independent variables can be written in a similar
manner.
A polynomial series in one or more variables is complete if it is of a high enough degree
and if no terms are omitted in the series. Thus, polynomials of Eqs. (5.1), (5.5) and (5.8)
are complete.
Coefficients aij or aijk in a polynomial series representing the field variable are referred to
as the generalized co-ordinates. These generalized co-ordinates are independent
parameters that specify or fix the magnitude of the prescribed distribution of the
polynomial. On the other hand, the nature of the prescribed distribution is determined by
the order of the polynomial selected. These co-ordinates have no direct interpretation but
are rather the linear combinations of the physical nodal degrees of freedom. In addition,
they are not identified with any particular node.
5.3 ONE-DIMENSIONAL ELEMENTS

Generally, line elements are used with different order of interpolation function for one-
dimensional finite element modelling. Various properties of linear line elements and related
interpolation functions are discussed in this section.
5.3.1 Line Element: Linear Interpolation Function

As discussed earlier, line element is used to discretize one-dimensional problems. Let the
field variable ɸ be a function of x, i.e., ɸ = ɸ(x). Let B and D be the points on the x-axis with
co-ordinates xi and xj, respectively and the field variable values be ɸi and ɸj. It is required
to evaluate the value of ɸ at any point between B and D. Let the co-ordinate of point G be
x, where the ɸ value is to be determined. From geometry in Figure 5.1,
65
Figure 5.1 Line element with linear interpolation function
Expression (5.11) can also be derived from the generalized polynomial expressed as
follows.
Let ɸ = ɸ(x) be expressed in linear form as ɸ(x) = a1 + a2 x, or
At x = x1, ɸ(x) = ɸ1 and at x = x2, ɸ(x) = ɸ2. Hence, Eq. (5.14) can be expressed in the
following form.
66
Combining Eqs. (5.14) and (5.17),
is obtained; where
Figure 5.2 Line element representation

In a similar manner, location of any point xp within the element as shown in Figure 5.2 can
be expressed as
Derivatives of the field variable ɸ can be obtained as
Linear variation of Ni and Nj over the element is shown in Figure 5.3.
Example 5.1
Find the value of the field variable ψ and its first derivative at a place x = 0.6. Consider xi=
0, xj = 2, ψi = 1.2 and ψj = 1.5.
Solution:
Length of the element, L = xj- xi = 2. Value of Ni at x = 0.6 is Ni = 1 – x/L = 1 – (0.6/2) = 0.7.
Value of Nj at x = 0.6 is Nj = x/L = 0.6/2 = 0.3.
Thus, the value of ψ(x = 0.6) = Ni ψi + Nj ψj = 0.7(1.2) + 0.3(1.5) = 1.29.
The first derivative of the field variable over the element is dψ/dx = (ψj – ψi)/L = (1.5 -
1.2)/2 = 0.15. It may be noted that the first derivative is constant throughout the element.
67
Figure 5.3 Variations of Ni and Nj over the element
Figure 5.4 Length co-ordinates
5.3.1.1 Integration formula for length co-ordinates

When the interpolation functions Ni and Nj are linear functions of length co-ordinates only,
they can be treated as length co-ordinates. For example, consider Figure 5.4 where
distance ij = L, p is an intermediate point, ip = Lj, pj = Li. Now Ni and Nj can be written as Ni
= Li/L and Nj= Lj/L. If L = L, then Ni = 1. If Lj = L then Nj = 1 which is true. Hence, (Li/L) and
(Lj/L) can be treated as length co-ordinates. Integration of length co-ordinates over length
of an element can be done with the aid of following formula (Heubner, 1975)
Eq. (5.21) is applicable only if Ni and Nj are linear functions of x and the exponents a and b
are integers.
Example 5.2
Evaluate
Solution:
The integral I can be written as . Here, a = 1, b = 1, therefore, I = (1! 1! L)/(1 + 1

+ 1)! =L/6.
68
5.3.2 Quadratic Interpolation Function

For a line element with linear interpolation function, 2 nodes are sufficient. However, 3
nodes are needed for a line element with quadratic interpolation function. Similarly, if n is
the order of interpolating polynomials for a line element, n + 1 nodes are needed.
For an element shown in Figure 5.5, variation of a field variable ɸ over the element can be
written as
The boundary conditions over the element are

At node 1: ɸ = ɸ1 and s = 0
At node 2: ɸ = ɸ2 and s = L/2
At node 3: ɸ = ɸ3 and s = L
Figure 5.5 Line element–quadratic type

Putting the boundary conditions in the polynomial Eq. (5.22),
is obtained. Thus,
From Eqs. (5.22) and (5.25), the polynomial ɸ can be expressed as
Polynomial ɸ can also be written as
Comparing Eqs. (5.27) and (5.28), it can be seen that
69
Example 5.3
Find the value of ɸ and its derivative at a point P(s = 0.25 L) shown in Figure 5.3 given that
ɸ1= 4, ɸ2 = 4.5 ɸ3 = 6.
Figure 5.6 Example 5.3
Figure 5.7 Line element with 4 nodes

Solution:
Variation of field variable over the element is ɸ = [N]{ɸ}, where {ɸ}T = [4 4.5 6].
Shape functions N1,N2,N3 at the point P are required to be computed. Using Eq. (5.31),
N1(s = 0. 25 L) = 1 – 3(0. 25L)/L + 2(0. 25 L)2/L2 = 3/8
N2(s = 0. 25 L) = 4(0. 25 L)/L – 4(0. 25 L)2/L2 = 3/4
N3(s = 0. 25 L) = – (0. 25 L)/L + 2(0. 25 L)2/L2 = –1/8
Hence, ɸ(s = 0.25 L) = Σ Ni ɸi = 4.125
Derivative of the field variable at the given point P(0.25 L) is dɸ/ds = Σ ɸi (dNi /ds). From
Eq. (5.32), [dNi /ds] at the given point P is [–2/ L, 2/L, 0]. Hence, dɸ/ds = – (4)(2)/L + (4. 5)
2/L = 1/L.
70
5.3.3 Cubic Interpolation Function

The cubic interpolation function for a line element shown in Figure 5.7 can be derived in a
similar manner as that of quadratic function. In this case, 4 nodes are needed and the
polynomial for ɸ will be of the form
The boundary conditions are as follows,
Putting these conditions in polynomial in Eq. (5.33),
is obtained. Thus, the polynomial can be written as
From Eq. (5.37),
From Eqs. (5.36) and (5.38), the polynomial ɸ becomes
The polynomial ɸ can also be written as
Comparing Eqs. (5.40) and (5.41), it can be shown that
71
The first derivative of these functions Ni with respect to s are
Example 5.4
Evaluate the field variable ɸ at a point P(s = L/6) for a line element with cubic interpolation
function and also its first derivative at the same point, given that {ɸ}T = [4 4.5 6 9].
Solution:
From Eqs. (5.43) and (5.44), at s = L/6; [N] and [dN/ds] values are obtained as [0.3125
0.9375 –0.3125 0.0625] and [–2.875 2.625 0.375 –0.125]/L, respectively. Hence, ɸ(L/6) =
[N]{ɸ} = 4.15625 and (dɸ/ds) = [dN/ds]{ɸ} = 1.4375/L.
5.3.4 Lagrangian Form of Interpolation Function

It has been shown in the preceding section that one has to resort to matrix inversion to
derive shape functions Ni. Such a process can be quite cumbersome for higher order
polynomials. Same functions can be derived by using Lagrangian polynomial easily for any
order.
Figure 5.8 Line element

The Lagrangian polynomial of order n is given by
It can be easily seen that Ni(x) = 1 for x = xi and Ni(x) = 0 for x = xm, m ≠ i.
a) Linear Function:
For linear function, n = 2. Thus,
72
Considering Figure 5.8, s = x – x1 and L = x2 – x1. Thus, from Eqs. (5.46) and (5.47), N1 =
(s + x1– x2)/{–(x2 – x1)} = 1 – (s/L) and N2 = s/L. It may be noted that identical expression
has been derived earlier in Eq. (5.18).
b) Quadratic Function:
For quadratic function, n = 3. By using Eq. (5.45),
can be obtained. Further, it can be shown from Figure 5.9 that
Thus, the interpolation function can be written from Eqs. (5.48), (5.49), (5.50), (5.51) and
(5.52) as
Figure 5.10 Line element with linear

Figure 5.9 Line element with 3 interpolation function
nodes
It can be noticed that Eqs. (5.53), (5.54) and (5.55) are identical to Eq. set (5.31). In a
similar manner higher order interpolation function can be derived easily using Lagrangian
interpolation polynomial.
73
5.3.5 Further Higher Order Elements in One-Dimension

Further higher order interpolation functions are discussed here in one-dimension with more
number of nodes in each element. Procedure of deriving interpolation functions and its
properties is very similar to that derived for quadratic type interpolation functions earlier.
Here a special indexing scheme is used. The scheme would be initially demonstrated for
linear, quadratic and cubic type line elements before explaining further to higher order
elements (Heubner, 1975).
Consider the linear type line element shown in Figure 5.10.
Let L1, and L2 be length co-ordinates defined as
Nαβ(L1,L2) are interpolation functions, where α + β = n for nth order polynomial.
With the use of Eq. (5.58), interpolation functions are derived for various line elements.
Linear Interpolation Function: From Figure 5.10.
Figure 5.11 Quadratic type line element

Quadratic Interpolation Function: Refer to Figure 5.11.
74
Cubic Interpolation Function: Referring to Figure 5.12, α = 3, β = 0 at node 1; α = 2, β =

1 at node 2; α = 1, β = 2 at node 3; and α = 0, β = 3 at node 4.
Figure 5.12 Line element with cubic interpolation function
75
Quartic Interpolation Function: Quartic type line element has 5 nodes in one element as
shown in Figure 5.13. Using the formula
Figure 5.13 Line element with quartic interpolation function
76
Thus,
At Node 1:
At Node 2:
At Node 3:
At Node 4:
At Node 5:
Quintic Interpolation Functions: Quintic type line element has got 6 nodes in one
element as shown in Figure 5.14.
From Figure 5.14, using the formula
Figure 5.14 Line element with quintic interpolation function
77
Similarly,
Then, the interpolation functions can be obtained as follows.

At Node 1:
At Node 2:
At Node 3:
At Node 4:
At Node 5:
At Node 6:
78
In a similar way, further higher order interpolation functions for other one-dimensional
elements and its properties can be easily derived.
5.4 TWO-DIMENSIONAL ELEMENTS

Triangular, rectangular and quadrilateral elements are most widely used for two-
dimensional finite element modelling. Various properties of these elements and related
interpolation functions are discussed in this section.
5.4.1 Triangular Element: Linear Interpolation Function in Cartesian Co-ordinates

The triangular element was formulated by Turner et al. (1956). It is commonly used
element shape for two-dimensional problems. Triangular elements can be used when the
field variable is a function of two independent variables, namely x and y. Variation of field
variable ɸ over the domain of an element can be written in polynomial form as
where α1, α2 and α3 are the coefficients to be evaluated.

Let the nodes of the element be designated as 1, 2, 3, as shown in Figure 5.15, in counter-
clockwise direction with known co-ordinates (x1, y1), (x2, y2) and (x3, y3), respectively.
Let
where [P] = [1 x y] and {α}T = [α1 α2 α3].

If ɸ has values ɸ1, ɸ2 and ɸ3 at nodes 1, 2 and 3,
Combining Eqs. (5.84) and (5.86),
is obtained. Field variable ɸ can also be written as
Here, [N] = [P] [G]–1 and
79
Figure 5.15 Triangular element

As shown in Figure 5.16, variation of Ni, I = 1, 2 and 3 is such that Ni at node i is 1 and
zero at the other nodes. Independent variables x and y can be expressed in terms of nodal
co-ordinates as
where {x}T = [x1 x2 x3] and {y}T = [y1 y2 y3].
80
Figure 5.16 Variations of N1, N2 and N3 over the triangular domain
Example 5.5
The triangular element used for ground water flow simulation is as shown in Figure 5.17.
The nodal co-ordinates are (x1 = 1, y1 = 2), (x2= 4, y2 = 0.5) and (x3 = 3, y3 = 4). The nodal
values of hydraulic heads {ɸ} at different nodes are [3.5, 2.2, 4.4], respectively. Find the
value of hydraulic head ɸ at point (2.5, 2.5).
Figure 5.17 Element for example 5.5

Solution:
The value of ɸ at point p, ɸp can be written as
81
Example 5.6
A point sink (well) is located at a point p(x = 2.5, y = 2.5) whose strength Q is 0.2 m3/min in
a region as given in Example 5.5. Distribute the strength of the sink proportionately to the
nodes 1, 2 and 3.
Solution:
From Example 5.5, values of N1, N2 and N3 are 3.25/9, 2/9 and 3.75/9, respectively at point
P. Contribution of sink at node 1 is N1Q = 3.25(0.2)/9 = 0.07222 m3/min. Similarly, at node
2 it isN2Q = 0.0444 m3/min. and at node 3 the contribution is N3Q = 0.08333 m3/min. Note
that (N1 + N2 + N3)Q = 0.2.
First Derivatives of the Field Variable
It was shown in Eq. (5.88) that ɸ = N1 ɸ1 + N2 ɸ2 + N3 ɸ3. Then, the first derivatives of the
field variable ɸ with respect to x and y can be written as
Now Therefore,
Further, it can be deduced from Eq. (5.95) that
Thus,
82
In a similar manner,
Example 5.7
Find the values of (∂ɸ/∂x) and (∂ɸ/∂y) for the data given in Example 5.5.
Solution:
5.4.2 Triangular Element—Area

Co-ordinates
Generation of interpolation
functions for a triangular element
can be greatly simplified if the
area co-ordinates are chosen. Let
A be the area of the triangle 123
shown in Figure 5.18. If p is the
point inside the triangle, A1 = area
(p23), A2 = area (p13), A3 = area
(p12). Further, by letting L1, L2 Figure 5.18 Area co-ordinates
and L3 to be the area co-
ordinates, L1 = A1/A, L2 = A2/A and L3 = A3/A. If point p lies on edge 23, A1 = 0,
hence L1 = 0. If point p coincides with point 1, L1 = 1. These two conditions satisfy that
L1qualifies for N1. In a similar way, it can be shown that L2 and L3 satisfy conditions for N2
andN3. Field variable ɸ can, thus, be written as
This is true only for linear interpolation functions. Derivative of field variable with respect to
x and y can be written as
5.4.3 Integration Formula for Triangular Elements

Following formula can be used to evaluate integrals usually required in finite element
formulations using triangular elements (Connors and Brebbia, 1976)
83
where N1, N2 and N3 are the linear functions of (x, y) and i, j, k are integers.
Example 5.8
Evaluate I = ∫∫N2dA over the triangular element.
Solution:
This integral can be written as ∫∫ N11 N 22 N30 dA. Here, i = 0, j = 1 and k = 0. Using Eq.
(5.106), I = (0! 1! 0!)(2A)/(0 + 1 + 0 + 2)! = A/3.
Example 5.9
Evaluate I = ∫∫N1 N2 N2 dA over the triangular element.
Solution:
This integral can be written as ∫∫N11 N22 N30 dA. Here, i = 1, j = 2 and k = 0. Using Eq.
(5.106), I = (1! 2! 0!)(2A)/(1 + 2 + 0 + 2)! = A/30.
5.4.4 Triangular Element—Quadratic Function

If the procedure given in section 5.3.5 is followed to derive shape functions for a triangular
element with quadratic interpolation functions, evaluation of resulting [G]–1 would be very
tedious. Hence, a different strategy is to be used. A convenient method of obtaining
interpolating polynomial of higher order was suggested by Silvester (1969) by considering
triple index numbering scheme: Nαβγ(L1, L2, L3) such that α + β + γ = n, n is the order of the
interpolating polynomial; and L1, L2, L3 are the area co-ordinates, as shown in Figure
5.19(Huebner, 1975).
Figure 5.19 Triangular element with triple index numbering scheme
84
Integration of area co-ordinates over the area of an element can be done with the aid of
the formula (Huebner, 1975)
Example 5.10
Evaluate following integrals over a triangle with 6 nodes where Ni is quadratic in nature.
Solution:
85
5.4.5 Triangular Element—Cubic Interpolation Function

It can be shown by using the triple index numbering scheme as shown in Figure 5.20 that
the cubic interpolation functions for 10 nodded triangular elements are
Figure 5.20 Triangular element with 10 nodes
86
Example 5.11
Evaluate and at point P (1, 4) for a

triangular element shown in Figure 5.21 using
cubic interpolation function.
Solution:
For the given data, ai, bi, ci, i = 1,2,3 and A can be Figure 5.21 Example problem 5.11
obtained from Eq. (5.90), (5.91), (5.92) and (5.93)
as a1 = 16, a2 = 0, a3 = 0; b1 = –4, b2 = 6 and b3 = –2; c1 = –2, c2 = –1, c3 = 3 and A = 8.
Hence, L1, L2, L3 are
5.4.6 Two-Dimensional Rectangular Elements

Consider a rectangular element as shown in Figure 5.22. Variation of field variable ɸ over
the domain of the rectangular element can be written in polynomial form as
where αi, i = 1,2,3,4 are the coefficients to be evaluated. By letting values of ɸ to be ɸ1, ɸ2,
ɸ3and ɸ4 at nodes 1, 2, 3 and 4, respectively, matrix equation can be written as
where {ɸ}T = [ɸ1 ɸ2 ɸ3 ɸ4], {α}T = [α1 α2 α3 α4].
87
Figure 5.22 Rectangular element

Further, {α} = [G]-1 {ɸ}, where
Thus,
Value of N1 at node 1 is 1 and 0 at other nodes. Similarly, value of N2 at node 2 is 1 and 0

at other nodes. At any point within the element, N1(p) + N2(p) + N3(p) + N4(p) = 1 should be
satisfied. Derivatives of interpolation functions can be written as
Field variable ɸ can be expressed in terms of nodal ɸ values as
Since ɸ is at a particular node, it is not varying with respect to x and y. Hence, the first
derivative with respect to x and y are
88
Similarly,
Sometimes, it is more convenient to use non-dimensional interpolation functions of the

form
where Ni is a function of (ξ, η) or Ni

= N(ξ, η) at node i, ξ and η are the
local co-ordinates such that -1 ≤ ξ ≤
1, -1 ≤ η ≤ 1, ξi and ηi are the local
co-ordinates of the nodes such that
ξ = ±1 and η = ±1. For example, ξ 1
= -1 and η1 = -1 at node 1 and ξ2 = 1
and η2 = -1 at node 2; ξ3 = 1 and η3
= 1 at node 3 and ξ4 = -1 and η4 = 1
at node 4, as shown in Figure 5.23.
Figure 5.23 Rectangular element, local co-ordinates
Thus,
Local co-ordinates are often referred to as natural or intrinsic co-ordinates. Derivatives of

interpolation functions with respect to ξ and η are
5.4.7 Rectangular elements—Lagrangian Form in natural and Cartesian co-ordinates

Interpolation functions for rectangular elements with sides parallel to the global axes can
be developed using Lagrangian interpolation function. Consider a line segment of length 2
units with 3 nodes as shown in Figure 5.24(a). By applying Lagrangian interpolation
function,
These functions in global co-ordinates can be converted into natural co-ordinates by noting
that
Figure 5.24a Line segment with 3 nodes in

x–direction
Figure 5.24b Line segment

with 3 nodes in y-direction
89
Similarly, choosing η as natural co-ordinate in the y-direction as shown in Figure 5.24(b),
By using Eqs. (5.135),

(5.136), (5.137) and (5.138),
interpolation functions for a
rectangular element with 9
nodes can be constructed.
Interpolation functions at
different nodes of the element
can now be written as node 1
of the quadrilateral that is
formed by using nodes 1 and
3 of the line elements shown
Figure 5.25 Two-dimensional Lagrangian type element in Figure 5.25. Hence,
In a similar way, other interpolation functions can be derived.

Rectangular elements—Lagrangian
Interpolation Function in Cartesian
Co-ordinates
Consider the 9 nodded rectangular
elements (Figure 5.26) with
Lagrangian interpolation function in
Cartesian co-ordinates.
The Lagrangian interpolation function,
Ni in Cartesian co-ordinates can be
derived in a very similar manner as
done for rectangular elements with
Lagrangian form in natural co- Figure 5.26 Rectangular elements with 9 nodes
ordinates as discussed earlier.
Interpolation functions can be written
90
as
5.4.8 Isoparametric elements

Domains with curved irregular boundaries are often encountered in FEM analysis where
straight edged elements are difficult to use in discretization. Elements with curved edges
as shown in Figure 5.27 can be used advantageously for accurate representation of the
domain. Curve-edged elements of quadrilateral type were first introduced in FEM by Taig
(1961). Later many other researchers extended the idea of curved elements. Curved-
edged elements are often formulated by transforming simple geometric shapes in some
local co-ordinate system into distorted shapes in the global Cartesian co-ordinate system
and then by evaluating element properties for the resulting curve-sided element (Huebner,
1975). Mapping functions are used to transform simple geometric shapes to curved one.
If the functional representation of field variable and mapping functions are expressed by
interpolation functions of same order, the curve-edged element used is referred to as an
isoparametric element. If element geometry is described by lower order polynomial than
used for the field variable, the element is termed as sub-parametric element. On the other
hand, element is called super-parametric element if the element is described by higher
order polynomial compared to the field variable. Out of three categories of curve-edged
elements, isoparametric element is most commonly used. Hence, properties of
quadrilateral isoparametric elements are discussed here. To demonstrate properties of
isoparametric element in two-dimensions, consider a parent element of four nodded
rectangular elements and its corresponding curve edged isoparametric element as shown
in Figure 5.27.
As discussed earlier, field variable ɸ can be expressed in terms of nodal ɸ values as
Nodes in the ξ—η plane may be mapped into corresponding nodes in the x—y plane by
defining
91
Since ɸ is specified as a function of the local co-ordinates ξ, and η, it is necessary to

express derivatives of ɸ as a function of ξ and η. This can be done as follows.
Figure 5.27 Element with curved edge and its mapping to natural co-ordinate system
Further, x and y are functions of local co-ordinates ξ and η such that x = x(ξ, η) and y = y(ξ,
η). Thus, following transformations can be used to formulate element matrices.
Eqs. (5.147) and (5.148) can be written in matrix form as
92
These derivatives as given by Eqs. (5.145) and (5.146) are possible only when inverse of
the Jacobian, [J]-1exists. This inverse exists provided interior angles of the quadrilateral
element are less than 180o. Numbering of nodes must follow the counter–clockwise
direction to produce positive |J| for the coordinate system shown in Figure 5.27.
Since, can be written as
Figure 5.28 Cubic interpolation function in one-dimension

In the above discussion, two-dimensional quadrilateral element has been used to illustrate
the concept of isoparametric element. The same concept can be extended to other type of
elements or for one-dimensional and three-dimensional domains.
5.4.9 Lagrangian Interpolation Functions for two-Dimensional elements

Here, the Lagrangian interpolation function elaborated in Section 5.4.7 is further
discussed.
The Lagrangian interpolation function can be expressed, with reference to Figure 5.28, as
Using the expression (5.160), interpolation functions for nodes 1, 2, 3, 4 can be written as
93
Figure 5.29a Natural co-ordinate system in ξ

direction
Interpolation functions in natural co-ordinates can

be derived using the above relations, with Figure 5.29b Natural co-ordinate
reference to Figure 5.29(a) as follows. system in η-direction
Similar interpolation functions in natural co-ordinates in the η direction (Figure 5.29b) can
be described as
94
Using the above expressions, interpolation functions for a Lagrangian element with 16
nodes (Figure 5.30) can be derived as
Figure 5.30 Lagrangian element with 16 nodes
95
5.4.10 Two-Dimensional serendipity elements

It can be seen from Sections 5.4.4 and 5.4.5 that internal nodes are necessary to maintain
completeness of polynomial for higher order
triangular element. Further, it can be inferred
from Section 5.4.9 that a number of identical
nodes on two opposite sides are required for
higher order Lagrange elements. However,
nodes on sides cannot be added without
introducing additional interior nodes. A sample
of some Lagrange elements is shown in Figure
5.31. Polynomials are cubic in both directions
as shown in Figure 5.31(a). On the other hand,
polynomial is quadratic in η direction (Figure
5.31b) and is linear in η direction (Figure 5.31c).
It may be remarked that Lagrange polynomials
are never complete and they possess
geometric isotropy only when equal number of
nodes are present in both directions.
It is possible to generate interpolation functions
dependent on nodal values placed on element
boundary (Zienkiewicz and Taylor, 1989).
These elements, known as the serendipity
elements, were derived by chance. They are
very useful since their construction can be
formalized to a concise numerical algorithm.
Here step by step construction of interpolation
function is demonstrated through an 8-node
quadratic element in natural co-ordinates (ξ, η).
To facilitate the construction, following one-
dimensional interpolation functions are used for
s = ξ or η.
Figure 5.31 Sample Lagrange
elements (a) Cubic–cubic (b) Cubic–
quadratic (c) Cubic – linear
96
The basic 4-node rectangle and the targeted

8-node element with node numbering are
illustrated in Figure 5.32 in ξ – η co-
ordinates. For the basic 4-node linear
element, interpolation functions are
Figure 5.32 Serendipity elements (a)

It is observed from Figure 5.33 that N5(ξ, η)/2
Basic 4-node element (b) Targeted 8-
needs to be subtracted from N1(ξ, η) to
node element
construct 5-node element. Here the revised
N1(ξ, η) can be then evaluated as
Thus, the field variable

To extend this construction to 6-node element, it is observed from Figure 5.34 that N5/2
andN6/2 need to be subtracted from N2 and N6/2 from N3, where N6 = n2(1)(ξ)n2(2)(η), which
yields
97
Figure 5.33 Generation of 5-node serendipity element
Similarly,
Figure 5.34 Generation of 6-node serendipity element
98
By redefinition,
Continuing the process, 8-node quadratic element can be generated with interpolation
functions
The above equations for linear and quadratic elements in compact form can be listed as
follows.
1. Linear element
2. Quadratic element
Corner nodes:
Mid-side nodes:
5.5 THREE-DIMENSIONAL ELEMENTS

Generally, three-dimensional elements maintain C0 continuity. Thus, continuity of field
variable along faces of an element is the basic requirement. Also, it needs to be ensured
that nodes on element interfaces uniquely define variation of function on interfaces. As
compared to the basic two-dimensional triangle and quadrilateral elements, the analogous
elements in three-dimensions are tetrahedral and hexahedral. In this section,
characteristics of basic three-dimensional elements such as tetrahedral and prism
elements are discussed.
99
5.5.1 Tetrahedral Elements

Four-nodded tetrahedral is the simplest element in three-dimensions (Figure 5.35).
Tetrahedral elements can be used when field
variable is a function of three independent variables,
namely x, y and z.
Variation of field variable ɸ over the domain of
element can be written as
where [P] = [1 x y z] and {α}T = [α1 α2 α3 α4].

By letting nodes of the element designated as 1, 2, 3
Figure 5.35 Tetrahedral element and 4 with known co-ordinates (x1, y1, z1), (x2, y2, z2),
(x3, y3, z3) and (x4, y4, z4), respectively, it can be
shown that
By combining, Eqs. (5.198) and (5.199),
is obtained. The field variable ɸ can also be written as
where [N] = [P] [G]–1 with
100
Other constants can be obtained through a cyclic permutation of subscripts 1, 2, 3 and 4.

By using Eqs. (5.198), (5.202) and (5.203), it can be shown that
Therefore,
Let p be any point within the tetrahedral chosen as shown in Figure 5.36. Let V1, V2, V3
and V4be volumes of tetrahedrals p234, p134, p124 and p123, respectively. Then the
volume co-ordinates Li can be expressed as
In this case, Ni = Li = Li(x, y, z) are linear interpolation functions.
Figure 5.36 Tetrahedral element—volume co-ordinates

Derivatives of field variable ɸ can be written as
Integration formula over an element for linear tetrahedral element can be shown to be
(Huebner, 1975)
101
where ! is for factorial, i, j, k and l are integers, N1, N2, N3 and N4 are linear functions of x,
y, z.
Example 5.12
Nodal co-ordinates of tetrahedral element are (x1 = 0, y1 = 0, z1 = 0), (x2 = 1, y2 = 1, z2 = 3),
(x3= 2, y3 = 2, z3 = 0), (x4 = 0, y4 = 3, z4 = 0). Find linear interpolation functions and value of
field variable ɸ at a point (1.5, 1.5, 2), given that {ɸ}T = [3 2 2.5 3.5].
Solution:
From Eq. (5.207),
Then, [N] = [P] [G]–1. Thus, N1 = (18 – 3x – 6y – 3z)/18; N2 = 6z/18; N3 = (9x – 3z)/18; N4 =
(–6x + 6y)/18;
At point (1.5, 1.5, 2), [N] = [–0.0833 0.667 0.4167 0], therefore, ɸ = [N] {ɸ} = 2.124.
Example 5.13
Nodal co-ordinates of tetrahedral element are: (x1= 0, y1= 0, z1= 0), (x2= 1, y2= 1, z2= 3),
(x3= 2, y3= 2, z3= 0), (x4= 0, y4= 3, z4= 0). Determine linear interpolation functions in terms
of volume co-ordinates and find value of ɸ at a point (1, 1, 2), given that {ɸ}T = [3 2 2.5
3.5].
Solution:
From Example 5.12, volume of tetrahedral is V = 3. Various required volumes are:
102
Thus, L1 = V1/V = 1/6; L2 = V2/V = 2/3; L3= V3/V = 1/6; and L4 = V4/V = 0;
Finally, [N] = [L] = [1/6, 2/3, 1/6, 0]; Thus, ɸ = [L] {ɸ} = 2.25.
Example 5.14
Evaluate ∂ɸ/∂x, ∂ɸ/∂y, ∂ɸ/∂z at point p (1.5, 1.5, 2) for the data given in Example 5.12.
Solution:
From Example 5.12,
N1 = (18 – 3x – 6y – 3z)/18; N2 = (0 + 0 + 0 + 6z)/18;
N3 = (9x + 0 + 0 – 3z)/18; N4 = (–6x + 6y + 0 + 0)/18; and 6V = 18. Further,
{b}T = [–3 0 9 –6]; {c}T = [–6 0 0 6]; {d}T = [–3 6 –3 0] and {ɸ}T = [3 2 2.5 3.5]
From Eq. (5.211),
From Eq. (5.212),
From Eq. (5.213),
Example 5.15
Evaluate following integrals over tetrahedral element.
Solution:
From the integration formula (5.214),
103
5.5.2 Tetrahedral elements: Quadratic Interpolation Function

Quadratic interpolation functions for tetrahedral elements can be easily derived by
following the procedure, which is analogous to the one used for triangular elements
(Silvester, 1969). Thus, referring to Section 5.4.4 the interpolation functions for the
tetrahedral element shown in Figure 5.37 are
Here Li = Li (x, y, z) = (ai+ bix + ci y + diz)/6V, where ai, bi, ci, di and V are as defined by Eq.
(5.205) and (5.206).
5.5.3 Tetrahedral elements: cubic Interpolation Function

Cubic interpolation functions for tetrahedral elements can be easily derived by following
the procedure, which is analogous to the one used for triangular elements (Silvester,
1969). Thus, referring to Section 5.4.5 the interpolation functions for the tetrahedral
element shown in Figure 5.38 are
5.5.4 Three-Dimensional elements—prismatic elements

A variety of prismatic elements such as triangular prisms, rectangular prism and
hexahedral can be used in the three-dimensional discretization. Shape of a typical
undistorted prismatic element is shown
104
Figure 5.37 Tetrahedral elements with quadratic interpolation function
Figure 5.38 Tetrahedral element with cubic interpolation function

in Figure 5.39. By choosing the co-ordinate system and node numbering as shown,
interpolation functions for different nodes can be written as
105
Figure 5.39 Rectangular prism element
First derivatives of interpolation functions with respect to x, y and z are as follows.
106
5.5.5 Three-Dimensional elements in local co-ordinates

Interpolation functions for a prismatic element in the local (or natural or intrinsic)
dimensionless co-ordinates ξ, η and ζ in x, y and z directions can be defined in a manner
similar to the one used for two-dimensional system. Let ξ = (x – xc)/a; η = (y – yc)/b and ζ =
(z – z)/c in Figure 5.39, where –1 ≤ ξ ≤ 1; –1 ≤ η ≤ 1; –1 ≤ ζ≤ 1. Then, interpolation function
for any node i can be written as
where ξ i, η i, ζ i are the local co-ordinates. Or,
Further,
For a distorted element, x = x(ξ, η, ζ), y = y(ξ, η, ζ) z = z(ξ, η, ζ). Then,
107
In matrix form, Eq. (5.231) can be written as
5.5.6 Three-Dimensional serendipity elements

By following procedure and analogous notations described in Section 5.4.10, interpolation
functions for three-dimension rectangular prismatic elements shown in Figure 5.40 can be
derived. The origin of the ξ,η,ζ, co-ordinate system is at the centroid of the element. Here,
these functions are listed in compact forms.
Figure 5.40 Rectangular prismatic serendipity elements (a) 8-node element (b) 20-node
element
1. Linear element
108
2. Quadratic element
Corner nodes:
Typical mid-side node:
with the remaining mid-side node expression obtained by changing variables.
5.6 CLOSING REMARKS

Different types of finite elements in linear and higher order forms suitable for a variety of
problems are presented in this chapter. Characteristics of various polynomials are also
discussed and their properties are presented. With the basic idea of polynomial
interpolation functions and Lagrangian functions, general procedures for constructing finite
elements and characteristics have been discussed. Using the linear interpolation
concepts, element properties for line, triangular, rectangular, tetrahedral and prism
elements have been elaborated. Using the higher order interpolation concepts, element
properties for one-dimensional line element, two-dimensional triangular and rectangular
elements are discussed. Each case has been explained in local and natural co-ordinate
systems. Also, the basics of isoparametric element and its properties are introduced.
Further, two-dimensional and three-dimensional serendipity elements are derived.
In finite element analysis, use of higher order elements can reduce the number of
elements to be used for a particular problem but more computational time is required for
the higher order elements. Depending upon the problem to be solved, suitable elements
can be chosen for discretization, and element properties can be derived for solution of the
problem. Details of interpolation functions and element properties given in this chapter can
be used as general guidelines for all types of elements used for discretization.
EXERCISE PROBLEMS
1. Using the linear interpolation function for line element, find the value of the field
variable ɸ and its first derivative at a place x = 0.8 given that xi = 0, xj = 5, ɸi = 2.5
and ɸj = 7.0.
2. Using the integration formula for linear type line element, evaluate
3. Temperature distribution in a steel plate is simulated using the linear type triangular
element with the nodal co-ordinates of (x1 = 1, y1 = 1), (x2 = 8, y2 = 0.5) and (x3 = 4,
y3 =5). The nodal values of temperature {ɸ} at different nodes are {25, 27, 23},
respectively. Find the value of temperature ɸ at point (3.5, 3.5).
4. The triangular element is used for ground water flow simulation. The nodal co-
ordinates are (x1 = 1, y1 = 1), (x2 = 8, y2 = 0.5) and (x3 = 4, y3 = 5). The nodal values
of hydraulic heads {ɸ} at different nodes are {12.5, 12.2, 12.8}, respectively. A point
source (recharge well) is located at a point p(x = 3.5, y = 3.5) whose strength Q is
0.3 m3/min Distribute the strength of the source proportionately to the nodes 1, 2
and 3.
5. Find the values of (∂ɸ/∂x) and (∂ɸ/∂x) for the data given in problem 3.
6. Using the integration formula for linear type triangular element evaluate I = ∫∫ N12
N2N3 dA over the triangular element.
109
7. Evaluate (∂Ns/∂x) and (∂Ns/∂y) at a point P(1, 4) for the triangular element with
quadratic interpolation function.
8. The co-ordinates of the linear tetrahedral element are: (x1 = 2, y1 = 2, z1 = 0), (x2 =
3, y2= 3, z2 = 4), (x3 = 3, y3 = 3, z3 = 0), (x4 = 5, y4 = 4, z4 = 0). Determine the linear
interpolation functions and the value of field variable ɸ at a point (2, 2,1.5), given
that {ɸ}T = [5.5 7 6.7 3.5].
9. Evaluate ∂ɸ/∂x, ∂ɸ/∂y, ∂ɸ/∂z at point p(2, 2, 3) for the data given in problem 7.
10. Using the integration formula for linear type tetrahedral element, evaluate the
following integrals over the tetrahedral element: (1)
11. Using the quadratic interpolation function for line element, find the value of the field
variable ɸ and its first derivative at a place x = 0.8 given that xi = 0, xj = 5, ɸi = 2.5
and ɸj = 7.0.
12. Calculate the field variable ɸ at a point P(s = L/3) for a line element with cubic
interpolation function and also its first derivative at the same point, given that {ɸ}T =
[2 3 5 5 7.5].
13. Using the integration formula for quadratic type triangular element with 6 nodes,
evaluate I = ∫∫N12 N2 N3 dA over the triangular element.
14. The temperature distribution in a steel plate is simulated using the quadratic type
triangular element with the nodal co-ordinates of (x1 = 1, y1 = 1), (x2 = 8, y2 = 0.5)
and (x3= 4, y3 = 5). Find the value of (∂Ns/∂x) and (∂Ns/∂y) at a point P (3.5, 3.5).
15. Assuming cubic interpolation functions, for the problem in question 4, determine
(∂Ns/∂x) and (∂Ns/∂y) at a point P(3, 4).
16. Using the special indexing scheme for line element, derive the interpolation
functions for an element with 8 nodes.

Connors, J. J., and Brebbia C. A., (1976). Finite Element Techniques for Fluid Flow,
Newnes-Butter worths, London.
Chapra S. C., and Canale R. P. (2002). Numerical Methods for Engineers, 4th ed. Tata
McGraw-Hill, New Delhi.
Desai, C. S., and Abel J. F., (1987). Introduction to the Finite Element Method – A
Numerical Method for Engineering Analysis, Wads Worth, California.
New York.
Reddy J. N. (1985). An Introduction to the Finite Element Method, McGraw-Hill
International, New York.
Segerlind, L. J. (1987). Applied Finite Element Analysis, John Wiley & Sons, New York.
Silvester, P. (1969). Higher order polynomial triangular finite elements for potential
problems, International Journal of Engineering Science, 7:849–861.
Taig, I. C. (1961). Structural analysis by the matrix displacement method, English Electric
Aviation Report, SO-17.
110
Turner, W. J., Clough, R. W., Martin, H. C., and Topp L. S. (1956). Stiffness and deflection
analysis of complex structures, Journal of Aeronautical Science,23:805–823.
Nonlinear Problems, John Wiley & Sons, New York.
Zienkiewicz, O. C. (1980). Finite Element Method, 3rd ed., Mac-Graw Hill, New York.
Zienkiewicz, O. C., and Taylor R. L. (1989). The Finite Element Method, Vol 1, Mac-Graw
Hill, London.
111
6 One Dimensional Finite Element Analysis

6.1 INTRODUCTION
Many practical problems can be modelled by employing one-dimensional finite elements
(also referred to as line elements). The one-dimensional discretization can give good
results when the primary unknown quantity (i.e., unknown continuous field to be modelled)
vary mainly in one direction and remains constant in the other two dimensions. Typical
examples of situations, where the one-dimensional analysis may be adequate, are framed
(skeletal) structures subjected to external forces; torsion of circular shafts; flow through
pipes, seepage through narrow, confined soil mass; heat conduction across a fire wall;
flow analysis in channels etc. When FEM is employed for one-dimensional analysis, 2-
node elements are used extensively. The FEM analyses of various one-dimensional
problems are illustrated in this chapter with various formulations and examples.
When FEM is employed for one-dimensional analysis, different types of 2-node elements,
as given in Table 6.1, are extensively used. All the elements summarized in Table 6.1
[except space frame elements (Gere 1965)] are elaborated in the following sections.
6.2 LINEAR SPRING

The linear spring–the simplest kind of line element–is considered first because it provides
a simple, yet generally instructive, tool to illustrate the basic concepts of the FEM. For the
illustration, consider the two spring assemblage as shown in Figure 6.1.
It is required to calculate the displacements at points (nodes) B and C, force each spring
and for the reaction at A by employing FEM.
This problem can be solved by following the steps outlined in Chapter 4.
Step 1: Discretize and select element types
The two spring assemblage shown in Figure 6.1 is discretized by employing two-node
spring elements, as shown in Figure 6.2(a).
The discretized model has three nodes–two externally applied nodal forces F2 = FB and
F3= Fc, and two spring elements. A typical spring element is shown in Figure 6.2(b).
A typical spring element has two nodes. The variation of displacement along x direction is
expressed in terms of axial displacements u1 and u2 at nodes 1 and 2 of the element,
respectively. In Figure 6.2(b), f1 and f2 are the elemental nodal forces. Note the sign
convention for nodal forces and displacements. The lower case x represents the local or
element co-ordinate system. It will be assumed that tensile forces are positive and
compressive forces are negative.
Table 6.1 Summary of 2-node finite elements used in one-dimensional analysis
112
Figure 6.1 Two spring assemblage

Step 2: Select approximation functions
A line can be defined uniquely by specifying two points on it. Hence, displacement function
for u(x) is assumed to vary linearly along the x-axis of the element because the element
has two nodes (and hence, two degrees of freedoms (DOF)). Therefore, assume
Figure 6.2a Finite element discretization of two spring assemblage
Figure 6.2b Typical spring element
113
Figure 6.3 Assumed variation of displacement field

where a1 and a2 are unknown coefficients. The u(x) can be expressed as a function of
nodal displacements u1 and u2 by employing conditions (Figure 6.3):
Shape Functions
N1 and N2 are called ‘shape functions’ (also known as interpolation or basic functions),
shown in Figure 6.4, because Ni's express shape of the assumed displacement function
over the domain of the element when the i-th element degree of freedom has unit value
and all other degrees of freedom are zero. A property of shape functions is that their sum
equals unity. Such a property is essential to ensure proper simulation of rigid body
displacements. Hence,
By expressing Eq. (6.3) in matrix notation,
114
Figure 6.4 Shape functions

is obtained where
Step 3: Define the strain-displacement and constitutive relationships

Tensile force T produces a total elongation (deformation) δ of the spring. T and δ are
related through Hooke's law for the linear spring by
where δ = u2 – u1 and K is the spring constant (stiffness of the spring).

Step 4:
Derive element equations
Since the physics of a spring is simple to understand, the direct equilibrium method can be
employed to derive element equations expeditiously. A spring element subjected to tensile
(positive) force T is shown in Figure 6.5.
By comparing Figures 6.2(b) and 6.5, it can be easily seen that
For the equilibrium of generic element (in the x direction), therefore,
It can be shown from Eqs. (6.7), (6.8) and (6.9) that
115
Figure 6.5 Spring element in equilibrium

Alternatively, in the matrix notation,
i.e., the element equation for a generic element is
Step 5: Assemble the element equations to obtain the global equations and introduce
boundary conditions
By employing Eq. (6.11) to elements 1 and 2, equations
and
can be obtained. Note that throughout the text, the superscript in parentheses refers to the
element number.
To derive global (or the system) equations, consider the free-body diagram of individual
spring elements as shown in Figure 6.6. Here, the nodal forces are shown to be consistent
with the element force-sign convention.
Figure 6.6 Free-body diagram of spring assemblage

The concept of continuity (compatibility) is employed implicitly in the assembly process as
shown in Figure 6.7. This concept requires that the body remains together and that no
tearing occurs anywhere in the system.
116
Figure 6.7 Inter-element compatibility

Therefore, elements 1 and 2 must remain connected at the common node 2 throughout
the deformation process. This continuity requirement yields
Furthermore,
In the above Eqs. (6.14) and (6.15), u1, u2 and u3 are displacements of nodes 1, 2 and 3,
respectively. Assembly can be performed in two ways:
Assembling by nodal equilibrium conditions; and
Assembling by direct superposition (direct stiffness method).
Assembling by Nodal Equilibrium Conditions
From Figure 6.6, the nodal equilibrium conditions at nodes 1, 2 and 3 can be written,
respectively, as
In view of Eqs. (6.12), (6.13), (6.14) and (6.15), Eq. (6.16) can be written as
Alternatively, in the matrix notation,
or
117
where = total or assembled stiffness matrix.
Assembling by Direct Superposition (Direct Stiffness Method)

Element equations for elements 1 and 2 are given by Eqs. (6.12) and (6.13), respectively.
These element matrices cannot be added directly together because they are not
associated with the same DOF's. An easy way of adding these matrices is to rewrite them
in an expanded form as
By adding Eqs. (6.21) and (6.22) together, the global (assembled equations) matrix can be
obtained as
From Eq. (6.16), Eq. (6.23) can be written as
118
For problems involving a large number of DOF's, it is tedious (and a waste of computer
resources) to expand each element matrix to the order of total stiffness matrix. To avoid
this expansion, a shortcut is employed in the assembly process where the columns of
each element matrix are labelled according to the DOF's associated with them as follows:
The system is then constructed simply by directly adding terms associated with DOF's
in (1) and K(2) into their corresponding identical DOF locations in as
The {q} and {F} are then simply arranged in the order of DOF locations as in . Thus,
The determinant of can be obtained as
It can be noticed that is symmetric and singular. Since is singular and F1 is an

unknown nodal force (reaction), Eq. (6.23) cannot be solved for {q}. The mathematical
singularity of is related to the physical reality that the spring assemblage cannot stand in
the space uniquely without having proper boundary conditions.
Boundary Conditions
There are two types of boundary conditions: (i) geometric, forced, essential or kinematics
boundary condition (also known as Dirichlet conditions in the context of seepage), and (ii)
the natural (sometimes referred to as Neumann conditions in the context of seepage)
boundary conditions. In the context of structural analysis, the latter type of boundary
condition may be, say, zero bending moment at the hinge support of a beam. The
assembled equations need to be modified for forced boundary conditions only. (There is
no need to satisfy natural boundary conditions. They will be satisfied automatically.)
119
Forced-boundary conditions are further classified into two categories of homogeneous and
non-homogeneous boundary conditions. The homogeneous or the zero-boundary
conditions occur at locations that are completely restrained (i.e., prevented from moving).
On the other hand, non-homogeneous boundary conditions occur where non-zero values
are specified for some selected DOF (e.g. support settlement). In general, forced-
boundary conditions can be treated mathematically by partitioning the global equations as
follows:
Here, 1 contains the unconstrained DOF's and 2 contains specified (constrained) DOF's.
By expanding Eq. (6.27),
are obtained. For homogeneous-boundary conditions, = {0}. Therefore,
For the two spring assemblage problem of Figure 6.2(a), suppose forced-boundary
condition is specified at node 1 (i.e., u1 is specified). Then, the assembled equations can
be partitioned as
For homogeneous-boundary conditions (i.e., u1 = 0),
120
is obtained. Note that Eq. (6.35) could have also been obtained directly by deleting the row
and column of Eq. (6.23), corresponding to the zero-displacement DOF.
To illustrate the case of non-homogeneous–boundary conditions, let u1 = δ, where δ is a
known (specified) displacement.
In view of Eq. (6.28), Eq. (6.32) can be written as
Hence, the row and column corresponding to the non-zero DOF of assembled equations
cannot be simply deleted when dealing with non-homogeneous–boundary conditions.
Step 6: Solve for the unknown DOF
Eqs. (6.35) and (6.36) are the global equations modified for boundary conditions. By
solving Eq. (6.35) u2 and u3 can be obtained as follows.
Step 7: Solve for secondary quantities

The secondary quantities of interest in this problem are the reaction at node 1 and the
element forces. The reaction at node 1, F1, can be found from either Eq. (6.29) or Eq.
(6.31) depending on the type of boundary condition. Finally, element forces are
determined by back-substitution, applied to each element equation.
Figure 6.7a Spring assembly in series

Step 8: Interpret the results
For spring assemblage, the computed global nodal forces and element nodal forces can
be shown on a free-body diagram for easy interpretation.
Illustrative Example 1
For the spring assemblage shown in Figure 6.7(a), obtain
the assembled stiffness matrix;
the displacements of nodes 2 and 3;
the reaction forces at nodes 1 and 4; and
the forces in each spring.
121
Nodes 1 and 4 are fixed.

Element equation for spring element is given by
The global is given by
The global equation has the form
Boundary conditions for this problem are u1 = u4 = 0.

By partitioning the global equation (or deleting the first and the fourth rows and columns
from ) and by setting F2 = 0 and F3 = 1,000, the modified equations are obtained as
Figure 6.7b Free body diagram of series spring assemblage
By solving for u2 and u3,
The global nodal forces at nodes 1 and 4 can then be computed as
122
Use element equations to obtain element forces.
The free body diagram of the spring assemblage is shown in Figure 6.7(b).
6.2.1 Expressions for Equivalent Spring Constant and Nodal Forces

In situations where springs are arranged parallel to each other or in series, an equivalent
element can be formed. This equivalent element can be employed to reduce the size of
the problem at hand. Expressions for equivalent spring constants and nodal forces for
parallel springs and springs in series are derived below.
6.2.1.1 Parallel Springs

Figure 6.8(a) shows a spring assembly in parallel conditions. The element equations can
be derived as given below.
Figure 6.8a Spring assembly in parallel

The global equations can be obtained by following the assembly procedure as:
By comparing the global equations with equations for a generic spring element, it can be
seen that
123
6.2.1.2 Springs in Series

Figure 6.8(b) shows spring system in series. The element equations can be obtained as
follows
The global equations can be obtained by following the assembly procedure as:
By condensing u3 (refer to Chapter 10 on computer implementation of FEM for static

condensation process) from the global equations,
Figure 6.8b Spring system in series
By comparing Eqs. (6.40) and (6.41) with equations for a generic element, it can be seen
that
124
6.3 TRUSS ELEMENT

Lattice transmission towers, steel truss bridges or roof support systems can be modelled
accurately by employing truss elements. In this section, the element stiffness matrix is
derived for a linear-elastic truss (or bar) element. Local and global co-ordinate systems are
introduced to facilitate the formulation. Each element has its own local co-ordinate system
which is designated by lower case letters, x, y, z whereas the global co-ordinate system
(i.e., reference co-ordinate) is indicated by upper case letters X, Y, Z (Figure 6.9(a)).
Similarly, the displacement components with respect to local co-ordinate system are
indicated by the lower case letters u, v, w and those in the global co-ordinate system are
shown by the upper case letters U, V, W. Moreover, the element level matrices and
vectors are shown by lower case and upper case letters, respectively, when referred to
local and global co-ordinate system. Thus, ke and Ke represent element stiffness matrix in
the local and global co-ordinate systems, respectively. However, the element DOF vector
is referred to as {q1e} in the local co-ordinate system, and as {q1e}, in the global co-ordinate
system. (Note that in the previous section on analysis of spring assemblage, the local co-
ordinate system was assumed to coincide with the global co-ordinate system. Hence, both
{qle} and {qe} had the same meaning).
6.3.1 Plane Truss

A typical example of a plane truss is shown in Figure 6.9(a).
Figure 6.9a A typical example of plane truss
125
Figure 6.9b Two-node truss element subjected to tensile force T

Note the global co-ordinate system (X, Y) and local co-ordinate systems (x, y) for elements
I and j. There is only one global co-ordinate system to represent the entire structure,
whereas each element has its own local co-ordinate system. The generic element of
length L from this plane truss is shown in Figure 6.9(b).
The following assumptions are made to formulate element equations for a truss member.
Displacements and forces pointing in the positive directions of axis are positive.
Tensile forces are positive.
Truss (bar) is linearly elastic and has a constant cross-sectional area. (Later, a truss
member having varying cross-section will be considered.)
By definition, a truss member cannot sustain shear force. In other words, f1y = f2y = 0.
Consequently, effect of transverse displacement (along local y-axis) is ignored.
Hooke's law applies: that is σx = Eεx.
Thus, it can be shown easily that
where A is the cross-sectional area of bar, E is the modulus of elasticity of bar material
and δ = deformation = u2 – u1.
Eq. (6.45) leads to
where K = (AE/L). Hence, a truss member and a spring member are analogous as shown
in Figure 6.9(c).
By following the procedure adopted for spring element, element equations (in local co-
ordinate system) for a truss member can be obtained easily. Thus, approximation function
can be written as
Figure 6.9c A truss member and spring member
Similarly, the strain–displacement relation and the constitutive law can be expressed as
126
respectively, where K = (AE/L).

Equilibrium condition for the element (in x direction) gives
which leads to element equations.
Assembly process and other steps are similar to those for the spring element.
For the three-bar assemblage shown in Figure 6.10, determine (a) the assembled stiffness
matrix, (b) the displacements of nodes 2 and 3, and (c) the reactions at nodes 1 and 4.
By using Eq. (6.51), the element stiffness matrices are obtained as
Figure 6.10 Three-bar assemblage

Therefore, the assembled equations are
127
Applying homogeneous boundary conditions u1 = u4 = 0, the above assembled equation

set (6.52) reduces to
By solving the above 2 × 2 equation set, the unknown displacements are obtained as
Back substitution of Eq. (6.54) in Eq. (6.52) yields
6.3.2 Element Equations by Minimizing Potential Energy

One of the alternative methods often used to derive element equations for an elastic body
is based on the principle of minimizing potential energy which is one of the energy (or
variational) methods. In this text, the method of minimizing potential energy will be referred
to as the energy method. This method is widely used in finite element formulations,
especially those involving two- and three-dimensional domains.
As mentioned in the Chapter 4, the energy method is based on the idea of finding
equilibrium states of the body associated with stationary values of a scalar quantity. This
scalar quantity is the potential energy, πp, in structural analysis.
The potential energy πp is given by
where the strain energy, ø, is the capacity of internal forces (or stresses) to do work
through strains in the body. Work done by the external forces, Ω, on the other hand, is the
capacity of forces such as body forces, surface traction forces, and applied nodal forces to
do work through deformations of the body. Expressions of ø are given below for the one-,
two- and three-dimensional deformations (Refer to Zienkiewicz, 1980 for details.)
One-Dimensional Deformations in the x Direction
It can be shown that
where σx and εx are, respectively, the stress and the strain along x. In the matrix notation,
Eq. (6.57) can be shown as
128
Here, {σ} = {σx} and {ε} = {εx}.

Two-Dimensional Deformations in the X–Y Plane
where σx, σy are normal stresses in the X and Y directions, respectively εx, εy are normal
strains in the X and Y directions; and τxy and γxy are the shear stress and strain,
respectively, in the X-Y plane. In the matrix notation,
where
Three-Dimensional Deformations
The strain energy expression for a body experiencing three-dimensional deformation is
given by
Here, σx, σy and σz are normal stresses in the X, Y and Z directions, respectively; εx, εy and
εz are normal strains in the X, Y and Z directions; and τxy, τyz, τzx and γxy, γyz, γzx are,
respectively, shear stresses and strains in the X, Y and Z direct strains. In the matrix
notation,
where
6.3.2.1 General Forces Acting on one-Dimensional Truss Element

For the forces shown in Figure 6.11(a), the Ω is given by
129
Where xb = body force (weight) per unit volume
Tx = surface loading or traction per unit surface area
V = volume
S1 = part of the surface on which surface loading acts
M = number of points at which joint forces are applied.
Figure 6.11a Bar element subjected to loads
6.3.2.2 Derivation of Stiffness Matrix by Energy Method

By ignoring body forces and surface tractions, πp for a truss element is obtained as
By using Eq. (6.47), the axial displacement function can be expressed in terms of the
shape functions and nodal displacements as
The axial stress–strain relationship is given by linear Hooke’s law
130
where
Eq. (6.64) can be written in matrix notation as
By substituting Eqs. (6.68) and (6.71) into Eq. (6.73),
is obtained. Here,
By expanding Eq. (6.75), following equation can be derived
By applying the principle of minimum potential energy, the equilibrium state corresponds to
Therefore,
Hence, the element equation for a truss member in the local co-ordinate system is
131
Therefore,
Note that A or E have not been assumed to be constant. Therefore, any variation in A or E
with respect to x-direction can be included in [ke].
In view of Eq. (6.69),
By substituting Eq. (6.82) into Eq. (6.81), the element matrix for a truss member having
constant A and E can be derived to be
Therefore, element equations take the form
Figure 6.11b Bar element with varying cross section
For the bar element with varying cross-section given in Figure 6.11(b), derive the element
matrix. Consider the problem shown in Figure 6.11(b) for which E is constant and A varies
linearly,
The element stiffness matrix can be derived as
132
Hence, the element equations are given by
6.3.3.3 Equivalent Nodal Loading Due to Surface Forces and Body Forces
It was shown previously that
Therefore,
From the principle of minimum potential energy,
133
After assembling,
6.3.3 Local and Global Element Equations for a Bar in the X–Y Plane
Eq. (6.80) relates the displacement components in the local direction to the forces in the
local x direction. Therefore, Eq. (6.80) is called ‘the local element equation’. In plane truss
problems, it is often necessary to relate global displacement components of an element to
its respective global force components. This relationship is called the ‘global element
equation’.
The local element equations are
where
It is desired to write the global element equations as
where
134
It is now required to establish a relationship between [ke] and [Ke]. For this purpose, the
relationships between {qle} and {qe}, and the relationship between {fe} and {Fe} have to be
established first.
As shown in Figure 6.12(a), the components of displacement vector in the local x – y
and the global X–Y directions are, respectively, u, v and U, V. It can be seen that
Similarly, by looking at Figure 6.12(b), it can be shown that
Application of Eq. (6.94) to the displacement components associated with local nodes 1
and 2 results in
Eqs. (6.95) and (6.96) can be written in matrix form as
135
Figure 6.12a Co-ordinate transformation
Figure 6.12b Components of force in the x–y and X–Y directions

By letting
and
Eq. (6.100) can be written in view of Eq. (6.80) as
where
136
Here, C=cosθ and S=sinθ.

6.3.4 Computation of Stress for a Bar in the X–Y Plane
The axial tensile stress (σe) in element ‘e’ is given be
It can be extracted from Eq. (6.80) that
By combining Eqs. (6.99), (6.104) and (6.105),
is obtained where
For the plane truss shown in Figure 6.13,
determine the displacement components at nodes,
137
determine the stresses in each bar, and

verify the nodal equilibrium at node 1.
Solution:
Let the direction of x for each element is taken in the direction from node 1 to other node.
Once the direction is chosen, the angle θ is established as positive when measured
counter clockwise from the positive X to x. It is convenient to construct a table as an aid in
determining each element stiffness matrix.
Figure 6.13 Plane truss–Example 4
138
Therefore,
Eq. (6.106) gives
139
Hence, axial stress in each element is given by
and
By summing the forces in the global X and Y directions,
6.4 SPACE TRUSS

Derivation of the global stiffness matrix for a space truss is analogous to the plane truss
case. An element of constant cross-section as shown in Figure 6.14 is considered here.
Element equilibrium equation and stiffness matrix for an element are given in Eqs. (6.84)
and (6.83), respectively.
140
Figure 6.14 (a) Two-node space truss element subject to tensile force T (b) Global
displacements
Displacement u and the force f in the local co-ordinate have the components u(x) = (U, V,
W) and f = (Fx, Fy, Fz) in the global co-ordinates (X, Y, Z). It remains now to find the
transformation matrix [T*] to transform from the local to the global co-ordinate system.
It can be seen from Figure 6.15 that an arbitrary vector has the components
Figure 6.15 Co-ordinate transformation for space truss element

and the direction cosine
cosθx, cosθy, and cosθz are components of a unit vector in the x-direction. From Figure
6.14(b)and Eq. (6.108), it can be shown that
141
which in matrix notation can be written as
Further, the displacement vectors u1 and u2 can be shown form Figure 6.14(b) and Eq.
(6.108) to be
Multiplying Uj by cosθx, Vj by cosθy, and Wj by cosθz, adding them and using Eq. (6.109),
the following equation is obtained.
Eq. (6.113) can be written in the matrix notation as
Substituting Eq. (6.114) in Eq. (6.84), and this in turn in Eq. (6.111), the following equation
can be derived.
Then matrix [Ke] can be obtained as
142
where
Force and stress in space truss element can be obtained in an analogous way to
computation of stress for a bar in the X-Y plane
For the space truss shown in Figure 6.16,
Determine the displacement components at node A in terms of AE.
Forces in each bar
For all bars AE = constant.
Note: This problem is statically determinate. By taking the equilibrium of joint A, the
answers for the forces in bars are:
TAB = 15,094.77 N; TAC = 13,856.95 N; TAD = 15,577.84 N. All forces are tensile.
Solution:
Let the direction of x for each element is taken in the direction from node 1 to the other
node as shown in Figure 6.16(b) above. The direction cosine of each member is shown in
the table below.
Element 1 is from node i = 1 to node j = 2.
143
Figure 6.16 Space truss–example
From Eq. (6.116), the matrix equation for each element in global co-ordinates is
144
145
By assembling the matrices of the three members, the nodal equations are obtained as
Application of boundary conditions, U2 = 0, V2 = 0, W2 = 0, U3 = 0, V3 = 0, W3 = 0, U4 = 0,

V4= 0, W4 = 0, yields the first three rows (1 – 3) and the first three columns as
146
Thus,
By solving above equation

AEU1 = 0.0940 × 104; AEV1 = –2.8942 × 104; AEW1 = –1.4315 × 104.
Again from the assembled equation
substituting the values of U1, V1 and W1

F2x = 9,038.0 N; F2y = 12,048.0 N; F2z = 0.
The resultant force for element 1, = 15,060. N

Similar operations for elements 2 and 3 yields the values of the element forces as
F3x = 0.0 N; F3y = 12,198.0 N; F3z =6,506.0 N and = 13,825.0 N
F4x = –9,036.0 N; F4y = 10,843.0 N; F4z = 6,506.0 N and = 15,542.0 N.

An alternate way of obtaining the forces is to use the transformation of displacement
components from the global axis to local axis and use the fundamental definition of a force
in a truss element. Consider the element 1, as an example.
For this element the nodes are from 1 to 2 and the node 2 is pinned.
Therefore, u2 = 0.0 and
u1 = U1, cosθx + V1 cosθy + W1 cosθz Thus u1 can be determined as
AEu1 = [(0.0940 × 0.6) + (–2.8942 × 0.8) + (–1.4315 × 0.0)] × 104
6.5 ONE-DIMENSIONAL TORSION OF A CIRCULAR SHAFT

The one-dimensional torsional elements can be used to analyze an assemblage of circular
shafts. Procedure is explained in the following steps.
Step 1: Select element type
A two-node torsion element is shown in Figure 6.17(a). The T1 and T2 are the torques at
nodes 1 and 2, respectively, whereas ø1 and ø2 are angular rotations at nodes 1 and 2,
respectively, of the element. The sign convention is shown in the Figure 6.17(a).
147
Figure 6.17a One-dimensional torsion element

Step 2: Select approximation function
Let the rotation about x-axis at a section located at distance x units from node 1 can be
written as
Because ø(x = 0) = ø1 and ø(x = L) = ø2,
where
Step 3: Define constitutive law and gradient of the unknown

Constitutive Law
The constitutive law is given by
where G = shear modulus and J = polar moment of inertia of the cross-sectional area
about the axis of the element. Alternatively, the stress–strain relationship can be
expressed as
Here, A represents the cross-sectional area.

Gradient or unknown rotation
It can be shown easily that
148
where

The total potential energy of a torsion element is given by
By substituting Eqs. (6.121) and (6.123) into (6.124), it can be shown that
Hence, element equations are given by
A circular shaft as shown in Figure 6.17(b) is subjected to torques T2 and T3 as shown in
the diagram. By employing one-dimensional torsion elements, compute angular rotations
at nodes 2 and 3 and reactive torques at nodes 1 and 4.
Solution:
The element matrices can be obtained as follows.
149
Figure 6.17b Circular shaft
The system can be assembled as
By applying homogeneous boundary conditions ø1 = ø4 = 0, following equation is obtained.
After solving, the solution is obtained as,
6.6 ONE-DIMENSIONAL STEADY STATE HEAT CONDUCTION

A 2-node heat flow element as shown in Figure 6.18 can be utilized to analyze one-
dimensional steady state heat conduction problems. Note that convection is neglected in
the formulation. The solution is obtained in the following steps.
Sign convention: Heat flow entering a control volume is positive. Outgoing heat flow is
negative.
150
Let q be the heat flow (per unit area) and t be the temperature. Then, qi, i = 1, 2 are the
nodal heat flow rates and ti, i = 1, 2 are the nodal temperatures.
Figure 6.18 One-dimensional heat flow Element

Let the temperature at a section located at distance x units from node 1 be
Note that
Step 3: Define gradient of the unknown and the constitutive law

Gradient of Unknown Temperature
The gradient is obtained as
Constitutive Law: Fourier's Law of Heat Conduction

According to Fourier's law, the heat flow in the x direction is given by
where and Kxx = Thermal conductivity in x-direction [KW/m.°C in SI units]. The Kxx
measures the amount of heat energy (W.hr) that flows through a unit length of a given
substance in a unit time (hr.) to raise the temperature 1 degree (°C).
Since physics of the element can be understood easily for the one-dimensional heat flow
element, direct equilibrium approach can be employed to derive element equations. By
employing Eq. (6.131), the heat flowing in the x-direction at node 1 is
151
From the principle of conservation of energy, the net heat flow into the system must be
zero for a steady state condition.
The furnace wall shown in the Figure 6.19 consists of 25 cm of fire brick [K1 = 0.012 W/(cm
°C)], 10 cm of insulation brick [K2 = 0.0014 W/(cm °C)], and 20 cm of red brick [K3 =
0.0086 W/(cm °C)]. The specified inner and outer temperatures are 500°C and 150°C,
respectively. Determine the internal temperature distribution.
Figure 6.19 Cross section of furnace wall and its finite element model
Solution:
This problem can be treated as a one-dimensional model. A constant arbitrary heat
conduction area of unit cm2 is assumed for all elements.
Element conduction matrices are given by
152
Assembly
Figure 6.20 Internal temperature distribution

By applying the non-homogeneous boundary conditions, the following equation can be
obtained.
Hence,
Figure 6.20 shows the temperature distribution.
6.7 ONE-DIMENSIONAL FLOW THROUGH POROUS MEDIA

Analysis for one-dimensional flow through porous media is similar to the heat flow analysis
(e.g., ground water flow, seepage). A 2-node element as shown in Figure 6.21 is the
153
simplest possible element can be used to analyze one-dimensional seepage problems.

Formulation is given in the following steps.
Sign convention: Fluid flow entering a control volume is positive. Outgoing fluid flow is
negative. Let vx, q and h be the flow velocity, the volumetric flow rate and the hydraulic
head, respectively. Then, qi, i = 1, 2 and hi, i = 1, 2 shown in Figure 6.21 are the nodal fluid
flow rates and the nodal hydraulic heads, respectively.
Let the hydraulic head at a section located at distance x units from node 1 be
Figure 6.21 One-dimensional fluid flow element

Note that
Step 3: Define gradient of the unknown and the constitutive law

Gradient of hydraulic head
The gradient is obtained as:
Constitutive Law: (Darcy's law)

According to the Darcy's law; the fluid velocity vx in the x direction is given by
where Kxx is the coefficient of permeability in the x-direction. Consequently, the volumetric
flow rate q can be obtained from
Here, A represents cross-sectional area of the element.

154
The element equations can be derived easily by adopting the direct equilibrium approach.
From Eq. (6.146), fluid flow entering the element in the x-direction at node 1 is
where
From the principle of conservation of mass, the net fluid flow into the system must be zero
for a steady state condition.
or
Figure 6.22 One-dimensional flow through porous media and its finite element model
In a laboratory, experiment of one-dimensional flow through porous media over the section
shown, the following data are known.
Determine the distribution of hydraulic head over the length of the section, the flow rate
and fluid velocities in each element.
Solution:
Element matrices are given by
155
Assembly
Hence,
or
The hydraulic head distribution is shown in Figure 6.23.
Flow rate = 1.333 cm3/s.

By making use of Eq. (6.145), fluid velocity in each element can be obtained as
Figure 6.23 Hydraulic head distribution
156
6.8 ONE-DIMENSIONAL IDEAL FLUID FLOW THROUGH PIPES (INVISCID

FLUID FLOW)
The one-dimensional ideal fluid flow analysis is analogous to analysis for one-dimensional
flow through porous media.
Let A = cross-sectional area, vx = fluid velocity in the x direction øi (I = 1, 2) = velocity
potentials at nodes 1 and 2, and qi (i = 1, 2) = fluid flow rates at nodes 1 and 2 in Figure
6.24. It can be shown that
Hence, set Kxx = 1 in the one-dimensional porous media problem to get the element
equations. Then,
where .
Figure 6.24 A finite element to analyze one-dimensional ideal fluid flow through pipes
6.9 BEAM ELEMENT

A beam is a long, slender structural member generally subjected to transverse loading that
produces significant bending effects as opposed to twisting or axial effects. A continuous
beam can be modelled by employing 2-node beam elements. To derive equations of a
beam element, it is important to briefly review classical beam theory.
157
Figure 6.25 (a) Beam subjected to transverse loading, (b) Beam theory sign conventions
6.9.1 Review of Beam Theory

Consider the simply supported beam shown in Figure 6.25. The beam is of length L with
axial co-ordinate x and transverse local co-ordinate y.
In Figure 6.25, fy(x) = transverse loading, v = transverse displacement, V = shear force,
and M= bending moment.
The equilibrium of the differential element in the y direction is given by
Therefore, in the limit Δx → 0.
The moment equilibrium about x-axis results in Σ M at A = 0. Thus,
Referring to (Crandall et al. (1978)) and Figure 6.25, it can be shown for a shallow beam,
that
where ρ = radius of curvature, E = Young's modulus of elasticity and I = moment of inertia.
158
Figure 6.26 Neutral axis before and after deformation

The axial strain developed at any point on a cross section, located at distance x from the
left support is given by
where u is axial displacement. The u at any point on a cross section, on the other hand,
can be shown to be of the form (refer to Figure 6.27)
Thus,
Figure 6.27 u variation at a typical cross section
159
Stress–strain relation (Hooke's law)

From the Hooke's law, the axial stress σx is related to εx by
Note that tensile stresses are considered positive. Therefore, from Eqs. (6.165) and
(6.166), moment M about z-axis can be expressed as
Here, I = moment of inertia of the cross-sectional area about the z-axis. It is to be noted
here that EI has not been assumed to be constant. By combining Eqs. (6.159), (6.161) and
(6.167), Eq.
can be obtained. Thus, the governing differential equation for bending of a beam is
If EI is constant, the above Eq. (6.169) takes the form
If fy(x) = 0, the governing equation becomes
General solution of Eq. (6.171) is given by
where C1, C2, C3, C4 are unknown, constant terms. They can be determined by applying
6.9.2 Finite Element Formulation of a Beam Element

A finite element formulation of a beam element is presented below in a step-by-step
manner.
Step 1: Discretize and select element type
A continuous beam can be discretized into number of 2-node beam elements. A typical
beam element is shown in Figure 6.28 with positive nodal displacements, rotations, forces
and moments.
160
Figure 6.28 Typical beam element

In the formulation, deformation due to bending is measured in terms of a transverse
displacement v and a rotation θ about z-axis. Hence, there are two DOF vi and θi at i-th
node of the element. As a result, the formulation pertains to C1 continuum (in all previous
cases, formulation for C0 continuum was considered). Let element DOF vector be given by

Solution of the homogeneous Eq. (6.171) can be chosen as basis for selecting
approximation function for a beam element. In conjunction with the solution given by Eq.
(6.172), following four geometric boundary conditions need to be satisfied.
Therefore, the four unknown constants Ci, i = 1, 2, 3, 4 in Eq. (6.172) can be obtained in
terms of element DOF by making use of boundary conditions of Eq. (6.168). By
substituting Ci's back into Eq. (6.172), equation describing variation of v (the primary
unknown quantity) over x in terms of element DOF, as
Here, [N] = [N1 N2 N3 N4]
where
161
N1, N2, N3 and N4 are the shape functions for a beam element. In the literature, Ni's are
called beam functions, Hermite polynomials, or cubic splines. Variations of Ni, i = 1, 2, 3, 4
over the length of element are sketched in Figure 6.29.
Here, it may be noted that
Figure 6.29 Shape functions for a beam element

At node 1, v = v1. Hence, from Eq. (6.177), it can be seen that at x = 0. N1 = 1 and Ni = 0,
i= 2, 3, 4.
Similarly, at node 2, v = v2, and N3 = 1 and Ni = 0 for i = 1, 2, 4.
At node 1, . Thus, at x = 0, , i = 1, 3, 4.
Similarly, at node 2, θ = θ2 and , i = 1, 2, 3.

Step 3: Define strain-displacement relation and constitutive law
Strain-Displacement Relation
From Eqs. (6.165) and (6.175), the strain-displacement relation can be defined, in terms of
{qe}, as
where
162
Here,
Alternatively, [B] can be shown as
where
Constitutive Law
Following expressions can be easily derived from Eqs. (6.166) and (6.178):
Figure 6.30 A beam element subjected to distributed and concentrated point loads
In the formulation presented below, the element equations are derived by employing the
principle of minimum potential energy. (Note that equations can also be derived by
employing direct equilibrium approach. Refer to texts (e.g., Logan (1993); Gere (1965)) for
details.)
Consider the beam element shown in Figure 6.30 for deriving element equations.
Following notations are used in Figure 6.30.
fy(x) is distributed load
Pn are concentrated point load applied at distance xn from node 1 (n = 1,2,3,…)
163
Mm are concentrated moment applied at distance xm from node 1 (m = 1,2,3,…)

Equating the work done by these loads undergoing assumed displacement field v(x) with
the work done by the nodal loads fiy0 and nodal moments mi0 (i = 1, 2) undergoing
displacements vi and rotations θi (i = 1,2), respectively, expression for the equivalent load
vector {f0e} can be derived. These equivalent load vectors are subsequently used in the
analysis.
The total potential energy of a beam element is given by
In Eq. (6.183), fiy and mi are (nodal components), respectively, the shear force and the
bending moment at the i-th node. Moreover, v′ indicates slope of v with respect to x. In
other words, v′ represents rotation.
By letting
and by substituting Eqs. (6.176), (6.178) and (6.182) into Eq. (6.183),
can be obtained. Here,
and
164
The {f0 e} in Eq. (6.188) is equivalent concentrated nodal load vector . By

minimizing the potential energy πp, the element equations are obtained as
Note that variation in E and I along x can be considered easily in Eq. (6.186). For constant
EI, Eq. (6.186) simplifies to
By substituting Eq. (6.181) in Eq. (6.190) and performing the matrix multiplication, [ke]
takes the form
All the terms of the element stiffness matrix [ke] can be evaluated explicitly by substituting
Ni's from Eq. (6.177). The resulting matrix is given below.
Step 5: Assemble element equations and apply boundary conditions

By applying Eq. (6.189) to all the elements of a discretized continuous beam, the system
(or global) equations can be obtained as
Here, [K], [F] and {F0} are assembled stiffness matrix, the load vector consisting of
loads/moments applied directly at the joints (it also includes support reactions), and the
assembled, equivalent nodal load vector. Note that the size of the assembled matrix is
reduced after applying the geometric (or forced) boundary conditions.
Step 6, 7 and 8: Solve for primary and secondary unknowns and interpret results
The assembled equations, which have been modified by applying boundary conditions,
can be solved for unknown displacement components contained in {q}. Once the primary
quantities ({q}) are known, the secondary quantities like reactions and element forces can
be computed by applying Eqs. (6.193) and (6.189), respectively. Shear force and bending
moment diagrams can then be constructed for design purposes.
6.9.3 Illustrative Examples
Example 1
Consider a beam subjected to a uniformly distributed load of intensity w per unit length (in
the –y direction). Derive equivalent nodal force vector.
165
Solution:
By applying Eq. (6.188), the equivalent concentrated nodal force vector is given by
Example 2
Consider a beam subjected to concentrated load P (in –y direction) at a distance ‘a’ from
node 1. Let L = a + b. Derive equivalent nodal force vector.
Solution:
By applying Eq. (6.188), the equivalent nodal force vector can be computed easily as
166
Example 3
Consider a fixed beam subjected to distributed and/or concentrated loads. Derive
equivalent nodal force vector.
Solution:
Note that for a fixed beam, {qe} = 0. Then, from Eq. (6.189), the actual (or corrected)
element nodal forces are given by
In other words,
Therefore, equivalent concentrated nodal forces can be treated as fixed-end reactions

(forces and moments) when both ends of the beam element are fixed. Fixed-end reactions
for various transverse loading cases can be found in Gere (1965).
Example 4
Analyze beam shown in the figure by using FEM.
Solution:
Consider the beam to be made up of the two elements 1 and 2.
The bou
ndary conditions are
167
V2 and θ2 are required to be computed. Element matrices are given by
After performing assembly, following equation is obtained.
Substituting l = L/2 and applying boundary conditions, following equation is obtained.
Thus, from the assembled equation set,
168
From the element equations,
Letting l = L/2 and simplifying the element force vector is obtained as
Similarly,
Above results can be represented graphically as shown below.
Example 5
Analyze beam system shown by using FEM. Consider P = wL, M = PL = wL2 and EI =
constant.
169
Solution:
Consider the beam to be made up of two elements and
The boundary conditions are V1 = θ1 = V2 = V3 = θ3 = 0
Element matrices are given by
After performing assembly, following equation is obtained.
170
Now applying boundary conditions and solve for θ2
or
From the assembled equation set,
or
By making use of the element equation, the element force vectors are obtained as
171
Above results can be represented graphically as shown below.
Example 6
Analyze beam shown in the figure by using FEM. Consider constant EI and
Solution:
Consider beam elements and as well as spring element to discretize the domain.
The boundary conditions are V1 = θ1 = V3 = θ3 = V4 = 0
The element equations are obtained as shown below.
Element
172
Element
Element
The assembled equations are given below.
173
Force vector for each beam member as well as force in the spring can be computed by
making use of respective element matrices.
6.10 ANALYSES OF PLANE FRAMES AND GRIDS
Many civil engineering structures like buildings and bridges consist of frames and/or grids.
In this section, finite element analysis procedures will be formulated for such structures.
Generally, a plane frame is a series of slender members rigidly connected to each other
such that the original angles made between elements at their joints remain unchanged
from one element to another at the joints. In addition, the element centroids, as well as the
applied loads, lie in a common plane. Since applied loads may have components along an
element’s local x-axis, unlike beam element, a frame element may be subjected to flexural
as well as axial effects.
A grid, on the other hand, is a structure on which loads are applied perpendicular to the
plane of the structure, as opposed to plane frame, where loads are applied in the plane of
the structure. Since applied loads do not have any component along a grid element's local
x-axis, no axial force can be developed in a grid member.
A typical plane frame and grid are shown in Figure 6.31.
6.10.1 Plane Frame Analysis
174
Elements in plane frames are subjected to both lateral loading and axial loading.
Therefore, axial deformations need to be included in the analysis. As described earlier, all
the distributed and point loads (which are not applied at the nodes) are replaced by
equivalent nodal loads. Local effective nodal forces (actual element nodal forces plus
equivalent concentrated loads), which act on an arbitrarily oriented member of a plane
frame, are shown in Figure 6.32. The plane frame structure consisting of prismatic
members is assumed to lie in the X–Y plane. The local axes x and y are located along the
plane frame element’s length (node 1 → 2) and transverse to the element, respectively,
and the global axes X and Y are chosen with respect to whole structure.
Figure 6.31 Typical plane frame and grid structures
Figure 6.32 A typical member in a plane frame

By considering strain energy in axial deformation (i.e., strain energy considered in the
formulation of truss element), it can be shown that the axial effects are governed by
175
where C1 = (AE/L). Here, f1x and f2x are axial forces at nodes 1 and 2 in the local x
direction. The u1 and u2, on the other hand, are axial displacement components along local
x at nodes 1 and 2, respectively.
Similarly, by considering strain energy in bending, (strain energy expression considered in
the formulation of beam element), equations governing the transverse effects (bending
and shear) can be derived to be
where C2 = (EI/L3). Here, fiy and mi are the transverse shear force and bending moments,
respectively at i-th node.
By combining the axial effects from Eq. (6.194) and flexural effects from Eq. (6.195),
following equation can be derived for a plane frame element.
In other words, a plane frame member's governing equations in the local direction are
given by
By looking at Figure 6.12(a), it can be easily proved that
where C = cos α, S = sin α. The C and S can be obtained from the global co-ordinates of
the element's nodes as
where
176
Note that since the local z-axis coincides with the global Z axis, no transformation is
required for the rotation θ along z-axis.
and since z coincides with Z,
Application of Eq. (6.198) to nodes 1 and 2 results in
or,
where
Similarly, by applying Eqs. (6.200) and (6.201) to local nodes 1 and 2, the following
equation Eq. (6.205) is obtained.
is obtained where
In view of Eqs. (6.197), (6.198) and (6.203), Eq. (6.205) can be written as
Therefore, the global equation for a plane frame element is given by
where
177
After performing the triple matrix product, Eq. (6.209) can be written as
where
Thus, for each element,
which can be written in the global co-ordinate system. Moreover, the effective element
force vector can be written as
In Eq. (6.213), the {Fee} and {F0e} are, respectively, the element force vector (which
contains details like shear force and bending moments) and the equivalent nodal force
vector (corresponding to transverse loading) in the global co-ordinates. By combining Eqs.
(6.212) and (6.213), it can be shown that
By applying Eq. (6.214) to all the plane frame elements in the structure and assembling
the resulting equations
can be formed, where

{F} = external nodal force vector;
{F0} = equivalent nodal force vector due to transverse loads;
[K] = assembled stiffness matrix; and
{q} = assembled (global) displacement vector.
Summary of plane frame analysis: The plane frame analysis can be summarized in the
following steps.
For each element, determine
[ke], {f0e}
[Ke] = [T]T [ke] [T]
{F0e} = [T]T{f0e}
178
By assembling element equations, obtain system equations

Apply boundary conditions
Solve for {q}
Solve for unknown reactions
For each element, determine
local effective forces by employing
Local actual forces by employing
These steps are demonstrated through an illustrative example given below.
Illustrative Examples
Example 1
Analyze the plane frame shown in the figure by using FEM. Consider E = 200 GPa, I = 1 ×
10−4m4 and A = 1 × 10−2 m2.
Solution:
Let element axis be directed from node 1 to node 2 for element and from node 2 to
node 3 for element . Various coefficients required in analysis are given below in the
tabular form.
For analysis of plane frame, it is useful to write global element equations in symbolic,
partitioned form. Thus,
179
After Assembly,
Here,
By substituting boundary conditions, following equation is obtained.
Calculation of element force vectors

Element 1
180
Element 2
181
Example 2
Analyze the plane frame shown in the figure by using FEM. Consider E = 200 GPa, I = 2
×10−4 m4 and A = 1 × 10−2 m2.
182
Solution:
Let element axis be directed from node 1 to node 2 for element 1 and from node 2 to node
3 for element 2. Various coefficients required in analysis are given below in the tabular
form.
Write global element equations in symbolic, partitioned form. Thus,
After assembly,
183
Here,
By substituting boundary conditions, following equation is obtained.
Calculation of element force vectors

Element 1
184
Element 2
185
6.10.2 Grid Analysis

In this section, element equations are derived for a grid element. A typical grid element
used to discretize a grid structure is shown in Figure 6.33.
Since a grid structure is assumed to lie in the X–Z plane, the local y-axis coincides with the
global Y-axis. The degrees of freedoms at the i-th node of a grid element, in the local
186
directions, are a deflection viy (normal to the grid), a torsional rotation θix about the x axis,
and a bending rotation θiz about the z-axis. The nodal effective (actual plus equivalent
concentrated) forces at the i-th node consists of a transverse force fiy normal to the grid, a
torsional moment mix about the x-axis, and a bending moment miz about the z-axis.
As mentioned before, a grid member is not subjected to axial force components. Loading
in the direction perpendicular to the plane of grid may induce a torsion and bending in a
grid element. Thus, one can consider strain energy in torsion and flexure to derive
equations for a prismatic grid element.
By considering strain energy in torsion, it can be shown (from the one-dimensional torsion
element formulation) that
Figure 6.33 A typical grid element
where C3 = . On the other hand, by minimizing potential energy expression for flexure
(refer to formulation of beam element), the following equation
is obtained where C2 = (EI/L3).
187
By combining Eqs. (6.218) and (6.219), the local element equations for a grid element can
be obtained as
Or,
By adopting the procedure used to obtain Eq. (6.207), the local degrees of freedoms can
be expressed in terms of global degrees of freedoms by putting
where
In Eq. (6.226), the global displacement components Vi, θiX and θiZ, respectively, are the
deflection along Y, rotation about X and rotation about Z axes at the i-th node.
Note that α is measured in the X–Z plane from the local x to global X (and not from the
global X to local x, as done for plane truss and frame) in the counter clockwise direction.
By following the formulation of a plane frame element, the global equations for a grid
element can be derived as
where
188
By performing the triple matrix multiplication in Eq. (6.228), the global stiffness matrix for a
grid element is obtained as
For the grid member also, the steps summarized for plane frame analysis in Section 6.10.1
can be followed.
Illustrative Example
Analyze the grid shown in the figure by using FEM. Consider
E = 200 GPa, G = 80 GPa, I = 4 × 10−4 m4, J = 2 × 10−4 m4
Solution:
Consider element axes for both the elements as shown in the figures. Note that local x-
axis for each element is defined as directed from node 1 to node 2 of the element
189
Element properties and direction cosines are shown in the tables presented below.
Element matrices in the global directions are shown below.
After assembly, following equation is obtained.
By applying boundary conditions the assembled equations reduce to
Element forces are calculated as below.

Element 1
190
Element 2
191
6.11 FURTHER ONE-DIMENSIONAL APPLICATIONS

6.11.1 Flow Network Analysis
Flow of fluid through pipes is often required to be examined for municipal water supply,
sewage networks, chemical flows in factories etc., where usually a network consisting of a
large number of pipes, pumps, valves etc. are involved. A large number of pipe elements
of uniform cross sections and pipe junctions and other elements are usually utilized in a
flow network. Flow in a pipe network can be considered as one-dimensional as the flow is
predominant in the direction of flow. Simple pipe networks can be analyzed using
analytical/approximation based methods like Hardy-Cross method and Newton Raphson
192
method (Streeter et al., 1998). However, for complex pipe networks FEM can be very
effectively used by considering networks as a system of one-dimensional finite elements
consisting of the pipe elements. Here, application of FEM in fluid flow networks is
discussed.
Figure 6.34 Pipe element

Consider a typical pipe element shown in Figure 6.34 which can be used to analyze simple
network without any booster pumps like the one shown in Figure 6.35. The flow takes
place only due to the existing pressure gradient. For simplicity of problem formulation, the
flow in these pipes is assumed to be laminar. Problem here, is to find rate of flow in each
of the links and pressure at each node. This network can be treated as a collection of finite
element by considering each flow path between two junctions as an element. The network
shown in Figure 6.35 is having 6 nodes and 9 elements. Element characteristic would then
be pressure head loss versus flow rate relationships. If the flow is assumed to be fully
developed laminar flow and the flow paths are circular pipes of constant cross-sectional
area, the pressure head loss H1 – H2 between nodes 1 and 2 due to flow rate is given by
(Baker, 1983)
where Q is flow through pipe, L is the length of the pipe, μ is the dynamic viscosity of the
fluid, and D is the internal diameter of the pipe.
For a typical element shown in Figure 6.34, the flow entering the element at the nodes 1
and 2 in terms of the pressure heads can be written as
for node 1 and
Figure 6.35 Flow network
193
for node 2.
In other words,
where
In matrix notation, these equations can be written as
For the configuration shown, in Figure 6.35, the element to node connection relationship is
as follows:
Element, e Node, i Node, j
1 1 2
2 2 4
3 1 3
4 4 6
5 3 5
6 5 6
7 1 4
8 3 4
9 3 6
where e denotes element, 1, 2–denote connecting nodes of the element e, Q1(e) is the flow
entering the element at node 1, Q2(e) is the flow entering the element at node 2.
For elements 1 to 9, the element equations are written for the network in Figure 6.35.
194
The nodal flow continuity equations can be set up as follows.
195
Here, ci, i = 1,2,…,6 are consumptions at the nodes or the rate of outflow from the network
with proper sign. On the other hand, Qi, i = 1,6 are flow through the pipe. The ci can be
termed as loads on the network. The assembled matrix is as shown below, which is of the
form [K]{H} ={q}, where [K] is the fluid stiffness matrix, {H} = column vector of nodal
pressure heads and {q} is the column vector of nodal consumptions.
In the above equations,
196
Note that the matrix shown above is applicable for only laminar flow and this does not
include the pressure losses associated with the fittings, valves and pumps that are the part
of the most fluid network. These pressure losses are normally taken into account by
introducing an equivalent length of pipe (Streeter et al., 1998), that will cause the same
loss.
In many practical applications, the fluid flow in network is turbulent. It is still possible to
define an element as a length of fluid carrying conduit, but the element equations are no
longer linear. The pressure head-loss versus flow rate relationship is given by Darcy-
Weisbach, in the following form (Streeter et al., 1998).
Here, H1 – H2 is the pressure head drop in meters between section 1 and 2 which are L
(meter) and f is the friction factor. As before, Eq. (6.246) can be used to write two
equations for flow at the nodes in terms of pressure at the nodes. Flow ‘entering’ the
element at its ends are obtained from Eq. (6.246) as
where h=|H1−H2|=pressure head loss.

Now the flow Q is,
where Te is the transmittance of the element in turbulent flow case, which is defined as
Then from Figure 6.34, the turbulent flow equations are
for node 1,
for node 2. In the matrix form, above equations can be written as
197
As in the laminar flow case, for flow network in Figure 6.35 the individual element
equations from Eq. (6.251) can be written as follows.
The global matrix can be written as follows:
where
198
The global assembly of these element matrices is given above. The individual elements in
the global matrix on the left hand side are not independent of the pressure head H. These
matrix elements are to be updated often in each iteration. The number of iterations
depends on the convergence criteria. Hence, values of pressure heads are required to be
determined by an iterative process. At the first instance, pressure heads at different nodes
are assumed and the global transmittance matrix is formulated. The ensuing set of linear
equations can be solved for pressure head and the transmittance matrix can be updated
again. This procedure can be repeated till convergence is achieved.
Find the flow rates in each of the elements (pipes) in the network as shown in Figure 6.35.
The diameters of the pipe are 10, 10, 10, 10, 10, 10, 12, 12, 12 mm, respectively. The
lengths are 1,000, 1,000, 1,000, 1,000, 1,000, 1,000, 1,500, 1,500, 1,500 mm,
respectively. The flow rates at each node out of the network are [+125, –25, –25, –25, –25,
–25] cubic cm/min. The liquid flowing is water. If the pressure head at node 1 is 2 m
expressed in terms of the height of liquid that is flowing, find the pressure heads, (H), at
each of the remaining nodes and the flow rates in the pipe. The friction factor may be
taken as 0.02.
Solution:
The given data has been tabulated as follows.
The problem has been solved by using MATLAB program as given in Appendix G (Section
G.1). Results are as shown below. Pressure head values at different nodes in (m) are:
Node H (m)
1 2.00000
2 1.99999
3 1.99995
199
4 1.99997
5 1.99999
6 1.99995
Discharge values in different pipes, on the other hand, are
Pipe No. Discharge (lit/s)
1 0.00061
2 0.00019
3 0.00065
4 0.00031
5 0.00027
6 −0.00016
7 0.00082
8 −0.00029
9 0.00028
6.11.2 Electrical Network Analysis

In the previous section, the fluid flow network was discussed, wherein flow path was
considered as a finite element of fluid network. Similarly, the direct-current electrical
networks can be considered as a finite element network. The current carrying member of
the electrical network can be taken as a finite element. The Ohm's law can be used for
establishing the element characteristics. Procedure of finite element formulation is very
similar to those used for fluid flow network. Voltages V1 and V2 play the same role as the
nodal pressures and a current I replaces the flow rate Q.
Basic relationship between the current flowing and voltage difference as given by Ohms'
law is (see Figure 6.36)
200
Figure 6.36 A typical element in an electrical network
Figure 6.37 Basic electrical circuit

where Ge is the conductance of the element e and Ie is the current flowing in the resistance
(element) e, Re is the resistance of the element, Ii(e) is the current entering the element e
through node i and Ij(e) is the current entering the element e through node j, such that Ii(e) =
Ie and Ij(e) = −Ie.
Consider the basic electrical circuit shown in Figure 6.37. Element equations for each
resistor can be written as follows.
Element No. 1 (connecting nodes are 1 and 2): The element equation is
201
Here, Gi = 1/Ri, i = 1,2,…6.

All these element equations are assembled into the global form. There are 6 elements and
5 nodes. Thus, the global assembly will be 5 by 5 size matrix given as
From Figure 6.37, it may be noted that
From these equations, the right hand side of the global matrix can be written as
Thus,
202
Proper boundary conditions must be substituted before this problem can be solved.
Ensuing set of linear equations can be solved for voltage at various other nodes.
For the electrical network shown in Figure 6.37, the voltage difference between nodes 1
and 5 is 10 V. The resistance at various resistors are: 250, 100, 200, 300, 400, 500 Ohm,
respectively. Calculate the voltage at various nodes and the current in each element?
Solution:
It can be seen in Figure 6.37, that are 6 elements and 5 nodes. By substituting G1 = 1/250,
G2 = 1/100, G3 = 1/200, G4 = 1/300, G5 = 1/400 and G6 = 1/500, V1 = 10 V and V5 = 0, in
Equation (6.273), following ensuing equation is obtained.
This equation set has been solved by using the Matlab program in the Appendix G,
Section G.2. The current values are as follows.
Element No. Currents
1 0.01012
2 0.00675
3 0.00337
4 0.00578
5 0.00434
6 0.01012
On the other hand, voltages at nodes are as follows.
203
Node No. Voltage
1 10.00000
2 7.46988
3 6.79518
4 5.06024
5 0.00000
6.12 SUMMARY OF ELEMENT MATRICES FOR ONE-DIMENSIONAL FINITE

ELEMENTS
Element matrices of one-dimensional finite elements for various cases considered can be
summarized as follows.
Element stiffness matrix for a truss/spring element:
(in the local co-ordinate system)
Element stiffness matrix for a plane truss element:

(in the global co-ordinate system)
Element stiffness matrix for a space truss element in the global co-ordinate system:
204
Element stiffness matrix for a torsion element:
Element conductivity matrix for a heat flow element:
Element permeability matrix for a fluid flow element:
Element stiffness matrix for a beam element:
Element stiffness matrix for a plane frame element:

205
The C, S and L in above equations are:
Element stiffness matrix for a grid element:

206
where
Element matrix for an element in flow network analysis
where for laminar flow and for turbulent flow.

Element matrix for a resistor element
6.13 CLOSING REMARKS

In this chapter, one-dimensional finite element analysis of engineering problems is
described. The direct as well as minimum potential energy methods in Chapter 4 are used
in the analysis. Linear type line elements are used in the problem analysis. Engineering
problems of axial bar, plane and space truss, beam, frame, grid, pipe flow, electrical
networks, porous media flow, torsion problem and heat conduction are discussed as
sample applications. Few sample problem analysis are also demonstrated. Sample Matlab
programs are presented in the Appendix, which can be used to solve similar types of
207
problems. Even though the problems discussed in this chapter are related to one-
dimensional FEM analysis using direct and variational approach, method of weighted
residual approach can also be applied for various one-dimensional problems in a very
similar way as discussed in Chapter 4. The two-dimensional FEM analysis of engineering
problems will be discussed in the next chapter, i.e. Chapter 7.
EXCERCISE PROBLEMS
1. Compute the assembled 8 × 8 matrix for the spring assemblage shown in the figure.
2. Analyze the composite axial bar shown in the figure when

F2 = 10 kN;
F2 = 10 kN and U4 = 2 mm; and
F2 = 10 kN and U1 = U4 = 2 mm. Consider Est = 200 GPa, Ast = 2 × 10−2 m2; Eal = 70 GPa, Aal =
1 × 10−2 m2
3. Analyze the plane truss structure shown in the figure using finite element method.
Let E = 200 GPa, A = 1 × 10−2 m2 and L = 1 m.
4. Analyze the plane truss shown in the figure by using finite element method.
208
Let E = 200 GPa, A = 2 × 10−2 m2
5. Analyze the space truss shown in the figure by using finite element method. Let E = 200
GPa, A = 4 × 10−2 m2
6. Analyze the shaft shown in the figure by using finite element method. Let G = 70 GPa, J
= 4 × 10−2 m4, L = 2 m.
7. Analyze the temperature distribution through the composite wall shown in the figure
using finite element method. Also determine the head flux entering and exiting the system.
8. Analyze the one-dimensional flow through porous media over the section shown in the
figure by using finite element method. Also compute the flow rate and fluid velocity in each
element. Let A = 4 cm2.
209
9. Analyze the beam system shown in the figure by using finite element method.
10. Analyze the beam supported by a vertical spring by using finite element method.
11. Analyze the beam system shown in the figure by using finite element method.
12. Analyze the plane in the figure by using finite element method.
13. Analyze the plane frame shown in the figure by using finite element method.
210
14. Analyze a horizontal elastic member supported by a wire by using finite element
method.
15. Analyze the grid system shown by using finite element method.
16. Find the flow rates in each of the elements (pipes) in the network as shown in Figure
6.35. The diameters of the pipe are 10, 10, 10, 10, 10, 10, 12, 12, 12 cm, respectively. The
lengths are 100, 100, 100, 100, 100, 100, 150, 150, 150 m, respectively. The flow rates at
each node out of the network are [+50, −10, −10, −10, −10, −10] lit/sec. The liquid flowing
is water. The friction factor may be taken as 0.02. If the pressure head at node 1 is 15 m,
expressed in terms of the height of liquid that is flowing, find the pressure heads, (H), at
each of the remaining nodes and the flow rates in the pipe. [Hint: The problem can be
solved by suitably modifying the MATLAB program given in Appendix G, Section G.1.]

Backstrom, G. (1999). Fluid Dynamics by Finite Element Analysis, Studentlitteratur, Lund.
Baker, A. J. (1983). Finite Element Computational Fluid Mechanics, McGraw Hill,
Hemisphere, New York.
Bear, J. (1972). Dynamics of Fluids in Porous Media. Elsevier, New York.
Chandrupatla, T. R. (2004). Finite Element Analysis for Engineering and Technology,
University Press, Hyderabad.
Connor, J. J., and Brebbia, C. A. (1976). Finite Element Techniques for Fluid Flow,
Butterworth, London.
Chung, T. J. (1978). Finite Element Analysis in Fluid Dynamics, McGraw Hill, New York.
211
Crandall, S. H., Dahl, N. C., and Lardner T. J. (1978). An Introduction to the Mechanics of
Solids, Mechanics of Solids, McGraw Hill, New York.
Desai, C. S., and Abel J. F. (1987). Introduction to the Finite Element Method—A
Numerical Method for Engineering Analysis, Wadsworth, California.
Gere, J. M. (1965). Analysis of Framed Structures, Princeton, New Jersey.
New York.
Logan, D. L. (1993). A First Course in Finite Element Analysis, PWS Publishing, Boston.
Mohr, G. A. (1992). Finite Elements for Solids, Fluids and Optimization, Oxford University
Press, New York.
Pironneau, O. (1989). Finite Element Methods for Fluids, John Wiley and Sons,
Chichester.
Reddy, J. N., and Gartling, D. K. (1994). The Finite Element Method in Heat Transfer and
Fluid Dynamics, CRC Press, London.
Shames, I. H., and Dym, C. L. (1995). Energy and Finite Element Methods in Structural
Mechanics, Wiley Eastern, New Delhi.
Stasa, F. L. (1985). Applied Finite Element Analysis for Engineers, CBS Publishing Japan,
New York.
Streeter, V. L., Wylie, E. B., and Bedford, K. W. (1998). Fluid Mechanics, WCB/McGraw
Hill, Boston.
Visser, W. (1985). A finite element method for the determination of non-stationary
temperature distribution and thermal deformations, Proceeding of Conference on Matrix
Methods in Structural Mechanics, Air Force Institute of Technology, Ohio.
212
7 Two Dimensional Finite Element Analysis

7.1 INTRODUCTION
One-dimensional finite element analysis of various engineering problems was discussed in
the previous chapter. It was discussed in the introductory chapter that even though most
engineering problems are three-dimensional in nature, a two-dimensional analysis can
give sufficiently accurate and satisfactory results in many occasions. Since three-
dimensional analysis is generally cumbersome and time consuming, majority of engineers
and scientists prefer two-dimensional analysis to the maximum extent possible. Two-
dimensional finite element analysis of various engineering problems is discussed in this
chapter. Several practical engineering problems can be defined conceptually in two-
dimensions and can be represented in terms of two-dimensional governing equations.
Generally, two-dimensional FEM analysis and discretization can render good results when
the primary unknown quantities vary mainly on two directions while keeping constant in the
third direction.
Application of finite element analysis to two-dimensional problems will be discussed in this
chapter. In particular, the following formulations will be considered.
1. Triangular element for two-dimensional fluid flow analysis
2. Triangular element for two-dimensional stress analysis
3. Triangular element for axisymmetric analysis
4. Quadrilateral element for two-dimensional stress analysis
5. Isoparametric element for two-dimensional stress analysis
6. Application of method of weighted residual for various partial differential equations
Because the scalar functionals (strain energy) are easy to define, especially for two- and
three-dimensional analyses, they will be applied to the fluid flow and stress analysis
initially. Further, method of weighted residual is used for the solution of various partial
differential equations.
7.2 TWO-DIMENSIONAL FLOW THROUGH POROUS MEDIA (SEEPAGE FLOW)
Formulation presented in this section is focused on the two-dimensional fluid flow analysis.
However, the same formulation is applicable straightforwardly to two-dimensional torsion
and two-dimensional steady state heat flow (in the absence of convection). At steady
state, the governing differential equation for these problems can be written as
Table 7.1 Analogy Between Two-Dimensional Seepage, Heat Flow and Torsional
Analyses
213
The primary unknown field ɸ is function of (X,Y). The ɸ and other terms appearing in Eq.
(7.1) are explained in Table 7.1 in the context of two-dimensional seepage, heat flow and
torsional analyses.
Eq. (7.1) can be solved by modelling the problem with two-dimensional finite elements.
The two-dimensional elements are thin plate-like (plane) elements such that two co-
ordinates define a position on the element surface.
The ɸ appearing in Eq. (7.1) is given approximately by
for two-dimensional seepage analysis, where h indicates elevation (or fluid head), p
represents pore pressure and w is the specific weight of the fluid (water). The velocity
head term v2/(2g) has been customarily omitted from the above equation.
The two-dimensional steady state seepage problems can be categorized into (i)
unconfined seepage problem and (ii) confined seepage problem.
Unconfined seepage problem
Seepage through an earthen dam can be classified as unconfined seepage.
Unconfined seepage is distinguished from the confined seepage by the presence of a free
or phreatic surface. At the phreatic surface, the potential ɸ equals the fluid head h
measured from a datum. Moreover, there cannot be flow in the direction normal to the free
surface. (Also, note that flow cannot occur through an impervious bed. In other words, for
the example shown in Figure 7.1, ∂ɸ/∂Y = 0 at the impervious bed.) Mathematically, these
two conditions can be written at free surface as
where n indicates normal to the free surface. Both the conditions mentioned above need to
be satisfied simultaneously. The mixed boundary condition problem introduces non-
linearity in the system and an iterative approach is required to solve the problem. Such a
case is discussed later in the chapter.
Confined seepage problem
In the case of a confined seepage, the flow occurs in the absence of a free surface,
through a saturated media subjected to prescribed boundary conditions (usually ɸ). The
confined steady-state seepage requires a linear analysis for solution. An example of
confined seepage flow is shown in Figure. 7.2.
214
Figure 7.1 Unconfined seepage through a dam

The porous, saturated media in Figure 7.2 can be modelled by employing a variety of two-
dimensional elements. A typical constant strain triangular (CST) element used in
discretization is shown in the figure. The finite element formulation for CST element is
elaborated below.
7.2.1 Step-by-step Formulation for the CST Element for Two-Dimensional Confined
Seepage Analysis
In the formulation presented below, steps outlined in Chapter 3 are followed.
A typical i-th CST element from the discretized media is shown in Figure 7.3.
Figure 7.2 Confined seepage flow
215
Figure 7.3 CST element

The CST element has 3-nodes. Choose any apex as node 1, number the remaining 2-
nodes in the counter-clockwise direction, starting from node 1. At each node, there is one
primary unknown fluid potential ɸ. Thus, the nodal DOF (degree of freedom) for the CST
element for seepage analysis are ɸ1, ɸ2 and ɸ3 at nodes 1, 2 and 3, respectively. Let the
nodal co-ordinates in the X –Y plane be defined as (X1, Y1), (X2, Y2), and (X3, Y3).
Unknown continuous field ɸ can be expressed in terms of nodal quantities ɸi, i = 1, 2, 3.
Hence, three terms can be chosen uniquely to form an approximation function. In many
two-dimensional situations, Pascal's triangle is used as the basis in choosing an
approximation function of the form
Here, dij represents a constant term. Pascal's triangle is shown in Figure 7.4.
By selecting the first three terms from the Pascal's triangle, the approximation function can
be constructed as
where d00, d10 and d01 are unknown constants. Letting
for convenience, Eq. (7.4) can be written as
Application of the above equation at nodes 1, 2 and 3 results in
216
Figure 7.4 Pascal's triangle

Alternatively, in the matrix form,
Then,
where |P| is the determinant of [P] and
The [P] can be obtained to be
To further simplify the expression for |P|, consider Figure 7.5.
Figure 7.5 Area calculation for a CST element
217
From Figure 7.5,
Thus,
Similarly, it can be proved that
It can be seen from Eqs. (7.10) and (7.14) that
By substituting the above equations into Eq. (7.6) and by collecting the coefficients of ɸ1,
ɸ2and ɸ3, the following equation can be derived.
Here,
The N1, N2 and N3 are called the shape functions for the CST and [N] denotes the shape-
function matrix (Ni are also called local area co-ordinates).
Properties of shape functions
It can be easily verified that
1. Ni = 1 at the i-th node and Ni = 0 at the other 2-nodes (i = 1, 2, 3).
2.
Variations of Ni, i = 1, 2, 3, over an element are shown in Figure 7.6.
218
By referring to the second property of shape functions, it can be noted that at any point
(X,Y),
Figure 7.6 Shape functions for a CST element

Alternatively, from the divided areas A1, A2 and A3 in Figure 7.7(a), it can be proved that
In many finite element books, Ni are designated as local, area or triangular co-ordinates.
The triangular co-ordinate system (N1, N2, N3) is defined such that the nodal co-ordinates
of nodes 1, 2 and 3 are respectively (1, 0, 0), (0, 1, 0) and (0, 0, 1), as shown in Figure
7.7(b).
Step 3. Define constitutive law and gradients.
Gradient of Unknown Field
Let εx and εy be the gradients (slopes) of ɸ in the X and Y directions, respectively. Then,
219
Figure 7.7 (a) Another interpretation of shape functions and (b) Triangular co-ordinate
system
From Eqs. (7.6) and (7.22), it can be easily seen that
Thus, the gradients (strains) are constant over the element and, hence, a 3-node triangular
element is referred to as constant strain triangle (CST) element.
By substituting Eq. (7.17) into Eq. (7.22), it can be shown that
Constitutive Relations (Darcy's Law)

Noting that flow takes places from a higher fluid potential to a lower one, the velocities vx,
vy of the seepage flow in the X and Y directions are given, respectively, by
Therefore,
or
where
220
As mentioned earlier, energy method is applied to derive element equations. Application of

the energy method requires a scalar functional πp to be defined. The πp for seepage
analysis is given by
Assuming unit thickness in the Z-direction, Eq. (7.31) can be written, in view of Eqs. (7.25)
and (7.29), as
where
Minimization of πp results in the element equations
Note that [B] and [D] are constant matrices. Therefore, Eq. (7.34) can be written as
Steps 5, 6 and 7: Assemble element equations and apply boundary conditions, solve for
primary unknowns, calculate secondary unknowns
Assembling element equations given by Eq. (7.35) for the whole system, equation
can be derived, where [K], {q} and {F} represent the assembled permeability matrix, the
potential head vector and the externally applied fluid flux vector, respectively.
After applying geometric boundary conditions (specified potential heads corresponding to
Dirichlet conditions) Eq. (7.37) can be solved for the unknown {q}. The secondary
quantities can be obtained by equations developed previously. For example, flow velocity
components can be computed from Eqs. (7.26) and (7.27).
221
Illustrative examples dealing with confined seepage are discussed later in this chapter.
7.3 TWO-DIMENSIONAL STRESS ANALYSIS

Two-dimensional finite elements can be used for two-dimensional stress analysis
problems. A summary of pertinent equations from the theory of elasticity is presented
below in order to better understand finite element formulation for two-dimensional stress
analysis.
7.3.1 Review of Theory of Elasticity

Three-Dimensional Elastic Body (Crandall et al., 1978)
Consider an elastic body in space. Any particle within the elastic body may experience
displacements U, V, W in the X, Y, Z directions, respectively. For such displacement
components, under the assumption of isotropic, elastic material properties, following
expressions are well established.
1. Equations relating strains to displacements
Normal strains:
Shear strains:
2. Stress-strain equations
Normal strains:
Shear strains:
3. Generalized Hooke's law
222
7.3.2 Application of Three-Dimensional Equations for Two-Dimensional analysis

In reality, a physical system has two-dimensional domain. However, wherever possible,
engineering approximations are introduced on the basis of physics of the problem to
reduce the dimension of a problem at hand. For example, in the previous chapter, one-
dimensional analysis was presented for a truss, beam and a frame member under the
assumption of slender body (i.e., small cross sectional dimensions compared to member
length). Similarly, in many three-dimensional situations (even when all three dimensions
are comparable), the domain of the problem can be considered to be a two-dimensional
one. Three common situations are: (i) plane stress, (ii) plane strain and (iii) axisymmetric
solids. Typical examples of plane stress, plane strain and axisymmetric situations are
shown in Figure 7.8. These three analyses procedures are elaborated next.
7.3.2.1 Plane Stress Situation

It is that state of stress in which the normal stress and the shear stresses directed
perpendicular to the plane X-Y (in which loads are applied) are assumed to be zero. This
approximation can be applied when two dimensions of a physical system are larger than
the third one (say, thickness). A typical example of plane stress situation is a thin plate
with a hole subjected to in-plane tensile loading. Note that if the loads act normal to the
plate, plate bending theory (e.g. Bathe, 2001) is required in the analysis.
Figure 7.8 Typical examples of plane stress, plane strain and axisymmetric situation
For the plane stress condition, it is assumed that
Hence, it can be seen from Eq. (7.42) that
223
and
where {σ} and {ε} are
and
7.3.2.2 Plane Strain Situation

It is a state of strain in which the strain normal to the X-Y plane (the plane in which loads
are applied), i.e., εz, and the shear strains γxz and γyz are assumed to be zero. The plane
strain approximation can be applied when the third dimension (say length) is much larger
than the dimensions of the cross section, and the loading is uniform over the length. A
typical example where the plane strain approximation may be adequate is a gravity dam.
For plane strain case, following strains are assumed to be zero.
In view of Eq. (7.42), it can be seen from the above equation that
or,
where
224
7.3.2.3 Axisymmetric Situation

As the name ‘axisymmetric’ suggests when the geometry, boundary conditions, material
properties and loading are identical with respect to axis of symmetry (i.e., independent of
angle θ), the three-dimensional problem can be reduced to an analogous two-dimensional
problem. Typical examples, where the axisymmetric analysis may be sufficient, are
pressure vessels, pistons, rotors, water tanks and circular column foundations. General
loading will be considered later in Chapter 8.
By using cylindrical co-ordinate system (r, θ, Z) and denoting the displacement
components in radial, tangential and vertical directions as U, V and W (see Figure 7.9), it
can be shown for axisymmetric case that
Here, εr, εθ and εz are the components of strain in the radial, tangential and vertical
directions, respectively.
Figure 7.9 Axisymmetric analysis

From Eqs. (7.53) and (7.42), it can be easily shown that
7.3.3 CST Element for Plane Stress and Plane Strain Analyses
Steps outlined in Chapter 3 are followed in the finite element formulation.
225
A typical i-th CST element used to discretize a two-dimensional domain is shown in Figure
7.10.
A typical CST element has 3-nodes. Unknown continuous field consists of displacement
components U and V in the X and Y direction, respectively.
By following the formulation of CST element for seepage analysis, it can be shown that
Figure 7.10 Typical CST element for plane stress/strain analysis

where N1, N2, N3 are shape functions for the CST element and U1, U2, U3, V1, V2, V3 are
nodal displacement components. Therefore,
or
Here,
Step 3: Define Strain-displacement relations and constitutive law

Strain–Displacement (Gradients) Relations
226
Therefore,
Moreover,
where
Note that {ε} is a constant vector. Therefore, strains are constant.

Constitutive Relations
where [D] is given by Eq. (7.47) for plane stress conditions and by Eq. (7.52) for plane
strain conditions. By substituting Eq. (7.61) into Eq. (7.63),
can be obtained.
By employing energy method, the element equations can be derived easily. The potential
energy πp is defined for two-dimensional analysis as
Here, {X}, {fee} and {T} represent body force (per unit volume) vector, the element nodal
force vector and the surface traction (per unit surface area) vector, respectively. In view of
Eqs. (7.61) and (7.64), Eq. (7.65) can be written as
227
where
and {P} represents the element nodal force vector.

Minimization of πp results in the element equations
Element Stiffness Matrix

Since [B] and [D] are constant matrices, for an element with constant thickness t, Eq.
(7.67) becomes
Assume t = 1 for plane strain analysis.

Body Forces
The equivalent nodal loads due to body forces are given by
where
and Xb and Yb are the body forces (weight densities) in the X and Y directions,
respectively.
Hence,
Consider
Note that
228
By definition, co-ordinates of the centroid are

Thus, it can be observed that
Similarly, by integrating other terms in Eq. (7.73),
The above equation indicates that each node shares 1/3 of total body force.
Element Surface Forces
The equivalent nodal loads due to surface forces are given by
Steps 5, 6 and 7: Assemble element equations and apply boundary conditions, solve for
primary unknowns, calculate secondary unknowns
Assembling the contributions from all elements of discretized domain, equation
can be derived, where [K], {q} and {F} represent the assembled stiffness matrix, the nodal
displacement vector and the externally applied, equivalent nodal load vector, respectively.
After applying geometric boundary conditions (specified displacement components), Eq.
(7.80) can be solved for the unknown {q}. The secondary quantities can be obtained by
229
equations developed previously. For example, the principal stresses can be computed
from
where σx, σy, τxy and be calculated from Eq. (7.64) (θp in the above equation indicates
principal angle).
Two-dimensional model of an anchor plate of a communication tower's guy cable is shown
in the illustrative figure. The anchor consists of a triangular steel plate, which is subjected
to force of 20 kN, as shown in the figure. Analyze the anchor plate. Thickness of the plate
is 8 mm. Consider E = 200 GPa and v = 0.3.
Solution:
Consider reference axes X and Y as shown in the figure. Further for illustration, consider
one CST element to model entire plate, with nodes as shown in the figure.
From Eq. (7.11),
Considering the state of plane stress, the elasticity matrix [D] is obtained from Eq. (7.47)
as
230
Further, Eq. (7.69) can be written in partitioned form as
where
Due to homogeneous boundary conditions at nodes 2 and 3,
Thus, only ḵ11 is required for computation of Q1.

By substituting values of t, A,D11,D12, D33,b1 and a1, the following equation is obtained.
By making use of Eq. (7.64),
231
Since only one CST element is used to model the entire plate, the stress field is constant
throughout the plate.
Figure 7.11 Typical triangular element to analyze axisymmetric solids
7.3.4 Triangular Element for Axisymmetric Analysis

Step-by-step formulation of a triangular element is presented below.
A typical 3-node, triangular element from the discretized domain of an axisymmetric solid
is shown in Figure 7.11. An axisymmetric solid is modelled by rotating such elements
through 360°.
At each node, there are two DOF corresponding to the displacements in the radial and
vertical directions. Thus,

where shape functions N1, N2 and N3 are given by
Alternatively,
232
or
Here,
Step 3: Define strain–displacement relations and constitutive law

Strain-Displacement (Gradients) Relations
It can be shown from Eqs. (7.53) and (7.86) that
i.e.,
where
Note that the strain vector is no longer constant. Also, [B] involves singular terms when r =
0.
Constitutive Relations
where [D] is given by Eq. (7.54). By substituting Eq. (7.89) into Eq. (7.91),
can be obtained.
233
By minimizing the potential energy, element equations can be derived as
Here,
and
Integrations appearing in Eqs. (7.94) and (7.95) can be evaluated by several ways.
1. Integrate all the terms explicitly. However, the explicit evaluation may be tedious
and complicated due to (1/r) terms in the integrands.
2. Use numerical integration. Usually, a Gauss-quadrature procedure is employed for
numerical integration. Refer to Appendix D for details.
3. Integrations can be obtained approximately by evaluating integrands at the centroid
of an element. This approach is reasonable when ‘small’ elements are employed in
discretization. At the centroid,
and thus, say, [Ke] can be approximated as
where [ ] is evaluated at the centroid.

Steps 5, 6 and 7: Follow steps outlined for plane stress/plane strain analyses by using
CST element
Axisymmetric formulations with both axisymmetric as well as non-axisymmetric forces
have been further elaborated later.
7.3.5 Some remarks on Triangular Elements

In order to improve accuracy of results, mesh can be refined by reducing the size of 3-
node, triangular elements. This approach is often known as h-version of finite element
analysis. Alternatively, in the p-version, instead of refining the mesh, higher order
elements (where strains can vary linearly, parabolically, cubically, etc.) can be used to
improve accuracy. Higher order elements are presumed to closely model the strain and
stress distribution in a continuum.
Triangular elements having 6- and 10-nodes are shown in Figure 7.12. Since the strains
vary linearly and quadratically for 6- and 10-node elements, they are, respectively, referred
to as linear strain triangle (LST) and quadratic strain triangle (QST) elements.
234
7.3.6 Four-Node Rectangular Element for Plane Problems

In this section, a step-by-step, finite element formulation of a 4-node rectangular element
is presented.
A rectangular element shown in Figure 7.13 can be used to discretize a domain having a
regular geometric shape. However, in practice, quadrilateral elements are used more
extensively. Note that a rectangular shape is a degenerated quadrilateral.
Figure 7.12 Higher order triangular elements
Figure 7.13 Typical, 4-node, rectangular element

A typical 4-node, rectangular element from the discretized domain is shown in Figure 7.13.
The element DOF vector for this element is given by

Since there are 4-nodes (and, hence, four unknowns per global direction), one can select
only four terms from the Pascal's triangle (Figure 7.14).
Assume
235
By eliminating ai's and bi's, it can be shown that
where
Figure 7.14 Criterion for selecting approximation function

The displacement vector {U (X, Y)} can be written as
In other words,
where
Step 3: Define strain–displacement relations and constitutive law

It can be easily shown that
236
Let
Then,
where
Note that, unlike in the CST formulation, the strains are not constant.
Constitutive Relation
As shown earlier,
where 3 × 3 elasticity matrix [D] is given by Eq. (7.47) for plane stress condition and Eq.
(7.52) for plane strain conditions.
Step 4. Derive element equations.
By following the procedure adopted for the CST formulation, in conjunction with energy
method, the element equations are given by
where
237
Here, {X} represents the body force vector, {P} indicates the element nodal force vector
and {T} represents the surface traction vector.
Analytical evaluation of integrations in Eqs. (7.111) and (7.112) is tedious. Therefore,
numerical integration is usually employed. The most widely used integration scheme in the
FEM is the Gauss-quadrature. The Gauss-quadrature is defined in a natural (or intrinsic)
co-ordinate system. Therefore, the integrals expressed in the X-Y co-ordinate system
need to be transformed to the natural co-ordinates ξ-η. The numerical integration is
discussed later in the context of isoparametric formulation.
Step 5, 6 and 7: Follow procedures outlined previously
7.4 ISOPARAMETRIC FORMULATION

Term ‘isoparametric’ is derived from the use of the same shape functions (or interpolation
functions) to define the element's geometric shape as those used to define the variation of
unknown field within an element. Isoparametric element equations are formulated using a
‘natural’ (or ‘intrinsic’) co-ordinate system defined by the element geometry. The intrinsic
co-ordinates normally vary from −1 to +1, regardless of the element's dimensions. Thus,
use of such a co-ordinate system becomes very attractive, especially when all the
matrices need to be integrated numerically.
The concept of isoparametric formulation is explained below in the context of a line (bar)
element and is extended later to a two-dimensional, quadrilateral element.
7.4.1 Two-Node Isoparametric Line Element (Bar Element)

Consider a 2-node bar element of length L as shown in Figure 7.15.
Refer to the finite element formulation presented in the Chapter 6 for a bar element. In this
section, this formulation is revisited in the context of natural co-ordinate system. For the
element shown in Figure 7.15, the nodal co-ordinates can be written in global, local or
intrinsic co-ordinate system.
At node 1: X = X1; x = 0; ξ = −1.
At node 2: X = X2; x = L; ξ = +1.
X global co-ordinate (X1 ≤ X ≤ X2)
x local co-ordinate (0 ≤ x ≤ L)
ξ natural or intrinsic co-ordinate (−1 ≤ ξ ≤ 1)
Co-ordinate Transformation
Global co-ordinate (X) is related to natural co-ordinate (ξ) by
Therefore,
238
or
where
Variation of shape functions over an element are shown in Figure 7.16.

Approximation Function
Figure 7.15 Two-node, isoparametric line element
Figure 7.16 Shape-function variation with respect to natural co-ordinate

From chain rule of differentiation,
239
Hence,
Constitutive relation:
Stiffness matrix:
Note that X = X1 + x and dX = dx. From calculus
where is called the ‘Jacobian matrix’ of the co-ordinate transformation. | | is

determinant of . For one-dimensional case, is a scalar, equal to L/2. From Eq. (7.114),
Therefore,
240
It can be seen from the above equations that each element of the stiffness matrix is of the
form . In general, these integrations can be performed by employing Gauss-

quadrature procedure. (Refer to Appendix D for details.) However, for a linearly varying
function, these expressions can be derived explicitly. Thus, it can be seen for 2-node
element that
7.4.2 Four-Node Isoparametric Element for Plane Problems (Quadrilateral Element)

Formulation presented for rectangular element is extended for a more general 4-node
element in this section.
A typical quadrilateral element from the discretized domain is shown in Figure 7.17.
Intrinsic co-ordinates ξ, η are also shown in this figure. Note that regardless of the size or
shape of the quadrilateral element, the intrinsic co-ordinates always vary from −1 to +1.
Thus, a given quadrilateral element is effectively mapped into a rectangular element in the
natural co-ordinate system.
Figure 7.17 Quadrilateral element in the X-Y and the ξ-η co-ordinate systems
By considering four terms from the Pascal's triangle shown in Figure 7.14, X and Y co-
ordinates can be expressed in terms of ξ and η
Co-ordinates of all nodes in the global and natural system are:
241
Hence, the application of Eq. (7.126) to all the nodes yields
By solving the above equation set for ai's and by substituting the results in Eq. (7.126),
is obtained, where
or
where ξi and ηi are natural co-ordinates of i-th node.

Similar expression for Y can be derived as
Thus, the co-ordinate transformation is given by
or
242
Figure 7.18 Shape functions for a quadrilateral element

Notes:
1. Ni(i = 1, 2, 3, 4) is equal to one at node i and equal to zero at all other nodes.
2. For all values of ξ and η,
Variations of Ni, i = 1, 2, 3, 4 over an element are shown in Figure 7.18.

In an isoparametric formulation, the same shape functions are employed to describe the
variations in the global co-ordinates and the unknown field. Thus,
Alternatively,
243
where
where
Here,
The chain rule of differentiation yields
244
In the above equations, and are unknowns. They can be obtained, in terms of
and , as
Where
Hence
The constitutive relation is given by
where [D] is given by Eqs. (7.47) and (7.52) for plane stress and plane strain conditions,
respectively.
Derivation of Element Equations
By minimizing the potential energy, the element equations
can be obtained, where
Assume t = 1 for plane strain situation. Computations of [Ke], {fb} and {fs} are discussed
next.
1. Stiffness matrix:
From calculus, for an arbitrary function f(x, y)
245
Hence,
Where
Here,
Therefore,
It can be seen from the derivation that each element of [Ke] is of the form
These expressions can be evaluated by employing Gauss-quadrature procedure for area

integration, as shown in Appendix D.
2. Body forces:
One can use Gauss-quadrature rule to compute {fb}.

3. Surface forces:
Consider edge η = 1 with overall length L.
Gauss-quadrature rule can be used easily to compute {fs}.

For the element shown, evaluate k11 of the element stiffness matrix [Ke]. Assume plane
strain condition. Let E = 200 GPa, v = 0.25. The global co-ordinates (in m) are shown in
the figure.
246
Solution:
From Eq. (7.155), the expression for k11 is obtained as
Where
Therefore,
Hence,
Where
1. 1 × 1 Gauss-quadrature
4. Exact value
247
A cantilever beam of span 9 m is subjected to a tip load of 40 kN. Analyze the beam by
employing 4-node isoparametric elements under plane stress condition. Assume E = 200
GPa, v = 0.25 and width = 0.3 m.
Solution:
The finite element representation of the problem is shown in the figure above. Node
numbers are indicated by Arabic numbers, while element numbers are denoted by
encircled Arabic numbers. The problem has been solved by using program sfeap given in
the Appendix F.
The nodal displacements are tabulated below.
NODE U -DISP (m) V -DISP (m)
1 .0000000E+00 .0000000E+00
2 .0000000E+00 .0000000E+00
3 .0000000E+00 .0000000E+00
4 –.1940316E–04 –.6343784E–04
5 .0000000E+00 –.6298111E–04
6 .1940316E–04 –.6343784E–04
7 –.3112135E–04 –.2198400E–03
8 .0000000E+00 –.2197260E–03
9 .3112135E–04 –.2198400E–03
10 –.3502455E–04 –.4232532E–03
11 .0000000E+00 –.4231881E–03
12 .3502455E–04 –.4232532E–03
248
It can be noted that nodes 1, 2 and 3 are completely restrained and, thus, both U and V
displacements are zero at these nodes. Further, the U displacement is zero for all nodes
on the neutral axis (viz., nodes 5, 8 and 11). Also, as expected, U displacements at the
nodes are anti-symmetric with respect to the reference neutral axis. On the other hand, V
displacements at the nodes are symmetric with respect to the neutral axis. Stresses at the
centroids of the elements are tabulated below.
7.5 FINITE ELEMENT SOLUTION OF PARTIAL DIFFERENTIAL EQUATIONS

BY METHOD OF WEIGHTED RESIDUAL
In this section, the finite element formulations of some important steady-state partial
differential equations governing the field problems in two dimensions are presented. In
particular, Laplace equation, Poisson equation and Helmholtz equation are considered.
Laplace equation is an elliptic equation used to characterize steady-state systems (Chapra
and Canale, 2002). Thus, the two-dimensional form of Laplace equation can be used to
determine the steady-state distribution of an unknown function in two spatial dimensions.
The Laplace equation can be used to model a variety of engineering problems involving
potential of an unknown such as heat conduction problems, seepage problems, aquifer
problems, distribution of an electric field, potential flow problems, etc.
Poisson equation is another class of partial differential equations similar to Laplace
equation with a constant term such as a source or sink. The two-dimensional form of
Poisson equation can be used to determine steady-state distribution of an unknown
function in two spatial dimensions with a source or sink term in the domain. The Poisson
equation can be used to model a variety of engineering problems involving heat transfer
problems, torsion of constant cross section member, steady-state aquifer problems with
source or sink, electrostatics, magnetostatics, irrotational flow, etc.
Helmholtz equation is another class of steady-state partial differential equations closely
related to the quasi-harmonic equations. The Helmholtz equation can be used to model a
variety of engineering problems involving standing waves on a shallow body of water
(seiche motion) (Taylor et al., 1969), electromagnetic waves (Arlett et al., 1968), acoustic
vibration (Kinsler and Fray, 1950), fluid structure interaction (Brebbia and Walker, 1979),
etc. For example, the Helmholtz equation with appropriate boundary conditions can be
used to model the fluid vibrating in a closed volume or the propagation of electromagnetic
waves in a waveguide filled with a dielectric material (Huebner, 1975).
These equations can be solved by using method of weighted residual technique or
classical variational principle. Here, the Galerkin’s based method of weighted residual has
been used in the formulation of these equations for illustration. A brief discussion of the
FEM based application of classical variational principle is also presented.
7.5.1 Governing Equations and Boundary Conditions
1. Laplace Equation
For homogeneous isotropic media field problems, the Laplace equation in two-dimensions
can be written as
249
where ɸ is the potential function, x and y are the Cartesian co-ordinates.

2. Poisson’s Equation
The Poisson's equation can be used to represent the steady state of an unknown variable
such as heat transfer with a heat source, groundwater flow with recharge or pumping,
electrostatics or magnetostatics with a charge density, a torsion problem with an initial
twist etc. For homogeneous media problems the Poisson equation in two dimensions can
be written as
where ɸ is the potential function and C is the source or sinks term in the domain
considered.
3. Helmholtz Equation
General form of Helmholtz equation in two dimensions can be written as (Brebbia and
Walker, 1979)
where Kx and Ky are media properties (for example, in case of heat transfer problems, Kx
and Ky represent the thermal conductivities), λ is a constant and Q is a term such as
source or sinks in the domain considered.
The commonly used boundary conditions are Dirichlet and Neumann type boundary
conditions. The boundary conditions can be generally represented as (refer Figure 7.19)
where n is the unit outward normal to the boundary, α is the transfer coefficient (e.g., heat
transfer) and (ɸ - ɸ0) denotes the difference between the potential within the medium; and
p are known values and Γ = Γ1 + Γ2.
Figure 7.19 Definition of problem domain

7.5.2 FEM Formulation
It can be seen from the above equations that the Helmholtz, Poisson and Laplace
equations describe similar class of steady-state field problems with some variations from
each other. When Kx = Ky = 1; λ = Q = 0 in Eq. (7.164), Laplace equation is obtained. On
the other hand, for Kx = Ky = 1; λ = 0; Q = C, Poisson equation is obtained from Eq.
(7.164). Here, the detailed FEM formulation of Helmholtz equation is presented. The FEM
formulation for other equations can be easily derived in a similar way.
1. Helmholtz’s Equation
Let the domain be divided into n CST elements. By using the Galerkin's approach, first
approximate the unknown exact solution ɸ as
250
where N is the shape function or interpolation function and {ɸ}i is the array of field variable
at specified nodes of the element of the discretized domain.
Applying the Galerkin's criteria, the Helmholtz Eq. (7.164) can be reduced to a set of linear
equations as follows:
The first term can be integrated by parts as
Similarly, the second term can be written as
By using Eq. (7.168), the original equation can be written as
By substituting for ɸ using the relationship (7.166) and rearranging, following equation is
obtained.
Element Eq. (7.170) can be written in a matrix form as
where
251
As shown in Section 7.2 and Figure 7.3, using CST elements or any other two-dimensional
elements and assembling the element matrices and applying the boundary conditions, a
linear system of equations can be formed, which are solved for the unknowns.
2. Laplace Equation
For the Laplace Eq. (7.162), in a similar way as demonstrated for Helmholtz equation, the
element equation can be derived. As given in Section 7.2 and Figure 7.3, using the CST
element and shape functions given in Eq. (7.18), the element property matrix can be
written with kx = ky = 1 as
Here, p1, p2 and p3 are known values, if any.

Eq. (7.173) can be written in the compact form
Eq. (7.174) can be assembled into global matrix over all elements and can be solved to
get the potential values ɸ at different nodes, after application of the global boundary
conditions.
3. Poisson Equation
For the Poisson Eq. (7.163), the element equation can be derived in a similar way as
demonstrated for Helmholtz equation. As given in Section 7.2 and Figure 7.3, using the
CST element and shape functions given in Eq. (7.18), the element property matrix can be
written as
Eq. (7.175) can be written in the form
Eq. (7.176) can be assembled into global matrix over all elements and got solved to get
the potential values ɸ at different nodes.
7.6 FEM FORMULATION BASED ON VARIATIONAL PRINCIPLE

In the previous section, the FEM formulation for various field problems using Galerkin's
based method weighted residual technique has been presented. In this section, the
application of variational principle for a field problem is demonstrated by considering a
non-homogeneous steady-state field problem. The governing differential equation for non-
homogeneous steady-state media problem such as seepage flow problems, can be
expressed as
Here, ɸ is the potential function and k represents the media property such as hydraulic
conductivity. The boundary conditions can be generally represented as (refer Figure 7.19)
252
where kx and ky are the hydraulic conductivities in x and y directions and n is the unit
outward normal to the boundary.
As discussed in the previous section, the governing equation can be reduced to a system
of equations by using the FEM technique. In the previous section, the method of weighted
residual approach for the FEM formulation has been used. Here, the variational
formulation approach for the derivation of the element matrices is used. From Eq. (7.177)
and (7.178), using equivalent variational formulation, equation reduces to minimization of
the functional (Huebner, 1975)
over the whole domain Ω.

In the variational approach, the functional has to be minimized so that
Here, Galerkin's technique can also be used to formulate the problem, which is very similar
to that given in Section 7.5 for homogeneous steady-state problems.
As given in Section 7.2 and Figure 7.3, the CST element and shape functions given in Eq.
(7.18) has been used so that
with usual notation as given in Section 7.2.

From the variational principle, Eq. (7.180) can be written as
Therefore,
253
Combining the above three Eqs. (7.183–7.185) and substituting boundary prescribed
values p1,p2, p3 etc. if any, the following element matrix will be obtained.
For isotropic case kx = ky = k. Eq. (7.186) can be written in the form,
Eq. (7.187) can be assembled into global matrix over all elements and solved to get the
potential values ɸ at different nodes after application of boundary conditions.
7.7 FINITE ELEMENT SOLUTION OF STOKES FLOW EQUATIONS

In the preceding sections, the FEM formulations of various types of partial differential
equations such as Laplace equation, Poisson equation and Helmholtz equations were
considered. In this section, the simplest form of two-dimensional steady-state flow called
Stokes flow or creeping flow, which has a large number of applications in civil, mechanical,
metallurgy and chemical engineering is considered. In many engineering fluid flow
problems, the Reynolds number, which represents the ratio of inertia force to viscous force
in a fluid motion, can be very small (less than one) and, hence, the inertia forces are
insignificant compared to viscous force and can be omitted from the concerned
momentum equations (Navier Stokes equations) (Granger, 1985). Here, the flows with
small Reynolds number such as slow-moving flows and the flows of very viscous fluids
such as polymer processing flows are considered.
7.7.1 Problem Statement

The governing equations for Stokes or creeping flows in two-dimensions are given by
(Granger, 1985) the following momentum equations.
and the continuity equation is
Here, p is the pressure, u and v are velocity components in the x and y directions and μ is
coefficient of dynamic viscosity. Boundary conditions may include specified pressure,
velocity or velocity gradient. The u, v and p are solved within the problem domain
considered.
254
7.7.2 FEM Solution

Here, the Galerkin's method based on the weighted residual approach is used. The
domain (Figure 7.20) is divided into n CST elements (Figure 7.21). Using the Galerkin's
approach, the unknown exact solution ɸ is approximated as
where N is the shape function or interpolation function and {ɸ}i is the array of field variable
at specified nodes of the element of the discretized domain, say here, (u, v and p).
Considering a two-dimensional domain A bounded by curve S and applying Galerkin's
approach on the Eq. (7.188), the following equation is obtained.
As demonstrated in the previous sections, after integration by parts, the first term can be
converted as
Similarly, the second term can also be written.

Hence, Eq. (7.192) can be written as
Here, lx = nx and ly = ny, as defined early.
Figure 7.20 Problem domain for Stoke's Flow
255
Figure 7.21 FEM Discretization of the Problem Domain (not to scale)

Similarly, Eq. (7.189) can be written as
Using the Galerkin's approach, the continuity Eq. (7.190) can be written as
or,
Consider the Stokes flow problem in a rectangular domain. The domain and boundary
conditions are shown Figure 7.20. The domain is divided into CST elements as shown in
Figure 7.21 and the numbering is done. As discussed earlier, the interpolation function can
be written as
256
Figure 7.22 (a) Triangular element

For the shape shown in Figure 7.22 with 1 unit as the side length of the triangle and 1/6
unit as height of the triangle, the various element properties can be written as follows.
For the following element shape:
257
Figure 7.22 (b) Triangular element
For the first elemental shape shown in Figure 7.22(a), Eqs. (7.193), (7.194) and (7.196)
can be shown to be
258
Where
For the second elemental shape shown in Figure 7.22(b), Eqs. (7.193), (7.194) and
(7.196) can be shown to be
Using the above equations, the system of elemental equations can be assembled in the
form,
259
where {ɸ} represent the field variables (u, v and p). After substitution of boundary
conditions, the system of equations can be iteratively solved to find the unknown velocities
and pressure.
7.8 ILLUSTRATIVE EXAMPLES

Based on the Galerkin FEM formulation given in Section 7.5, a computer program has
been developed using the FORTRAN for the Laplace equation. The program can be used
for homogeneous and non-homogeneous media problems. The program is tested with the
problem mentioned below. The detailed program listing is given in Appendix G.
Example 1–Potential Variation in Porous media

Determine the potential variation in a homogeneous isotropic porous media of length 100
m and width 30 m. As shown in Figure 7.23, the potential at one side of the porous media
is 300 m and the opposite side is 100 m. Other two sides of the section are considered as
impervious.
Solution:
The problem is solved as a two-dimensional problem using the Laplace equation with the
FEM. The domain is discretized using 24 CST elements and 20 nodes as shown in Figure
7.23. This problem is solved using the FORTRAN Computer program given in Appendix G,
Section G3. The input file used and results are also given Appendix G. Results are plotted
in Figure 7.24(a)and (b). Results are compared with analytical solution ɸ = (ɸ2 – ɸ1) x/L +
ɸ1; where ɸ is the head at distance x, ɸ1 is the head at the beginning, ɸ2 is the head at the
end and L is the total length.
Example 2–Flow net analysis below a Dam

Consider the flow beneath a concrete dam on a homogeneous isotropic permeable
foundation. As shown in Figure 7.25, there is an impermeable boundary at the upstream
and the downstream. Boundary conditions with the potential function and stream function
are shown in the figure. Determine the potential and stream function distribution in the
considered permeable foundation and draw the flow net.
Figure 7.23 Flow domain and FEM discretization for Example 1
260
Figure 7.24 (a) Potential variation–Example 1
Figure 7.24 (b) Potential variation in the aquifer media–Example 1
Figure 7.25 Problem domain and boundary conditions - Example 2

Solution:
Consider the Laplace equation in potential function ɸ and stream function Ψ. Both the
equations are solved separately with the given boundary conditions. The domain in Figure
7.25is discretized using 44 CST elements and 39 nodes, as shown in Figure 7.26. This
problem is solved using the FORTRAN Computer program given in the Appendix G. The
261
input file used and results are also given in the same Appendix. Results are plotted in the
form a flow net in Figure 7.27.
Example 3–Flow Beneath a Spillway-Case with three Zones

Here, the potential variation beneath a spillway is analyzed. There are three zones with
permeabilities 1 m/hr for zone 1, 2 m/hr for zone 2 and 3 m/hr for zone 3. Potentials on the
upstream and the downstream are respectively 20 m and 2 m. Domain size is 100 m × 20
m. On both the sides and bottom, the media is impermeable and hence . At the
interzonal faces, there are two cut-off walls of size 5 m × 1 m as in Figure 7.28.
Figure 7.26 FEM discretization–Example 2
Figure 7.27 Flow net-flow beneath a dam - Example 2
262
Figure 7.28 FEM discretization-flow beneath a spillway - Example 3

The FEM discretization using CST elements is shown in Figure 7.28. The domain is
discretized using 96 triangular elements and 69 nodes as in Figure 7.28. The input data
can be prepared in the similar way as in the previous case studies. This problem is solved
using the FORTRAN Computer program given in Appendix G. The potential variation at
various internal nodes is given in Table 7.1.
Example 4–Seepage Problems in Earth Dams

Location of free surface is of paramount importance in the case of earth dams,
embankments, canals etc. For homogeneous media problems, even though some
approximate solution like Casagrande's solution exists, exact location of free surface for
heterogeneous or zoned media is still a challenge. The finite element formulation given in
Section 7.5 is applied here for the determination of free surface of zoned media. The top
boundary must satisfy two constraints: (1) The surface is a no flow boundary and (2) The
head at each point on the boundary must be equal to its elevation. With the above
conditions and using the FEM formulation for the zoned media, following additional steps
are used in the free surface determination.
1. Assume an initial free surface curve y = f (x).
2. Discretize the domain. Identify the free surface nodes.
3. Using the formulation given in Section 7.5, determine the potential at the nodes.
4. The boundary head values on the nodes on the assumed free surface are
compared against the assumed y co-ordinates values of the respective nodes.
Table 7.1 Flow Beneath a Spillway–Potential Variation
263
5. If the difference obtained in Step 4 is within the permissible tolerance, the obtained
potentials are assumed as free surface.
6. If the discrepancy is too large, go for another iteration with Y co-ordinates of free
surface node as the potential values obtained in Step 3.
The procedure is repeated in each iteration, until the discrepancies are within the
permissible limit. In each iteration, the mesh used should be adjusted to relocate the new
free surface position.
Example: The location of free surface in a zoned earth dam is required to be determined.
There are three zones with different permeabilities. It is assumed that permeabilities of
zone 1, 2 and 3 are 2 m/hr, 1 m/hr and 3 m/hr, respectively. Upstream head is 10 m and
the downstream head is 2 m. Bottom face,
264
Figure 7.29 FEM Discretization–Free Surface in a Zoned Earth Dam–Example 4

where the dam is placed is impermeable. On both the impermeable face and the free
surface, . Dimensions are as shown in Figure 7.29.
Discretization of the domain is as shown in Figure 7.29. A total of 26 CST elements and 24
nodes are used for discretization. Number of prescribed boundary nodes are 5. A
computer program, in a very similar way that is given in the Appendix G, can be
constructed for this problem also. Results along with the location of free surface is plotted
in Figure 7.30.
Example 5–aquifer Problem

Here, a steady ground water recharge problem is analyzed using the FEM formulation for
Poisson equation given in Section 7.5. A domain of 2,400 m length and 1,200 m width is
considered. There is a recharge basin of size 200 m × 100 m at the central portion. The
recharging rate is 0.2 m/day/m. The thickness of the aquifer is assumed to be 100 m. The
transmissibility is 1,000 m2/day. Configuration of the aquifer is shown in Figure 7.31. At the
center of each side of the aquifer, the potential is known. At the center of BC, ɸ = 97 m. At
the center of each DE and HA, ɸ = 90 m. At the center of FG =, ɸ = 92 m. The aim is to
compute the head variation at various points in the domain. Sides AB, CD, DE, EF, GH
and HA are impermeable so that prevails on these sides.

For a steady state recharge problem in porous media, the governing differential equation
can be written as (Bear, 1972)
Figure 7.30 Free Surface Profile in a Zoned Earth Dam - Example 4
265
Figure 7.31 Aquifer recharge - Problem Domain (not to scale)

where T is the transmissibility of the porous media, R is the source or sink term. For
isotropic media, Tx = Ty = T. Therefore, the equation becomes
Here, the FEM formulation for the Poisson equation described in Section 7.5 is used. The
domain is discretized using CST elements as shown in Figure 7.32. The domain is divided
into 136 CST elements and 95 nodes. Number of recharging nodes are 9. It can be seen
in Figure 7.32 that the recharge zone is discretized with a finer mesh and other regions
using a coarser mesh. A computer program for the Poisson equation can be developed in
a very similar way as demonstrated for the Laplace equation. The difference is only with
respect to the recharge term. Using a similar computer program given in Appendix G (for
Poisson equation after adding the recharge effects), the above problem is solved. The
potential variation at some selected points is given in Table 7.2.
266
Figure 7.32 FEM Discretization-Aquifer Problem (not to scale)

Table 7.2 Aquifer Problem-FEM Results
267
7.9 CLOSING REMARKS

In this chapter, two-dimensional finite element analysis of engineering problems are
described. The energy method, method of weighted residual based Galerkin approach and
variational principles in FEM described in Chapter 4 are used in the analysis. Various
types of elements such as linear type triangular elements, rectangular elements and
quadrilateral elements are used in the analysis of various problems. Basic concepts of
isoparametric formulation are also elaborated. FEM formulation for stress-strain analysis,
Laplace equation, Poisson equation, Helmholtz equation and Stokes flow equations are
discussed in this chapter. Engineering problems of flow through porous media, stress
268
analysis, seepage below hydraulic structures, earth dam seepage analysis and aquifer
recharge are discussed using the FEM. A sample two-dimensional computer program
based on FORTRAN is also presented and used in the solution of some engineering
problems. This program can be used to solve any similar types of problems, as
demonstrated. Even though the problems discussed in this Chapter are related to two-
dimensional FEM analysis using energy method, method weighted residual and variational
approaches, the three-dimensional FEM analysis can be done in a similar way.
EXERCISE PROBLEMS
1. Two-dimensional model of an anchor of a communication tower's guy cable is shown in
Figure E1. The anchor consists of a triangular steel plate, which is subjected to force P as
shown in the figure. Thickness of the plate is 8 mm. The maximum displacement of the
node, where the force is applied, has to be restricted to 0.1 mm. Compute the largest force
P that can be supported by the anchor. Use only one appropriate CST element to model
the plate for expeditious analysis. Consider E = 200 GPa and v = 0.25.
Figure E1 Exercise Problem 1

2. The 8-node square element of size 10 cm × 10 cm shown in Figure E2 is subjected to
uniformly distributed pressure load of magnitude 10 MPa, horizontal point load of 10 kN
applied at the centroid of the element and vertically downward uniformly distributed body
force of magnitude 100,000 kN/m3. Determine the equivalent nodal forces at nodes 1, 3
and 6. Consider thickness of the element to be 2 cm.
269

3. Compute co-efficient k34 of 8 × 8 element property matrix for the 4-node rectangular
element shown in Figure E3. Consider state of plane stress. Consider thickness of the
element to be 2 cm. Consider E = 200 GPa and v = 0.25.

4. Derive a 2-node element for axisymmetric analysis of a long cylinder by considering
state of plane strain in the length direction of the cylinder. Elaborate all steps clearly.
5. Analyze cross section of a long culvert shown in Figure E4 by using 4-node elements.
Use program sfeap. Consider state of plane strain the length direction of the culvert.
Assume E= 20 GPa, v = 0.30 and uniform thickness of 0.3 m.
270

6. For the corner flow (flow between two parallel plates) shown in the Figure E5, equal
amounts of inflow and outflow giving a zero net flow rate takes place. Using the FEM
formulation for Laplace equation (in stream function) and suitable mesh with linear
triangular elements, determine the stream function variation across the corner line AB.
(FORTRAN program given in Appendix G can be used for solution).
Figure E5 Problem 6
7. For the earth dam with the assumed free surface shown in Figure E6, using the free
surface algorithm and FEM formulation given for the Laplace equation, determine the final
position of the free surface. Develop a computer program for the given algorithm. Use a
suitable mesh with linear triangular elements.
271
Figure E6 Problem 7
8. For the flow beneath weir problem shown in the Figure E.7, determine the potential
variation. The FEM formulation for Laplace equation and the FORTRAN program given in
Appendix G can be used for the solution.
Figure E7 Problem 8
9. For the aquifer problem with a recharging river and 2 pumping wells shown in Figure E8,
calculate the head distribution using a suitable mesh. The stream recharges 0.2
m3/(day.m) and pump P1 pumps 800 m3/day and P2 pumps 600 m3/day.
10. For the Stokes flow problem given in Figure 7.20, assume the domain is of dimensions
6 m × 36 m and determine the velocity distribution in the domain.
Figure E8 Problem 9
Arlett, P. L., Bahrani, A. K., and Zienkiewicz, O. C. (1968). Application of Finite Element
Method to the Solution of Helmholtz's Equation, Proc. IEE, 115(12):1762–1766.
Baker, A. J. (1983). Finite Element Computational Fluid Mechanics, McGraw
Hill/Hemisphere, New York.
272

Brebbia, C. A., and Walker, S. (1979). Dynamic Analysis of Offshore Structures, Newnes-
Butterworths, London.
Chapra S. C., and Canale R. P. (2002). Numerical Methods for Engineers, 4th ed., Tata
Connor, J. J., and Brebbia, C. A. (1976). Finite Element Techniques for Fluid Flow,
Crandall, S.H., Dahl, N.C. and Lardner, T.J. (1978). An Introduction to the Mechanics of
Solid, McGraw Hill, New York.
Desai, C.S., and Abel J.F., (1987), Introduction to the Finite Element Method–A Numerical
Method for Engineering Analysis, Wads Worth Publishing Co., California.
Gere, J. M. (1965). Analysis of Framed Structures, Princeton, New Jersey.
Granger R. A. (1985). Fluid Mechanics, Holt, Rinehart & Winston Publications, New York.
New York.
Kinsler, L. E., and Frey, A. R. (1950). Fundamentals of Acoustics, John Wiley and Sons,
New York.
Logan, D. L. (1993). A First Course in Finite Element Analysis. PWS Publishing, Boston.
MacCamy, R. C., and Fuchs, R. A. (1945). Wave Forces on Piles: A Diffraction Theory, US
Army Coastal Engineering Center, Tech. Mem. No. 69.
Press, New York.
Chichester.
Stasa, F. L. (1985). Applied Finite Element Analysis for Engineers, CBS Publications
Japan, Ltd., New York.
Srinivas K. (1988). Finite Element Analysis of Wave Forces on Offshore Gravity
Structures, M. Tech Thesis, Supervisor: Prof. B.V. Rao, Department of Civil Engineering,
IIT Bombay.
Streeter, V. L., Wylie, E. B., and Bedford, K.W. (1998). Fluid Mechanics, WCB/ McGraw
Hill, Boston.
Taylor, C., Patil, B. S, and Zienkiewicz, O. C. (1969). Harbor Oscillation: A Numerical
Treatment for Undamped Natural Modes, Proc. Inst. Civil Eng., 43:141–155.
Zienkiewicz, O. C. (1980). Finite Element Method, 3rd ed., McGraw Hill Book, New York.
273
8 Three Dimensional Finite Element Analysis

8.1 INTRODUCTION
Two-dimensional finite element analysis of various engineering problems was discussed in
the previous chapter. Analysis of axisymmetric solids for axisymmetric loading was also
discussed in the previous chapter by using 3-node triangular elements. In this chapter,
analysis of axisymmetric solid is considered in detail for axisymmetric as well
asymmetrical loading by using semi-analytical approach. Loads are expanded in the
circumferential direction using Fourier series and thus the finite element discretization is
still considered in the two-dimensional domain. Three-dimensional stress analysis is also
considered by using 8-node element. Energy methods have been used in the formulations.
Illustrative examples are also presented to demonstrate the formulations.
8.2 AXISYMMETRIC SOLIDS
This section deals with analysis of axisymmetric solids (solids of revolution) subjected to
axisymmetric and non-axisymmetric (asymmetric) loadings. Solid is formed by rotating a
plane about an axis of rotation, z (see Figure 8.1) and is assumed to have material
properties and boundary conditions independent of the circumferential co-ordinate, θ. The
axisymmetric solid structures have wide applications in civil and mechanical applications,
like water tanks, cooling towers, circular columns, silos, rotors, pressure vessels, etc.
Though the structure is three-dimensional, it can be mathematically decomposed into a
series of two-dimensional problems. If the loading is axisymmetric, the problem reduces to
a two-dimensional problem analogous to the plane stress/strain case. For the non-
axisymmetric case, loading is represented by the Fourier series in the circumferential
direction, and for each harmonic, n, a two-dimensional problem is solved. Final solution is
determined by adding solutions of each harmonic.
A brief description of the Fourier series for determination of the Fourier coefficients of the
loading is presented followed by the isoparametric finite element formulation for structures
under axisymmetric and non-axisymmetric loadings.
8.2.1 Determination of the Fourier Coefficients
Assume that a pressure distribution, p(θ), along the circumference of the solid can be
expressed in a finite Fourier series as
274
Figure 8.1 Cross-section of a cylinder showing typical finite elements

where (2N + 1) number of coefficients (a0, a1, a2, …., aN; b1, b2, ….,bN) are to be
determined; an and bn are the amplitudes of the corresponding harmonic. Coefficient b0 is
immaterial since the term bn sin nθ gets eliminated for n = 0. Each term of the series is
called a harmonic and n is the harmonic number.
Multiplying both sides of Eq. (8.1) first time by cosmθ and next by sinmθ and integrating
along the circumference for each case gives
Following orthogonality conditions for sine and cosine can be used.
From Eq. (8.2), the amplitude coefficients an and bn can be determined as
275
It may be noted that only a0 exists for the axisymmetric loading. On the other hand, all the
Fourier coefficients exist for non-axisymmetric loading. Coefficients an and bn are
coefficients of the harmonics, which are symmetric and anti-symmetric, respectively, with
respect to θ = 0. Values of an are the amplitudes at θ = 0 and that of bn are the amplitudes
at θ =90°. If the closed form answers of Eq. (8.3) are not available, integration is evaluated
by numerical means. For example, the trapezoidal or Simpson's rule for the numerical
integration may be used.
It now remains to consider determination of number of terms N for the convergence. For
this purpose, coefficients of the Fourier expansion need to be examined. If the coefficients
are small, the chosen number of terms N is considered to be adequate. The smallness of
the last three coefficients in each series (i.e., aN, aN-1, aN-2 and bN, bN-1, bN-2) may be
assessed for the truncation of the Fourier series.
8.2.2 Isoparametric Finite Element Formulations
Axisymmetric structure is formed by rotating curve(s) about z-axis as shown in Figure 8.1.
Finite elements are now the nodal rings whose cross-section can be the usual two-
dimensional elements discussed in the previous chapters. The figure shows typical 3-node
and 4-node elements in the θ = 0 plane in a cylinder. In the cylindrical co-ordinates at each
node of the element, (u, w, v) are displacements in the r (radial), z (axial) and θ
(circumferential) directions for non-symmetric loading, and (u, w) for symmetric loading.
Though the formulations are valid for materials having θ as symmetry axis, only isotropic
materials will be considered here. Isoparametric formulations for these two cases will be
discussed separately.
8.2.2.1 Axisymmetric Loading
For definiteness, a 4-node element as shown in Figure 8.2 is considered as an example,
but the formulation is general. The displacement field for this element is given by
Here, {q} is the vector of nodal displacements (u1 w1 u2 w2 u3 w3 u4 w4). For isoparametric
formulation, displacements and geometry are interpreted using the same shape functions
276
Figure 8.2 Four-node element for axisymmetric loadings
Strain-displacement relations for this problem can be written as
277
Stress-strain relations for isotropic material in cylindrical co-ordinates are
where
The 2×2 element stiffness matrix [kij] can be evaluated in the usual manner as
Here determinant of the Jacobian matrix is obtained as discussed for plane stress/strain
case. For numerical integration, the terms r in matrix [Bi] and integrand of [kij] are
evaluated at the Gauss points to avoid singularities.
8.2.2.2 Non-axisymmetric Loading
Loads applied to geometrically and materially symmetric structures are seldom distributed
in an axisymmetric form. The problem is the three-dimensional one and all three
displacement components u, w and v exist. An arbitrary non-axisymmetric loading may be
represented by its Fourier series expansion as
278
Figure 8.3 Symmetric and anti-symmetric loadings

fr, fz and fz are the load components in r, θ and z-directions. In Eq. (8.9), the single over bar
denotes the amplitudes of the force components symmetric about θ = 0 and the double
over bars denote anti-symmetric parts. Examples of symmetric and anti-symmetric
loadings are shown in Figure 8.3.
Resulting displacements have the corresponding variation
Summations are over the integral values of n from zero to infinity. Introduction of the
negative sign in the terms for v has the effect of yielding the same stiffness matrix for both
symmetric and anti-symmetric components. Amplitudes of displacement components are
primary unknowns, which are functions of r and z only and do not depend on θ. Thus, what
was originally a three-dimensional problem, is reduced to a two-dimensional problem due
to orthogonality of the trigonometric functions. Furthermore, the symmetric and anti-
symmetric components of the displacements are also uncoupled because of orthogonality.
For isotropic material, the Fourier harmonics are uncoupled. This implies that different
numerical values of n present different problems that do not interact. Moreover, the
linearity of the problem enables the super-position of separate solutions. Thus, the need
for a division into finite elements in the θ direction is replaced by the need to superpose
separate solutions.
279
Since the symmetric and anti-symmetric components of the displacements are uncoupled,
only symmetric components are considered here for nth harmonic (n ≠ 0). Again a 4-node
element, shown in Figure 8.4, is considered for definiteness.
Displacement field for this case is written as
Figure 8.4 Four-node element for non-axisymmetric loadings

Amplitudes of displacements are interpolated from the nodal amplitudes
which is
Substitution of Eq. (8.13) into Eq. (8.11) results in
280
which can be written as
The strain–displacement relations for the problem is
Strains can now be expressed in terms of the nodal amplitudes by substituting Eq. (8.14)
into Eq. (8.15) as
281
Stress-strain relationship for an isotropic material is given by
282
Noting that
the 3×3 element stiffness matrix can be evaluated in the usual manner as
where is the determinant of the Jacobi matrix.
r in the matrix and in the integrand of [kij] is evaluated at Gauss points by using
. Integration with respect to ξ and η is done similar to plane strain/stress

problems.
8.2.2.3 Traction on the Edge
Let the edge 2–3 be subjected to an externally applied distributed load{T} as shown in
Figure 8.5. For axisymmetric loading, the surface nodal forces {Q} are evaluated as
283
Figure 8.5 Element with traction on edge 2–3
In Eq. (8.19), dl is along edge 2–3 is given by Eq. (8.5), the over bar on the shape
functions indicates that they are to be evaluated on the edge ξ = 1 and .
Since variables in the integrand of Eq. (8.19) are independent of θ, the integration results
in
Similarly, for non-axisymmetric loading, expression for the surface nodal force {Qn} for nth
harmonic can be written as
where is given by Eq. (8.13) and

8.2.2.4 Boundary Conditions
As can be observed from strain-displacement relations Eq. (8.6) and Eq. (8.15) that term
(1 / r) must be eliminated for r → 0, otherwise the strains become unbounded (Belytschko,
1972). This yields the following boundary conditions, which are applicable to those nodes
lying on the z-axis.
Furthermore, the rigid modes must be suppressed in the global stiffness matrix for
axisymmetric loading. Because of the axial symmetry, translation in the z-direction is the
284
only possible rigid body mode. It can be restrained by prescribing w on a single node lying
on the r-axis.
For the harmonic number n = 1, imposition of the boundary conditions for the points
(nodes) on the z-axis needs special considerations. Designating
from Eq. (8.23) the boundary
conditions for n = 1 can be written as
The first condition is not difficult to impose. It is more appropriate to refer the second one
as a ‘constraint equation’; the imposition of which is not straight forward. This can be
achieved by defining a transformation for the nodal unknowns of the nodes lying on the z-
axis as
The element stiffness is, therefore, formulated with nodal unknowns for the
nodes on the z-axis and for nodes not located on the z-axis. The conditions
are then imposed in addition to other boundary conditions.
8.2.2.5 Illustrative Examples
Illustrative examples on application of formulation presented above have been presented
next.
Example 8.1
A hollow column of length 0.8 m having inner radius 0.2 m and outer radius of 0.4 m is
loaded by a uniformly distributed load of 1.0 GPa as shown in the figure. The column rests
on ground at the base and is free at the top. Analyze this column using 8-node
isoparametric elements as shown. Assume E = 20 GPa, v = 0.20 and restrain node 1 in
the radial direction.
285
Solution:
This example has been analyzed by using program sfeap discussed in Appendix E on the
basis of the formulation presented in this section. The example corresponds to
axisymmetric solid with axisymmetric loading. The output for nodal displacements has
been presented below.
NODE U (m) W (m)
1 .0000E100 .0000E+00
2 –.1180E –02 .0000E+00
3 –.2296E–02 .0000E+00
4 –.2794E–02 .0000E+00
5 –.3271E–02 .0000E+00
6 –.1991E–02 .9670E–02
286
7 –.2581E–02 .9965E–02
8 –.3686E–02 .1013E–01
9 –.1488E–02 .1992E–01
10 –.2361E–02 .1991E–01
11 –.3247E–02 .1986E–01
12 –.3633E–02 .1993E–01
13 –.4022E–02 .2002E–01
14 –.2261E–02 .2992E–01
15 –.2933E–02 .2997E–01
16 –.4051E–02 .2990E–01
17 –.1506E–02 .4015E–01
18 –.2352E–02 .3992E–01
19 –.3179E–02 .3983E–01
20 –.3559E–02 .3995E–01
21 –.3945E–02 .4002E–01
Note that U and W, respectively, represent radial and vertical displacements. As expected,
the maximum displacements occur at the nodes located on the top free surface.
Example 8.2
The circular tank of inner radius 5 m having thickness 0.25 m and height 4 m is shown in
Figure Example 8.2(a). It is partially filled with water up to 2 m as shown. It is assumed
that the wall is fixed at the base and free at the top. Assume E = 20,000 kN/m2, v = 0.15, p
= γw(L1— z), L1 = 2m and γw = 9.81 kN / m3. Analyze the tank by employing equal 8-node
isoparametric elements as shown in the finite element idealization of Figure Example
8.2(b).
287
Example 8.2
Solution:
This example has also been solved by using program sfeap detailed in Appendix E. Both
the geometry and loads are axisymmetric. The nodal displacements obtained from
program sfeapare tabulated below.
NODE U (m) W (m)
1 .0000E+00 .0000E+00
2 .0000E+00 .0000E+00
3 .0000E+00 .0000E+00
4 .4394E–02 .2790E–02
5 .4234E–02 –.2744E–02
6 .1151E–01 .3260E–02
7 .1139E–01 –.1037E–04
288
8 .1137E–01 –.3279E–02
9 .1773E–01 .2411E–02
10 .1752E–01 –.2631E–02
11 .2159E–01 .9413E–03
12 .2156E–01 –.2541E–03
13 .2139E–01 –.1448E–02
14 .2255E–01 –.5590E–03
15 .2234E–01 –.2796E–03
16 .2102E–01 –.1766E–02
17 .2096E–01 –.5792E–03
18 .2084E–01 .6009E–03
19 .1780E–01 –.2506E–02
20 .1766E–01 .1054E–02
21 .1383E–01 –.2800E–02
22 .1378E–01 –.8379E–03
23 .1373E–01 .1117E–02
24 .9918E–02 –.2756E–02
25 .9847E–02 .9027E–03
289
26 .6476E–02 –.2530E–02
27 .6437E–02 –.9809E–03
28 .6429E–02 .5634E–03
29 .3694E–02 –.2250E–02
30 .3667E–02 .2153E–03
31 .1529E–02 –.1986E–02
32 .1512E–02 –.1032E–02
33 .1517E–02 –.8110E–04
34 –.1342E–03 –.1784E–02
35 –.1332E–03 –.2892E–03
36 –.1467E–02 –.1652E–02
37 –.1465E–02 –.1028E–02
38 –.1456E–02 –.4061E–03
39 –.2637E–02 –.1586E–02
40 –.2618E–02 –.4418E–03
41 –.3761E–02 –.1553E–02
42 –.3747E–02 –.9892E–03
43 –.3734E–02 –.4277E–03
290
8.3 EIGHT-NODE ISOPARAMETRIC ELEMENT FOR THREE-DIMENSIONAL STRESS

ANALYSIS
Formulation presented for 4-node isoparametric element presented in the previous chapter
for two-dimensional stress analysis has been extended here to 8-node element for three-
dimensional stress analysis. A typical element from the discretized domain is shown in
Figure 8.6. Intrinsic co-ordinates ξ, η and ζ are also shown in this Figure. Note that
regardless of the size or shape of the element, the intrinsic co-ordinates always vary from
–1 to +1. Thus, a given 8-node element is effectively mapped into a cube in the natural co-
ordinate system. At each i-th node, Ui, Vi and Wi are degrees of freedom, where i = 1,2, …,
8.
Figure 8.6 Eight-node element in the X-Y-Z and the ξ-η-ζ co-ordinate systems
Following relations can be used for co-ordinate transformation
By following the process used in the previous chapter, the co-ordinates can be expressed
as
291
where
In an isoparametric formulation, same shape functions are employed to describe variations
in the global co-ordinates and the unknown field. Thus,
Alternatively,
where
and
292
Strain–Displacement Relation
or,
where
The chain rule of differentiation yields
293
or
In the above equations, are unknowns. They can be

obtained in terms of by inverting the matrix.
The constitutive relation is given by
where the constitutive matrix [D] is given by Eq. (7.42).

Derivation of Element Equations
By minimizing the potential energy, the element equations
can be obtained, where
Computations of [K e], {fb} and {fs}are discussed next.

1. Stiffness matrix
2. From calculus, for an arbitrary function f(x, y, z)
294
Hence,
where
It can be seen from the derivation that each element of [Ke] is of the form
These expressions can be evaluated by employing Gauss quadrature procedure.

3. Body forces
One can use Gauss quadrature rule to compute {fb}.

4. Surface forces
5. Surface forces can be evaluated by extending formulation discussed in the
previous chapters.
A cantilever beam of span 120 cm is subjected to an end-load of 40 kN as shown in the
figure below. Analyze the beam by employing 8-node brick elements. The beam is 10 cm
deep and 2 cm wide. Assume for all members, Young's modulus, E, as 20,000 kN/m2 and
Poisson's ratio v = 0.3.
295
The finite element representation of the problem is also shown in the figure. Node
numbers are indicated by numbers outside the circle, while element numbers are denoted
by encircled Arabic numbers.
Solution:
Nodes 1, 10, 19 and 28 are constrained to represent fix boundary. Generally, three-
dimensional stress analysis using finite element is performed using computer program.
The problem described above has been analyzed by using program sfeap detailed in the
Appendix E. The outputs for displacements at different nodes are presented below.
296
It can be seen from the output that V displacements of all the nodes on the top surface are
equal to those at the analogous nodes on the bottom surface of the beam. Also, the U
displacements at the nodes on the top surface are equal and opposite to those at the
297
analogous nodes on the bottom surface of the beam. Such a trend can be attributed to the
fact that plane X–Z is plane of anti-symmetry. Planes of symmetry and anti-symmetry and
ensuing boundary conditions are further elaborated in the next chapter.
8.4 CLOSING REMARKS
Three-dimensional stress analysis has been presented in this chapter. For axisymmetric
loading, it has been shown that the three-dimensional problem can be effectively reduced
to two-dimensional one by expanding the loading using Fourier series. Two-dimensional
analysis can be performed for each term in the Fourier series by taking one term at a time.
The final solution can then be obtained by superposing results obtained for individual
terms. Isoparametric formulations have been presented for axisymmetric solids.
Formulation for an 8-node isoparametric element has also been presented for analysis of
three-dimensional solids. Formulations have been implemented in computer program
sfeap, which has been discussed later. Few illustrative examples have been considered to
demonstrate applicability of the formulations.
EXERCISE PROBLEMS
1. Analyze the column shown in the Illustrative Example 1 when it is loaded by a triangular
load of zero at inner radius (r = 0.2 m) to 0.4 GPa at the outer radius (r = 0.4 m).
2. Consider an assembly of two heavy circular shafts. The top shaft has inner radius of 0.5
m and outer radius of 0.75 m, while the bottom one has inner radius of 0.5 m and outer
radius of 1.0 m. Shafts are restrained at the top and bottom surfaces to movement in the
vertical direction. A uniform load of 1 GPa is acting as shown in Figure E1. Analyze the
assembly employing 8-node isoparametric elements. Assume E = 20 GPa, v = 0.20 and
restrain node 1 in the radial direction.
Figure E1 Assembly of two shafts

3. Analyze the tank shown in the Illustrative Example 2 when the water tank is completely
filled to the top.
4. The circular tank of inner radius 5 m having thickness 0.25 m and height 4 m is shown
in Figure E2(a). It is assumed that the wall is fixed at the base and free at the top. The tank
is subjected to the wind pressure p as p = p0 (–0.387 + 0.338 cosθ), p0=0.6×v2
where v is the wind velocity in m/sec.
Assume E = 20,000 kN/m2, v = 0.15 and v = 30 m/sec.
298
Employing equal 8-node isoparametric elements as shown in the finite element idealization
of Figure E2(b) analyze the tank for
p = p1 = –0.387 p0
p = p2 = 0.338 p0 cosθ
Present the results for p = p1 + p2.
Figure E2 Water tank subjected to wind pressure p

5. A circular tank of inner radius 4 m has a stepwise variation of thickness whose cross
section is shown in Figure E3. The tank is assumed to be fixed at the base and free at top.
Analyze the tank under hydrostatic loading when it is filled to the top with water. Note that
the triangular load is not shown in the figure. Assume E = 20,000 kN/m2 and v = 0.15 and
γw = 9.81 kN/m3.
299
Figure E3 Stepwise variation of tank wall

6. Compute the nodal force vector for a typical element considered in the illustrative
example of Section 8.3.1 for a constant body force in the x-direction of magnitude 10
kN/m3.
7. Compute the nodal force vector for the element shown in Section 8.3.1 (illustrative
example) for uniformly distributed pressure loading of 10 kN/cm2 on surface formed by
nodes 26, 27, 36 and 35 of element number 8.
300
8. For an 8-node isoparametric element used to model three-dimensional solid cube of

side 5 cm, the nodal displacements are
U1 = 0.0 cm, V1 = 0.0 cm, W1 = 0.0 cm; U2 = 0.0 cm, V2= 0.0 cm, W2 = 0.0 cm
U3 = 0.0 cm, V3 = 0.0 cm, W3= 0.0 cm; U4 = 0.0 cm, V4= 0.0 cm, W4= 0.0 cm U5 = 0.001 cm,
V5 = 0.001 cm, W5 = 0.001 cm; U6 = 0.002 cm, V6 = 0.001 cm, W6 = 0.001 cm U7 =0.001
cm, V7 = 0.002 cm, W7 = 0.001 cm; U8 = 0.001 cm, V8 = 0.001 cm, W8 = 0.002 cm
Evaluate state of strain and the state of stress at the centroid of the element. Consider E
=200 GPa and v = 0.3.
9. Analyze the illustrative example considered in Section 8.3.1 for uniform axial load at the
free end of magnitude 40 kN. Use program sfeap with the discretization, boundary
conditions and material data considered in the illustrative example. Compare results with
one-dimensional axial bar.
Belytschko, T. (1972). Finite elements for axisymmetric solids under arbitrary loading with
nodes on origin, AIAA Journal, 11(9):1357–1358.
Chandrupatla, T. R., and Belegundu, A. D. (2003). Introduction to Finite Elements in
Cook, R. D. (1981). Concepts and Applications of Finite Element Analysis, 2nd ed., John
Wiley and Sons, New York.
Krishnamoorthy, C. S. (1996). Finite Element Analysis Theory and Programming, 2nd ed.,
Tata McGraw Hill, New Delhi.
Wilson, E. L. (1965). Structural analysis of axisymmetric solids, AIAA Journal, 3(12):2269–
2274.
Zienkiewicz, O. C., Taylor, R. L., and Zhu, J. Z. (2007). The Finite Element Method Its
Basis and Fundamentals, 6th ed., Elsevier, Oxford.
301
9 Computer Implementation of FEM

9.1 GENERAL
Computer implementation of the FEM is discussed in this chapter. FEM can be applied
more elegantly by reducing the size of the problem. To this effect, one-dimensional and
two-dimensional analyses were discussed in the earlier chapters, wherein dimension of
three-dimensional problems were reduced by making engineering approximations on the
basis of geometry, loads and boundary conditions. The size of the problem can be further
reduced when certain conditions are met. Reduction in size of the problem has been
discussed in this chapter by (i) considering application of conditions of symmetry and anti-
symmetry and (ii) employing the static condensation procedure. Computer implementation
of the FEM has been discussed with the help of simplified finite element analysis program
(sfeap). Various storage schemes for the global stiffness matrix and application of
boundary conditions in computer code are also discussed.
9.2 USE OF SYMMETRY AND ANTI-SYMMETRY CONDITIONS IN REDUCING A
PROBLEM
In many problems, conditions of symmetry (or anti-symmetry) can be employed to reduce
the size of the problem. Symmetry symbolizes correspondences in size, shape, position of
loads; material properties and boundary conditions that are on opposite sides of a dividing
line or plane, which are often referred to as a line or a plane of symmetry (or anti-
symmetry).
Condition of symmetry: All displacement components normal to the plane of symmetry
are zero.
Condition of anti-symmetry: All displacement components in the plane of anti-symmetry
are zero.
Note that the discussion on conditions of symmetry and anti-symmetry is not restricted to
the structural analysis. These conditions can be applied to any finite element analysis.
Conditions of symmetry and anti-symmetry are applied symbolically in Figure 9.1 to
reduce the size of problem.
Illustrative example
The bar elements 1, 2, 7 and 8 of the plane truss shown below have axial rigidity of
400,000 kN and bar elements 3, 4, 5 and 6 have axial rigidity of × 200,000 kN. The
vertical plane perpendicular to the plane truss passing through nodes 2, 3 and 4 is the
plane of symmetry because identical geometry, material, loading and boundary conditions
occur at the corresponding locations on the opposite side of this plane. In other words, the
geometry, material, loading and boundary conditions on the opposite side of the plane are
‘mirror images’. The loads and the area in the plane of symmetry must be halved in the
reduced structure. Furthermore, all the displacements normal to the plane of symmetry
must be set to zero in the reduced structure. Thus, U2 = U3 = U4 = 0. All the boundary
conditions applied to the reduced plane truss and the resulting displacements and forces
are shown in Figure 9.2.
302
Figure 9.1 Conditions of symmetry and anti-symmetry

It can be seen from the figure that analysis performed on the reduced structure yields the
same displacements and element forces as those prevalent in the actual structure, with
one exception compared to the actual plane truss–forces in elements 3 and 4 in the
reduced truss are half in magnitude. This difference is expected because the cross-
sectional areas of these elements were halved in the analysis.
It is possible to compute displacements and member forces in the real structure by
analyzing only a quarter portion of the actual truss, as shown in Figure 9.3. However, two
separate analyses are required [Cases (A) and (B)].
303
Figure 9.2 Analysis of a plane truss by employing conditions of symmetry

Results from these two analyses are tabulated in Table 9.1. Final results, which are
presented in Figure 9.2, can then be obtained by superimposing results from Cases (A)
and (B), by appropriately employing the conditions of symmetry and anti-symmetry.
9.3 STATIC CONDENSATION

An application of FEM to a problem results in simultaneous equation set
If size of K is too large to fit into the CPU memory of the computer at hand, it is necessary
to reduce the size of the matrix to solve the problem. Static condensation is a process of
eliminating or condensing some unknown DOF (known as secondary or slave DOF) and
re-cast the original Eq. set (9.1) in terms of fewer unknowns (termed as primary unknowns
or master DOF).
304
Figure 9.3 Further reduction in the size of the problem

Table 9.1 (a) Nodal displacements from the analysis of the quarter structure
Table 9.1 (b) Element forces stemming from the analysis of the quarter structure
9.3.1 Applications of Static Condensation

The concept of static condensation is utilized in the following.
305
Forming a ‘super element’. Super elements are used quite frequently in the analysis of
complicated structures or geometries, where several smaller elements are condensed to
reduce the number of unknowns without sacrificing the accuracy.
Sub-structure analysis. For example, soil structure interaction problems.
9.3.2 Static Condensation Procedure
To eliminate the secondary DOF, matrix Eq. (9.1) is partitioned as
where K11(s×s), K12(s×p), K 12(p×s) and K22(p×p) are sub-matrices of matrix K (n×n). The
T and
, on the other hand, are the sub-vectors of the partitioned vectors and such that
contains secondary DOF and contains primary DOF. Note that n = s + p.
The aim of performing static condensation is to re-cast Eqs. (9.1) into reduced equation set
as
By expanding the partitioned Eqs. (9.2), the following equations are obtained.
Hence, from Eq. (9.4),
By substituting Eq. (9.6) in Eq. (9.5), the following equations can be derived.
=>
=>
where
Given
306
Derive
In order to use Eq. (9.5), Eq. (9.10), the 3rd and 4th rows must be inter-changed with the
1st and 2nd rows as well as the 3rd and 4th columns with the 1st and 2nd columns so that
the symmetric matrix in Eq. (9.11) can be partitioned into the form given by Eq. (9.2).
Basically, the original matrix has to be re-arranged such that the ‘secondary’ variables
(i.e., the variables to be eliminated) x3 and x4 appear in the first sub-vector, , and the
‘primary’ variables (i.e., the variables to be retained) x1 and x2 appear in the second sub-
vector, .
After re-arranging the equations, the following equation is obtained.
so that
From Eq. (9.10), it can be noticed that the following matrix operations are required.
307
Finally,
and
9.4 COMPUTER IMPLEMENTATION OF FEM–SFEAP

In order to illustrate computer implementation of FEM, simplified Finite Element Analysis
Program (sfeap) has been developed in FORTRAN language. Following different elements
have been incorporated in the program.
1. 2-node axial bar/plane truss element
2. 2-node space truss element
3. 2-node beam/plane frame element
4. CST element for two-dimensional stress analysis
5. 3- to 9-node element for two-dimensional stress analysis
6. Axisymmetric isoparametric element for axisymmetric load (Circumferential
harmonic = 0)
7. Axisymmetric isoparametric element for asymmetrical load (Circumferential
harmonic > 0)
8. 8-node isoparametric element for three-dimensional stress analysis
In addition, few dummy routines are kept, which can be used to add user defined
elements. The program has been developed such that steps followed in the method are
easily understood while going through the organization of the program. Details of the
program including overall flowchart, description of various sub-routines, major arrays and
variables are provided more conveniently in the Appendices E and F. The program can be
used to learn the implementation of method on computer and also for solving a variety of
problems. In order to facilitate data preparation, a graphical user interface has been
developed. The entire software has been given in the accompanying CD. Source code of
the program has also been given in the CD for teaching and research purpose.
308
Storage scheme for the assembled matrix and application of the boundary conditions in
program sfeap are discussed next.
9.5 STORAGE SCHEMES FOR GLOBAL STRUCTURAL STIFFNESS MATRIX

There are several ways in which the global stiffness matrix can be stored in the computer
memory.
1. Full form: The matrix is stored in its full (natural) form. This requires large storage
and running time.
2. Upper triangular form: Since the global stiffness matrix [K] is symmetric, it is
necessary to store only the elements in the upper triangle of the [K] matrix.
If N is the number of equations, this scheme requires ½ N × (N + 1) storage locations.

Note that the element Kij will be found at the [1/2{j(j −1)} + i]-th location.
In other words, K(I, J) ⇒ S (L), where L = 0.5 × J × (J - 1) + I
3. Banded form: Bandedness of the stiffness matrix has little to do with the
formulation of finite elements. It results from an efficient nodal numbering scheme.
Refer to Figure 9.4.
As a thumb rule, number the joints (nodes) in a grid in the narrow direction to minimize the
bandwidth.
MBAND = (maximum node difference + 1) × NDF
Elements of the stiffness matrix that correspond to the second numbering scheme can be
stored in the upper triangular, banded form as shown in Eq. (9.27).
309
Program sfeap stores the global stiffness matrix in the banded form.
4. Skyline form: This also results from an efficient nodal numbering. Consider the
stiffness matrix shown in Eq. (9.28).
310
Figure 9.4 Banded assembled equations
311
Stiffness matrix shown in Eq. (9.28) can be written in a one-dimensional array as shown in
Eq. (9.29).
9.6 APPLICATION OF BOUNDARY CONDITIONS

There are two ways to handle boundary conditions associated with a constrained node.
1. Specify a very large number on the diagonal associated with the corresponding
displacement component.
2. Do not assemble the equation corresponding to the specified displacement
component.
The second approach has been discussed next as the same has been employed in
program sfeap.
Consider example of two span continuous beam shown in Figure 9.5.
The total DOF for this example are
Specifying the unknowns in this manner is not convenient for computer implementation.
Thus, let the DOF be referred to as [r1 r2 r3 r4 r5 r6]T.
Then, the boundary conditions are
For the computer implementation, these DOF's are given a zero ID number as follows.
DOF ID#
r1 0
r2 0
312
r3 0
r4 1
r5 0
r6 2
Figure 9.5 Application of boundary conditions

Note that r4 and r6, respectively, correspond to unconstrained DOF θ2 and θ3. Thus, it is
seen that only four coefficients of the system level matrix are required to be assembled by
assigning location numbers corresponding to unconstrained DOF to reduce number of
storage locations in the computer's random access memory (RAM). For illustration,
consider the element stiffness matrix for element 2. Coefficients of this matrix are required
to be assembled at locations 3, 4, 5 and 6 as shown below. Further, since r3 = r5 = 0, all
elements of rows marked 3 and 5, as well as columns marked 3 and 5 are not required to
be assembled and, thus, the same have been renumbered as 0. The original rows marked
4 and 6 as well as columns marked 4 and 6 are re-numbered as 1 and 2, respectively.
313
Figure 9.6 Application of boundary conditions (a) two-dimensional domain; (b) DOF prior
to application of boundary conditions and (c) unknown DOFs
Consider the example of two-dimensional analysis using four CST elements shown in
Figure 9.6 (a).
For CST element, there are two DOF per node. These DOF have been numbered
sequentially as shown in Figure 9.6 (b). Note that U1 = V1 = U2 = V4 = 0. Thus, equations
pertaining to these DOF are not required to be assembled. The resulting re-numbered
DOF, which are required to be computed, are shown in Figure 9.6 (c).
In program sfeap, the above mentioned steps are implemented as follows.
1. Initialize all elements of array ID with dimensions NDF × Number of nodes
(NUMNP) to zero.
2. Read boundary conditions. Note that constrained DOF are assigned value unity in
the input whereas free (unconstrained) DOF are assigned value zero.
314
3. Generate equation numbers. Start the counter from 1 and increment by 1 for each
unconstrained DOF sequentially.
Finally, assembly takes place at locations 1 through 6, and six simultaneous equations in
banded form are solved by using Gauss elimination method discussed in Appendix A.
Further details of computer implementation of FEM through program sfeap are given in
Appendices E and F.
9.7 CLOSING REMARKS
Details on application of conditions of symmetry and anti-symmetry have been given in this
chapter. It has been emphasized that with the help of such boundary conditions, the size
of the problem can be reduced significantly. Static condensation process has also been
elaborated in the chapter, which can be used to reduce size of the problem and also to
form a super-element. The salient features of program sfeap are presented through which
computer implementation of FEM can be studied. Various storage schemes used for
assembled matrix are discussed and the process of application of boundary conditions
and assembly are explained. Further details of program sfeap are given in Appendices E
and F.
EXERCISE PROBLEMS
1. It has been decided to analyze the plane truss shown in Figure E1(a) by using
symmetric and anti-symmetric components of loads. By applying appropriate boundary
conditions at nodes 4 and 5, the reduced half portion of the truss has been analyzed for
four cases. Cases 1 and 2 correspond to unit horizontal load applied at node 2, as shown
in Figure E1(b). The displacements at all joints and member forces are summarized in
Tables E1.1 and E1.2, respectively, for Cases 1 and 2. On the other hand, Cases 3 and 4
correspond to unit vertical load applied at node 3, as shown in Figure E1(c). The
displacements at all joints and member forces are summarized in Tables E1.3 and E1.4,
respectively, for Cases 3 and 4. Identify each case and using the data of these four cases,
compute, for the truss shown in Figure E1(a),
1. displacements at all nodes; and
2. force in all elements.
Consider the axial rigidity AE of all the elements to be 10,000 kN in the actual truss.
315
Figure E1(a)
Figure E1(b)
Figure E1(c)
316
317
2. Matrix Eq. [K]4×4 {q}4×1 = {f}4×1 has been defined where
Condense q2 and q3, and modify the equation to [K*]2×2 {q*}2×1 = {f*}2×1 where {q*} = [q1 q4]T.

Bathe, K. J. (1982). Finite element procedures in engineering analysis, Prentice Hall India,
New Delhi.
Crandall, S. H., Dahl, N. C., and Lardner, T. J. (1978). An Introduction to the mechanics of
solid, McGraw-Hill, New York.
Desai, C. S., and Abel, J. F. (1972). Introduction to the finite element method, Van
Nostrand Reinhold, New York.
Gere, J. M. (1965). Analysis of framed structures, Princeton, New Jersey.
Logan, D. L. (1993). A first course in finite element analysis, PWS Publishing, Boston.
318
10 Further Applications of Finite Element Method

10.1 INTRODUCTION
Typical finite element formulations for various engineering applications were discussed in
the earlier chapters. Details presented in the preceding chapters are sufficient to provide
essential ideas and procedures for the development of finite element formulations and
computer models for most engineering problems in one, two and three dimensions.
However, each complex engineering problem differs in one way or another and there are
many additional ideas and difficulties that need to be addressed, which are related to
formulation of the problem using finite element method (FEM), development of computer
models and applications to field problems.
A few more advanced important engineering problems and their related FEM formulations
and applications are discussed in this chapter. Discussions are focused on demonstration
of applications of FEM to complex engineering problems. There are a number of books on
the topics covered in this chapter. Detailed description of these topics is beyond the scope
of the current book. However, it is envisaged that the readers will be able to get general
‘feel’ of the topics and can use the details presented herein as a step in the direction of
further development of FEM formulation and modelling of the engineering problems
concerned with the area of their interest.
10.2 FINITE ELEMENT ANALYSIS OF PLATES
10.2.1 Introduction
A plate is a three-dimensional structure bounded by two parallel planes separated by a
distance, h, denoting thickness of the plate. The x–y plane at z = 0, which is equidistant
from two bounding planes, is called the middle surface of the plate. In the linear elastic
plate theory for isotropic material, the problem uncouples into in-plane and out-of-plane
problems, depending on whether the loading is parallel to the x–y plane or normal to it.
Finite element formulation for the parallel loading has already been presented in Chapter 7
for plane stress or plane strain situation. In this section, only the bending of plates is
presented when the loading is normal to the x–y plane. An interested reader can refer to
the book on plates by Timoshenko and Woinowsky-Krieger (1959) and Reddy (1999) for
detailed studies.
If the ratio h/L where L is the least dimension on the plan in x–y plane, is less than1/10, it
is normally classified as a thin plate. On the other hand, the plate is classified as a thick
plate if the ratio exceeds 1/10. Procedure for deriving the pertinent finite element
formulation for plate theory will follow the pattern presented in Chapter 6 for the beam
theory. First, the plate theories will be presented. This is followed by the finite element
formulation for thick and thin plates.
319
Figure 10.1 Forces on a plate element: (a) Element detail and (b) Forces on the middle
plane of the plate
10.2.2 Review of Plate Theories
Consider an infinitesimal plate element in the x–y plane as shown in Figure 10.1 (a) and
(b). The x–y plane is located at the middle surface of the plate and the thickness of the
plate is assumed to be h (– h/2 ≤ z ≤ h/2). Load q(x, y)(N/m2) acts normal to the plate.
10.2.2.1 Equilibrium Equations
The stress resultants, moments (Mx,Mxy,My) and the shear forces (Qx,Qy) shown in Figure
10.1 (a) and (b) are defined as
Moments and shear forces have dimensions of Nm/m and N/m, respectively. Mx, My are
the bending moments and Mxy is the twisting moment.
Moments and shear forces on the positive faces are expressed in the Taylor series,
retaining only the first two terms as
As presented in the beam theory, the static equilibrium equations are written by
considering force equilibrium in the z direction and moment equilibrium equations about
the x and y directions, respectively, and neglecting higher order terms as
320
10.2.2.2 Thin-plate (Classical plate) Theory

Geometry: Following assumptions are made for deformation in the classical plate theory
(CPT).
The middle surface of the plate is unstrained.
The normal to the middle surface remains normal after the deformation. This assumption
Kirchhoff's hypothesis, is analogous to the one made in the thin-beam theory, where the
plane normal to the neutral axis remains plane and normal after deformation.
Normal displacement, w, is independent of z, which implies w = w(x, y).
Deflection (the normal component of the displacement vector) of the mid-plane is small
compared with the thickness of the plate. The slope of the deflected surface is also very
small so that the square of the slope can be neglected in comparison with unity.
The strain-displacement relations for linear elasticity in three dimensions are given by
where u, v and w are the displacements in x, y, and z directions, respectively.

However, the displacements u and v can be expressed from assumption (ii) as
Thus, the strain components can be written as
Stress–strain relations and manipulations: Stress–strain relations for the plate theory
are assumed to be those for the plane strain case (see Chapter 7), since εz = 0. They are
321
From Eq. (10.6), these relations can be written as
Since plate is assumed to be thin, the normal stress σz is assumed to be negligible.

Substituting Eq. (10.8) into Eq. (10.1), the stress resultants can be evaluated after simple
integration as
where D is the plate rigidity given by

Plate rigidity D plays similar roll to the beam rigidity EI. For a beam of unit thickness and
depth h, the moment of inertia I = h3/12. It can be seen that D = EI if v = 0.
Shear forces can be evaluated from the equilibrium Eqs. (10.3) to be
where the Laplacian
4
where the bi-harmonic operator, , is given by
322
It may be mentioned here that though γxz = 0 and γyz = 0, it is assumed that τxz and τyz
exist. Shear stresses cannot be evaluated from the theory of elasticity. In practice, guided
by elementary beam theory, they are approximated as
Though there is an inconsistency in the theory, it produces acceptable results for thin
plates.
Because equation for the deflection w is a fourth order partial differential equation, two
boundary conditions on the edge are required. There are three moments and two shear
forces. Therefore, the situation is different compared to the beam theory. Kirchhoff derived
the consistent boundary conditions (Timoshenko and Woinowsky-Krieger, 1959) using
from the calculus of variation given below for a rectangular plate of dimension a × b.
10.2.2.3 Thick-plate (Shear deformation) Theory

Geometry: Shear deformation is neglected in the thin-plate theory resulting in a governing
differential equation in one dependent variable w(x, y). This is justifiable for thin plates.
However, the shear stresses exist for thick plates. Thus, there will be deformations on the
x and y faces. When there is variation of shear forces along the x and y directions, there
will be a stretch in the x and y directions, resulting in the longitudinal and shear strains and
stresses. Thus, it is desirable to include the effect of shear deformation when the plate is
thick. This will require increase in number of dependent variables to three to account for
the shear deformation. Following assumptions are made for deformation in the first-order
shear deformation plate theory (SDT).
The middle surface of the plate is unstrained.
323
The normal to the middle surface before deformation remains straight but is not
necessarily normal to the middle surface after the deformation.
Normal displacement, w, is independent of z, which implies w = w(x, y).
Deflection (the normal component of the displacement vector) of the mid-plane is small
compared with the thickness of the plate. The slope of the deflected surface is also very
small such that the square of the slope can be neglected in comparison with unity.
Here, the procedure followed by Mindlin (1951) is described, which is simple and
mathematically elegant. The assumed displacement distributions are taken as (see Figure
10.2)
where w is the transverse displacement, βx is the rotation of the normal to the undeformed
middle surface in the x–z plane and βy is the rotation of the normal to the undeformed
middle surface in the y–z plane. Note that in the Kirchhoff's hypotheses βx = −∂w / ∂x and
βy = ∂w; / ∂y.
Strain components can be obtained by substituting the assumed displacements Eq. (10.
12) in Eq. (10.4) as
Stress–Strain Relations and Manipulations: Stress–strain relations given in Eq. set

(10.7) can be written as
324
Figure 10.2 Rotations of the normal about x and y axes
s
Shear stresses τxz and τyz in Eq. set (10.14) represent average stresses over the sections.
Stress resultants can now be written from Eq. (10.1) as
In order to account for the non-uniform shear stress distribution over the cross-section, the
shear correction factor K is introduced. The correction factor K is normally assumed to be
5/6 (statics) or π/12 (dynamics).
325
Substitute Eq. (10.15) into Eq. (10.3) to obtain the displacement equations of motions.
From the first of the Eq. (10.3),
which simplifies to
From the second of Eq. (10.3),
After the manipulation of this equation, the second of Eq. (10.3) can be written as
From the third of Eq. (10.3),
which after some manipulation can be written as
Thus, the three coupled equations for variables w,βx, and βy are
326
10.2.3 Finite Element Formulations

Finite element formulation is straightforward for a thick plate and is presented first followed
by formulation for thin plate.
10.2.3.1 Thick Plate
Only energy due to bending is stored for a thin plate, since shear deformation is neglected.
However, energy due to shear deformation has also to be included for a thick plate. The
normal and shear stresses and strains are uncoupled for an isotropic material. Therefore,
energy due to each of them can be added separately. Bending strains εxx,εyy, γxy and the
transverse shear strains γyz,γxz are
The constitutive relations are given by
The potential energy π, on the other hand, is given by
which can be expanded to
327
where K is the shear correction factor, p(x, y) is the distributed load and Fi are the
concentrated forces. After integrating over the thickness h, Eq. (10.23) yields
where
{εb} and {εs} contain only the first-order derivatives of the dependent variables βx,βy and w.
Procedure presented earlier for the plane stress/strain problems can be followed to
formulate the stiffness matrix of the thick plate element.
Assuming the displacement and rotation distribution as
where wi,θyi,θxi are the nodal values, Ni (ξ,η) are the shape functions associated with the i-
th node, and Ne is the number of nodes in that element. Eq. (10.26) can be written as
328
where
Therefore,
where
where
and
Minimizing π of Eq. (10.24), = 0, and noting and are

symmetric matrices yields
or
where
Once the formulation for [K] and {R} are completed, analysis of the plate structure follows
the standard procedure.
329
It may be mentioned here that this formulation is restricted to thick plates, since w, βx, and
βyare treated as independent quantities. The differential operators [L]b and [L]s contain only
the first-order derivatives of the shape functions, which requires only C0 continuity. From
Eqs. (10.29) and (10.24), bending stiffness matrix and shear
stiffness matrix where
and
From these equations, it is inferred that the relative magnitude of the shear stiffness
becomes of order 1/(h/L)2 times that of bending stiffness, as plate thickness is reduced.
For example, if h/L = 0.01, 1/(h/L)2 = 104. Here, L is a characteristic length. This means
that if thick plate formulations are used to analyze thin plates, the solution will be governed
according to the shear stiffness. However, shear (transverse) strains are zero and shear
stiffness is neglected in thin-plate theory. The dominance of the shear stiffness in C0
formulation in thin-plate range is known as ‘shear locking’; that is, the element becomes
stiffer and deforms only in the shear mode and flexural deformations are negligible. To
remedy the problem, some techniques are proposed (Zienkiewicz and Taylor, 1989): (a)
impose as constraints and develop a constrained variational
formulation (penalty function approach; discreet Kirchhoff's hypothesis); (b) use reduced
integration technique making shear stiffness singular through a selective integration
scheme or (c) formulate element stiffness using thin-plate theory as discussed in the
following section.
10.2.3.2 Thin Plate
From Equations (10.6) and (10.8), one can write strain–displacement and stress–strain as
330
The potential energy π is given by
which can be expanded to
Substituting Eq. (10.31) in Eq. (10.32) and integrating over the thickness of the plate, the
first term results in
Assume the displacement distribution for w, the only degree of freedom (DOF) at any
node, as
Here, wi are the nodal displacements, Ni are the shape function associated with node i and
Ne is the number of nodes in that element.
Substituting Eq. (10.34) in the curvature vector {K} yields
331
Thus, the potential energy can be written as
Minimizing the potential energy of Eq. (10.36), following equations are obtained.
Evaluation of stiffness matrices [K]b and [K]s in Eq. (10.29) and [K]p in Eq. (10.37) requires
integration. It can be seen from Eqs. (10.28) and (10.35) that the integrands of [K]b and
[K]s contain first-order derivatives of shape functions while those of [K]p contains second-
order derivatives. As pointed out in Chapter 5, the shape functions need to be continuous
only at the interface (C0 continuity) for evaluation of [K]b or [K]s. However, evaluation of [K]p
also requires the continuity of the derivative (C1 continuity) at the interface. For one-
dimensional problem, the beam functions (Hermit polynomials), which satisfy C1 continuity
at the nodes have been encountered in Chapter 6. However, it is difficult in two dimension
to construct shape functions, which not only satisfy C1 continuity but are also
computationally efficient.
FEM is a numerical method where the domain is an assemblage of elements. Procedure
to form the global matrix for the domain is to evaluate the potential energy π(e) of each
element and then summing total energy π of the system as
and then minimizing π. Summation of Eq. (10.38) presumes that the displacements and
rotations are continuous at the inter-element boundaries. If the continuity is not
maintained, the gaps (overlays or separations) will form and the solution may not
converge. This aspect is discussed in the next section.
332
Desirable Requirements of the Shape Functions: Continuum has infinite DOF and it
possesses minimum potential energy. Discretizing the domain by mesh of finite elements
limits the DOF and renders the system more stiff, resulting in higher potential energy π.
The displacement-based finite elements formulation yields an upper bound for π. The
energy approaches the true value of π from above, as shown in Figure 10.3. The true
minimum energy may never be reached even by refining the mesh, unless the shape
functions are solution of the governing equations. The shape functions should satisfy
some requirements to ensure that the numerical solution converges towards the true
minimum, as the mesh is refined.
Requirement 1: Interpolation function of an element must be able to represent the rigid
body displacement. Rigid body displacements are those displacement modes that an
element is able to undergo as a rigid body without stresses being developed.
Constant terms in the shape functions will ensure this requirement.
Requirement 2: Displacement functions must be capable of representing the constant
generalized strain states within elements as elements get smaller. Generalized strains
mean strains and/or curvatures.
An element is said to be complete if displacement functions satisfy these two
requirements.
Linear terms in the polynomial satisfy this requirement for strains while second-order terms
satisfy the requirement for curvatures. Reasoning for this requirement can easily be
explained.
Figure 10.3 Convergence of potential energy
Figure 10.4 Evaluation of potential energy
333
Area under a curve y(x) is evaluated by dividing the curve into small intervals ∆xe, as
shown in Figure 10.4 and integrating as
Now the potential energy π is given by
which, for one-dimensional problem of axial force can be written as
Similar reasoning can be made for higher dimensions. The state of strain in each element
approaches constant value, when elements of the assemblage approach small size. The
energy can then be found by the summation, as shown in one dimension.
Requirement 3: Elements must be compatible.
This requirement means that displacement within elements and across element
boundaries must be continuous. When only the translational DOF are defined at the
element nodes, only continuity of displacements must be preserved. However, when the
rotational DOF are also defined in addition to the displacement DOF, as in the case of
plate bending element, it is also necessary to satisfy the zero and first-order derivatives of
the displacements. If this condition is not satisfied, displacements across thickness of plate
will not be continuous at edges. Compatibility will also imply that elements have the same
number of nodes, co-ordinates and interpolation functions at the inter-element boundaries.
Elements that satisfy the above three requirements are called compatible or conforming
elements. If these requirements are violated, the element is referred to as incompatible or
non-confirming elements.
Physically, compatibility ensures that no gaps or overlaps occur between elements.
Requirement 4: An element should have no preferred direction. In other words, an
element should be geometrically invariant (geometrically isotropic or spatially isotropic).
334
Figure 10.5 The pascal triangle

Geometric invariance is achieved if polynomial in shape function is complete, that is, it
includes all the terms as shown in Pascal triangle of Figure 10.5. However, invariance can
still be achieved if the corresponding orders of terms are selected on either side of the axis
of symmetry. This requirement ensures that the polynomial expansion for element remains
unchanged under a linear transformation from one Cartesian co-ordinates to another
(Dunne, 1968).
For bending of thin plates, difficulties of finding compatible displacement functions have
led to use of non-confirming elements. Usually, a complete slope continuity is ignored for
such elements, which still satisfy other requirements. It is essential that the non-
conforming elements pass the patch test. It is a simple numerical test of element validity. A
numerical test (Irons and Razzaque, 1977) is performed by imposing nodal displacements
on an arbitrary patch of element, corresponding to any state of constant strain. If nodal
equilibrium is simultaneously achieved without any external nodal forces and if constant
stress field is obtained, the convergence is assured as mesh is refined. For further details
on the patch test, refer (Zienkiewicz and Taylor, 1989).
In the following section, two plate bending elements for thin plates, one non-conforming
and the other for conforming elements, will be presented.
Shape Functions for Plate Bending Elements: Equations necessary to derive plate
bending elements are derived in the preceding sections. The pertinent equations to form
element stiffness matrices are Numerous finite elements for plate bending have been
developed over the years (Hrabok and Hrudey, 1984). In this section only two shape
functions, which can be used to evaluate the stiffness matrix using Eq. (10.40), are
presented.
335
Figure 10.6 Plate element (a) Rectangular plate geometry (b) Nodal DOF
Non-conformal Shape Function: One of the popular rectangular plate bending element
(Melosh, 1963) with 12-DOF is considered here. Figure 10.6 shows a rectangular plate
bending element with nodal displacements wi and the rotations θxi and θyi (i = 1,2,3,4)
where θx = ∂w /∂y and θy = –∂w / ∂x. The origin has been taken at (0, 0). Since the element
has 12-DOF, a 12-term polynomial is selected as
The polynomial is complete up to the third-order terms (see Figure 10.5) and is also
geometrically isotropic. Differentiating Eq. (10.41) with respect to x and y, vector {u} can
be written as
336
From Eq. (10.42), curvatures and twist can be determined as
From Eqs. (10.41) and (10.43), it is observed that the interpolation function satisfies
requirement 1 for rigid body displacement and requirement 2 for constant curvature as the
element becomes smaller. Thus, this element is complete. Consider the compatibility of
this element. Along edge 1–2 (y = 0)
In Eq. (10.44), it is noted that w(x, y) is cubic in x and the slope ∂w / ∂x is the same as in
beam element.
Constants a1, a2, a4, and a7 can be evaluated from the nodal displacements and rotations
w1,θy1, w2 and θy2 as was done in the derivation of beam functions (Hermite polynomial).
The resulting beam functions are given in Chapter 6, in Eq. (6.172) and are also listed later
on in Eq. (10.48).
Thus w and ∂w / ∂x are continuous along the edge 1 – 2. The slope ∂w/∂y is cubic in x
involving four constants a3, a5, a8 and a11. However, only two DOF (θx1 and θx2) remain to
determine these constants.
The slope ∂w/∂y is not uniquely defined and the slope discontinuity occurs. Therefore, this
element is non-conforming element and the solution may not render minimum potential
energy.
Evaluating Eq. (10.42) at each of the four nodes of the element, the following equation is
obtained.
{ qe } = [G]{a}
where
337
From Eq. (10.45),
is obtained.
Substitution of Eq. (10.46) in Eq. (10.41) yields
w(x, y) = [P]{a} = [P][G]-1 {qe} = [N]{qe}, thus the shape function matrix is given by
The stiffness matrix is then obtained from Eq. (10.40).

Conforming Shape Function: It is possible to construct polynomials for a confirming 16-
DOF rectangular element by adding twist term θxy = ∂2w / ∂x∂y at each node of the four
node element, considered in non-conforming element described above. However, some
polynomials may not produce an invertible matrix [G]-1 as done in Eq. (10.46).
An alternate way of forming the shape functions ensuring continuity of w and its normal
derivatives ∂w / ∂x and ∂w / ∂y along element boundary is to employ product of Hermite
polynomials (Bogner et al., 1968). The Hermite polynomials given in Eq. (6.177) in
Chapter 6are listed here for completeness.
where Ls is the length of the element in s direction.

The appropriate cubic interpolation polynomial in the expansion of w is
338
Eq. (10.49) can be written in matrix form as
The stiffness matrix is then obtained from Eq. (10.40).
10.3 DYNAMICS WITH FINITE ELEMENT METHOD

10.3.1 Introduction
In the earlier chapters, it was assumed that the loads on the structure are static and do not
vary with time in the finite element equations [K]{q} = {F}. These equations are accurate
enough if the load {F} has a frequency less than one-quarter of the lowest natural
frequency of the structure. However, loads due to moving vehicles, machinery,
earthquake, blasts and so on have fairly high frequency content. For a structure subjected
to any of these loads, it is essential to perform a dynamic analysis and include inertial
effects. There is little difficulty in extending the static equilibrium equations to the dynamic
case by including mass and damping matrices in the analysis. In the following sections,
governing equations and solution techniques are presented.
10.3.2 Governing Equations

The direct approach to formulate the finite element equations using the linear spring
system described in Chapter 6 is presented first. Then, the energy approach for the
general case is discussed.
10.3.2.1 Linear Spring Model

Consider a two-DOF oscillator model of Figure 10.7 (a) and its free-body diagram shown in
Figure 10.7 (b).
Equations of motions for the system are merely expressions of the equilibrium of the
individual masses, as follows (Clough and Penzien, 1975).
339
Thus, Eq. (10.51) can be written as
The above equation can be expressed in matrix form as
where {q}T = [u1 u2], [M] is the diagonal mass matrix, [C] is the damping matrix, and [K] is
the stiffness matrix.
Using the principle of virtual work, general finite element equations are formed.
Figure 10.7 Two-DOF oscillator: (a) Model; (b) Free-body diagram
10.3.2.2 Finite Element Formulation

In Chapter 7, for an element the displacement vector {U} and the force vector {Fe} were
given in Eqs. (7.57) and (7.68) as
where {U} is the displacement, {N} represents interpolation function matrix, {X} is the body
force vector, {T} represents the surface traction vector, {P} contains external nodal forces,
and {qe} contains the nodal displacement values.
340
The body force vector has to be modified to include the inertia and damping forces.
The element force vector { e
} is now written as
In Eq. (10.55), m is the mass per unit volume and c is the damping per unit volume.
Thus, [Ke]{qe} = {Fe} of equation (7.69) modifies to
Rewriting this equation, the dynamic equilibrium equation takes the form
[M e] is the element consistent mass matrix, [C e] is the element consistent damping matrix
and [K e] is the consistent stiffness matrix. [K e], [B] and [D], were defined in earlier
chapters.
Assembling the contributions from all elements of the discretized domain, the global
dynamic equilibrium equation can be written as
In the sequel, it will be assumed that the matrices [M], [C], and [K] are symmetric of size n
× n, and the vectors {q} and {F(t)} are of size n × 1
In practice, determination of the damping matrix [C] is difficult and it is seldom known. Only
the following case will be considered.
where α and β are determined experimentally (Clough and Penzein, 1975).

Equation (10.56) can be written as
Equation (10.58) is of the same form as equation (10.56). In the sequel, the hat is
suppressed and consider the form as
Equation (10.59) can be numerically solved by using either mode superposition method or
direct time integration method. The mode superposition method is presented first followed
341
by a time integration method [Bathe (2001), Huebner et al. (1995), Clough and Penzien,
(1975)].
10.3.3 Mode Superposition Method
The method is applied in two steps. In the first step, the eigenvalues (frequencies) and the
eigenvectors (mode shapes) of the homogeneous Eq. (10.59) ({F(t)} = 0) are found. These
modes are then superposed to solve for the forced vibration problems.
10.3.3.1 Free Vibrations

For the case when {F(t)} = 0, Eq. (10.59) is the homogeneous differential equation given
by
To find the natural frequencies (eigenvalues) and the natural or principal modes
(eigenvectors) of Eq. (10.60), the nodal vector {q(t)} is expressed as
where is the unknown vector of amplitudes and ω (rad/sec) is the natural

frequency. Substituting Eq. (10.61) in Eq. (10.60) results in an eigenvalue problem
Eq. (10.62) has a non-trivial solution only when the determinant of Eq. (10.62) is zero, that
is,
Since, [K] and [M] are assumed to be of order n × n, Eq. (10.63) yields n eigenvalues ω2r (r
= 1,2,…, n). There will be a corresponding mode shape {(φ}r (r = 1,2,…, n) associated with
each frequency ωr. Frequencies ωr will be positive if [K] and [M] are positive definite, which
is generally the case. It is seldom practical to write the polynomial expansion for the
determinant of Eq. (10.63) for large matrices. A number of standard computer programs
are available in open literature as library subroutines for the purpose. In many cases, only
a few modes {(φ}r (r = 1,2,…, p, p ≤ n) are required. There are several iterative solution
procedures to solve for these modes (Clough and Penzien 1975; Bathe 2001).
An important property of the mode shapes, which is very useful for solving forced vibration
problems, is their orthogonality with respect to the weight [M]. The orthogonality condition
can be stated as
where {φ}r and {φ}s are mode shapes corresponding to frequencies ωr and ωs,
respectively.
The orthogonality relation for the stiffness matrix [K] can be established from Eq. (10.62)
as
342
Mr and Kr are called generalized mass and stiffness, respectively and ωr is the natural
frequency of the r-th mode.
Sometimes the mode shapes are normalized as which makes the
generalized mass and the generalized stiffness

In the next section, only the mode shapes {φ}r will be employed to solve transient
problems.
10.3.3.2 Transient Response

The orthogonality properties [Eqs. (10.64) and (10.65)] of the mode shapes now may be
used to uncouple the non-homogeneous equation of motion, Eq. (10.59). The solution
vector {q(t)} is expressed by superposing the modes as
and {Y(t)} is the vector of unknown modal amplitudes. Note that [Ф] transforms the
generalized coordinates {Y(t)} to the physical coordinates {Q(t)}.
By substituting Eq. (10.66) in Eq. (10.59) and then pre-multiplying the resulting Eq. by [Ф]T,
the following equation set is obtained.
Using the orthogonality Eqs. (10.64) and (10.65), Eq. (10.67) reduces to a set of
uncoupled ordinary differential equations shown below.
where the generalized force Fr(t) = {(φ}Tr{F(t)}. Thus, the problem statement can be stated
as
Initial conditions are
It can be noted that Eq. (10.68) is a governing differential equation for a single-DOF
system.
Solution can be obtained using the Duhamel integral as
where the constant C1 and C2 are to be determined from the initial conditions. The integral
in the above equation may have to be evaluated numerically.
Once all p solutions of Eq. (10.68) are determined, the complete response is obtained by
superposing the modes as follows.
343
The mode superposition method is favored if only a few modes are needed to describe the
response. For example, an earthquake usually has only a few dominant frequencies in its
frequency spectra. However, the method has also been used for a concentrated force
(Green's function) in wave propagation problems (Datta and Shah, 2008), employing few
modes p(p < n).
10.3.4 Direct Time Integration Method

When a response for a short duration is sought, direct integration is an effective method of
choice. It is easy in this procedure to incorporate the full damping matrix [C] and the
conditions of Eq. (10.57) are not necessary. Of course, the computation time will increase.
A numerical step-by-step procedure is used in this method to integrate Eq. [10.56(b)] in
time directly without using any transformation matrix. It is based on deriving recursion
relations that relate the values of {q}, and at time t to those at time t + ∆t, where ∆t
is a small time interval. Starting at time t = 0 and incorporating the initial conditions, the
scheme marches out at the time step ∆t.
The algorithm continues to determine the solution at the discrete time t + ∆t, t + 2∆t,t+3∆t,
…. etc., until a desired time length is reached. It may be noted that the method includes
contribution of all n modes discussed in the previous section.
Direct integration techniques are either explicit or implicit. If the recursion relations are
substituted in the dynamic equations of equilibrium, Eq. [10.56(b)], at time t, the method is
explicit. On the other hand, if the recursion relations are substituted at time t + ∆t, the
method is implicit. Detailed discussion of these methods is beyond scope of this book.
They can be found in advanced finite element textbooks (Bathe, 2001). In the following,
both techniques are presented, the central difference method for the explicit case followed
by the Newmark Beta method for the implicit case.
10.3.4.1 Central Difference Method

Dynamic equilibrium relations in Eq. [10.56(b)] are a system of ordinary differential
equations. Any convenient finite difference expressions for kinematic variables can be
used to construct the algorithm. The central difference algorithm uses finite difference
expressions for velocity and acceleration at time t as
The dynamic equilibrium equations, Eq. [10.56(b)], at time t are
where subscript indicates values at time t.

Substituting relations for t and t of Eq. (10.72) into Eq. (10.73), the recursive
algorithm can be written of the form
344
Eq. (10.74) can be expressed in a compact form shown below.
where the effective stiffness matrix [ ] and the effective force { (t)} are given by
The algorithm starts at t = 0 incorporating the initial conditions {q}0 and 0. The initial
acceleration vector 0 can be found from Eq. [10.56(b)] using these conditions. The
displacement {q}−∆t needed to start the algorithm is evaluated from Eq. (10.72) as
Computation then continues using Eq. (10.74) and Eq. (10.75).

If the size of ∆t in Eq. (10.74) is selected larger than ∆tcr, the response may become
unstable and it may grow without bound as the time increases. To obtain a stable solution
the step size must be
where ωn is the largest frequency and Tn the shortest period of the equation (10.63) having
matrices [M] and [K] of size n × n.
The central difference method is said to be conditionally stable since it requires the use of
a time step ∆t smaller than a critical time step ∆tcr.
10.3.4.2 Newmark Beta Method

The Newmark Beta method is based on the following assumptions on displacement and
velocity.
Eq. (10.78) is solved for t+∆t , which is then substituted in Eq. (10.79) to obtain t+∆t . The
resulting equations are
Dynamic equilibrium equations, Eq. [10.56(b)], at time t + ∆t are
where the subscript indicates values at time t∆t.
345
Substitution of relations for t and t of Eqs. (10.80) and (10.81) into Eq. (10.82) results
in the following recursive algorithm.
Eq. (10.83) can be expressed in a compact form as
where the effective stiffness and the effective force are
The algorithm starts at t = 0 incorporating the initial conditions {q} and 0. 0 can be
found from Eq. [10.56(b)] using these conditions. Then, the algorithm can march forward
from Eq. (10.84).
Performance of the Newmark Beta method has been studied extensively for stability
(Bathe 2001; Zienkiewicz and Taylor 1989) and is not discussed here further. A good
choice of parameters for an implicit method that is conditionally stable is α = 0.5 and β =
0.25. This is called the constant average acceleration method, as seen from Eqs. (10.78)
and (10.79). For this choice of α and β there are no amplitude errors in any sine-wave
motion, regardless of the frequency. However, the periods of the sine-wave motion are
overestimated. The error magnifies as ∆t /T increases (Cook, 1981). The time integration
method is favored for short duration loads and for impulsive loads. For example, blast
loading has a wide frequency band and impact load has white noise as its frequency
spectra.
10.4 NON-LINEAR ANALYSIS
Final assembled equations arising from the application of FEM can be broadly represented
in one of the following forms.
Eq. set (10.85) arises from application of FEM to static (time independent) or steady-state
problems. Eq. sets (10.86) and (10.87), on the other hand, arise from the time-dependent
(transient) situations. However, these time-dependent equation sets can be transformed to
346
an equivalent ‘pseudo static’ set by using appropriate numerical techniques (e.g. Section
10.3.4). Thus, the final equation set considered in an analysis is generally of the form
described by Eq. set (10.85).
If [K] is a constant coefficient matrix, the ensuing linear simultaneous equations can be
solved by using methods such as those described in Appendix A. However, in certain
cases, [K] = [K (q)]. For example, consider case of stress analysis of structural
system/continuum. When the constitute model itself is non-linear (due to non-linear
material behavior), or when original geometry deforms considerably with respect to the
original geometry (i.e., geometric non-linearity) under application of loads, the assembled
matrix will no longer be a constant coefficient matrix. In such cases of [K] = [K (q)], and the
resulting equations are non-linear.
Due to presence of non-linearity, principle of superposition cannot be used. The ensuing
equations are required to be solved incrementally, in an iterative manner. Moreover,
distinction has to be made between the deformed and the original undeformed state. Final
(equilibrium) equations are written with respect to the deformed geometry by using
reference system either as the original undeformed configuration (Lagrangian system) or
the current deformed configuration (updated Lagrangian or Eulerian system). Detailed
discussion of non-linear analyses is beyond the scope of this book. They can be found in
advanced finite element textbook like (Zienkiewicz and Taylor, 1989; Bathe, 2001).
Typical aspects of finite element formulation for non-linear stress analysis are discussed
next.
10.4.1 Finite Element Formulation for Non-Linear Analysis
Equilibrium conditions between internal and external forces must be satisfied irrespective
of geometric or material non-linearity. If continuous displacement field in a system is
represented by the nodal displacement vector stemming from the application of FEM,
equilibrium equations can be obtained from the virtual work principle as
where is the sum of internal and external forces and is equivalent external nodal
force vector (vector in the context of usual linear finite element analysis). On the other
hand, is strain-displacement relationship matrix and is stress vector, which depend on
the type of finite element used and dimensionality of the problem. In general, , and
depend on displacements and, thus, the set of Eq. (10.88) is non-linear.
Residuals can be visualized as nodal forces required to bring the assumed
displacement pattern into equilibrium and, thus, the residuals should be reduced to zero as
much as possible.
Eq. set (10.88) is generally cast in the incremental form for ease in formulating iterative
process, as
where is known as the tangent stiffness matrix, and is a combination of usual linear
stiffness matrix and non-linear stiffness matrices arising from large displacements and/or
stress levels.
Solution of the ensuing non-linear equation set is discussed next.
10.4.2 Solution of Non-Linear Equations
347
Typical methods available in the literature to solve simultaneous non-linear equations are
listed below.
Newton–Raphson method
Modified Newton–Raphson method
Initial Stiffness method
Secant method
Arc-line method
Arc-line method with line search
Pre-conditioned conjugate gradient techniques
All these methods have their relative merits and demerits. For the purpose of illustration,
the Newton–Raphson method has been briefly discussed here.
When the Newton–Raphson method is used, system of non-linear equations is
represented as a set of functions for which roots are desired, that is,
These functions are expanded in a Taylor series about some arbitrary point, which
represents an initial estimate of solution. Linear solution of the problem is generally
considered as the initial estimate . Subsequently, the estimate is improved iteratively as
where
From Eq. (10.89),
and, thus, the incremental nodal displacement vector can be written as
Here, is the tangential stiffness matrix evaluated on the basis of i.

The iterative process is repeated until the required convergence criterion is satisfied.
Application of full load in a single step may cause solution procedure to diverge. Hence,
load is generally applied in increments. Such incremental process is sometimes
computationally cheaper and it also provides a complete load-displacement history.
The Newton–Raphson method is illustrated in Figure 10.8 with a single load increment,
where the applied load is ψ0 and the equilibrium solution is qEQN. On the other hand, K is
usual linear stiffness term.
Concepts in geometric and material non-linear analyses are illustrated next with the help of
simple illustrative examples.
348
Figure 10.8 The Newton–Raphson procedure
Example 1
Consider the geometrically non-linear spring shown in the figure. If K (u) = 200 +
20u2N/mm and P = 1,140 N, find the displacement u.
Solution:
For the single-DOF system depicted above, the restoring force is k.u = 200 u + 20 u3.
Thus, the equivalent of Eq. set (10.88) can be shown to be
Taking differentiation with respect to u, the following equation is obtained.
By comparing the above Eq. (2) with Eq. set (10.89), [KT] = 200 + 60u2.
Eqs. (10.91) through (10.94) can be rewritten for this example as
349
The initial estimate u0 is obtained from the linear solution, which is obtained by setting
nonlinear stiffness term to zero, as u0 = 1140/200 = 5.7 mm.
Repeated application of Eq. (6) and then Eq. (3) for i = 0,1,2, … yields following results.
Similarly, for
Thus, it can be seen that the displacement is converging to 3 mm. It can be confirmed that
the restoring force for 3 mm displacement is 200× 3 + 20× 33 = 1,140 N, which is the
magnitude of externally applied force P.
Example 2
A single-DOF system was considered in the previous example to illustrate geometric non-
linear analysis. In this example, a two-DOF geometrically non-linear spring system is
considered to illustrate solution process for a set of non-linear simultaneous equations. For
the two spring assemblage shown below, stiffness of each geometrically non-linear spring
is as described in the previous example. Thus, k1 = 200 + 20uB2 and k2 = 200 + 20(uc −
uB)2. Consider PB = 580 N and PC = 560 N. Compute the displacements at nodes B and C.
Solution:
By following the formulation given in Chapter 6, the assembled matrix can be shown (after
the application of boundary condition uA = 0) to be
From the above equation set, an equivalent form of Eq. (10.88) can be shown to be
350
By substituting expressions for K1 and K2 in the above equation and by using Eq. (10.89),
it can be shown that
where
The initial estimate is obtained from the linear equation set obtained from Eq. (1) by
considering only linear terms in K1 and K2 as
The residual force vector ψ (qi) required in the i-th iteration can be found from Eq. (2) by
substituting uBi and uCi as values of uB and uC, respectively.
Progress of numerical computations has been shown in the following table.
It can be seen from the table that as the iterative process progresses, (i) the residual
forces reduce [column (2)], (ii) elements of incremental solution vector become smaller
(column (4)) and (iii) solution converges to values uB= 3 mm and uC = 5 mm.
Example 3
The bar assemblage shown in Figure (a) has linear geometric properties and idealized bi-
linear stress strain properties, as shown in Figure (b). Consider cross sectional area as
A1= 500 mm2 and A2 = A3 = 300 mm2. Analyze the assemblage by using Newton–Raphson
method.
351
Solution:
Consider the discretization shown in the Figure (c).
Here, the direction cosines are

cos θ1= 0; sin θ1 = 1; cos θ2 = –0.6; sin θ2= 0.8; cos θ3 = 0.6; sin θ3 = 0.8
Due to bi-linear stress–strain curve and unequal force distribution in elements, it would be
useful to find effective stiffnesses of three bar elements in terms of appropriate elasticity
constants. Effective modulus of elasticity will be identical for elements 2 and 3, as they
have identical area and are placed symmetrically with respect to the vertical axis (which is
also the direction of load). However, element 1 has different stiffness and orientation as
compared to elements 2 and 3. Thus, it can have different modulus of elasticity, depending
on the induced strain levels. Thus, the stiffness coefficients for all the three elements are
computed as follows.
352
Element matrices in the local co-ordinates are
On the other hand, the global element matrices are obtained from formulation presented in
Chapter 6 as
Note that U2 = V2 = U3 = V3 = U4 = V4 = 0. After application of these boundary conditions,

the ensuing 2 × 2 assembled matrix is obtained as shown below.
Here, F1Y is the vertical load applied at node 1.

From the above equation (and also from the vertical plane of symmetry), it can be
concluded that U1 = 0. Thus, the only non-trivial equation is
353
Note that the numerical values of E1 and E2 depend on the strain levels.
Residual force can be computed by considering the nodal equilibrium of node 1, as shown
in the following figure.
Fe (1), Fe (2) and Fe (3) are magnitudes of forces in elements 1, 2 and 3, respectively, induced
due to application of vertical downward force F at node 1. By applying equilibrium
condition, following equations are obtained.
Here,
Eqs. (7) have been obtained from the given bi-linear stress–strain relations. Strain terms
can be obtained from the formulation presented in Chapter 6 as
Thus, an equivalent of Eq. (10.88) for this example can be written as
where the stress components are computed from Eqs. (7).

The initial estimate of the displacement can be obtained from Eq. (5) by setting E1 = E2 =
200 kN/mm2 and F1Y = –516 kN as
In the next iteration, the residual is estimated by computing stress components from Eqs.
(7) and substituting the same in Eq. (9). The tangent stiffness matrix is simply considered
to be K, defined in Eq. (5). Then, the incremental and the total updated displacements are
354
computed, respectively, by employing Eqs. (10.94) and (10.91), respectively. For example,
consider i = 0. From Eq. (8),
Then, stresses are computed from Eqs. (7) to be
The tangent stiffness is obtained from Eq. (5) as
On the other hand, the residual force computed from Eq. (9) is obtained to be
Subsequently, the incremental displacement and the total displacement are computed,
respectively, from Eqs. (10.94) and (10.91) as
Progress of the iterative process has been summarized below in the tabular form.
It can be observed from the table that the converged value of vertical displacement at
node 1 isV1 = –25.0245 mm. It should be noted that a large load increment (like the one
used in the current example) can be used only for elastic system, where loading and
unloading paths are the same. For an inelastic system, however, smaller load increments
must be used to trace load-displacement history.
10.4 NON-LINEAR ANALYSIS

Final assembled equations arising from the application of FEM can be broadly represented
in one of the following forms.
Eq. set (10.85) arises from application of FEM to static (time independent) or steady-state
problems. Eq. sets (10.86) and (10.87), on the other hand, arise from the time-dependent
(transient) situations. However, these time-dependent equation sets can be transformed to
an equivalent ‘pseudo static’ set by using appropriate numerical techniques (e.g. Section
355
10.3.4). Thus, the final equation set considered in an analysis is generally of the form
described by Eq. set (10.85).
If [K] is a constant coefficient matrix, the ensuing linear simultaneous equations can be
solved by using methods such as those described in Appendix A. However, in certain
cases, [K] = [K (q)]. For example, consider case of stress analysis of structural
system/continuum. When the constitute model itself is non-linear (due to non-linear
material behavior), or when original geometry deforms considerably with respect to the
original geometry (i.e., geometric non-linearity) under application of loads, the assembled
matrix will no longer be a constant coefficient matrix. In such cases of [K] = [K (q)], and the
resulting equations are non-linear.
Due to presence of non-linearity, principle of superposition cannot be used. The ensuing
equations are required to be solved incrementally, in an iterative manner. Moreover,
distinction has to be made between the deformed and the original undeformed state. Final
(equilibrium) equations are written with respect to the deformed geometry by using
reference system either as the original undeformed configuration (Lagrangian system) or
the current deformed configuration (updated Lagrangian or Eulerian system). Detailed
discussion of non-linear analyses is beyond the scope of this book. They can be found in
advanced finite element textbook like (Zienkiewicz and Taylor, 1989; Bathe, 2001).
Typical aspects of finite element formulation for non-linear stress analysis are discussed
next.
10.4.1 Finite Element Formulation for Non-Linear Analysis

Equilibrium conditions between internal and external forces must be satisfied irrespective
of geometric or material non-linearity. If continuous displacement field in a system is
represented by the nodal displacement vector stemming from the application of FEM,
equilibrium equations can be obtained from the virtual work principle as
where is the sum of internal and external forces and is equivalent external nodal
force vector (vector in the context of usual linear finite element analysis). On the other
hand, is strain-displacement relationship matrix and is stress vector, which depend on
the type of finite element used and dimensionality of the problem. In general, , and
depend on displacements and, thus, the set of Eq. (10.88) is non-linear.
Residuals can be visualized as nodal forces required to bring the assumed
displacement pattern into equilibrium and, thus, the residuals should be reduced to zero as
much as possible.
Eq. set (10.88) is generally cast in the incremental form for ease in formulating iterative
process, as
where is known as the tangent stiffness matrix, and is a combination of usual linear
stiffness matrix and non-linear stiffness matrices arising from large displacements and/or
stress levels.
Solution of the ensuing non-linear equation set is discussed next.
356
10.4.2 Solution of Non-Linear Equations

Typical methods available in the literature to solve simultaneous non-linear equations are
listed below.
Newton–Raphson method
Modified Newton–Raphson method
Initial Stiffness method
Secant method
Arc-line method
Arc-line method with line search
Pre-conditioned conjugate gradient techniques
All these methods have their relative merits and demerits. For the purpose of illustration,
the Newton–Raphson method has been briefly discussed here.
When the Newton–Raphson method is used, system of non-linear equations is
represented as a set of functions for which roots are desired, that is,
These functions are expanded in a Taylor series about some arbitrary point, which
represents an initial estimate of solution. Linear solution of the problem is generally
considered as the initial estimate . Subsequently, the estimate is improved iteratively as
where
From Eq. (10.89),
and, thus, the incremental nodal displacement vector can be written as
Here, is the tangential stiffness matrix evaluated on the basis of i.

The iterative process is repeated until the required convergence criterion is satisfied.
Application of full load in a single step may cause solution procedure to diverge. Hence,
load is generally applied in increments. Such incremental process is sometimes
computationally cheaper and it also provides a complete load-displacement history.
The Newton–Raphson method is illustrated in Figure 10.8 with a single load increment,
where the applied load is ψ0 and the equilibrium solution is qEQN. On the other hand, K is
usual linear stiffness term.
Concepts in geometric and material non-linear analyses are illustrated next with the help of
simple illustrative examples.
357
Figure 10.8 The Newton–Raphson procedure
Example 1
Consider the geometrically non-linear spring shown in the figure. If K (u) = 200 +
20u2N/mm and P = 1,140 N, find the displacement u.
Solution:
For the single-DOF system depicted above, the restoring force is k.u = 200 u + 20 u3.
Thus, the equivalent of Eq. set (10.88) can be shown to be
Taking differentiation with respect to u, the following equation is obtained.
By comparing the above Eq. (2) with Eq. set (10.89), [KT] = 200 + 60u2.
Eqs. (10.91) through (10.94) can be rewritten for this example as
358
The initial estimate u0 is obtained from the linear solution, which is obtained by setting
nonlinear stiffness term to zero, as u0 = 1140/200 = 5.7 mm.
Repeated application of Eq. (6) and then Eq. (3) for i = 0,1,2, … yields following results.
Similarly, for
Thus, it can be seen that the displacement is converging to 3 mm. It can be confirmed that
the restoring force for 3 mm displacement is 200× 3 + 20× 33 = 1,140 N, which is the
magnitude of externally applied force P.
Example 2
A single-DOF system was considered in the previous example to illustrate geometric non-
linear analysis. In this example, a two-DOF geometrically non-linear spring system is
considered to illustrate solution process for a set of non-linear simultaneous equations. For
the two spring assemblage shown below, stiffness of each geometrically non-linear spring
is as described in the previous example. Thus, k1 = 200 + 20uB2 and k2 = 200 + 20(uc − uB)2.
Consider PB = 580 N and PC = 560 N. Compute the displacements at nodes B and C.
Solution:
By following the formulation given in Chapter 6, the assembled matrix can be shown (after
the application of boundary condition uA = 0) to be
From the above equation set, an equivalent form of Eq. (10.88) can be shown to be
359
By substituting expressions for K1 and K2 in the above equation and by using Eq. (10.89), it
can be shown that
where
The initial estimate is obtained from the linear equation set obtained from Eq. (1) by
considering only linear terms in K1 and K2 as
The residual force vector ψ (qi) required in the i-th iteration can be found from Eq. (2) by
substituting uBi and uCi as values of uB and uC, respectively.
Progress of numerical computations has been shown in the following table.
It can be seen from the table that as the iterative process progresses, (i) the residual
forces reduce [column (2)], (ii) elements of incremental solution vector become smaller
(column (4)) and (iii) solution converges to values uB= 3 mm and uC = 5 mm.
Example 3
The bar assemblage shown in Figure (a) has linear geometric properties and idealized bi-
linear stress strain properties, as shown in Figure (b). Consider cross sectional area as
A1= 500 mm2 and A2 = A3 = 300 mm2. Analyze the assemblage by using Newton–Raphson
method.
360
Solution:
Consider the discretization shown in the Figure (c).
Here, the direction cosines are

cos θ1= 0; sin θ1 = 1; cos θ2 = –0.6; sin θ2= 0.8; cos θ3 = 0.6; sin θ3 = 0.8
Due to bi-linear stress–strain curve and unequal force distribution in elements, it would be
useful to find effective stiffnesses of three bar elements in terms of appropriate elasticity
constants. Effective modulus of elasticity will be identical for elements 2 and 3, as they
have identical area and are placed symmetrically with respect to the vertical axis (which is
also the direction of load). However, element 1 has different stiffness and orientation as
compared to elements 2 and 3. Thus, it can have different modulus of elasticity, depending
on the induced strain levels. Thus, the stiffness coefficients for all the three elements are
computed as follows.
361
Element matrices in the local co-ordinates are
On the other hand, the global element matrices are obtained from formulation presented in
Chapter 6 as
Note that U2 = V2 = U3 = V3 = U4 = V4 = 0. After application of these boundary conditions,

the ensuing 2 × 2 assembled matrix is obtained as shown below.
Here, F1Y is the vertical load applied at node 1.

From the above equation (and also from the vertical plane of symmetry), it can be
concluded that U1 = 0. Thus, the only non-trivial equation is
362
Note that the numerical values of E1 and E2 depend on the strain levels.
Residual force can be computed by considering the nodal equilibrium of node 1, as shown
in the following figure.
Fe (1), Fe (2) and Fe (3) are magnitudes of forces in elements 1, 2 and 3, respectively, induced
due to application of vertical downward force F at node 1. By applying equilibrium
condition, following equations are obtained.
Here,
Eqs. (7) have been obtained from the given bi-linear stress–strain relations. Strain terms
can be obtained from the formulation presented in Chapter 6 as
Thus, an equivalent of Eq. (10.88) for this example can be written as
where the stress components are computed from Eqs. (7).

The initial estimate of the displacement can be obtained from Eq. (5) by setting E1 = E2 =
200 kN/mm2 and F1Y = –516 kN as
In the next iteration, the residual is estimated by computing stress components from Eqs.
(7) and substituting the same in Eq. (9). The tangent stiffness matrix is simply considered
to be K, defined in Eq. (5). Then, the incremental and the total updated displacements are
363
computed, respectively, by employing Eqs. (10.94) and (10.91), respectively. For example,
consider i = 0. From Eq. (8),
Then, stresses are computed from Eqs. (7) to be
The tangent stiffness is obtained from Eq. (5) as
On the other hand, the residual force computed from Eq. (9) is obtained to be
Subsequently, the incremental displacement and the total displacement are computed,
respectively, from Eqs. (10.94) and (10.91) as
Progress of the iterative process has been summarized below in the tabular form.
It can be observed from the table that the converged value of vertical displacement at
node 1 isV1 = –25.0245 mm. It should be noted that a large load increment (like the one
used in the current example) can be used only for elastic system, where loading and
unloading paths are the same. For an inelastic system, however, smaller load increments
must be used to trace load-displacement history.
10.6 HYDRODYNAMICS SIMULATION OF SHALLOW WATER FLOW

10.6.1 Introduction
Flow regime in rivers, channels, estuaries and coastal regions is a transient real world
phenomenon. Solution to these problems can be obtained through physical models,
analytical solutions or numerical models. Physical model is an attempt to reduce the
complex problem into small observable scale in laboratory. Physical models are generally
difficult to make, time consuming to prove (possibility of scale effects, etc.) and expensive.
Analytical solutions refer to closed form solutions of governing equations. For rivers,
channels or estuary problems, very few analytical solutions are available for simplified
domain and boundary conditions. A numerical model is an attempt to represent system by
solving governing equations that describe natural processes. Very complex geometry and
364
boundary conditions (peculiar to river or estuary problems) can be handled in numerical

modelling (Vreugdenhil, 1994) using FEM. If numerical models are used for simulation
related to river, estuary or coastal region, choice is between one-dimensional model and
the one that incorporates two- or three-spatial dimensions (Hinwood and Wallis, 1975).
One-dimensional models are easier to build, economical and fairly adequate for practical
purposes. Due to complexity of estuary or river hydrodynamics, however, one-dimensional
models may not give sufficiently accurate results. Hence, two-dimensional or three-
dimensional models are often required for a complete analysis of estuary problems.
Problems related to numerical diffusion, dispersion and oscillation, demands of computer
time and memory for simulation are some of the challenges in development of numerical
models for two-dimensional and three-dimensional analyses (Vreugdenhil, 1994). Here,
the two-dimensional finite element formulation for estuary or coastal hydrodynamics is
presented with detailed formulation and application.
10.6.2 Governing Equations and Boundary Conditions

For many coastal and estuarine flow problems, the vertical velocity component is relatively
small in comparison with the horizontal velocity components U, V. Hence, the continuity
equation and Navier–Stokes equations can be integrated over the depth and solved
numerically to give the depth averaged velocity field (Martin and McCutcheon, 1998). Such
flow field can generally be assumed to occur in unstratified or weakly stratified, estuaries,
harbors, bays and coastal seas, where the wind shear is also relatively small. Shallow
water equations in non-conservative form are used here for hydrodynamics of estuaries.
Pertinent equations are described below (Gray and Lynch, 1979; Liggett, 1994).
(1) Continuity equation
(2) Momentum equation in X direction
(3) Momentum equation in Y direction
Here, τ = g(U2 + V2)1/2/C2H; f = Coriolis parameter = 2 Ω sin θ; Ω = angular velocity of earth;

a = depth mean eddy viscosity; θ = degree of latitude; g = acceleration due to gravity; z =
H – h; H = total depth = h + ζ (see Figure 10.14); h = distance of bottom from reference
plane; ζ =surface elevation from reference line; C = Chezy's Coefficient; W = wind shear
component in the direction indicated by the suffix; t = time; U = depth averaged velocity in
x direction; and V = depth averaged velocity in y direction.
Initial and Boundary Conditions
Governing equations are solved by using initial and boundary conditions. The simplest
initial condition is that the fluid is at rest with a horizontal water level. This normally does
not happen in reality. Influence of this wrong initial data gradually fades out due to wave
damping by bottom friction.
365
Figure 10.14 Vertical section of a shallow water flow field

Surface and bottom conditions come in two kinds. The kinematic conditions, for example,
can be of the form that water particles do not cross either boundary. For the solid bottom,
this means that the normal velocity component vanishes. Mathematically, the condition
can be stated as
where Zb is the bottom level, measured from some horizontal reference level at the free
surface and W is vertical velocity. Here, computations may be complicated as the surface
may be moving by itself. Then, the relative normal velocity must vanish, as shown below.
Here h is the surface level measured from a horizontal reference level.

At the bottom, one can assume that the viscous fluid ‘sticks’ to the bottom, that is,
This is called the ‘no-slip’ condition.

At the free surface, continuity of stresses is assumed, that is, the stresses in fluid just
below the free surface are assumed to be the same as those in air just above. This means
that the surface tension is not taken into account. For pressure,
where pa is the atmospheric pressure. The absolute pressure level is not important so it
can be considered to be zero.
10.6.3 Finite Element Formulation

In the present study, FEM based on Galerkin's formulation is used. Neglecting the eddy
viscosity terms and spatial variations of atmospheric pressure and wind frictional effect,
the shallow water (governing) equations can be written as (Gray, 1977; Lynch and Gray
1979)
366
Here, U = depth averaged velocity in x direction; V = depth averaged velocity in y direction;

H = total depth = h + ζ (see Figure 10.14); h = distance of bottom from reference plane; C
=Chezy's Coefficient; and f = Coriolis parameter.
Unknown variables in shallow water equations are approximated in FEM as (Connors and
Brebbia, 1976).
where n is the number of nodes (n = 3, for triangular element) per element. The Chezy's
coefficient is taken as constant (but it can vary in some cases). Here, Ni are shape
functions. In the present study, simple three node triangular elements are used to
represent the domain as shown in Figure 10.9. The shape function details are given in
Section 10.5.3.1.
Variation of quantity can be expressed in matrix form as
Governing equation L (ɸ) is approximated in the context of weighted residual approach

(Connor and Brebbia, 1976) as
If Wi = Ni, method weighted residual is called Galerkin's weighted residuals method, that is,
Galerkin's method is applied to the continuity Eq. (10.145) and momentum Eq. (10.146),
Eq. (10.147) in the sequel.
Continuity Equation
Momentum Equation in X Direction
367
where
Momentum Equation in Y Direction
where . Dot indicates the time derivatives
and .
Time Domain Discretization
Eqs. (10.153), (10.155) and (10.157) can be written in the form
where R is function of variables U, V, H and spatial coordinate, ɸ is any unknown variable

U, V, or H. Eq. (10.158) can be written in finite difference form as
Eqs. (10.153), (10.155) and (10.157) are rewritten below in form depicted by Eq. (10.159).
Continuity Equation
Eq. (10.160) can be written in the following form
368
where
Momentum Equation in X Direction
Eq. (10.162) can be written as
where
Momentum Equation in Y Direction
Eq. (10.164) can be written as
where
369
Generally, explicit or implicit scheme is used for time integration. In explicit scheme, {R} is
considered at time t. At time t, values of variable U, V and H are known. Thus, values of U,
V and H can be computed explicitly. Time steps to be used in explicit method should be
very small to get good accuracy in solution.
Figure 10.15 Rectangular channel domain

Weighted average of {R} is considered at time t and time t + ∆t, when an implicit scheme is
used.
Here, w0 = 1– w1.
For implicit model, value of w1 can be taken as 0.7. For this value, system becomes
unconditionally stable (Reddy, 1993). Larger time step can be taken with small decrease in
accuracy of solution. Larger time step, in addition to decreasing accuracy of solution can
also introduce some unwanted, numerically induced oscillations into solution. Numerical
oscillation may occur, but they never become unstable for problems considered here.
By superimposing element matrices for all elements in proper order, the global matrix for
entire discretized domain can be obtained as follows.
Here, {x} is a matrix or column vector consist of H, U or V at time instant t + ∆t. On the
other hand, coefficient matrix [A] and column vector {R} are composed of values at time
instant t. Thus, H, U and V can be obtained at t + ∆t time step by knowing values at t by
using standard procedure such as Gauss elimination or iterative scheme discussed earlier.
10.6.3 Case Study
Based on the above formulation, an implicit FEM based model has been developed and is
applied to a straight simple channel (Figure 10.15) described by Gray and Lynch (1979).
The channel has a length of 871.65 Km and width of 160 km. Here, the unknowns are
horizontal velocities (U, V) and depth of water (H). Triangular elements have been used in
discretization as shown in Figure 10.16. Grid has 176 elements and 115 nodes. For
simulation, model is tested with constant bathymetry of 9.1435 meter. A sinusoidal tidal
forcing function period of 12 hours and amplitude of 0.091435 m has been applied to the
opened end of the channel. Cold start (U = V = 0, H = h) is taken as initial condition.
370
Normal flux to solid boundaries is taken as zero. Coriolis effect is neglected. Time step of
∆t = 350 sec is used. Solution obtained with models are shown in Figures 10.17 and 10.18
for surface elevation and velocities (U) after the steady-state has been reached. Small
oscillations are present in solution when compared with analytical solution given by Gray
and Lynch (1979). This is expected since the present model is based on numerical
approximation.
Figure 10.16 Finite element grid used in computation of the rectangular channel
Figure 10.17 Surface elevation computed by FEM model compared with analytical
solution
371
Figure 10.18 Velocity computed by FEM model compared with analytical solution
10.7 FEM–SOFTWARE AND WEB RESOURCES

10.7.1 Introduction
Due to tremendous development of FEM in various areas of Engineering and Sciences for
the past 50 years or so, a large number of software and packages are available
commercially and also as open sources. Even though few packages can solve problems in
more than one area, most FEM packages are area specific and can solve only particular
set of problems. Furthermore, while few packages can be employed to solve problems in
three-, two- and one-dimensional domains, many packages can be employed to solve
problems in one- or two-dimensional domains. Most of earlier packages did not have any
graphical user interface (GUI) for discretization and data preparation and, thus, the pre-
and post-processing had to be done separately. With recent advancements in computer
sciences, most software packages presently available in market have GUI with various
pre- and post-processing options to facilitate data preparation and interpretation of results.
In this section, some of the commonly used FEM based software and packages in various
areas of Engineering and Sciences are briefly described, with an aim to make reader
aware of available software packages. Relative merits or demerits of competing software
packages are not discussed here. Reader is referred to a comprehensive description on
software packages at
sitehttp://en.wikipedia.org/wiki/List_of_finite_element_software_packages.
In addition, websites of the developers and vendors of specific software packages can be
used to know about full capabilities and features to arrive at conclusion for selecting best
possible software package(s) for the intended applications.
10.7.2 FEM in Structural Engineering

As FEM was initially developed in the field of structural engineering, a large number of
software packages have been developed during the past five decades in the area. Many
software packages are available for structural design, planning and management.
Exclusive design oriented software packages are available for structural steel design,
reinforced cement concrete design, mechanical and aerospace structural design etc. In
372
this section, some of the commonly used software packages in structural engineering have
been briefly discussed.
STAAD
(http://www.bentley.com/en-US/Products/Structural+Analysis+and+Design)
STAAD family of FEM analysis and design packages (STAAD.Pro/STAAD.beava) can
be used for typical structures like multi-storied buildings, towers, culverts, plants, bridges,
stadiums, marine structures and so on. It can easily accommodate design and loading
requirements, in conformation of various international and national codes of practice.
RAM Products/ Bentley Structural
There are number of RAM modules (RAM Advanse / RAM Frame / RAM Concrete
/RAM Concept / RAM Steel) for analysis and design of variety of structures, structural
components: Structures such as trusses, continuous beams, frames of all types, retaining
walls, masonry walls, tilt-up walls, shear walls, footings, etc. Static and dynamic analysis
of two- and three-dimensional structures made up of common construction materials can
also be performed for a variety of loading conditions
Bentley Structural software can be used in structural design and construction
documentation for structures in steel, concrete and timber in two- or three-dimensional. It
gives the integrated analytical model with finite elements, nodes, boundary conditions and
member releases, loads and load combinations.
NISA Products
(http://www.nisasoftware.com/Products)
NISA is a FEM software for design and analysis to address the Automotive, Aerospace,
Energy and Power, Civil, Electronics and Sporting Goods industries. NISA include
integrated pre-/post- processing environment within a GUI and inter-operability with
leading commercial CAD software. It can be used for solid and surface modelling, finite
element mesh generation, various types of analyses such as that for stress, vibration,
seismic, electromagnetic, fatigue, CFD/fluid structure interactions and so on. Some
products (NISA/CIVIL, NISA DesignStudio) can be used for analysis, design and
detailing of variety of RCC and steel structures.
MSC Nastran
(http://www.mscsoftware.com/products/msc_nastran.cfm)
MSC Nastran is a general purpose finite element analysis solution for small to complex
assemblies. It provides a wide range of modelling and analysis capabilities, including
linear statics, displacement, strain, stress, vibration, heat transfer and so on.
NEi Fatigue
(http://www.nenastran.com/newnoran/fatigue)
NEi Fatigue allows Engineers to calculate fatigue life of components under dynamic
loading. Applications of NEi Fatigue include fatigue analysis for bodywork structures, axle
components, crankshafts, rotary blades for wind power stations and so on.
ANSYS
(http://www.ansys.com/products/structural-mechanics)
373
ANSYS structural mechanics solutions offer a broad spectrum of capabilities covering a

range of analysis types, elements, contact, materials, equation solvers and coupled
physics. It has a CAD interface for the direct use of CAD models without any modifications
or mediations.
ABAQUS
(http://www.simulia.com/)
ABAQUS can be used in many different engineering fields such as civil, structural,
automobile, mechanical, aerospace engineering and so on. It performs static and/or
dynamic analysis and simulation on structures. It can deal with bodies with various loads,
temperatures, contacts, impacts, and other environmental conditions. Pre- and post-
processing GUI interface available with the package can be used for visualization and
animations.
10.7.3 FEM in Geotechnical Engineering

Many software packages are available for geotechnical design, planning and
management. A variety of software packages for foundation design, tunnel design, earthen
dam and embankment design, retaining wall design and so on are available. In this
section, some of the commonly used software packages in geotechnical engineering have
been discussed.
STAAD.foundation / RAM Foundation
STAAD.foundation can automatically absorb the geometry, loads and results from a
STAAD.Pro model to design foundations. It keeps track of changes made in the
STAAD.Pro model and can merge reaction output to the existing foundation model.
RAM Foundation is an integrated module within the RAM Structural System that
performs design, evaluation and analysis of spread, continuous and pile cap foundations
and so on.
Midas GTS
(http://www.midas-diana.com/products/product_gts.asp)
Midas GTS is a Geotechnical and Tunnel Analysis System. It can be used to analyze
tunnel intersections, deep foundations, excavations, embankments and slope stability,
groundwater flow, effective stress analysis and vibration analysis for earthquake, blasting
and so on.
CRISP
(http://www.mycrisp.com/products/crisp2d.html)
CRISP can be used for a variety of geotechnical problems, including retaining structures,
embankments, tunnels, foundations, geo-textile reinforcement, soil nailing, slope stability,
borehole stability, construction sequence studies and so on.
PLAXIS
(http://www.plaxis.nl)
Plaxis is a range of finite element packages intended for two- and three-dimensional
analysis of foundation, deformation, stability, groundwater flow and so on in geotechnical
engineering.
374
10.7.4 FEM in Fluid Mechanics

Many software packages are available for fluid flow analysis of pipes, open channels,
rivers, estuaries, mechanical and aero systems and so on. Software packages are also
available for computational fluid dynamics analysis, groundwater analysis, surface flow
analysis, contaminant transport, etc. In this section, some of the commonly used software
packages in the field of Fluid Mechanics are briefly discussed.
ANSYS CFX (http://www.ansys.com/products)
ANSYS CFX is based on computational fluid dynamics (CFD) technology for simulations
and solutions of steady-state/transient analysis; laminar/turbulent flow analysis for a
variety of problems.
NEi Fluid Dynamics (http://www.nenastran.com/newnoran/fluidDynamics)
NEi Fluid Dynamics provides CFD analysis solutions and models of three-dimensional
fluid velocity, temperature and pressure for steady-state and transient applications.
Various simulations like laminar and turbulent fluid flow, low speed and high speed
compressible flow, forced-convection, natural convection and mixed flows and so on can
be performed by using the software.
NISA/3D-FLUID (http://www.nisasoftware.com/Portals)
NISA/3D-FLUID can be used for gas flow from the low Mach number up to hypersonic
speed. It allows specification of flow specific boundary conditions allows specification of
flow variables such as density, temperature and its derivative, species mass fractions, and
x, y, z momentums at any location in the computational domain.
FEFLOW (http://www.feflow.info/about+M520d624e0eb)
FEFLOW is a software package for modelling fluid flow and transport of dissolved
constituents and/or heat transport processes in the subsurface. It can be used for flow and
transport processes simulation in porous media, ranging from lab scale to continental
scale and geotechnical applications for tunnel construction, construction site dewatering,
dam seepage and so on.
FEFLOW-F3 (http://www.swstechnology.com/groundwater-software/groundwater-
modeling/feflow-f3)
FEFLOW-F3 (Finite Element subsurface FLOW system) is a ground water modelling
software and can be used for modelling of flow and transport processes in porous media
under saturated and unsaturated conditions.
FEMWATER (http://www.epa.gov/ceampubl/gwater/femwater/index.html)
It is a software package of groundwater modelling system (GMS) for different groundwater
models, including a three-dimensional finite element model of density-dependent flow and
transport through saturated–unsaturated porous media.
10.7.5 FEM in Thermal and Automobile Engineering

Many software packages are available for heat flow analysis and automobile design
dealing with wide spectrum of problems encountered in heat transfer, complex real world
automobile problems and so on. Some of the commonly used software packages are
briefly discussed here.
NEI Thermal Basic (http://www.nenastran.com/newnoran/thermalBasic)
375
NEI Thermal Basic (TMG) provides solutions for linear and non-linear, steady-state
and transient heat transfer processes including conduction, radiation, and free convection.
It includes modelling of conduction, convection, radiation and phase change and it
provides a range of thermal boundary conditions and solver controls.
NISA/ HEAT (http://www.nisasoftware.com/Products/MechanicalApplications)
NISA/Heat is a general purpose finite element program to analyze a wide spectrum of
problems encountered in heat transfer dealing with concentrated heat flows, distributed
heat flux, internal heat generation, radiation effects and so on. It can analyze non-linearity
due to time and temperature-dependent materials and boundary conditions, such as
temperature-dependent thermal conductivity and phase change (melting and freezing) and
so on.
LS-DYNA (http://www.easi.com/software-products-ls-dyna.aspx)
LS-DYNA software is a general purpose transient dynamic finite element program used
by automotive industries to analyze vehicle designs, especially for crash worthiness. It can
be used for non-linear analysis; rigid body dynamics; quasi-static simulations; FEM-rigid
multi-body dynamics coupling (MADYMO, CAL3D); real-time acoustics; design
optimization; multi-physics coupling; structural-thermal coupling, adaptive re-meshing and
so on.
10.7.6 FEM in Physics

Some of the commonly used software packages for applications in physics such as
acoustics, heat flow analysis and material studies and so on are briefly discussed below.
COMSOL Multi-physics (http://www.comsol.com/products/)
COMSOL Multi-physics (formerly FEMLAB) is a finite element analysis and solver
software package for various physics and engineering applications, especially coupled
phenomena, or multi-physics. The software has several modules like AC/DC Module,
Acoustics Module, CAD Import Module, Chemical Engineering, Earth Science Module,
Heat Transfer Module, Material Library, MEMS Module, RF Module, Structural Mechanics
Module.
Xenos (http://www.feildp.com/index.html)
Xenos (X-ray/electron numerical optimization suite) is a suite of two- and three-
dimensional finite element programs to handle X-rays and electrons and has a focus on
coupled or stand-alone applications for electric and magnetic field calculations, electron
beam design, Monte Carlo modelling of radiation transport and thermal analysis and so
on.
10.7.7 Multi-Field FEM Software

Some of the commonly used software packages for multi-field applications are discussed
below.
NISA (http://www.nisasoftware.com)
NISA is available to address the needs of the automotive, aerospace, energy and power,
oil and gas, electronics packaging and civil engineering industries (two- and three-
dimensional rigid frames and trusses). NISA provides solutions in the areas of stress
analysis, seismic analysis, vibration analysis, composite material analysis, fatigue
analysis, thermal analysis, PCB analysis, Computational Fluid Dynamics, electromagnetic
analysis, Civil structure analysis and so on.
376
Elmer (http://www.csc.fi/english/pages/elmer)
Elmer is an open source multi-physical simulation software. Elmer includes physical
models of fluid dynamics, structural mechanics, electromagnetics, heat transfer and
acoustics, for example. These are described by partial differential equations; which Elmer
solves by FEM. One can use basic element shapes in one-, two- and three-dimensional
domains with the Lagrange shape functions of k ≤ 2 and FEM solvers (without multiple
acceleration). Assembly and iterative solution can also be done in parallel.
FlexPDE6 (http://www.pdesolutions.com/)
FlexPDE6 is Multi-Physics Finite Element Solution Environment for Partial Differential
Equations. It can be used for one-, two- or three-dimensional Multi-Physics PDE problems
of heat flow, stress analysis, fluid mechanics, chemical reactions, electromagnetics,
diffusion, etc. It has a self-contained processing system that analyses problem description,
symbolically from Galerkin's finite element integrals, derivatives and dependencies. It
builds a coupling matrix and solves it and plots the results.
10.7.8 FEM in Other Fields

Other than the various fields mentioned above, FEM has been used in many other fields.
Many software packages are available for various areas like electrical engineering,
electronics, bio-medical engineering and so on. In this section, some of the commonly
used software packages in such fields are discussed briefly.
FEAP (http://www.nisasoftware.com)
FEAP has been developed for analyzing stress, random vibration, fatigue life, three-
dimensional convective fluid flow and thermal analysis of printed circuit boards (PCBs),
mounted components and electronic systems.
NEi Nastran (www.waveaxis.com/nei-nastran.html)
NEi Nastran provides solutions for bio-medical engineering to evaluate and optimize
many performance and reliability aspects of designs and manufacturing processes
encountered in the field. Design and deployment of cardio-vascular related systems such
as vascular stents, catheters, drug delivery systems, pacemakers, defibrillators and heart
valve replacements can be done using this software. Design and deployment of orthopedic
related devices such as knee replacement, hip implants, spinal implants, cartilage and
joint replacement, human modelling, virtual biomechanics and biocompatible materials
analysis, can also be performed.
10.8 CONCLUDING REMARKS

Many approximate numerical analysis methods have evolved in the last six decades for
solution of complex engineering problems due to advent of high speed computers. Out of
various available numerical techniques, finite difference method (FDM), finite element
method (FEM), boundary element method (BEM), finite volume method (FVM) and
meshless method have become popular among scientists and engineers. With
advancements in computational science and development of modern computers and high
speed processors, application of these numerical techniques has become much easier.
Out of these numerical techniques, one of the most important advances in the field of
numerical methods was the development of FEM in the 1950s.
FEM has become a very popular and practical tool for all type of computing in engineering
and science. As demonstrated in this book, FEM is robust and thoroughly developed
method. It is widely used in all engineering and science fields due to its versatility and
377
applicability to complex geometry and flexibility for many types of linear and non-linear
problems. Most practical engineering problems related to solids, structures, fluids and
materials are currently solved using well developed FEM packages.
As a numerical method, it appears that FEM has reached a point where no significant
break though may be expected in the near future. However, future growth of FEM will
involve broader applications to further engineering problems, further refinement in the
basic theorems and formulations and further efficient pre-processing and post-processing
tools. Various sophisticated tools are available for pre-processing of data and post-
processing of FEM simulation results. However, further efforts are required to efficiently
automate pre- and post-processing tasks for more user friendly and error free operations.
Tremendous growth in computer technology will undoubtedly continue to have significant
impact on further evolution of FEM. Improved efficiency achieved in computer hardware
and software tools such as micro/nano processors and large-scale parallel processing will
need significant improvements in finite element model development and applications.
From Engineer's view point, FEM can always be made more efficient and easier to use
with sophisticated pre- and post-processing tools. As FEM is applied to very large and
complex problems, it is very important that the solution process remains efficient and
economical. With continuing demand and economic pressures to further improve
innovations, efficiency and engineering productivity, this century will see further growth of
FEM in engineering design and analysis. FEM will remain more exciting and will be an
important part of Engineer's design and analysis toolkit.

Arlett, P. L., Bahrani, A. K., and Zienkiewicz, O. C. (1968). Application of Finite Element
Method to the Solution of Helmholtz's Equation, Proc. IEE, 115(12):1762–1766.
Arpita, M. (2008). FEM-GA based simulation-optimization models for groundwater
remediation, M.Tech. Thesis, Department of Civil Engineering, Indian Institute of
Technology, Bombay.
Baker, A. J. (1983). Finite Element Computational Fluid Mechanics, McGraw Hill,
Hemisphere, New York.
Bear, J. (1979). Hydraulics of Groundwater. McGraw Hill, New York.
Bogner, F. K., Fox, R. L., and Schmidt, L. A. (1968). The generation of inter-element-
compatible stiffness and mass matrices by the use of interpolation formulae, Proceedings
of 1st conference on matrix methods in structural mechanics, Wright-Patterson Air Force
Base, Dayton, OH, pp. 397–443.
Brebbia, C. A., and Walker, S. (1979). Dynamic Analysis of Offshore Structures, Newnes-
Butterworths, London.
Chapra, S. C., and Canale, R. P. (2002). Numerical Methods for Engineers, 4th ed., Tata
Clough, R. W., and Penzien, J. (1975). Dynamics of Structures, McGraw Hill.
378
Connors, J. J., and Brebbia, C. A. (1976). Finite Element Techniques for Fluid Flow,
Cook, R. D. (1981). Concepts and Applications of Finite Element Analysis, John Wiley and
Sons.
Datta, S. K., and Shah, A. H. (2008). Elastic Waves in composite Media and Structures,
CRC Press, Boca Raton.
Desai, C. S., and Abel J. F. (1987). Introduction to the Finite Element Method — A
Numerical Method for Engineering Analysis, Wads Worth, California.
Dunne, P. (1968). Complete polynomial displacement fields for the finite element method,
Aeronaut. J., 72: 246–247.
Freeze, R. A., and Cherry, J. A. (1979). Groundwater, Prentice Hall, Englewood Cliffs, NJ.
Gray, W. G. (1977). An efficient finite element scheme for two-dimensional surface water
computation, Finite Element in Water Resources, Vol. I, Brebbia, C. A., Gray, W. G. and
Pinder, G. F., (eds.), Pentech Press, London, pp. 4.33–4.49.
Gray, W. G., and Lynch, D. R. (1979). On the control of noise in finite element tidal
computations: a semi-implicit approach, Computer and Fluids,7:47–67.
Hinwood, J. B., and Wallis, I. G. (1975). Review of Models of Tidal Waters, J. Hydra. Eng.,
ASCE, 101(10).
Hrabok, M. M, and Hrudey, T. M. (1984). A review and catalogue of plate bending finite
elements, Computers and Structures, 19(3):479–495.
Huebner, K. H., Thornton, E. A., and Byron, T. G. (1995). The Finite Element Method for
Engineers, John Wiley and Sons.
Huyakorn, P. S., and Pinder, G. F. (1983). Computational Methods in Subsurface Flow.
Academic Press, New York, pp. 473.
Irons, B. M., and Razzaque, A. (1977). Experience with the patch test for convergence of
finite element method, in Mathematical Foundation of the Finite Element Method, Aziz, A.
K. (ed.). Academy Press, pp. 557–587.
Istok, J. (1989). Groundwater Modeling by the Finite Element Method, American
Geophysical Union, Washington.
Kinsler, L. E., and Frey, A. R. (1950). Fundamentals of Acoustics, John Wiley and Sons,
New York.
Krishnamoorthy, C. S. (1994). Finite Element Analysis: Theory and Programming, Tata
Liggett, J. A. (1994). Fluid Mechanics, McGraw Hill, Singapore.
Lynch, D. R., and Gray, W. G. (1979). A wave equation model for finite element tidal
computations, Computer and Fluids, 7:207–228.
MacCamy, R. C., and Fuchs, R. A. (1945). Wave Forces on Piles: A Diffraction Theory, US
Army Coastal Engineering Center, Tech. Mem. No. 69.
Martin, J. L., and McCutcheon, S. C. (1998). Hydrodynamics and Transport for Water
Quality Modelling, Lewis Publishers, London.
379
Melosh, R. J. (1963). Basis for derivation of matrices for the direct stiffness method, J. Am.
Inst. Aeronaut. Astron. 1(7):1631–1637.
Mindlin, R. D. (1951). Influence of rotatory inertia and shear on flexural motion of isotropic,
elastic plates, J. Appl. Mech., 31–38.
Press, New York.
Pinder, G. F., and Gray, W. G. (1977). Finite Element Simulation in Surface and
Subsurface Hydrology, Academic Press, New York.
Chichester.
Oden, J. T. (1969). A general theory of finite elements: II. Applications, Int. J. Numer.
Methods in Eng., 1(3).
Reddy, J. N. (1993). An Introduction to the Finite Element Method, McGraw Hill, New York.
Reddy, J. N. (1999). Theory and Analysis of Elastic Plates, Taylor and Francis, PA, USA.
Rice, J. (1993). Numerical Methods, Software, and Analysis, Academic Press, New York.
Stasa, F. L. (1985). Applied Finite Element Analysis for Engineers, CBS Publ. Japan, New
York.
Streeter, V. L., Wylie, E. B., and Bedford, K. W. (1998). Fluid Mechanics, WCB/ McGraw
Hill, Boston.
Sun, N. (1996). Mathematical Modeling of Groundwater Pollution, Springer Publishing
House, New York.
Taylor, C., Patil, B. S., and Zienkiewicz, O. C. (1969). Harbor Oscillation: A Numerical
Treatment for Undamped Natural Modes, Proc. Inst. Civil Eng., 43:141–155.
Timoshenko, S. P., and Woinowsky-Krieger, S. (1959). Theory of plates and shells,
Second ed. McGraw Hill International ed., Singapore.
Vreugdenhil, C. B. (1994). Numerical Methods for Shallow-Water Flow, Kluwer Academic
Publishers, London.
Wang, H., and Anderson, M. P. (1982). Introduction to Groundwater Modeling Finite
Difference and Finite Element Methods, W. H. Freeman and Company, New York.
Zheng, C., and Bennett, G. D. (1995). Applied Contaminant Transport Modeling-Theory
and Practice, Van Nostrand Reinhold, New York, NY.
Zienkiewicz, O. C., and Taylor, R. L. (1989). The Finite Element Methods, 4th ed., Vol. 1:
Basic Formulation and Linear Problems, Vol. 2: Solid and Fluid Mechanics; Dynamics and
Non-linearity McGraw Hill Company Limited, London.
380
Appendix A
REVIEW OF MATRIX ALGEBRA AND
MATRIX CALCULUS
A.1 REVIEW OF MATRIX ALGEBRA
Finite element method can be viewed as a means of forming matrix equation that
approximates behavior of a physical system. Hence, any finite element application results
in forming, manipulating and solving matrix equations. Therefore, matrix algebra pertaining
to finite element analysis is reviewed here.
A.1.1 Definition of a Matrix
A matrix is an m × n array of numbers arranged in m rows and n columns.
Size of [a] = m × n
ai j → element of [a] corresponding to i th row and j th column.
A.1.2 Types of Matrices
1. Column matrix (vector) – n = 1:
2. Row matrix (transposed vector) – m = 1:
3. Rectangular matrix (m ≠ n):
381
4. Square matrix (m = n):
5. Symmetric matrix:
Symmetric matrix is always a square matrix.

6. Diagonal matrix:
It is a square matrix, where only the diagonal elements are non-zero, i.e., aij = 0, if i ≠ j.
7. Identity matrix (unit matrix):

A diagonal matrix with ones on the main diagonal is called an identity matrix.
8. Transposed matrix:
It is obtained by interchanging rows and columns.
Transpose of [a] = [a] = [b] =

T
bij = aji
Note: For a symmetric matrix [a] = [a].
T
9. Null matrix:
All elements are zero in the null matrix.
382
10. Triangular matrix:

It can be either an upper or lower triangular matrix.
Upper triangular matrix:
Lower triangular matrix:
11. Banded matrix:
A.1.3 Equality of Matrices
Two matrices are said to be equal only when they are equal element by element.
A.1.4 Addition or Subtraction of Matrices
Matrices must be of the same size and elements must be added or subtracted term by
term.
or, in subscript (index) notation, it can be shown that

cij = aij + bij = bij + aij
Let
383
So
A.1.5 Scalar Multiplication
Let
Then
A.1.6 Matrix Multiplication
Two matrices [a] and [b] can be multiplied only if the number of columns in matrix [a]
equals the number of rows in matrix [b].
c11 = 4(–2) + 3(0) + (–1)(1) = –9

c12 = 4(0) + 3(1) + (–1)(1) = 2
c21 = 2(–2) + (0) + 1(1) = –3
c22 = 2(0) + 0(1) + 1(1) = 1
Hence, [c] =
In general, matrix multiplication is not commutative.
i.e., [a][b] ≠ [b][a]
A.1.7 Inverse of a Square Matrix
384
There are several ways of computing an inverse of a matrix. When the size of matrix is
small (m = n ≤ 4), inverse can be computed easily by employing cofactor method.
Co-factor or adjoint method:
where [c] = cofactor matrix.
Mij is the minor of aij and is defined as the determinant of the matrix obtained by deleting
the row and column belonging to aij from [a].
Inverse exists only if |a| ≠ 0 (non-singular matrix).
Then:
c11= (-1) (1 ×1 – 1 × 0) = 1
2
c12 = (-1) (2 × 1 – 1 × 0) = –2
3
c13 = (-1) (2 ×1 – 1 × 1) = 1
4
c21 = (-1) (2 × 1 – 1 × 1) = –1
3
c22= (-1) (2 ×1 – 1 × 1) = 1
4
c23 = (-1) (2 ×1 – 1 × 2) = 0
5
c31 = (-1) (2 ×0 – 2 ×1) = –1

5
c32 = (-1) (2× 0 – 1 × 1) = 2

5
c33 = (-1) (2 ×1 – 2 × 2) = –2
6
Hence
385
A.1.8 Some Useful Notes on Matrices
1.
2.
3.
4.
5. if λ is a scalar λ = λ
T
6.
7.
8. If [k] is symmetric then [A] [k][A] is symmetric.
T
9. Proof:
10.
11.
12.
13. If [a] is symmetric, [a] is also symmetric.
–1
14. Proof:
15.
A.2 MATRIX CALCULUS

Let the elements of [a] be functions of a parameter t.
Then, [b] = = [a],t implies bij=

and [b] =∫ [a]dt implies bij =∫aijdt
Example:
386
A.3 SOLUTION OF LINEAR SIMULTANIOUS EQUATIONS

An application of finite element method results in a set of simultaneous equations as
shown below.
In matrix notation,
It is assumed that |K| ≠ 0 and { f } ≠ 0. Under these conditions, the above system is a non-
homogeneous equation system. A non-homogeneous equation system can be solved by
using any of the following methods:
1. Inversion of co-efficient matrix
2. Gauss–elimination method
3. Cramer's rule
4. Gauss–Siedel method
5. Gauss–Jordan method
6. LU decomposition method
7. Jacobi method
The first two methods will be discussed here.
(a) Inversion of co-efficient matrix
By pre-multiplying both sides of the equation [K]{ q} = {f } with the inverse of the coefficient
matrix [K], following equation is obtained.
387
The method of inversion of coefficient matrix can be used for hand calculations of 2 × 2
and 3×3 matrices.
(b) Gauss-elimination method
This method is widely used in many commercially available finite element packages. It is
implemented in the teaching purpose finite element program SFEAP too. The method
involves triangularization of the coefficient matrix [K] and evaluation of the unknowns {q}
by back-substitution starting from the last equation.
Steps
1. Augment the n × n coefficient matrix [Kij] with the right hand side vector {f} to form n
× (n +1) matrix.
2. Inter-change rows if necessary to make the value of K11 the largest magnitude of
any coefficient in the first column.
3. Create zeros in the second through nth rows in the first column by subtracting
Ki1/K11 times the first row from the i-th row (i = 2, ………,n).
4. Repeat for each i = 2 to n − 1:
1. Inter-change rows if necessary to make Kii the largest magnitude in the i-th column
by inter-changing rows (consider rows only i to n).
2. After completion of step 4.a, for each row j = i +1, … n,
At the completion of Step 4, the co-efficient matrix [Kij]n×n is transformed into an

upper triangular matrix.
Solve for qn from the nth equation as qn = Kn.n+1/knn.
Solve for qi (i = n − 1, n – 2, .., 1) from n – 1 through the first equation in turn by
using equation
For computer implementation of Gauss-elimination method, the above steps can be

summarized as
Note: when i > n, summation is interpreted as zero.

Example: Solve the equations given below by employing Gauss elimination method.
388
Step 1: n = 3
Step 2: Not necessary to interchange rows.

Steps 3 and 4:
Steps 5 and 6:
Summary of equations:
Step 1: Choose ranges of i, j, l.
Step 2: Compute updated Kij :
389
For l = 1, i = 2, j = 1, 2, 3, 4:
For l = 1, i = 3, j = 1, 2, 3, 4:
For l = 2, i = 3, j = 2, 3, 4:
Step 3: Compute qi's

i = n = 3:
i = n – 1 = 2:
390
i = n – 2= 1:
A FORTRAN–77 code for this method is given next.
C ROUTINE TO SOLVE [A]{X} = {C} BY USING GAUSS-ELIMINATION
DIMENSION A(10,10), C(10)
READ (5,*)N
READ (5,*) ((A(I,J),J = 1,N),I = 1.N)
READ (5,*)(C(I), I = 1.N)
WRITE(6,5)
5 FORMAT(/1X,’MATRIX-A’/1X,10(‘=’)/)
DO 10 I = 1.N
10 WRITE(6,15)(A(I,J), J = 1, N)
15 FORMAT(10F12.5)
WRITE(6,25)(C(I),I = 1,N)
25 FORMAT(//1X,’VECTOR-C’/1X,10(‘=’)//(10F12.5))
C FORWARD ELIMINATION
391
DO 20 L= 1, N − 1
DO 20 I = L + 1,N
TEMP = A(I,L)/A(L,L)
DO 30 J = L+ 1,N
30 A(I,J) = A(I,J)- TEMP*A(L,J)
20 C(I) = C(I)-TEMP*C(L)
C BACK – SUBSTITUTION…
C(N) = C(N)/A(N,N)
DO 40 II = 1, N–1
I = N-II
SUM = 0.0
DO 50 J = I + 1, N
50 SUM = SUM+A(I,J)*C(J)
40 C(I) = (C(I)-SUM)/A(I,I)
WRITE(6,35)(C(I),I = 1,N)
35 FORMAT(//1X,’SOLUTION VECTOR’ /1X,15(‘=’)//(10F12.5)/)
STOP
392
END
393
Appendix B
ELEMENTS OF CALCULUS OF
VARIATIONS
B.1 GENERAL
Governing equations in engineering are generally second (or fourth) order ordinary or
partial differential equations. This requires the function to be continuous having derivatives
up to order two (or four). Description of a problem is not complete until correct associated
boundary conditions are specified. If there is a function of the functions (functional) for the
governing equation, the governing equations as well as the appropriate boundary
conditions can be derived through the technique of Calculus of Variations (CV). The
functional requires the functions to be continuous having lower order derivatives. Rich
history of the CV of the last 300 years can be found in the book by Goldstine (1980).
One of the major advantages of CV is that it reduces the requirement of the order of
derivatives. In the Finite Element Method, this advantage is exploited to derive a system of
linear algebraic equations (designated as equilibrium equations) for an approximate
solution.
The stiffness matrix for a differential equation was derived using the variational approach
in Example 4.1 of Chapter 4. However, it may be mentioned that if the differential equation
is dispersive (dissipative) as in Example 4.2, no functional can be constructed. In such a
case, resort is taken to one of the weighted residual approach. Galerkin's method is one of
the widely used approaches, which is described in the Appendix C. The variational
technique is briefly described here.
B.2 THE EULER–LAGRANGE EQUATION

For a one-dimensional problem, consider functions whose inputs are functions and whose
outputs are real number as:
where x is an independent variable, u(x) a dependent variable and a prime denotes

differentiation with respect to x. The function u(x) is continuous having derivatives up to
order two and satisfies the boundary conditions
394
Figure B.1 Admissible comparison paths: (a) Single path, and (b) Family of paths
A function like I [u(x)] is actually called a functional, this name is used to distinguish I [u(x)]
from ordinary functions whose domain consist of ordinary variables. Function F(x; u, u′, u″)
in the integral is to be viewed as an ordinary function of the variables x, u, u′ and u″.
It is required to find u(x) such that I [u(x)] is stationary (minimum or maximum).
Let u(x) be the true solution which makes I [u(x)] stationary and U(x) be a comparison or
competing path as shown in Figure B.1(a).
Then U(x) = u(x) + η(x) where the neighboring curve η(x) is continuous having a derivative
up to order two and satisfies the boundary conditions
However, there are infinitely many comparison paths U1(x), U2(x), U3(x),…., etc. as shown
in Figure B.1(b). This dilemma can be circumvented by introducing a real parameter α,
such that
It can be noticed that Eq. (B.4) represents a family of curves, which includes u(x) when α =
0. Thus, a new problem can be posed as
Thus, the problem reduces to finding minimum or maximum (stationary) for a function of a
real variable α. A necessary condition from the calculus of single variable to find
maxima/minima is
After differentiating I(α) and setting α = 0 the sought maxima or minima can be recovered.
Using the chain rule the differentiation of Eq. (B.5) is evaluated as
395
Noting that and setting α = 0
The governing differential equation and consistent boundary conditions can be recovered
by integrating by parts Eq. (B.8). Integrating by parts gives
Substituting Eq. (B.9) in Eq. (B.7), the resulting equation can be written as
1. Euler−Lagrange Equation:
Since η(x) is arbitrary, the integral in Eq. (B.10) yields the governing differential equation
as
2. Boundary Conditions:
The boundary terms in Eq. (B.10) yield the consistent boundary conditions
396
Note that one term from each row has to be specified in the above boundary conditions. If
F(x; u, u′) is not a function of u″, Euler−Lagrange equation is
The boundary conditions are
EXAMPLE B.1:
In pure bending of a prismatic beam having bending rigidity EI the functional for the
bending energy can be written as
where w is the lateral displacement of q is the distributed load.

Find the differential equations and the boundary conditions.
Now where w is the dependent variable. Eq. (B.11) can be

written as
Noting
For boundary conditions it is necessary to specify,
397
B.3 THE VARIATIONAL OPERATOR

The Euler−Lagrange equations were obtained by taking a competing path in the previous
section. In this section, variational calculus is introduced, which is very useful in deriving
formally the governing equations in the finite element analysis.
In the rigid body dynamics, one encounters the variational operator, δ, when the work is
described as δW = F. δr, where δW is the virtual work, F is the force vector and δr is the
virtual displacement. The adjective virtual means not being so in actual fact but being so in
essence or effect.
Here, the variational operator will be used in a broader sense or generalized sense. The
quantity u(x), u(x, y), or u(x, y, z) can be temperature, electric potential, potential flow
distribution, displacement etc., and apply the δ operator to it. It may be noted that the δ
variation can be applied only to the dependent variables and not to the independent
variables. For example, the total variation for u(x, y) can be shown to be
Figure B.2 Differences between ∆h, dh and δh
In order to contrast the differential and variation, consider a one-dimension case of the
function h(x) shown in Figure B.2. At point a the change in h(x), ∆h, resulting from a
change in the independent variable x, ∆x = dx, the differential dh = h′(a)dx, and δh are
shown in the figure. It can be seen that ∆h and δh are a very close approximation but not
the same, δh stands by itself and does not depend on the independent variable x.
The true path, u(x), is now varied by giving a small variation δu (refer to Figure B.1 (a) in
which η(x) is replaced by δu(x)). From Eq. (B.4),
U(x) = u(x) + αη(x) = u(x) + δu, which implies αη(x) = δu.
In the following, a commutative property with respect to differentiation, of the δ operator is

established.
398
The variational principle is now applied to functional I(u) where u is considered as being a
function of two independent variables x and y.
Noting δx = 0 and δy = 0,
and using Green's identity the last term can be written as
where C is the contour of integration.

For stationary value of I, δI = 0, which yields the equilibrium equation and boundary
conditions as
Boundary conditions: .
EXAMPLE B.2
Find the governing differential equation and boundary condition for the functional
399
where
(a) Using Eq. (B.15)
(b) Solving formally:
Note f (x,y) is a function of independent variables only and, therefore, δf = 0.
But
The first integral yields the differential equation: and the second integral after
applying Green's identity, yields: .

Mikhlin, S. G. (1964). Variation Methods in Mathematical Physics, Macmillan Co., New
York.
Goldstine, H. H. (1980). The History of the Calculus of Variations from 17th through 19th
Century, Springer−Verlag, New York.
400
Appendix C
EXAMPLE ILLUSTRATING USE OF
GALERKIN'S METHOD
C.1 GENERAL
Methods of weighted residual are used when differential equations (that describe the
behavior of physical system) are known. Of all the weighted residual methods, Galerkin's
method is employed extensively in the context of finite element analysis. The use of the
method is demonstrated by employing an illustrative example.
C.2 EXAMPLE
A physical phenomenon is described by the differential equation
such that
1. Derive a 2-node line element having length L, by employing Galerkin's method. Assume
approximate solution of the form
where
2. Solve the differential equation by employing (i) two; (ii) three and (iii) four such elements
in each case. Use equal length elements in each case.
Solution:
1. For an i-th generic element having length L, the Galerkin's equations are given from Eq.
(2.9) as
401
Therefore, from Eq. (C.1),
can be obtained. By substituting
and the expression for u into Eq. (C.2), following equations can be derived.
Then for m = 1,
Note that N1 = 1 at x = 0, N1 = 0 at x = L, and u′ = u′1 at x = 0. Hence, from the above

equation,
can be formulated. Similarly, for m = 2, following equation can be derived.
Noting N2 = 0 at x = 0, N2 = 1 at x = L and u′ = u′2 at x = L, equation
402
can be obtained from Eq. (C.7). Therefore, from Eqs. (C.6) and (C.8),
2. (i) Solution of the differential equation by using two elements:
By assembling the element equations, the global equations take form
from which,
4.3334u2 = 0.25, or
u2 = 0.0577.
403
(ii) Solution of the differential equation by using three elements:
By assembling element contributions and enforcing boundary conditions, the global

equations
(iii) Solution of the differential equation by using four elements
404
By assembling element contributions and enforcing boundary conditions, the global

equation set takes form
405
Appendix D
REVIEW OF GAUSS QUADRATURE
PROCEDURE FOR NUMERICAL INTEGRATION
D.1 GENERAL
It was shown in Chapter 7 that integrals, which may need to be evaluated numerically, are
of the form
In finite element analysis, ‘Gauss Quadrature’ procedure is used extensively for numerical
integration.
D.2 FUNDAMENTAL CONCEPT

Consider a simple curve shown in Figure D.1. Application of trapezoidal rule is shown in
Part (a) of the figure to evaluate area under the curve between the limits of a and b. The
area of the trapezoid is given by
where w = b – a = width of the trapezoid. the above equation can be alternatively written
as
By employing trapezoidal rule, the area under the curve between the limits a and b is
approximated as the area of the trapezoid. However, this approximation may give rise to
significant errors, especially when large trapezoids are considered.
By using Gauss Quadrature, on the other hand, instead of choosing A and B at the end of
interval of interest, two intermediate points C and D are chosen, as shown in Part (b) of
Figure D.1. A trapezoid is then constructed by drawing a straight line through these points
and extending it to the end of the interval. Thus, a part of the trapezoid lies outside the
curve (upper corners in the figure), while part of the curve lies outside the trapezoid. By
appropriately choosing points C and D, it is possible to balance the two areas so that the
area in the trapezoid equals the area under the curve. The resulting approximation can
then give exact integration. The Gauss Quadrature rule essentially consists of a simple
way of choosing C and D to get accurate integration. The C and D are termed as Gauss
points, and the integration is sought of the form given by Eq. (D.3). The W1 and W2 in the
equation are commonly referred to as weights associated with Gauss points.
406
Figure D.1 Two-point Gauss rule as derived from trapezoidal rule

In a more general form, if integration I is required by employing m Gauss points, it can be
shown that
where m is the number of Gauss points and Wk indicates weight with the k th Gauss point.
A general m point Gauss Quadrature procedure is sketched in Figure D.2. The locations of
Gauss points and associated weights, for different Gauss Quadrature, on the other hand
are presented on Table D.1.
Note that generally, Gauss Quadrature, which employs n Gauss points, is exact if the
integration is polynomial of degree 2n – 1 or less.
407
Figure D.2 m point Gauss quadrature

Table D.1 Table of Gauss points for integration from –1 to +1
Number of Gauss Points Locations of Gauss Points, ξ i Associated Weights, W i
1. ξ1 = 0 2.00000 00000 00000
2. ξ1,ξ2 = ±0.577350269189626 1.000
3. ξ1,ξ3 = ±0.77459666924148 5/9 = 0.5555555555556
ξ2 = 0.000000000000000 8/9 = 0.8888888888889
4. ξ1,ξ4 = ±0.86113 63115 94053 0.34785 48451 37454
ξ2,ξ3 = ±0.33998 10435 84856 0.65214 51548 62546
5. ξ1,ξ5 = ±0.90617 98459 38664 0.23692 68550 56189
ξ2,ξ4 = ±0.53846 93101 05683 0.47862 86704 99366
ξ3 = ±0.00000 00000 00000 056888 88888 88889
6. ξ1,ξ6 = ±0.93246 95142 03152 0.17132 44923 79170
ξ2,ξ5 = ±0.66120 93864 66265 0.36076 15730 48139
408
ξ3,ξ4 = ±0.23861 91860 83197 0.46791 39345 72691
7. ξ1,ξ7 = ±0.94910 79123 42759 0.12948 49661 68870
ξ2,ξ6 = ±0.74153 11855 99394 0.27970 53914 89277
ξ3,ξ5 = ±0.40584 51513 77397 0.38183 00505 05119
ξ4 = ±0.00000 00000 00000 0.41795 91836 73469
8. ξ1,ξ8 = ±0.96028 98564 97536 0.10122 85362 90376
ξ2,ξ7 = ±0.79666 64774 13627 0.22238 10344 53374
ξ3,ξ6 = ±0.52553 24099 16329 0.31370 66458 77887
ξ4,ξ5 = ±0.18343 46424 95650 0.36268 37833 78362
9. ξ1,ξ9 = ±0.96816 02395 07626 0.08127 43883 61574
ξ2,ξ8 = ±0.83603 11073 26636 0.18064 81606 94857
ξ3,ξ7 = ±0.61337 14327 00590 0.26061 06964 02935
ξ4,ξ6 = ±0.32425 34234 03809 0.31234 70770 40003
ξ5 = ±0.00000 00000 00000 0.33023 93550 01260
10. ξ1,ξ10 = ±0.97390 65285 17172 0.06667 13443 08688
ξ2,ξ9 = ±0.86506 33666 88985 0.14945 13491 50581
ξ3,ξ8 = ±0.67940 95682 99024 0.21908 63625 15982
ξ4,ξ7 = ±0.43339 53941 29247 0.26926 67193 09996
409
ξ5,ξ6 = ±0.14887 43389 81631 0.29552 42247 14753
Example
Evaluate
by using 1, 2 and 3 Gauss points.
Solution:
Exact value of I is
In order to use numerical integration, change of variable is required.
Therefore, for the given data,
Thus,
and
(i) 1– Gauss point (m = 1)
410
(ii) 2 – Gauss points (m = 2)
(iii) 3 – Gauss points (m = 3)
D.3 GAUSS QUADRATURE FOR TWO DIMENSION

Gauss quadrature rule was summarized in the previous section for line integration. It is
extended for area integration in this section.
Consider
By applying Gauss Quadrature for line integration, it can be shown that
Here, m and n are, respectively, the number of Gauss points in the ξ and η directions;
(ξk,η1) are the co-ordinates of Gauss points in the ξ–η plane; and Wk and Wl are Gauss
weights in ξ and η directions, respectively. Depending upon the number of Gauss points in
411
the ξ and η directions, various Gauss quadrature rules can be established (refer to Figure
D.3).
(i) 1 × 1 Gauss Quadrature:
From Eq. (D.6), for 1×1 Gauss Quadrature,
Figure D.3 Different Gauss quadrature rules for area integration

(ii) 2 × 2 Gauss Quadrature:
From Eq. (D.6) and Table D.1, the 2 × 2 Gauss Quadrature can be written as
(iii) 3 × 3 Gauss Quadrature:

From Eq. (D.6) and Table D.1, the 3×3 Gauss Quadrature can be written as
412
413
Appendix E
USER'S MANUAL FOR THE SIMPLIFIED
FINITE ELEMENT ANALYSIS
PROGRAM (SFEAP)
E.1 PREAMBLE
The simplified finite element analysis program (sfeap) can be used for static linear-elastic
analysis of plane and space trusses, beam and plane frames, two- and three-dimensional
problems and axisymmetric solids. There are eight element subroutines, which are listed
in Table E.1. They are numbered as ELMT1 to ELMT8 for referencing. The program was
coded using the theory presented in the text. A copy of the program in PDF format is given
at the end of this appendix and the executable and the FORTRAN versions of the program
are included in the CD.
The first three subroutines employ one-dimensional element; ELMT1 and ELMT2 use truss
elements while ELMT3 employs beam elements. The next two subroutines use two-
dimensional elements in rectangular Cartesian co-ordinates; ELMT4 uses 3-node constant
strain triangles and ELMT5 employs three to nine nodes isoparametric elements. ELMT6
and ELMT7 use two-dimensional, three to nine node isoparametric elements in cylindrical
(r, z) co-ordinates. The subroutine ELMT8 employs 8-noded, three-dimensional brick
elements.
Table E.1 List of Programs
Program Program
Number
ELMT1 Plane truss element
ELMT2 3 - D truss element
ELMT3 Plane beam/frame element
ELMT4 CST element for plane stress/strain problem
ELMT5 3 - 9 variable nodes isoparametric plane element
ELMT6 3 - 9 variable nodes isoparametric element for axisymmetric analysis

m=0
414
ELMT7 3 - 9 variable nodes isoparametric element for axisymmetric analysis

m>0
ELMT8 8 node brick element for 3-D stress analysis
ELMT9 ′Dummy subroutine′ for the user to write
ELMT10 ′Dummy subroutine′ for the user to write
E.2 PROGRAM LAYOUT

Program sfeap is organized as follows:
* SUBROUTINE (ICN=2) IS ACTIVE ONLY IN ELMT3, ELMT5, ELMT6 AND

DISTRL ELMT7.
** SUBROUTINE (ICN=4) IS CALLED BY EVERY ELEMENT MODULE. AT

STRESS PRESENT WE HAVE TO INPUT LS IN ELMT5, ELMT6 AND
ELMT7.
***SUBROUINE IS CALLED BY EVERY ELEMENT MODULE TO CALCULATE

RLOAD EQUIVALENT NODAL FORCES DUE TO SINKING SUPPORTS.
415
MAIN ROUTINE
FUNCTION
MAIN PROGRAM Reads control information and evaluates the space for arrays.
PROCESS Allocates Dynamic Memories to the arrays.
BANDW Computes the half bandwidth.
LOAD Read and write the concentrated nodal load data and compute the
global load vector stemming from the concentrated loads.
DISTRL Compute and assemble the equivalent, elemental load vectors due
to the distributed loads (if any).
ASSEMB Assemble element matrices.
BANSOL Solve the global equation by using Gauss elimination method.
DISP OR DISP1 Print the computed nodal displacements (post-processing).
STRESS Compute and print the ‘secondary’ unknowns i.e. compute element
stresses/forces (post-processing).
E.2.1 Global Variables and Arrays

Following FORTRAN variables and arrays are declared in the main program and process.
INTEGER VARIABLES
NUMNP Total no. of nodes
NUMEL Total no. of elements
NDF No. of nodal DOF (degrees of freedom 1 / 2 / 3)
NDM No. of dimension (1 / 2 / 3)
416
MNEL Maximum no. of nodes per element
NUMAT No. of material sets
NSN No. of supported nodes
NPROP No. of material data
ARRAYS
The FORTRAN program uses dynamic memory allocation. To achieve this feature
following integers are defined in the Main Program and passed on to the program Process.
MST=MNEL*NDF
NST=MST
NEQ2=NDF*NUMNP
N1=1
I2=1
N4=N1
N5=N1+NDM*NUMNP
I3=I2+NUMEL
I4=I3+NUMEL
I6=I4+MNEL*NUMEL
N8=N5+NPROP*NUMAT
I7=I6+NDF*NUMNP
N9=N8+NDF*NUMNP
N10=N9+MST*MST
N11=N10+NDM*MNEL
417
N12=N11+NDF*MNEL
N13=N12+MST
N14=N13+NEQ2
I14=I7+NUMAT
N15=N14+NEQ2*NEQ2
N16=N15+3*NUMEL
NTOT=N16+NEQ2+1000
IATOT= I14
Integer Arrays
IA(I2) = Stores material set numbers of each element

NELMAT(NUMEL)
IA(I3) = NELND(NUMEL) Stores no. of nodes per each element
IA(I4) = Stores element nodal connectivity data (i.e. global node

NP(MNEL,NUMEL) numbers corresponding to the local node numbers of
each element.)
IA(I6) = ID(NDF,NUMNP) Stores ID number or location number of each nodal DOF
IA(I7) = MEL(NUMAT) Stores element type number corresponding to each

material set
IA(I14) = NLD(NUMEL) Stores the element numbers which have distributed loads
Real Arrays (in double precision)
A(N1) = X(NDM,NUMNP) Stores global co-ordinates of all the nodes
418
A(N5) = Stores material data for each material set

D(NPROP,NUMAT)
A(N8) = U(NDF,NUMNP) Stores global displacements of each node
A(N9) = SL(MST*MST) Stores element stiffness matrix [k ]

e
A(N10) = XL(NDM,MNEL) Stores global co-ordinates of nodes of an element
A(N11) = UL(NDF,MNEL) Stores elemental global displacement component {q } e
A(N12) = RL(MST) Stores elemental level global load components {f }e
A(N13) = R(NEQ2) Stores global load components {F}
A(N14) = S(NEQ2,NEQ2) Stores global stiffness matrix, [K], in the banded form
A(N15) = STR(3,NUMEL) Stores the intensities of the distributed loads at the

nodes
E.2.2 Notes on Element Subroutines ELMTn

All the element subroutines have following common structures:
SUBROUTINE ELMTn(DL,NEL,XL,SL,UL,RL,ICN)
IMPLICIT REAL*8 (A-H, O-Z)
COMMON/BLK1/NUMNP,NUMEL,NDF,NDM,MNEL,NUMAT,NSN,NEQ,MBAND,N
PROP
COMMON/BLK3/MA,IELB,IELG,NST,NXSJ,MST
DIMENSION XL(NDM,1),SL(NST,1),DL(1),A(4)UL(1),RL(1)
GO TO (1, 2, 3, 4), ICN
1. This section (ICN=1) is activated by the subroutine READG to read all material
characteristics data that would be used at a later stage to compute stiffness matrix,
stress etc. Data generated in this section may be stored in the array
DL(NPROP,NUMAT).
2. This section (ICN=2) is activated by the subroutine DISTRL to calculate the
consistent nodal load vector RL(MST) due to traction forces, if any, acting on the
419
surface of an element. The intensity of the distributed load is brought in by the

array UL.
3. In this section (ICN=3), the element stiffness matrix SL(MST,MST) is generated by
using the data available in the array DL and XL.
4. The stresses (or stress resultants) are calculated in this section (ICN=4) by using
the data available in DL, XL, and UL.
NOTES:
Only the arrays DL, SL and RL are variables to be generated by this element subroutine.
Other arrays and parameters in the argument list are already specified and must not be
redefined by this subroutine. Array DL should also not be redefined for ICN > 1.
1. If a section is not required, it can be omitted by putting a RETURN statement in
that section.
Variables and arrays in the ELMTn subroutine
Following FORTRAN variables and arrays are used in an element subroutine ELMT n.
NUMNP No. of nodal points
NUMEL No. of elements
NDF No. of degrees of freedom per node
NDM No. of spatial dimension
MNEL Maximum no. of nodes per element
NUMAT No. of material characteristic sets
NSN No. of supported nodes
NEQ Total no. of equations for the entire structure
MBAND Half band width
NPROP Maximum number of material characteristics data associated with the element
type
MA Material characteristic number
420
IELB Element type number (in the LIBRARY)
IELG Current element number (in the entire structure)
NST Total number of DOF's in an element NST =NDF*NEL
NXSJ This is a switch. Its value is set to one if negative Jacobian is encountered.
Otherwise it is zero.
NEL No. of nodes actually connected to the current element
ISWICH This is a logical switch. If it is set. TRUE. in the main program, the Jacobian of
each element will be printed (can be used to debug data.)
DL Local array of material data. Information stored is used later to compute

stiffness, stresses etc.
XL Local array of global co-ordinates of element nodes
SL Local array to store element stiffness matrix. This array is initialized to zero
before the element subroutine is called.
UL Local array of displacements of element nodes in global axes. This array is

also used to pass on the distributed load data to the element subroutine.
RL Local array of equivalent nodal load vector. This array is initialized to zero
before the element subroutine is called.
ICN Switch used to direct control to appropriate part of the element subroutine. Its
value is automatically fixed depending upon the various level of the program.
E.2.3 Input Instructions

Sample examples for each element along with the input data are given in the Example
section followed by the listing of the FORTRAN source code.
Following input data are required for the simplified finite element analysis (sfeap). Format
to input the data are given in the parenthesis for each line. All the ‘integer’ input data are in
format (I5) whereas data for all the ‘real’ variables are in format (E10.0).
A. TITLE (15A4)
421
Columns 1 − 60 Title to be printed with the output.
B. CONTROL (815)
INFORMATION
Column Variables
1–5 NUMNP No. of nodes
6–10 NUMEL No. of elements
11–15 NDF No. of degrees of freedom per node
16–20 NDM No. of dimensions
21–25 MNEL Maximum no. of nodes per element
26–30 NUMAT No. of material sets
31–35 NSN No. of supported nodes (no. of nodes at which at least one
DOF is prescribed.)
36–40 NPROP No. of material data
C. CO-ORDINATES (one line per node) (2I5, 3E10.0) (2I5, 3E10.0)
Columns
1–5 Node number
6–10 Generation increment
11–20 X1* coordinate
422
*(X1, X2) = (X, Y) for programs ELMT1, ELMT3, ELMT4, ELMT5; (X1, X2, X3) = (X, Y, Z)
for ELMT2 and ELMT8 and (X1, X2) = (R, Z) for ELMT6 and ELMT7.
D. ELEMENT DATA (one line (13I5)

per node)
Columns
1–5 Element number
6–10 Generation increments for nodes (Element numbers will

be increased by one.)
11–15 Material set number
16–20 No. of nodes in the element
21–25 Node – 1 number
423
E. BOUNDRY CONDITIONS (one line per restrained node) (4I5)
Columns
1–5 Node number
6–10 DOF – 1 boundary code
Boundary code: 1 – restrained

0 – free
F. NODAL DISPLACEMENT DATA (one line per node) (I5,5x,3E10.0)
Columns
1–5 Node number
11–20 Displacement – 1
Terminate with a blank line

Note that the input displacement is ignored if the corresponding boundary code is zero.
G. MATERIAL PROPERTY (CHARACTERISTIC) DATA (2I5)
Columns
1–5 Material set number
5–10 Element type number
424
Each material set input line must be followed immediately by the material property
data required for the element type. Refer to the Input/output information for
elements section.
Any change in the material property data will require a new material set number. Thus, the
total number of material sets, NUMAT, has to be properly estimated.
H. NODAL LOAD DATA (one line per loaded node) (I5,5x,3E10.0)
Columns
1–5 Node number
11–20 Force – 1
21–30 Force – 2
31–40 Force – 3
Terminate with a blank line.
I. DISTRIBUTED LOAD DATA (one line per loaded element) (I5,5x,3E10.0)
Columns
1–5 Node number
11–20 Intensity – 1
Terminate with a blank line.
At present, the program can handle only normal distributed load for the beam element of
ELMT3, four node isoparametric elements of ELMT5, and four or higher than five nodes
elements for ELMT6 and ELMT7. Variation of the load for ELMT5 is assumed to be linear
acting on the edge ξ, of (ξ, η) or s = +1 of (s,t). For ELMT6 and ELMT7, intensities of the
normal stress at the local nodes 2 and 3 (and also node 6 if the element is quadratic) are
input.
E.2.3.1 Values of the variables in the ELMT n subroutine:
425
E.2.3.2 Input/output information for elements:

ELMT1: Plane Truss Element
MATERIAL DATA (one for each material set) (2E10.0)
Columns
1–10E Young's modulus
11–20 AR Area of cross-section
OUTPUT 1. Displacements in global axes.
2. Axial load P (stress resultant)
- tension if the sign is positive
- compression if the sign is negative.
ELMT2: 3-D Truss Element
Columns
1–10 E Young's modulus
426
OUTPUT 1. Displacements in global axes.
2. Axial load P (stress resultant)
- tension if the sign is positive
- compression if the sign is negative.
ELMT3: Plane Beam/Frame Element
MATERIAL DATA (one for each material (3E10.0,2I5)

set)
Columns
21–30 AI Second moment of area (moment of

inertia)
31–35 IH Release code* of hinge at end – 1
36–40 JH Release code of hinge at end – 2
* code = 1 : Release; code = 0: Fixed
1. Deflections in global axes.

[U1 V1 θ1 U2 V2 θ2] T
2. Stress resultants in local axes

[P1 V1 M1 P2 V2 M2] T
427
ELMT4: CST Element for Plane Stress/Strain Problem
MATERIAL DATA (one for each (3E10.0, I5)

material set)
Columns
11–20 ANU Poisson's ratio
21–30 TH Thickness
31–35 KODE KODE equal to zero for plane strain, 1 for plane
stress
OUTPUT: 1. Displacements in the global axes.
2. Stresses at the center. Note that the stresses are constant throughout an
element.
ELMT5: 3 – 9 Variable Nodes Isoparametric Plane Element
MATERIAL DATA (one for each (3E10.0,4 I5)

material set)
Columns
21–30 TH Thickness
428
31–35 KODE KODE equal to zero for plane strain, 1 for plane
stress
36–40 L Order of quadrature for stiffness
41–45 LS No. of stress points in each direction
46–50 LLD Order of quadrature for consistent loads
2. Stresses at the Gauss points.
ELMT6: 3 − 9 Variable Nodes Isoparametric Element for Axisymmetric Analysis m = 0
MATERIAL DATA (one for each material set) (2E10.0,3 I5)
Columns
ELMT7: 3 − 9 Variable Nodes Isoparametric Element for Axisymmetric Analysis m > 0
429
MATERIAL DATA (one for each material (2E10.0,4 I5)

set)
Columns
36–40 KODE* KODE is 1 for ‘core’ element, and zero

otherwise.
* When KODE is 1, the program assumes that the nodes of a core element are numbered
in such a way that the local nodes 1 and 4 (and also node 8 if applicable) are located on
the z– axis.
ELMT8: 8 Node Brick Element for 3 –H D Stress Analysis
Columns
430
E.3.3 EXAMPLES
Example 1
Analyze the plane truss shown in the figure above. The number in parentheses beside
each member indicates the cross-sectional area in square centimeters (cm2). The truss is
subjected to concentrated loads at joints E and F as shown. Moreover, the support at C
sinks by 0.36 cm. Assume Young's modulus, E, for all members as 20,000 kN/cm2.
encircled Arabic numbers.
Input Data
431
Example 2
Analyze the space truss shown in the figure below. The truss is subjected to a
concentrated load at joint A as shown. Assume for all members, Young's modulus, E, as
200 GPa 5 200*106 kN / m2 and area, A, as 0.03 m2.
432
Input Data
Example 3(a)
Analyze the plane frame shown in the figure below. The frame has simple supports at
nodes 1 and 6. A load of P = 100 kN is acting at node 4. For each element (member)
assume Young's modulus, E = 200 GPa, Area, A = 0.01 m2, and Moment of Inertia, I = 2 ×
10–4 m4.
433
Input Data
Example 3(b)
Analyze the plane frame shown in the figure below. The finite element representation of
the problem is shown in the figure. Node numbers are indicated by Arabic numbers, while
element numbers are denoted by encircled Arabic numbers.
The frame is fixed at nodes 1 and 3.
434
A load of 5 kN/m is acting on member 1. For each element (member) assume Young's
modulus, E = 200 GPa, Area, A = 0.01 m2, and Moment of Inertia, I = 2.0×10–4 m4.
Input Data
Example 4:
The figure below shows a 16 element model of a rectangular steel plate with one side fixed
and the opposite side under uniformly distributed tensile load of 1,000 N/m. Assume
Young's modulus, E as 20.0E + 06 N/m2, Poisson's ratio, v as 0.3 and thickness as 0.1 m.
Perform plane stress analysis.
435
Input Data
436
Example 5:
A cantilever beam of span 9 m is subjected to a tip load of 4 kN. Analyze the beam by
employing 4-node elements under plane stress condition. Assume Young's modulus, E as
2,000 kN/m2, Poisson's ratio, v as 0.2 and width as 1 m.
Input Data
437
Example 6:
A cylinder, shown in the figure, is subjected to uniform external pressure σrr = 1 GPa.
Assume Young's modulus E = 200 GPa, Poisson's ratio v = 0.2, and radii a = 0.5 m and b
= 1.0 m.
The finite element representation of the problem is also shown in the figure above. Node
The problem is purely axisymmetric and involves only one harmonic, m = 0. The rigid body
displacement in the z-direction is restrained at the nodes on the face z = 2.
Input Data
438
Example 7:
A solid cylinder, shown in the figure, is subjected to external pressure of form
where p0 = 1 GPa. Assume Young's modulus E =20 GPa, Poisson's

ratio v = 0.2, and radii a = 0.0 m and b = 1.0 m.
439
The finite element representation of the problem is also shown in the figure above. Node
The problem involves one harmonic, m = 1. The rigid body displacement in the z −
direction is restrained at the nodes on the face z = 2. Note that one side of elements 1
and 3 are on z − axis.
Input Data
Example 8:
A cantilever beam of span 120 cm is subjected to an end load of 40 kN as shown in the
figure below. Analyze the beam by employing 8 − node brick elements. The beam is 10 cm
deep and 2 cm wide. Assume for all members, Young's modulus, E, as 20,000 kN / cm2
and Poisson's ratio v = 0.3.
440
Input Data
441
E.3.4 Listing of SFEAP

C PROGRAM SFEAP
C Simplified Finite Element Analysis Program (SFEAP)
C
C Program SFEAP has been developed jointly by
C Drs. R. Paskaramoorthy, W. M. Karunasena, Yogesh Desai &
C Prof. Arvind Shah, Dept. of Civil Eng.
C corrected version January 8, 2000
C
C Version02 completed on October 15, 2008.
C Version02: In this version three variables, NNDS, NLN & NELDL
442
C are added to simplify making frames for Visual Basic.dotnet.

C Also the output is split in two - output and results.
C Trying to make ONE D arrays and no prior dimensioning.
C Revised this program on October 27, 2008 in working condition.
CC
Version03: Added 8node brick element as ELMT08
Q **************************************
C * *
C * SFEP: SIMPLIFIED FINITE ELEMENT PROGRAM FOR STATIC *
C * ANALYSIS *
C * CONTROL INFORMATION *
C * NUMNP = NO. OF NODES *
C * NUMEL = NO. OF ELEMENTS *
C * NDF = NO. OF DEGREES OF FREEDOM *
C * NDM = NO. OF DIMENSIONS *
C * MNEL = MAXM. NO. OF NODES/ELEMENT *
C * NUMAT = NO. OF MATERIAL SETS *
C * NSN = NO. OF SUPPORTED NODES *
C * NPROP = NO .OF MATERIAL DATA *
C * MST = MNEL*NDF *
C * NEQ2 = NDF*NUMNP *
C * *
C * ALL ARRAYS RESIDE IN THE ONE DIMENSIONAL ARRAY 'A' *
C * THEIR LOCATIONS ARE GIVEN BELOW *
C * A(NI) = X(NDM,NUMNP) *
C * IA(I2) = NELMAT(NUMEL) *
C * IA(I3) = NELND(NUMEL) *
C * IA(I4) = NP(MNEL,NUMEL) *
C * A(N5) = D(NPROP,NUMAT) *
C * IA(I6) = ID(NDF,NUMNP) *
C * IA(I7) = MEL(NUMAT) *
C * A(N8) = U(NDF,NUMNP) *
C * A(N9) = SL(MST*MST) *
C * A(N10) = XL(NDM,MNEL) *
C * A(N11) = UL(NDF,MNEL) *
443
C * A(N12) = RL(MST) *
C * A(N13) = R(NEQ2) *
C * IA(I14) = NLD(NUMEL) *
C * A(N14) = S(NEQ2,NEQ2) *
C * A(N15) = STR(3,NUMEL) *
C * ISWICH =.TRUE. THEN *
C * ACTIVATES,PRINTING OF JACOBIAN OF EACH ELEMENT *
C * corrected versión January 8, 2000 *
Q *************************************************
IMPLICIT REAL*8 (A–H,O–Z)
COMMON/BLK1/NUMNP,NUMEL,NDF,NDM,MNEL,NUMAT,NSN,NEQ,MBAND,NPROP
COMMON/BLK2/ISWICH
COMMON/BLK5/OM,NH
COMMON/BLK6/NNDS, NLN, NELDL
COMMON/BLK7/ N1,I2,N4,N5,I3,I4,I6,N8,I7,N9,N10,
*N11,N12,N13,N14,I14,N15,N16,N17,NTOT,IATOT
LOGICAL ISWICH
CHARACTER*4 HED(15)
ISWICH=.FALSE.
NH=0
C
C….READ & WRITE CONTROL INFORMATION
C
OPEN(5,FILE= ‘input.dat’)
OPEN(6,FILE=‘output.dat’)
OPEN(11,FILE=‘result.dat’)
READ(5,2) HED,NUMNP,NUMEL,NDF,NDM,MNEL,NUMAT,NSN,NPROP
2 FORMAT(15A4/9I5)
WRITE(6,3) HED,NUMNP,NUMEL,NDF,NDM,MNEL,NUMAT,NSN,NPROP
3 FORMAT(′′,15A4//
1 ‘ NO. OF NODES………………………………=’,I4/
2 ‘ NO. OF ELEMENTS…………………………=’,I4/
3 ‘ NO. OF DEGREES OF FREEDOM PER NODE…………………………=’,I4/
444
4 ‘ NO. OF DIMENSIONS…………………………=’,I4/
5 ‘ MAXM. NO. OF NODES/ELEMENT………………………………..=’,I4/
6 ‘ NO. OF MATERIAL SETS…………………………=’,I4/
7 ‘ NO. OF SUPPORTED NODES…………………………=’,I4/
8 ‘ NO. OF MATERIAL DATA…………………………=’,I4/
C
C…. SET POINTERS FOR ALLOCATION OF DATA ARRAYS
C
MST=MNEL*NDF
NST=MST
C
NEQ2=NDF*NUMNP
C
N1=1
I2=1
N4=N1
N5=N1+NDM*NUMNP
I3=I2+NUMEL
I4=I3+NUMEL
I6=I4+MNEL*NUMEL
N8=N5+NPROP*NUMAT
I7=I6+NDF*NUMNP
N9=N8+NDF*NUMNP
N10=N9+MST*MST
N11=N10+NDM*MNEL
N12=N11+NDF*MNEL
N13=N12+MST
N14=N13+NEQ2
I14=I7+NUMAT
N15=N14+NEQ2*NEQ2
N16=N15+3*NUMEL
NTOT=N16+NEQ2+1000
IATOT= I14
CALL PROCESS
CLOSE (5)
445
CLOSE (6)
CLOSE(11)
END
C****************************************************************
SUBROUTINE PROCESS
IMPLICIT REAL*8 (A-H,O-Z)
COMMON/BLK2/ISWICH
COMMON/BLK5/OM,NH
COMMON/BLK6/NNDS, NLN, NELDL
COMMON/BLK7/ N1,I2,N4,N5,I3,I4,I6,N8,I7,N9,N10,
*N11,N12,N13,N14,I14,N15,N16,N17,NTOT,IATOT
LOGICAL ISWICH
REAL*8, POINTER :: A(:)
INTEGER*4, POINTER :: IA(:)
INTEGER*4 ERROR
ALLOCATE(A(NTOT),STAT=ERROR)
IF (ERROR.NE.0) THEN
PRINT*, “ERROR, A NOT ALLOCATED. ABORTING.”
PRINT*, “ERROR=”,ERROR
STOP
END IF
ALLOCATE(IA(IATOT),STAT=ERROR)
IF (ERROR.NE.0) THEN
PRINT*, “ERROR, IA NOT ALLOCATED. ABORTING.”
PRINT*, “ERROR=”,ERROR
STOP
END IF
ISWICH=.FALSE.
CALL READG(A(N1),IA(I2),IA(I3),IA(I4),A(N5),IA(I6),IA(I7),A(N8),
*A(N10),A(N9),A(N11),A(N12))
C SUBROUTINE READG(X,NELMAT,NELND,NP,D,ID,MEL,U,XL,SL,UL,RL)
CALL BANDW(IA(I3),IA(I4),MBAND)
446
C SUBROUTINE BANDW(NELND,NP,MB)
N14=N13+NEQ
CALL LOAD(A(N13),IA(I6),A(N12))
C SUBROUTINE LOAD(R,ID,RL)
I14=I7+NUMAT
N15=N14+NEQ*MBAND
N16=N15+3*NUMEL
CALL DISTRL(A(N1),IA(I2),IA(I3),IA(I4),A(N5),A(N9),
*A(N10),A(N12),A(N13),IA(I6),IA(I7),IA(I14),A(N15))
C SUBROUTINE DISTRL(X,NELMAT,NELND,NP,D,SL,XL,RL,R,ID,MEL,NLD,STR)
N17=N14+NEQ*MBAND
CALL ASSEMB(A(N1),IA(I2),IA(I3),IA(I4),A(N5),A(N9),
*A(N10),A(N12),A(N14),A(N13),IA(I6),IA(I7),A(N11),A(N8))
C SUBROUTINE ASSEMB(X,NELMAT,NELND,NP,D,SL,XL,RL,S,R,ID,MEL,UL,U)
CALL BANSOL(A(N14),A(N13),NEQ,MBAND)
C SUBROUTINE BANSOL(S,R,NEQ,MBAND)
IF(NH.NE.1) CALL DISP(A(N13),A(N8),IA(I6))
C SUBROUTINE DISP(R,U,ID)
C
IF(NH.EQ.1) CALL DISP1(A(N13),A(N8),IA(I6),A(N1))
C SUBROUTINE DISP1(R,U,ID,X)
C
CALL STRESS(A(N1),IA(I2),IA(I3),IA(I4),A(N5),A(N8),A(N10), 1
A(N11),IA(I7),A(N12),A(N9))
C SUBROUTINE STRESS(X,NELMAT,NELND,NP,D,U,XL,UL,MEL,RL,SL)
END
C******************************************************************
SUBROUTINE READG(X,NELMAT,NELND,NP,D,ID,MEL,U,XL,SL,UL,RL)
C
C…. SUBROUTINE TO GENERATE THE MISSING DATA
C
DIMENSION X(NDM,1),NELMAT(1),NELND(1),NP(MNEL,1),XL(1),U(NDF,1), 1
D(NPROP,1),ID(NDF,1),MEL(1),SL(1),UL(1),RL(1)
447
C
C…. READ & WRITE NODE DATA
C
WRITE(6,280)
280 FORMAT(//10H NODE DATA//
1 5H NODE,11X,5H1-ORD,11X,5H2-ORD,11X,5H3-ORD)
C
NOLD=0
NXOLD=0
10 READ (5,20) N,NX,(X(I,N),I=1,NDM)
IF (NXOLD.EQ.0) GO TO 40
NUMINC=(N-NOLD)/NXOLD
IF(NUMINC.LE.1) GO TO 40
XNUM=NUMINC
NUMINC=NUMINC-1
DO 30 I=1,NDM
NN=NOLD
ST=(X(I,N)-X(I,NOLD))/XNUM
DO 30 K=1,NUMINC
NN=NN+NXOLD
30 X(I,NN)=X(I,NOLD)+K*ST
40 NOLD=N
NXOLD=NX
IF(N.EQ.NUMNP) GO TO 300
GO TO 10
20 FORMAT(2I5,3E10.0)
C
300 DO 200 J=1,NUMNP
WRITE(6,210) J,(X(I,J),I=1,NDM)
200 CONTINUE
210 FORMAT(I5,3F16.3)
C
C…. READ & WRITE ELEMENT DATA
C
WRITE(6,240)
448
MOLD=0
MXOLD=0
50 READ(5,60) M,MX,NELMAT(M),NEL,(NP(I,M),I=1,MNEL)
NELND(M)=NEL
IF(MXOLD.EQ.0) GO TO 70
NUMINC=(M-MOLD)
IF(NUMINC.LE.1) GO TO 70
NUMINC=NUMINC-1
DO 80 K=1,NUMINC
KK=K+MOLD
NELMAT(KK)=NELMAT(MOLD)
NELND(KK)=NELND(MOLD)
DO 80 I=1,NEL
80 NP(I,KK)=NP(I,MOLD)+MXOLD*K
70 MOLD=M
MXOLD=MX
IF(M.EQ.NUMEL) GO TO 310
GO TO 50
60 FORMAT(13I5)
C
310 DO 220 J=1,NUMEL
NEL=NELND(J)
WRITE(6,230) J,NELMAT(J),NELND(J),(NP(I,J),I=1,NEL)
220 CONTINUE
230 FORMAT(I5,I9,I5,1X,9(1X,I5))
240 FORMAT(///' ELEMENT DATA'//' ELMT MATERIAL NEL NOD-1 NOD-2', *1X,'NOD-3
NOD-4 NOD-5 NOD-6 NOD-7 NOD-8 NOD-9')
C
C…. READ & WRITE BOUNDARY CONDITIONS
C
DO 90 I=1,NDF
DO 90 J=1,NUMNP
U(I,J)=0.0
90 ID(I,J)=0
WRITE(6,120)
449
DO 130 J=1,NSN
READ(5,100) N,(ID(I,N),I=1,NDF)
WRITE(6,110) N,(ID(I,N),I=1,NDF)
130 CONTINUE
110 FORMAT(I5,6I8)
100 FORMAT(7I5)
120 FORMAT(//' BOUNDARY CONDITION CODES'// 1 1X,'NODE',3X,'1-STF',3X,'2-
STF',3X,'3-STF')
C
C…. GENERATE ID NO FOR THE ENTIRE STRUCTURE
C
NEQ=0
DO 161 J=1,NUMNP
DO 163 I=1,NDF
IF(ID(I,J).EQ.0) GO TO 162
ID(I,J)=0
GO TO 163
162 NEQ=NEQ+1
ID(I,J)=NEQ
163 CONTINUE
161 CONTINUE
C
C…. READ & WRITE THE NODAL DISPLACEMENTS IF SPECIFIED
C
NN=0
18 READ(5,15) N,(UL(I),I=1,NDF)
IF(N.EQ.0) GO TO 16
NN=NN+1
IF(NN.EQ.1) WRITE(6,21)
WRITE(6,22) N,(UL(I),I=1,NDF)
DO 17 I=1,NDF
17 U(I,N)=UL(I)
GO TO 18
15 FORMAT(I5,5X,3E10.0)
22 FORMAT(I5,3E15.4)
450
21 FORMAT(//' NODAL DISPLACEMENT DATA'//

* 1X,'NODE',9X,'1-DISP',9X,'2-DISP',9X,'3-DISP')
C
C….READ & WRITE MATERIAL DATA
C
16 DO 11 I=1,NPROP
DO 11 J=1,NUMAT
11 D(I,J)=0.0
C
DO 12 J=1,NUMAT
READ(5,13) MA,IELB
13 FORMAT(2I5)
MEL(MA)=IELB
WRITE(6,14) MA,IELB
14 FORMAT(///' MATERIAL SET NUMBER',I5,9X,' ELMT LIB NUMBER',I5/)
CALL ELMLIB(D(1,MA),NEL,XL,SL,UL,RL,1,IELB)
12 CONTINUE
RETURN
END
C*******************************************************************
SUBROUTINE BANDW(NELND,NP,MB)
C
C…. THIS SUBROUTINE CALCULATES THE HALF BAND WIDTH
C
DIMENSION NELND(1),NP(MNEL,1)
C
MB=0
DO 10 J=1,NUMEL
NEL=NELND(J)
NLAR=NP(1,J)
NSMA=NP(1,J)
DO 20 I=1,NEL
451
IF(NSMA.GT.NP(I,J)) NSMA=NP(I,J)
IF(NLAR.LT.NP(I,J)) NLAR=NP(I,J)
20 CONTINUE
MM=(NLAR−NSMA+1)*NDF
IF(MB.GT.MM) GO TO 10
MB=MM
MEL=J
10 CONTINUE
WRITE(6, 31) NEQ
31 FORMAT(/'NUMBER OF EQUATION, NEQ =' I5)
WRITE(6,30) MB,MEL
30 FORMAT(/' HALF BAND WIDTH =',I5,5X,' ELMT NUMBER =',I5/)
RETURN
END
C******************************************************************
SUBROUTINE LOAD(R,ID,RL)
DIMENSION R(1),RL(1),ID(NDF,1)
C
C…. INITIALIZE GLOBAL LOAD VECTOR
C
DO 10 I=1,NEQ
10 R(I)=0.0
C
C…. READ & WRITE NODAL LOAD DATA AND FORM GLOBAL LOAD VECTOR
C
NN=0
40 READ(5,50) N,(RL(I),I=1,NDF)
IF(N.EQ.0) GO TO 70
NN=NN+1
WRITE(6,60) N,(RL(I),I=1,NDF)
DO 80 I=1,NDF
452
II=ID(I,N)
IF(II.EQ.0) GO TO 80
R(II)=R(II)+RL(I)
80 CONTINUE
GO TO 40
50 FORMAT(I5,5X,3E10.0)
60 FORMAT(I5,6X,3E16.4)
20 FORMAT(//' NODAL LOAD DATA'//
1 5H NODE,16X,6H1-LOAD,10X,6H2-LOAD,10X,6H3-LOAD)
C
70 RETURN
END
C****************************************************************
SUBROUTINE DISTRL(X,NELMAT,NELND,NP,D,SL,XL,RL,R,ID,MEL,NLD,STR)
COMMON/BLK4/ICOUNT
DIMENSION X(NDM,1),NELMAT(1),NELND(1),NP(MNEL,1),D(NPROP,1),
*SL(1),XL(NDM,1),ID(NDF,1),LM(27),RL(1),R(1),MEL(1),NLD(1),STR(3,1)
C
NN=0
200 READ(5,10) IELG,P1,P2,P3
10 FORMAT(I5,5X,3E10.0)
IF(IELG.EQ.0) GO TO 100
NN=NN+1
15 FORMAT(/1X,'DISTRIBUTED LOAD'/
*1X,'ELMT',17X,'UDL-1',11X,'UDL-2',11X,'UDL-3')
NLD(NN)=IELG
STR(1,NN)=P1
STR(2,NN)=P2
STR(3,NN)=P3
WRITE(6,25) IELG,P1,P2,P3
453
25 FORMAT(I5,6X,3E16.4)
GO TO 200
C
C…. EVALUATE AND ASSEMBLE THE CONSISTENT LOAD VECTOR
C
100 IF(NN.EQ.0) RETURN
DO 30 L=1,NN
IELG=NLD(L)
MA=NELMAT(IELG)
IELB=MEL(MA)
NEL=NELND(IELG)
ICOUNT=L
DO 40 J=1,NEL
NG=NP(J,IELG)
DO 40 I=1,NDM
40 XL(I,J)=X(I,NG)
NST=NDF*NEL
C
C…. INITIALIZE RL
C
DO 50 I=1,NST
50 RL(I)=0.0
CALL ELMLIB(D(1,MA),NEL,XL,SL,STR(1,L),RL,2,IELB)
C
C…. FORM LOCATION MATRIX
C
DO 60 J=1,NEL
N=NP(J,IELG)
NL=NDF*(J-1)
DO 60 K=1,NDF
60 LM(NL+K)=ID(K,N)
C
C…. ASSEMBLE
C
DO 70 I=1,NST
454
IF(LM(I).EQ.0) GO TO 70
II=LM(I)
R(II)=R(II)+RL(I)
70 CONTINUE
30 CONTINUE
RETURN
END
C*******************************************************************
SUBROUTINE ASSEMB(X,NELMAT,NELND,NP,D,SL,XL,RL,S,R,ID,MEL,UL,U)
DIMENSION X(NDM,1),NELMAT(1),NELND(1),NP(MNEL,1),U(NDF,1),
1 D(NPROP,1),SL(1),XL(NDM,1),ID(NDF,1),
2 LM(54),S(NEQ,1),RL(1),R(1),MEL(1),UL(NDF,1)
C
C… INITIALIZE GLOBAL STIFFNESS MATRIX
C
DO 10 I=1,NEQ
DO 10 J=1,MBAND
10 S(I,J)=0.0
NXSJ=0.0 C
C
DO 20 L=1,NUMEL
MA=NELMAT(L)
IELB=MEL(MA)
IELG=L
NEL =NELND(L)
DO 70 J=1,NEL
NG=NP(J,L)
DO 71 I=1,NDM
71 XL(I,J)=X(I,NG)
DO 72 I=1,NDF
72 UL(I,J)=U(I,NG)
455
70 CONTINUE
NST=NDF*NEL
NST2=NST*NST
C
C….INITIALIZE RL & SL
C
DO 90 I=1,NST
90 RL(I)=0.0
DO 40 I=1,NST2
40 SL(I)=0.0
C
C…. FORM LOCATION MATRIX
C
DO 30 J=1,NEL
N=NP(J,L)
NL=NDF*(J-1)
DO 30 K=1,NDF
30 LM(NL+K)=ID(K,N)
C
C…. ASSEMBLE
C
DO 50 I=1,NST
IF(LM(I).EQ.0) GO TO 50
II=LM(I)
R(II)=R(II)+RL(I)
DO 60 J=1,NST
JJ=LM(J)-II+1
IF(JJ.LE.0) GO TO 60
K=NST*(J-1)+I
S(II,JJ)=S(II,JJ)+SL(K)
60 CONTINUE
50 CONTINUE
20 CONTINUE
IF(NXSJ.EQ.0) GO TO 45
456
WRITE(6,55)
55 FORMAT(//' ****PROG TERMINATED: NEG JACOBIAN ENCOUNTERED****'//)
STOP
45 RETURN
END
C*******************************************************************
SUBROUTINE BANSOL(S,R,NEQ,MBAND)
DIMENSION S(NEQ,1),R(NEQ)
C
C…. REDUCE STIFFNESS MATRIX
C
DO 60 N=1,NEQ
IF (S(N,1).GT.1.0D-40) GO TO 30
WRITE (6,20) N
20 FORMAT(///30H ZERO ON DIAGONAL FOR EQUATION,I4)
STOP
30 DO 50 M=2,MBAND
FACT=S(N,M)/S(N,1)
I=N+M-1
IF (I.GT.NEQ) GO TO 50
J=0
DO 40 K=M,MBAND
J=J+1
40 S(I,J)=S(I,J)-FACT*S(N,K)
50 S(N,M)=FACT
60 CONTINUE
C
C…. REDUCE LOAD VECTOR
C
70 DO 90 N=1,NEQ
DO 80 M=2,MBAND
I=N+M-1
80 R(I)=R(I)-S(N,M)*R(N)
457
90 R(N)=R(N)/S(N,1)
C
C…. BACK SUBSTITUTE
C
N=NEQ
100 N=N-1
IF (N.EQ.0) GO TO 120
DO 110 M=2,MBAND
I=N+M-1
R(N)=R(N)-S(N,M)*R(I)
110 CONTINUE
GO TO 100
C
120 RETURN
END
C*****7***********************************************************
SUBROUTINE DISP(R,U,ID)
COMMON/BLK2/ISWICH
COMMON/BLK5/OM,NH
LOGICAL ISWICH
DIMENSION R(1),U(NDF,1),ID(NDF,1)
ISWICH=.FALSE.
C
C…. WRITE NODE DISPLACEMENTS
C
WRITE (11,20)
20 FORMAT(//19H NODE DISPLACEMENTS//
1 5H NODE,16X,6H1-DISP,10X,6H2-DISP,10X,6H3-DISP)
C
DO 10 J=1,NUMNP
DO 30 I=1,NDF
458
II=ID(I,J)
IF(II.NE.0) U(I,J)=R(II)
30 CONTINUE
WRITE(11,40) J,(U(I,J),I=1,NDF)
10 CONTINUE
40 FORMAT(I5,6X,3E16.4)
RETURN
END
C*******************************************************************
SUBROUTINE DISP1(R,U,ID,X)
COMMON/BLK5/OM,NH
DIMENSION R(1),U(NDF,1),ID(NDF,1),X(NDM,1)
C
C…. WRITE NODE DISPLACEMENTS
C
WRITE (11,20) NH
20 FORMAT(//19H NODE DISPLACEMENTS,5X,5X,'HARMONIC NUMBER =',I3//
1 5H NODE,16X,6HR-DISP,10X,6HZ-DISP,10X,6HT-DISP)
FACT=DSQRT(2.0D00)
DO 10 J=1,NUMNP
DO 30 I=1,NDF
II=ID(I,J)
IF(II.EQ.0) GO TO 50
U(I,J)=R(II)
GO TO 30
50 IF(DABS(X(1,J)).GT.1.0D-03) GO TO 30
IF(I.NE.3) GO TO 30
U(1,J)=U(1,J)/FACT
U(3,J)=U(1,J)
30 CONTINUE
WRITE(11,40) J,(U(I,J),I=1,NDF)
10 CONTINUE
459
40 FORMAT(I5,6X,3E16.4)
RETURN
END
C*******************************************************************
SUBROUTINE STRESS(X,NELMAT,NELND,NP,D,U,XL,UL,MEL,RL,SL)
DIMENSION X(NDM,1),NELMAT(1),NELND(1),NP(MNEL,1),
1 D(NPROP,1),U(NDF,1),XL(NDM,1),UL(NDF,1),MEL(1),RL(1),SL(1)
C
DO 20 L=1,NUMEL
MA=NELMAT(L)
IELB=MEL(MA)
IELG=L
NEL=NELND(L)
DO 70 J=1,NEL
NG=NP(J,L)
DO 90 I=1,NDM
90 XL(I,J)=X(I,NG)
DO 70 K=1,NDF
70 UL(K,J)=U(K,NG)
C
20 CONTINUE
RETURN
END
C*******************************************************************
SUBROUTINE ELMLIB(DL,NEL,XL,SL,UL,RL,ICN,IELB)
C
C…. THIS SUBROUTINE CALLS THE APPROPRIATE ELEMENT
C
DIMENSION DL(1),XL(1),SL(1),UL(1),RL(1)
460
C
GO TO (1,2,3,4,5,6,7,8),IELB
1 CALL ELMT1(DL,NEL,XL,SL,UL,RL,ICN)
GO TO 100
GO TO 100
GO TO 100
GO TO 100
GO TO 100
GO TO 100
GO TO 100
GO TO 100
GO TO 100
100 RETURN
END
C*******************************************************************
SUBROUTINE ELMT1(DL,NEL,XL,SL,UL,RL,ICN)
C
C…. PLANE TRUSS ELEMENT
C
DIMENSION XL(NDM,1),SL(NST,1),DL(1),A(4),UL(1),RL(1)
C
GO TO (1,2,3,4),ICN
461
C
C…. READ MATERIAL PROPERTIES
C
1 READ(5,10) E,AR
WRITE(6,15)E,AR
DL(1)=E
DL(2)=AR
10 FORMAT(2E10.0)
15 FORMAT(′ YOUNGS MODULAS ……… =′,E13.4/
* ′ AREA OF CROSS SECTION .. =′,E13.4)
RETURN
2 RETURN
C
C…. FORM ELEMENT STIFFNESS MATRIX
C
3 DX=XL(1,2)-XL(1,1)
DY=XL(2,2)-XL(2,1)
XXL=DSQRT(DX**2+DY**2)
A(1)=-DX/XXL
A(2)=-DY/XXL
A(3)=-A(1)
A(4)=-A(2)
EAL=DL(1)*DL(2)/XXL
C
DO 30 I=1,NST
DO 30 J=I,NST
SL(I,J)=A(I)*A(J)*EAL
30 SL(J,I)=SL(I,J)
CALL RLOAD(SL,UL,NST,RL)
RETURN
C
C…. EVALUATE STRESS RESULTANTS
C
4 IF(IELG.EQ.1) WRITE(11,40)
DX=XL(1,2)-XL(1,1)
462
DY=XL(2,2)-XL(2,1)
XXL=DSQRT(DX*DX+DY*DY)
COS=DX/XXL
SIN=DY/XXL
DU=UL(3)-UL(1)
DV=UL(4)-UL(2)
EAL=DL(1)*DL(2)/XXL
P=EAL*(COS*DU+SIN*DV)
WRITE(11,50) IELG,P
50 FORMAT(I5,6X,E16.4)
40 FORMAT(/// ′ ELEMENT FORCES′//
*1X, ′ELMT′,17X, ′FORCE′)
RETURN
END
C*******************************************************************
SUBROUTINE RLOAD(SL,UL,NST,RL)
C
C…. FORM RHS VECTOR FOR SUPPORT SETTLEMENTS
C
DIMENSION SL(NST,1),UL(1),RL(1)
C
DO 110 J=1,NST
IF(DABS(UL(J)).LT.1.0D-10) GO TO 110
DO 115 I=1,NST
115 RL(I)=RL(I)-SL(I,J)*UL(J)
110 CONTINUE
RETURN
END
C*****7*************************************************************
C
C…. 3–D TRUSS ELEMENT
C
IMPLICIT REAL*8 (A−H,O−Z)
463
DIMENSION XL(NDM,1),SL(NST,1),DL(1),A(6),UL(1),RL(1)
C
GO TO (1,2,3,4),ICN
C
C…. READ MATERIAL PROPERTIES
C
1 READ(5,10) E,AR
WRITE(6,15)E,AR
DL(1)=E
DL(2)=AR
10 FORMAT(2E10.0)
15 FORMAT (′YOUNGS MODULAS ……… =′,E13.4/
* ′ AREA OF CROSS .. =′,E13.4/
RETURN
2 RETURN
C
C
3 DX=XL(1,2)-XL(1,1)
DY=XL(2,2)-XL(2,1)
DZ=XL(3,2)-XL(3,1)
XXL=DSQRT(DX**2+DY**2+DZ**2)
A(1)=-DX/XXL
A(2)=-DY/XXL
A(3)=-DZ/XXL
A(4)=-A(1)
A(5)=-A(2)
A(6)=-A(3)
EAL=DL(1)*DL(2)/XXL
C
DO 30 I=1,NST
DO 30 J=I,NST
464
SL(I,J)=A(I)*A(J)*EAL
30 SL(J,I)=SL(I,J)
RETURN
C
C…. EVALUATE STRESS RESULTANTS
C
DX=XL(1,2)-XL(1,1)
DY=XL(2,2)-XL(2,1)
DZ=XL(3,2)-XL(3,1)
XXL=DSQRT(DX**2+DY**2+DZ**2)
C1=DX/XXL
C2=DY/XXL
C3=DZ/xxl
DU=UL(4)-UL(1)
DV=UL(5)-UL(2)
DW=UL(6)-UL(3)
EAL=DL(1)*DL(2)/XXL
P=EAL*(C1*DU+C2*DV+C3*DW)
WRITE(11,50) IELG,P
50 FORMAT(I5,6X,E16.4)
40 FORMAT(/// ′ ELEMENT
RETURN
END
C*******************************************************************
C
C…. PLANE BEAM ELEMENT
C
COMMON/BLK4/ICOUNT
465
DIMENSION XL(NDM,1),SL(NST,1),DL(1),UL(1),RL(1),A(3,6),EK(3,3),
1 EKAA(3,6),FEF(4)
C
GO TO (1,2,3,4),ICN
C
C…. READ & WRITE MATERIAL PROPERTIES
C
1 READ(5,10) E,AR,AI,IH,JH
WRITE(6,15) E,AR,AI,IH,JH
DL(1)=E*AR
DL(2)=E*AI
DL(3)=IH
DL(4)=JH
10 FORMAT(3E10.0,2I5)
15 FORMAT (′ YOUNGS MODULAS………. = ′,E13.4/
* ′ AREA OF CROSS SECTION … =′,E13.4/
* ′ SECOND MOMENT OF AREA … =′,E13.4/
* ′ HINGE AT I-END ………. =′,I13/
* ′ HINGE AT J-END ………. =′,I13)
RETURN
C
C…. EVALUATE THE CONSISTENT LOAD VECTOR
C
2 IF(ICOUNT.EQ.1) WRITE(6,105)
105 FORMAT(///1X, ′FIXED END FORCES′//
*1X, ′ELMT′,9X, ′SF-1′,9X, ′BM-1′,9X, ′SF-2′,9X, ′BM-2′)
P1=UL(1)
P2=UL(2)
DX=XL(1,2)-XL(1,1)
XXL=DABS(DX)
RL(2)= (7.0*P1+3.0*P2)*XXL/20.0
RL(5)= (3.0*P1+7.0*P2)*XXL/20.0
RL(3)= (P1/20.0+P2/30.0)*XXL*XXL
RL(6)=-(P1/30.0+P2/20.0)*XXL*XXL
FEF(1)=-RL(2)
466
FEF(2)=-RL(3)
FEF(3)=-RL(5)
FEF(4)=-RL(6)
WRITE(6,115) IELG,(FEF(I),I=1,4)
115 FORMAT(I5,4E13.4)
cosA = DX/XXL
sinA = DY/XXL
RL(1) = -RL(2)*sinA
RL(2) = RL(2)*cosA
RL(4) = -RL(5)*sinA
RL(5) = RL(5)*cosA
RETURN
C
C
3 DX=XL(1,2)-XL(1,1)
DY=XL(2,2)-XL(2,1)
XXL=DSQRT(DX*DX+DY*DY)
COSA=DX/XXL
SINA=DY/XXL
EAL=DL(1)/XXL
EIL=DL(2)/XXL
EK(1,1)=EAL
EK(1,2)=0.0
EK(1,3)=0.0
EK(2,1)=0.0
EK(2,2)=4.0*EIL
EK(2,3)=2.0*EIL
EK(3,1)=0.0
EK(3,2)=EK(2,3)
EK(3,3)=EK(2,2)
C
IH=DL(3)
JH=DL(4)
CALL TRAN(A,IH,JH,COSA,SINA,XXL)
467
DO 30 I=1,3
DO 30 J=1,6
TEMP=0.0
DO 35 K=1,3
35 TEMP=TEMP+EK(I,K)*A(K,J)
30 EKAA(I,J)=TEMP
IF(ICN.EQ.4) GO TO 60
C
DO 50 I=1,6
DO 50 J=1,6
DO 50 K=1,3
50 SL(I,J)=SL(I,J)+A(K,I)*EKAA(K,J)
RETURN
C
C…. EVALUATE STRESSES
C
GO TO 3
60 CALL TRAN(A,IH,JH,1.0D00,0.0D00,XXL)
DO 70 I=1,6
DO 70 J=1,6
SL(I,J)=0.0
DO 70 K=1,3
70 SL(I,J)=SL(I,J)+A(K,I)*EKAA(K,J)
C
DO 80 I=1,6
RL(I)=0.0
DO 80 J=1,6
80 RL(I)=RL(I)+SL(I,J)*UL(J)
WRITE(11,90) IELG,(RL(I),I=1,6)
90 FORMAT(I5,5X, ′1′,3E16.4/10X, ′2′,3E16.4)
100 FORMAT(/// ′ ELEMENT FORCES ′//
*1X, ′ELMT′,2X, ′NODE ′,11X, ′AXIAL′,11X, ′SHEAR′,9X, ′BENDING′/
* 22X, ′FORCE′,11X, ′FORCE′,9X,′ MOMENT′/)
468
RETURN
END
C****************************************************************
SUBROUTINE TRAN(A,IHI,IHJ,COSA,SINA,XL)
C
C…. FORM ELEMENT TRANSFORMATION MATRIX
C
IMPLICIT REAL*8(A-H,O-Z)
DIMENSION A(3,6)
C
C…. NO HINGES
C
A(1,1)= −COSA
A(1,2)= −SINA
A(1,3)=0.0
A(1,4)= COSA
A(1,5)= SINA
A(1,6)= 0.0
A(2,1)= −SINA/XL
A(2,2)= COSA/XL
A(2,3)= 1.0
A(2,4)= SINA/XL
A(2,5)= −COSA/XL
A(2,6)= 0.0
A(3,1)= −SINA/XL
A(3,2)= COSA/XL
A(3,3)= 0.0
A(3,4)= SINA/XL
A(3,5)= −COSA/XL
A(3,6)= 1.0
C
IF (IHI.EQ.0.AND.IHJ.EQ.0) GO TO 60
IF (IHI.NE.0.AND.IHJ.EQ.0) GO TO 20
IF (IHI.EQ.0.AND.IHJ.NE.0) GO TO 40
C
469
C…. HINGES AT I AND J ENDS

C
DO 10 I=2,3
DO 10 J=1,6
10 A(I,J) =0.0
GO TO 60
C
C…. HINGE AT I END ONLY
C
20 DO 30 J=1,6
30 A(2,J) =-A(3,J)/2
GO TO 60
C
C…. HINGE AT J END ONLY
C
40 DO 50 J=1,6
50 A(3,J) =-A(2,J)/2
60 RETURN
END
C*****7*************************************************************
C
C…. CST ELEMENT FOR PLANE STRESS/STRAIN PROBLEM
C
COMMON/BLK2/ISWICH
LOGICAL ISWICH
DIMENSION XL(NDM,1),SL(NST,1),DL(1),RL(1),UL(1),EPS(3),SIG(3),
1 A(3),B(3)
ISWICH=.FALSE.
C
GO TO (1,2,3,4),ICN
470
C
C
1 READ(5,10) E,ANU,TH,KODE
10 FORMAT(3E10.0,I5)
WRITE(6,11) E,ANU,TH,KODE
11 FORMAT (′ YOUNGS MODULAS ………………………. = ′,E13.4/
* ′ POISSONS RATIO ………………………. =′,E13.4/
* ′ THICKNESS …………………………… =′,E13.4/
* ′ KODE (1 FOR PL STRESS, ZERO FOR PL STRAIN) =′,I13/)
DL(1)=E
DL(2)=ANU
DL(3)=TH
DL(4)=KODE
RETURN
2 RETURN
C
C…. FORM THE ELEMENT STIFFNESS MATRIX
C
3 IF(ISWICH) WRITE(6,210) IELG
210 FORMAT(// ′ ELEMENT NUMBER =′,I3)
C
C…. IF KODE=0, THEN PLANE STRAIN. OTHERWISE PLANE STRESS.
C
KODE=DL(4)
I=KODE
IF(I.NE.0) I=1
IF(I.EQ.0) I=2
D1=DL(1)*(1.+(1-I)*DL(2))/(1.0+DL(2))/(1.0-I*DL(2))
D2=D1*DL(2)/(1.+(1-I)*DL(2))
D3=DL(1)*0.5/(1.+DL(2))
C
CALL AB(XL,A,B,AR)
IF(ISWICH) WRITE(6,420) AR
420 FORMAT(′ AREA = ′,F10.5)
471
IF(AR.GT.1.0D-05) GO TO 25
NXSJ=1
WRITE(6,26) IELG
26 FORMAT(′ NEGATIVE AREA ENCOUNTERED FOR ELMT ′,I4)
25 DV=DL(3)*0.25/AR
D11=D1*DV
D12=D2*DV
D33=D3*DV
C
C…. FOR EACH J NODE COMPUTE DB=D*BJ
C
DO 50 J=1,NEL
DB11=D11*B(J)
DB12=D12*A(J)
DB21=D12*B(J)
DB22=D11*A(J)
DB31=D33*A(J)
DB32=D33*B(J)
C
C…. FOR EACH I NODE COMPUTE S=BI*DB
C
DO 60 I=1,J
SL(I+I-1,J+J-1)=SL(I+I-1,J+J-1)+DB11*B(I)+DB31*A(I)
SL(I+I-1,J+J )=SL(I+I-1,J+J )+DB12*B(I)+DB32*A(I)
SL(I+I ,J+J-1 )=SL(I+I ,J+J-1)+DB21*A(I)+DB31*B(I)
SL(I+I ,J+J )=SL(I+I ,J+J )+DB22*A(I)+DB32*B(I)
60 CONTINUE
50 CONTINUE
C
C…. COMPUTE LOWER TRIANGULAR PART BY SYMMETRY
C
DO 70 I=2,NST
DO 70 J=1,I
70 SL(I,J)=SL(J,I)
472
RETURN
C
C…. EVALUATE THE STRESSES
C
I=DL(4)
IF(I.NE.0) I=1
IF(I.EQ.0) I=2
D1=DL(1)*(1.+(1-I)*DL(2))/(1.0+DL(2))/(1.0-I*DL(2))
D2=D1*DL(2)/(1.+(1-I)*DL(2))
D3=DL(1)*0.5/(1.+DL(2))
CALL AB(XL,A,B,AR)
C
DO 13 I=1,3
13 EPS(I)=0.0
XX=(XL(1,1)+XL(1,2)+XL(1,3))/3.0
YY=(XL(2,1)+XL(2,2)+XL(2,3))/3.0
C
DO 14 J=1,NEL
EPS(1)=EPS(1)+B(J)*UL(J+J-1)
EPS(2)=EPS(2)+A(J)*UL(J+J)
14 EPS(3)=EPS(3)+A(J)*UL(J+J-1)+B(J)*UL(J+J)
C
SIG(1)=(D1*EPS(1)+D2*EPS(2))/(2.0*AR)
SIG(2)=(D2*EPS(1)+D1*EPS(2))/(2.0*AR)
SIG(3)=D3*EPS(3)/(2.0*AR)
WRITE(11,15) IELG,XX,YY,(SIG(I),I=1,3)
12 CONTINUE
100 FORMAT(/// ′ S T R E S S ′//
1 1X, ′ELMT ′,7X, ′1-ORD ′,7X, ′2-ORD ′,3X, ′XX-STRESS ′,3X,
2 ′YY-STRESS,3X, ′XY-STRESS ′)
RETURN
END
C*******************************************************************
473
SUBROUTINE AB(XL,A,B,AR)
DIMENSION A(3),B(3),XL(2,3)
C
A(1)=XL(1,3)-XL(1,2)
A(2)=XL(1,1)-XL(1,3)
A(3)=XL(1,2)-XL(1,1)
C
B(1)=XL(2,2)-XL(2,3)
B(2)=XL(2,3)-XL(2,1)
B(3)=XL(2,1)-XL(2,2)
AR=0.5*(A(1)*B(3)-A(3)*B(1))
RETURN
END
C*****7*************************************************************
C
C…. 3-9 VARIABLE NODED ISOPARAMETRIC PLANE ELEMENT
C
COMMON/BLK2/ISWICH
LOGICAL ISWICH
DIMENSION XL(NDM,1),SL(NST,1),DL(1),ZI(9),ET(9),WT(9),SHP(3,9),
1 XS(2,2),RL(1),UL(1),EPS(3),SIG(3)
C
GO TO (1,2,3,4),ICN
C
C
1 READ(5,10) E,ANU,TH,KODE,L,LS,LLD
10 FORMAT(3E10.0,4I5)
WRITE(6,11) E,ANU,TH,KODE,L,LS,LLD
474
11 FORMAT(′ YOUNGS MODULAS ………………………. = ′,E13.4/

* ′ POISSONS RATIO ………………………. = ′,E13.4/
* ′ THICKNESS …………………………… = ′,E13.4/
* ′ KODE (1 FOR PL STRESS, ZERO FOR PL STRAIN) = ′,I13/
* ′ ORDER OF QUADRATURE FOR STIFFNESS ……… = ′,I13/
* ′ NO. OF STRESS POINTS IN EACH DIRECTION …. = ′,I13/
* ′ ORDER OF QUAD FOR CONSISTENT LOAD ……… = ′,I13)
DL(1)=E
DL(2)=ANU
DL(3)=TH
DL(4)=KODE
DL(5)=L
DL(6)=LS
DL(7)=LLD
RETURN
C
C
2 P1=UL(1)
P2=UL(2)
LINT=DL(7)
CALL PGAUSL(LINT,ZI,ET,WT)
DO 105 K=1,LINT
CALL SHAPE(ZI(K),ET(K),XL,SHP,XS,XSJ,NEL,1)
SIGMA=SHP(3,2)*P1+SHP(3,3)*P2
DV=SIGMA*WT(K)*DL(3)
DO 120 I=1,NEL
RL(I+I-1)=RL(I+I-1)+SHP(3,I)*XS(2,2)*DV
120 RL(I+I)=RL(I+I)-SHP(3,I)*XS(2,1)*DV
105 CONTINUE
RETURN
C
C
3 IF(ISWICH) WRITE(6,210) IELG
475
210 FORMAT(// ′ ELEMENT NUMBER = ′,I3)

C
C…. IF KODE=0, THEN PLANE STRAIN. OTHERWISE PLANE STRESS.
C
KODE=DL(4)
I=KODE
IF(I.NE.0) I=1
IF(I.EQ.0) I=2
D1=DL(1)*(1.+(1-I)*DL(2))/(1.0+DL(2))/(1.0-I*DL(2))
D2=D1*DL(2)/(1.+(1-I)*DL(2))
D3=DL(1)*0.5/(1.+DL(2))
C
IF(IELG.EQ.1) LINT=0
L=DL(5)
IF(L*L.NE.LINT) CALL PGAUSS(L,LINT,ZI,ET,WT)
C
DO 30 K=1,LINT
IF(ISWICH) WRITE(6,420) ZI(K),ET(K),XSJ
420 FORMAT(′ FORMAT(′ ZI= ′,F10.5,5X, ′ET= ′,F10.5,5X, ′XSJ= ′,F10.5)
IF(XSJ.GT.1.0D-05) GO TO 25
NXSJ=1
WRITE(6,26) IELG
26 FORMAT(′ NEGATIVE JACOBIAN ENCOUNTERED FOR ELMT ′,I4)
25 DV=XSJ*WT(K)*DL(3)
D11=D1*DV
D12=D2*DV
D33=D3*DV
C
C
DO 50 J=1,NEL
DB11=D11*SHP(1,J)
DB12=D12*SHP(2,J)
DB21=D12*SHP(1,J)
476
DB22=D11*SHP(2,J)
DB31=D33*SHP(2,J)
DB32=D33*SHP(1,J)
C
C…. FOR EACH I NODE COMPUTE S=BI*DB
C
DO 60 I=1,J
SL(I+I-1,J+J-1)=SL(I+I-1,J+J-1)+SHP(1,I)*DB11+SHP(2,I)*DB31
SL(I+I-1,J+J )=SL(I+I-1,J+J )+SHP(1,I)*DB12+SHP(2,I)*DB32
SL(I+I ,J+J-1)=SL(I+I ,J+J-1)+SHP(2,I)*DB21+SHP(1,I)*DB31
SL(I+I ,J+J )=SL(I+I ,J+J )+SHP(2,I)*DB22+SHP(1,I)*DB32
60 CONTINUE
50 CONTINUE
30 CONTINUE
C
C
DO 70 I=2,NST
DO 70 J=1,I
70 SL(I,J)=SL(J,I)
RETURN
C
C
L=DL(6)
I=DL(4)
IF(I.NE.0) I=1
IF(I.EQ.0) I=2
D1=DL(1)*(1.+(1-I)*DL(2))/(1.0+DL(2))/(1.0-I*DL(2))
D2=D1*DL(2)/(1.+(1-I)*DL(2))
D3=DL(1)*0.5/(1.+DL(2))
C
477
DO 12 L=1,LINT
CALL SHAPE(ZI(L),ET(L),XL,SHP,XS,XSJ,NEL,2)
DO 13 I=1,3
13 EPS(I)=0.0
XX=0.0
YY=0.0
C
DO 14 J=1,NEL
XX=XX+SHP(3,J)*XL(1,J)
YY=YY+SHP(3,J)*XL(2,J)
EPS(1)=EPS(1)+SHP(1,J)*UL(J+J-1)
EPS(2)=EPS(2)+SHP(2,J)*UL(J+J)
14 EPS(3)=EPS(3)+SHP(1,J)*UL(J+J)+SHP(2,J)*UL(J+J-1)
C
SIG(1)=D1*EPS(1)+D2*EPS(2)
SIG(2)=D2*EPS(1)+D1*EPS(2)
SIG(3)=D3*EPS(3)
WRITE(11,15) IELG,XX,YY,(SIG(I),I=1,3)
12 CONTINUE
100 FORMAT(/// ′ S T R E S S ′//
1 1X, ′ELMT ′,7X, ′1-ORD ′,7X, ′2-ORD ′,3X, ′XX-STRESS ′,3X,
2 ′YY-STRESS ′,3X, ′XY-STRESS ′)
RETURN
END
C*******************************************************************
SUBROUTINE SHAPE(SS,TT,X,SHP,XS,XSJ,NEL,KEY)
C
C… SHAPE FUNCTIONS FOR ISOPARAMETRIC ELEMENTS
C… KEY=1 DERIVATIVES ARE WRT NATURAL COORDINATES
C… 2 DERIVATIVEA ARE WRT GLOBAL COORDINATES
C
DIMENSION SHP(3,1),X(2,1),S(4),T(4),XS(2,2),SX(2,2)
DATA S/-0.5D00,0.5D00,0.5D00,-0.5D00/
478
DATA T/-0.5D00,-0.5D00,0.5D00,0.5D00/
C
C… FORM 4-NODE QUADRILATERAL SHAPE FUNCTIONS
C
DO 100 I=1,4
SHP(3,I)=(0.5+S(I)*SS)*(0.5+T(I)*TT)
SHP(1,I)=S(I)*(0.5+T(I)*TT)
100 SHP(2,I)=T(I)*(0.5+S(I)*SS)
IF(NEL.GT.3) GO TO 20
DO 10 I=1,3
10 SHP(I,3)=SHP(I,3)+SHP(I,4)
20 IF(NEL.GT.4) CALL SHAP2(SS,TT,SHP,NEL)
C
C…. CONSTRUCT JACOBIAN AND ITS INVERSE
C
DO 130 I=1,2
DO 130 J=1,2
XS(I,J)=0.0
DO 130 K=1,NEL
130 XS(I,J)=XS(I,J)+SHP(I,K)*X(J,K)
XSJ=XS(1,1)*XS(2,2)-XS(1,2)*XS(2,1)
IF(KEY.EQ.1) RETURN
C
SX(1,1)= XS(2,2)/XSJ
SX(1,2)=-XS(1,2)/XSJ
SX(2,1)=-XS(2,1)/XSJ
SX(2,2)= XS(1,1)/XSJ
C
C… FORM GLOBAL DERIVATIVES
C
DO 140 I=1,NEL
TEMP =SHP(1,I)*SX(1,1)+SHP(2,I)*SX(1,2)
SHP(2,I)=SHP(1,I)*SX(2,1)+SHP(2,I)*SX(2,2)
140 SHP(1,I)=TEMP
RETURN
479
END
C****************************************************************
SUBROUTINE SHAP2(S,T,SHP,NEL)
DIMENSION SHP(3,1)
S2=(1.-S*S)/2
T2=(1.-T*T)/2
DO 100 I=5,8
DO 100 J=1,3
100 SHP(J,I)=0.0
C
SHP(1,5)=-S*(1.-T)
SHP(2,5)=-S2
SHP(3,5)= S2*(1.-T)
IF(NEL.LT.6) GO TO 107
C
SHP(1,6)= T2
SHP(2,6)=-T*(1.+S)
SHP(3,6)= T2*(1.+S)
C
SHP(1,7)= -S*(1.+T)
SHP(2,7)= S2
SHP(3,7)= S2*(1.+T)
C
SHP(1,8)= -T2
SHP(2,8)= -T*(1.-S)
SHP(3,8)= T2*(1.-S)
C
SHP(1,9)= -S*T2*4.0
SHP(2,9)= -T*S2*4.0
SHP(3,9)= 4.*S2*T2
C
480
C…. CORRECT CORNER NODES FOR INTERIOR NODE (LAGRANGIAN)

C
DO 106 J=1,3
DO 105 I=1,4
105 SHP(J,I)=SHP(J,I)-0.25*SHP(J,9)
C…. CORRECT MID SIDE NODES FOR INTERIOR NODE
DO 106 I=5,8
106 SHP(J,I)=SHP(J,I)-0.5*SHP(J,9)
C
C…. CORRECT CORNER NODES FOR PRESENCE OF MIDSIDE NODES
C
107 DO 108 J=1,3
SHP(J,1)=SHP(J,1)-0.5*(SHP(J,5)+SHP(J,8))
108 SHP(J,4)=SHP(J,4)-0.5*(SHP(J,7)+SHP(J,8))
RETURN
END
C*******************************************************************
SUBROUTINE PGAUSL(LINT,ZI,ET,WT)
C
C… GAUSS POINTS AND WEIGHTS TO CALCULATE THE EQUIVALENT NODAL
C… LOAD VECTOR DUE TO SURFACE TRACTION AT THE FACE ZI=+1
C
DIMENSION ZI(1),ET(1),WT(1)
GO TO (1,2,3),LINT
C
C… 1 PT INTEGRATION
C
1 ZI(1)=1.
ET(1)=0.
WT(1)=2.
RETURN
C
481
C
2 G = 1.0D00/DSQRT(3.0D00)
ZI(1)=1.0
ZI(2)=1.0
ET(1)= G
ET(2)=-G
WT(1)=1.0D00
WT(2)=1.0D00
RETURN
C
C
3 G = DSQRT(0.6D00)
ZI(1)=1.0
ZI(2)=1.0
ZI(3)=1.0
ET(1)=-G
ET(2)= 0.0
ET(3)= G
WT(1)= 5./9.
WT(2)= 8./9.
WT(3)= 5./9.
RETURN
END
C*******************************************************************
SUBROUTINE PGAUSS(L,LINT,ZI,ET,WT)
C
C… GAUSS POINTS AND WEIGHTS FOR TWO DIMENSIONS
C
DIMENSION LR(9),LZ(9),LW(9),ZI(1),ET(1),WT(1)
DATA LR/-1,1,1,-1,0,1,0,-1,0/,LZ/-1,-1,1,1,-1,0,1,0,0/
DATA LW/4*25,4*40,64/
LINT=L*L
482
GO TO (1,2,3),L
C
C… 1*1 INTEGRATION
C
1 ZI(1)=0.
ET(1)=0.
WT(1)=4.
RETURN
C
C
2 G = 1.0D00/DSQRT(3.0D00)
DO 21 I=1,4
ZI(I)=G*LR(I)
ET(I)=G*LZ(I)
21 WT(I)=1.0D00
RETURN
C
C
3 G = DSQRT(0.6D00)
H = 1.0D00/81.0D00
DO 31 I=1,9
ZI(I)=G*LR(I)
ET(I)=G*LZ(I)
31 WT(I)=H*LW(I)
RETURN
END
C*****7*************************************************************
C
C…. 3-9 VARIABLE NODED ISOPARAMETRIC ELEMENT FOR AXISYMMETRIC
C…. ANALYSIS M=0
C
483
COMMON/BLK2/ISWITCH
COMMON/BLK5/OM,NH
LOGICAL ISWITCH
*XS(2,2),RL(1),UL(1),EPS(4),SIG(4),DS(3,3),BJ(4,2),BI(4,2),DB(4,2)
C
GO TO (1,2,3,4),ICN
C
C
OM = 0.0
1 READ(5,10) E,ANU,L,LS,LLD
10 FORMAT(2E10.0,4I5)
WRITE(6,11) E,ANU,L,LS,LLD
11 FORMAT(′ YOUNGS MODULAS ……………………… = ′,E13.4/
* ′ POISSONS RATIO ………………………. = ′,E13.4/
* ′ ORDER OF QUADRATURE FOR STIFFNESS ……… = ′,I13/
* ′ NO. OF STRESS POINTS IN EACH DIRECTION …. = ′,I13/
* ′ ORDER OF QUAD FOR CONSISTENT LOAD ……… = ′,I13/)
RHO = 1.0
DL(1)=E
DL(2)=ANU
DL(3)=RHO
DL(4)=L
DL(5)=LS
DL(6)=LLD
PI=4.0*ATAN(1.0)
RETURN
C
C
2 P1=UL(1)
484
P2=UL(2)
LINT=DL(6)
DO 105 K=1,LINT
IF(NEL.GT.4) SIGMA=SIGMA+SHP(3,6)*UL(3)
RR=0.0
DO 125 J=1,NEL
125 RR=RR+SHP(3,J)*XL(1,J)
DV=2.0*PI*SIGMA*RR*WT(K)
DO 120 I=1,NEL
RL(I+I-1)=RL(I+I-1)+SHP(3,I)*XS(2,2)*DV
120 RL(I+I)=RL(I+I)-SHP(3,I)*XS(2,1)*DV
105 CONTINUE
RETURN
C
C
3 IF(ISWITCH) WRITE(6,210) IELG
210 FORMAT(// ′ ELEMENT NUMBER = ′,I3)
C
6 E=DL(1)
ANU=DL(2)
FACT=E/(1.0+ANU)/(1.0-2.0*ANU)
DS(1,1)=FACT*(1.0-ANU)
DS(2,2)=DS(1,1)
DS(3,3)=DS(1,1)
DS(1,2)=FACT*ANU
DS(1,3)=DS(1,2)
DS(2,1)=DS(1,2)
DS(2,3)=DS(1,2)
DS(3,1)=DS(1,2)
DS(3,2)=DS(1,2)
DS44=FACT*(0.5-ANU)
485
C
C…. INITIALIZE BI AND BJ MATRICES
C
DO 110 I=1,4
DO 110 J=1,2
BI(I,J)=0.0
110 BJ(I,J)=0.0
C
L=DL(4)
C
DO 30 K=1,LINT
IF(ISWITCH) WRITE(6,420) ZI(K),ET(K),XSJ
420 FORMAT (′ZI=′,F10.5,5X,′XSJ=′,F10.5)
NXSJ=1
WRITE(6,26) IELG
26 FORMAT (′ NEGATIVE JACOBIAN ENCOUNTERED FOR ELMT’,I4)
C
25 RR=0.0
DO 35 I=1,NEL
35 RR=RR+SHP(3,I)*XL(1,I)
DV=2.0*PI*RR*XSJ*WT(K)
RHO=DL(3)
DM=2.0*PI*RR*XSJ*WT(K)*RHO*OM*OM
C
C….FORM BJ MATRIX
C
JJ=1
DO 50 J=1,NEL
BJ(1,1)=SHP(1,J)
BJ(2,1)=SHP(3,J)/RR
486
BJ(4,1)=SHP(2,J)
BJ(3,2)=SHP(2,J)
BJ(4,2)=SHP(1,J)
C
C
DO 70 K2=1,2
DO 80 K1=1,3
DB(K1,K2)=0.0
DO 80 KKK=1,3
80 DB(K1,K2)=DB(K1,K2)+DS(K1,KKK)*BJ(KKK,K2)
70 DB(4,K2)=DS44*BJ(4,K2)
C
C…. FORM BI MATRIX
C
II=1
DO 60 I=1,J
BI(1,1)=SHP(1,I)
BI(2,1)=SHP(3,I)/RR
BI(4,1)=SHP(2,I)
BI(3,2)=SHP(2,I)
BI(4,2)=SHP(1,I)
C
C…. FOR EACH I NODE, COMPUTE S=BI*DB
C
DO 90 K1=1,2
IR=II+K1-1
DO 90 KK=1,2
IC=JJ+KK-1
IF(K1.EQ.KK) SL(IR,IC)=SL(IR,IC)-DM*SHP(3,I)*SHP(3,J)
DO 90 KKK=1,4
90 SL(IR,IC)=SL(IR,IC)+DV*BI(KKK,K1)*DB(KKK,KK)
60 II=II+NDF
50 JJ=JJ+NDF
30 CONTINUE
487
C
C
DO 75 I=2,NST
DO 75 J=1,I
75 SL(I,J)=SL(J,I)
RETURN
C
C
LS=DL(5)
GO TO 6
5 IF(LS*LS.NE.LINT) CALL PGAUSS(LS,LINT,ZI,ET,WT)
C
DO 12 L=1,LINT
DO 13 I=1,4
13 EPS(I)=0.0
RR=0.0
ZZ=0.0
C
DO 14 J=1,NEL
RR=RR+SHP(3,J)*XL(1,J)
ZZ=ZZ+SHP(3,J)*XL(2,J)
J1=J+J-1
J2=J+J
EPS(1)=EPS(1)+SHP(1,J)*UL(J1)
14 EPS(4)=EPS(4)+SHP(2,J)*UL(J1)+SHP(1,J)*UL(J2)
EPS(2)=EPS(2)/RR
C
SIG(1)=DS(1,1)*EPS(1)+DS(1,2)*EPS(2)+DS(1,3)*EPS(3)
488
SIG(4)=DS44*EPS(4)
WRITE(11,15) IELG,RR,ZZ,(SIG(I),I=1,4)
12 CONTINUE
15 FORMAT(I5,2E12.4,4E11.3)
100 FORMAT(/// ′ S T R E S S′//
1 1X,′ELMT′,7X,′R-ORD′,7X,′Z-ORD′,3X,′RR-STRES′,3X,
2 ′ TT-STRES′,3X,′ZZ-STRES′,3X,′RZ-STRES′)
RETURN
END
C*****7*************************************************************
SUBROUTINE TRANS(SL,NST,NEL,NH)
DIMENSION SL(NST,1),IEQ(3)
DATA IEQ/1,10,22/
C
IF(NH.GT.1) RETURN
FACT=DSQRT(0.5D00)
KK=3
IF(NEL.LT.8) KK=2
DO 30 K=1,KK
IR=IEQ(K)
C
C…. MODIFY THE COLUMNS
C
DO 10 I=1,NST
SK1=SL(I,IR)
SK2=SL(I,IR+2)
SL(I,IR)=(SK1+SK2)*FACT
10 SL(I,IR+2)=(SK2-SK1)*FACT
C
C…. MODIFY THE ROWS
C
DO 20 J=1,NST
489
SK1=SL(IR,J)
SK2=SL(IR+2,J)
SL(IR,J)=(SK1+SK2)*FACT
20 SL(IR+2,J)=(SK2/SK1)*FACT
30 CONTINUE
RETURN
END
C*******************************************************************
C
C…. 3-9 VARIABLE NODED ISOPARAMETRIC ELEMENT FOR AXISYMMETRIC
C…. ANALYSIS M > 0…
C
COMMON/BLK2/ISWITCH
COMMON/BLK5/OM,NH
LOGICAL ISWITCH
*XS(2,2),RL(1),UL(1),EPS(6),SIG(6),DS(3,3),BJ(6,3),BI(6,3),DB(6,3)
C
GO TO (1,2,3,4),ICN
C
C
1 READ(5,10) E,ANU,L,LS,LLD,KODE, NH
10 FORMAT(2E10.0,5I5)
OM = 0.0
RHO = 1.0
WRITE(6,11) E,ANU,L,LS,LLD,KODE,NH
11 FORMAT (′ YOUNGS MODULAS ………………………. =’,E13.4/
* ′ POISSONS RATIO ………………………. =’,E13.4/
* ′ ORDER OF QUADRATURE FOR STIFFNESS ……… =’,I13/
490
* ′ NO. OF STRESS POINTS IN EACH DIRECTION …. =’,I13/

* ′ ORDER OF QUAD FOR CONSISTENT LOAD ……… =’,I13/
* ′ KODE (=1 FOR CORE ELEMENTS ……………. =’,I13/
* ′ HARMONIC NUMBER………………………..=’,I13)
DL(1)=E
DL(2)=ANU
DL(3)=RHO
DL(4)=L
DL(5)=LS
DL(6)=LLD
DL(7)=KODE
PI=4.0*ATAN(1.0)
RETURN
C
C
2 P1=UL(1)
P2=UL(2)
LINT=DL(6)
DO 105 K=1,LINT
IF(NEL.GT.4) SIGMA=SIGMA+SHP(3,6)*UL(3)
RR=0.0
DO 125 J=1,NEL
125 RR=RR+SHP(3,J)*XL(1,J)
DV=PI*SIGMA*RR*WT(K)
II=1
DO 120 I=1,NEL
RL(II) =RL(II )+SHP(3,I)*XS(2,2)*DV
RL(II+1)=RL(II+1)-SHP(3,I)*XS(2,1)*DV
RL(II+2)=RL(II+2)+0.0
120 II =II+NDF
105 CONTINUE
491
RETURN
C
C
3 IF(ISWITCH) WRITE(6,210) IELG
210 FORMAT(// ′ ELEMENT NUMBER = ’, I3)
C
6 E=DL(1)
ANU=DL(2)
FACT=E/(1.0+ANU)/(1.0-2.0*ANU)
DS(1,1)=FACT*(1.0-ANU)
DS(2,2)=DS(1,1)
DS(3,3)=DS(1,1)
DS(1,2)=FACT*ANU
DS(1,3)=DS(1,2)
DS(2,1)=DS(1,2)
DS(2,3)=DS(1,2)
DS(3,1)=DS(1,2)
DS(3,2)=DS(1,2)
DS44=FACT*(0.5-ANU)
DS55=DS44
DS66=DS44
C
C…. INITIALIZE BI AND BJ MATRICES
C
DO 110 I=1,6
DO 110 J=1,3
BI(I,J)=0.0
110 BJ(I,J)=0.0
C
L=DL(4)
C
492
DO 30 K=1,LINT
IF(ISWITCH) WRITE(6,420) ZI(K),ET(K),XSJ
420 FORMAT (′ ZI=’,F10.5,5X,’ET=’,F10.5,5X,’XSJ=’,F10.5)
NXSJ=1
WRITE(6,26) IELG
26 FORMAT (′ NEGATIVE JACOBIAN ENCOUNTERED FOR ELMT’,I4)
C
25 RR=0.0
DO 35 I=1,NEL
35 RR=RR+SHP(3,I)*XL(1,I)
DV=PI*RR*XSJ*WT(K)
RHO=DL(3)
DM=PI*RR*XSJ*WT(K)*RHO*OM*OM
C
C…. FORM BJ MATRIX
C
JJ=1
DO 50 J=1,NEL
BJ(1,1)= SHP(1,J)
BJ(3,1)= SHP(3,J)/RR
BJ(4,1)= SHP(2,J)
BJ(5,1)=-SHP(3,J)*NH/RR
BJ(2,2)= SHP(2,J)
BJ(4,2)= SHP(1,J)
BJ(5,3)= SHP(3,J)/RR-SHP(1,J)
BJ(6,3)=-SHP(2,J)
C
C…. FOR EACH J NODE, COMPUTE DB=D*BJ
C
DO 70 KK=1,NDF
DO 80 K1=1,3
493
DB(K1,KK)=0.0
DO 80 KKK=1,3
80 DB(K1,KK)= DB(K1,KK)+DS(K1,KKK)*BJ(KKK,KK)
DB(4,KK) = DS44*BJ(4,KK)
DB(5,KK) = DS55*BJ(5,KK)
70 DB(6,KK) = DS66*BJ(6,KK)
C
C…. FORM BI MATRIX
C
II=1
DO 60 I=1,J
BI(1,1)= SHP(1,I)
BI(3,1)= SHP(3,I)/RR
BI(4,1)= SHP(2,I)
BI(5,1)=-SHP(3,I)*NH/RR
BI(2,2)= SHP(2,I)
BI(4,2)= SHP(1,I)
BI(5,3)= SHP(3,I)/RR-SHP(1,I)
BI(6,3)=-SHP(2,I)
C
C…. FOR EACH I NODE, COMPUTE S=BI*DB
C
DO 90 K1=1,3
IR=II+K1-1
DO 90 KK=1,3
IC=JJ+KK-1
IF(K1.EQ.KK) SL(IR,IC)=SL(IR,IC)-DM*SHP(3,I)*SHP(3,J)
DO 90 KKK=1,6
90 SL(IR,IC)=SL(IR,IC)+DV*BI(KKK,K1)*DB(KKK,KK)
60 II=II+NDF
50 JJ=JJ+NDF
30 CONTINUE
C
494
C…. FORM LOWER TRIANGULAR PART BY REFLECTION

C
DO 75 I=2,NST
DO 75 J=1,I
75 SL(I,J)=SL(J,I)
KODE=DL(7)
IF(NH.EQ.1.AND.KODE.EQ.1) CALL TRANS(SL,NST,NEL,NH)
RETURN
C
C
LS=DL(5)
GO TO 6
5 IF(LS*LS.NE.LINT) CALL PGAUSS(LS,LINT,ZI,ET,WT)
C
DO 12 L=1,LINT
DO 13 I=1,6
13 EPS(I)=0.0
RR=0.0
ZZ=0.0
DO 14 J=1,NEL
RR=RR+SHP(3,J)*XL(1,J)
14 ZZ=ZZ+SHP(3,J)*XL(2,J)
C
JJ=0
DO 55 J=1,NEL
BJ(1,1)= SHP(1,J)
BJ(3,1)= SHP(3,J)/RR
BJ(4,1)= SHP(2,J)
BJ(2,2)= SHP(2,J)
BJ(4,2)= SHP(1,J)
495
BJ(5,3)= SHP(3,J)/RR-SHP(1,J)
BJ(6,3)=-SHP(2,J)
C
DO 65 I=1,6
DO 65 K=1,3
65 EPS(I)=EPS(I)+BJ(I,K)*UL(JJ+K)
55 JJ=JJ+NDF
C
SIG(4)=DS44*EPS(4)
SIG(5)=DS55*EPS(5)
SIG(6)=DS66*EPS(6)
WRITE(11,15) IELG,RR,ZZ,(SIG(I),I=1,6)
12 CONTINUE
15 FORMAT(I5,2E12.4,6E11.3)
100 FORMAT(/// ′ S T R E S S ′//
1 1X, ′ ELMT ′,7X, ′ R-ORD ′,7X, ′ Z-ORD ′,3X, ′RR-STRES′,3X,
2 ′ TT-STRES ′,3X,′ ZZ-STRES ′,3X,′ RZ-STRES ′,3X, ′RT-STRES ′,3X,
3 ′ ZT-STRES′)
RETURN
END
C
C******************************************************************
C
C
C…. 8 NODE BRICK ELEMENT FOR 3-D STRESS ANALYSIS
C
496
COMMON/BLK2/ISWICH
LOGICAL ISWICH
DIMENSION XL(NDM,1),SL(NST,1),DL(1),RL(1),UL(1),EPS(6),SIG(6)
DIMENSION SHP(4,8),RG(8),SG(8),TG(8),WG(8)
ISWICH=.FALSE.
IFLAG=0
C
GO TO (1,2,3,4),ICN
C
C
1 CONTINUE
READ(5,10) E,ANU
10 FORMAT(2E10.0)
WRITE(6,11) E,ANU
11 FORMAT (′ YOUNGS MODULUS …………………………..=′, E13.4/
* ′ POISSONS RATIO …………………………..=′,E13.4/)
DL(1)=E
DL(2)=ANU
RETURN
C
2 RETURN
C
C
3 CONTINUE
E=DL(1)
ANU=DL(2)
DL1=E*(1.0D0-ANU)/(1.0D0+ANU)/(1.0D0-2.0D0*ANU)
DL2=ANU*DL1/(1.0D0-ANU)
DL3=E/2.0D0/(1.0D0+ANU)
IF(ISWICH) WRITE(11,310) IELG
310 FORMAT(//′ ELEMENT NUMBER =′,I3)
CALL PGAUSS3(RG,SG,TG,WG)
497
DO 320 L=1,NEL
CALL SHAPE3(RG(L),SG(L),TG(L),XL,SHP,XSJ,NDM,NEL,.FALSE.)
DO 330 I =1,NEL
I3=3*I
I1=I3-2
I2=I3-1
DO 330 J =I,NEL
J3=3*J
J1=J3-2
J2=J3-1
SL(I1,J1) = SL(I1,J1)+(DL1*SHP(1,I)*SHP(1,J)
* +DL3*(SHP(2,I)*SHP(2,J)+SHP(3,I)*SHP(3,J)))*XSJ
SL(I1,J2)= SL(I1,J2)+(DL2*SHP(1,I)*SHP(2,J)
* +DL3*SHP(2,I)*SHP(1,J))*XSJ
SL(I2,J2) = SL(I2,J2)+(DL1*SHP(2,I)*SHP(2,J)
* +DL3*(SHP(1,I)*SHP(1,J)+SHP(3,I)*SHP(3,J)))*XSJ
330 SL(I3,J3) = SL(I3,J3)+(DL1*SHP(3,I)*SHP(3,J)+DL3*(SHP(1,I)*
* SHP(1,J)+SHP(2,I)*SHP(2,J)))*XSJ
320 CONTINUE
C
C
DO 340 I = 1,24
DO 340 J = I,24
340 SL(J,I) = SL(I,J)
498
RETURN
C
4 CONTINUE
C
C
IF(IELG.EQ.1) WRITE(11,100)
CALL PGAUSS3(RG,SG,TG,WG)
DO 400 L=1,NEL
CALL SHAPE3(RG(L),SG(L),TG(L),XL,SHP,XSJ,NDM,NEL,.FALSE.)
DO 410 I=1,6
410 EPS(I)=0.0D0
XX=0.0D0
YY=0.0D0
ZZ=0.0D0
JJ=0
DO 420 J=1,NEL
C
XX=XX+SHP(4,J)*XL(1,J)
YY=YY+SHP(4,J)*XL(2,J)
ZZ=ZZ+SHP(4,J)*XL(3,J)
C
EPS(1)=EPS(1)+SHP(1,J)*UL(JJ+1)
EPS(4)=EPS(4)+SHP(2,J)*UL(JJ+1)+SHP(1,J)*UL(JJ+2)
C
420 JJ=JJ+NDF
C
SIG(1)=DL1*EPS(1)+DL2*EPS(2)+DL2*EPS(3)
499
SIG(4)=DL3*EPS(4)
SIG(5)=DL3*EPS(5)
SIG(6)=DL3*EPS(6)
C
WRITE(11,200) IELG,L,XX,YY,ZZ,SIG
C
400 CONTINUE
RETURN
C
100 FORMAT(5X,′ELEMENT STRESSES′//4X,′ ELEM ′,4X,′ GP ′,2X,
* 3X,′ 1-COORD ′,5X,′ 2-COORD ′,5X,′ 3-COORD ′,3X,′ 11-STRESS ′,
* 3X,′ 22-STRESS ′,3X,′ 33-STRESS ′,3X,′ 12-STRESS ′,3X,′ 23-STRESS ′,
* 3X,′ 31-STRESS ′,3X)
200 FORMAT(1X,I5,2X,I5,2X,9E12.4)
END
C
C*************************************************************
C
SUBROUTINE PGAUSS3(R,S,T,W)
C
C…..2 POINT GAUSS INTEGRATION
C
DIMENSION LR(8),LS(8),LT(8),R(1),S(1),T(1),W(1)
DATA LR/-1,1,1,-1,-1,1,1,-1/
DATA LS/-1,-1,1,1,-1,-1,1,1/
DATA LT/1,1,1,1,-1,-1,-1,-1/
G=1.0D0/SQRT(3.0D0)
DO 10 I=1,8
R(I)=G*LR(I)
S(I)=G*LS(I)
T(I)=G*LT(I)
10 W(I)=1.0D0
RETURN
END
500
C
C*************************************************************
C
SUBROUTINE SHAPE3(RR,SS,TT,X,SHP,XSJ,NDM,NEL,IFLAG)
C
C…..SHAPE FUNCTIONS FOR 8 NODE 3-D BRICK ELEMENT
C
LOGICAL IFLAG
DIMENSION SHP(4,1),X(NDM,1),R(8),S(8),T(8),SX(3,3),XS(3,3)
DATA R/-1.0D0,1.0D0,1.0D0,-1.0D0,-1.0D0,1.0D0,1.0D0,-1.0D0/
DATA S/-1.0D0,-1.0D0,1.0D0,1.0D0,-1.0D0,-1.0D0,1.0D0,1.0D0/
DATA T/1.0D0,1.0D0,1.0D0,1.0D0,-1.0D0,-1.0D0,-1.0D0,-1.0D0/
C
DO 10 I =1,8
SHP(4,I) =0.125D0*(1.0D0+RR*R(I))*(1.0D0+SS*S(I))*(1.0D0+TT*T(I))
SHP(1,I) =0.125D0*R(I)*(1.0D0+SS*S(I))*(1.0D0+TT*T(I))
SHP(2,I) =0.125D0*(1.0D0+RR*R(I))*S(I)*(1.0D0+TT*T(I))
SHP(3,I) =0.125D0*(1.0D0+RR*R(I))*(1.0D0+SS*S(I))*T(I)
10 CONTINUE
C
C…. CONTSTRUCTION OF JACOBIAN AND INVERSE
C
DO 30 I = 1,NDM
DO 30 J = 1,3
XS(I,J) = 0.0D0
DO 30 K=1,NEL
30 XS(I,J) =XS(I,J)+X(I,K)*SHP(J,K)
IF (NDM.LT.3) RETURN
XSJ = XS(1,1)*(XS(2,2)*XS(3,3)-XS(2,3)*XS(3,2))-XS(1,2)*(XS(2,1)
* *XS(3,3)-XS(2,3)*XS(3,1))+XS(1,3)*(XS(2,1)*XS(3,2)-XS(2,2)*
* XS(3,1))
IF (IFLAG) RETURN
SX(1,1) = (XS(2,2)*XS(3,3)-XS(2,3)*XS(3,2))/XSJ
SX(2,1) =-(XS(2,1)*XS(3,3)-XS(2,3)*XS(3,1))/XSJ
501
SX(3,1) = (XS(2,1)*XS(3,2)-XS(2,2)*XS(3,1))/XSJ
SX(1,2) =-(XS(1,2)*XS(3,3)-XS(1,3)*XS(3,2))/XSJ
SX(2,2) = (XS(1,1)*XS(3,3)-XS(1,3)*XS(3,1))/XSJ
SX(3,2) =-(XS(1,1)*XS(3,2)-XS(1,2)*XS(3,1))/XSJ
SX(1,3) = (XS(1,2)*XS(2,3)-XS(1,3)*XS(2,2))/XSJ
SX(2,3) =-(XS(1,1)*XS(2,3)-XS(1,3)*XS(2,1))/XSJ
SX(3,3) = (XS(1,1)*XS(2,2)-XS(1,2)*XS(2,1))/XSJ
C
C…. FORM GLOBAL DERIVATIVES
C
DO 40 I=1,NEL
TP1 = SHP(1,I)*SX(1,1)+SHP(2,I)*SX(2,1)+SHP(3,I)*SX(3,1)
TP2 = SHP(1,I)*SX(1,2)+SHP(2,I)*SX(2,2)+SHP(3,I)*SX(3,2)
SHP(3,I) = SHP(1,I)*SX(1,3)+SHP(2,I)*SX(2,3)+SHP(3,I)*SX(3,3)
SHP(1,I) = TP1
40 SHP(2,I) = TP2
RETURN
END
C********************************************************************
RETURN
END
RETURN
END
C******************************************************************
502
Appendix F
GRAPHICAL INTERFACE FOR THE SIMPLIFIED
FINITE ELEMENT ANALYSIS PROGRAM (SFEAP)
F.1 PROGRAMS
The simplified finite element analysis program (sfeap) can be used for static linear-elastic
analysis of plane and space trusses, beam and plane frames, two and three-dimensional
problems, and axisymmetric solids. There are eight Finite Element programs contained on
the accompanying CD-ROM. The programs are listed in Table F.1, and are numbered as
P1 to P8 for referencing. All the programs were coded using the theory presented in the
text.
F.1.1 Preamble
The first three programs use one dimensional element; P1 and P2 use truss elements
while P3 employs beam elements. The next two use two-dimensional elements in
rectangular Cartesian co-ordinates; P4 uses three nodes constant strain triangles and P5
employs three to nine nodes isoparametric elements. P6 and P7 use two-dimensional
three to nine nodes isoparametric elements in cylindrical (r, z) co-ordinates. The last
program P8 employs eight nodded three-dimensional brick elements.
The executables resulting from the finite element code written in FORTRAN 77 have been
“wrapped” in a graphical user interface (GUI) developed using Visual C# for ease of use.
The GUI requires at least Microsoft .Net (“dot” Net) framework 2.0 which can be
®
downloaded freely fromhttp://www.microsoft.com. FlexCell Grid Control is used for data

visualization and import/export features. Moreover, the GUI software uses the Microsoft ®
Windows GDI+ for drawing figures (more about drawing figures is discussed later on).
®
The software has been tested using 32 bit and 64 bit versions of Microsoft Windows XP.
® ®
It is recommended that the screen resolution be set to at least 1,024 × 768 pixels for
optimal visual experience.
All the programs are standard Microsoft Windows program. As such, it possesses the
® ®
standard controls including minimize, restore, and exit controls, as well as horizontal and
vertical scroll bars, if required. For each program data has to be entered in six separate
screens (frames). The screens are Control Information, Co-ordinates, Element Data,
Boundary Conditions and Nodal Displacement Data, Material Characteristic Data, and
Nodal and Distributed Load Data. All the screens will be explained in the section Executing
Programs. Note that all input (force, displacement, material properties, etc.) must use a
consistent set of units. Moreover, all inputs that are floating point quantities require the use
of a decimal point, and can, if desired, be entered in “E” format (e.g., 20,000.0 or 2.0E+04
or 2.0e+04). Each program has its own manual which can be viewed by Clicking the icon
Table F.1 List of Programs
Program Number Program
503
P1 Analysis of plane truss
P2 Analysis of three-dimensional truss
P3 Analysis of plane frames or beam
P4 Plane stress/strain analysis using CST element
P5 Plane stress/strain analysis using isoparametric element
P6 Axisymmetric analysis for circumferential harmonic = 0
P7 Axisymmetric analysis for circumferential harmonic > 0
P8 Three-dimensional stress analysis using brick elements
F.2 EXECUTING PROGRAMS

From the Start Menu → All Programs → SFEAP → <Clicking> the SFEAP icon will
open the Table of Contents screen as shown in Figure F.1. Manual buttons are next to the
corresponding program buttons. Six input screens of every program are very similar and
hence only the case of the first program P1 is discussed with comments for other
programs. It may be noted that every screen has an Exit button which allows the user to
exit the program.
An electronic copy of Appendix F will open by <Clicking> the button Appendix F, at the top
left corner of the Table of Contents screen.
<Clicking> on the first button, Analysis of Plane Truss, will open a screen similar to that
shown in Figure F.2. The manual for the program, in PDF*, can be seen by <Clicking> the
corresponding manual button in the Table of Contents. The manual will guide the user
through the entire process of executing the program.
Each manual is short. It is recommended that the user should read the applicable manual
and experiment with the sample data to get familiar with the programs. For practice, a
problem with output is given at the end of the manual.
1. CONTROL INFORMATION
Note that the only buttons present at this time are the Control Information on the top left
region and the Load Sample Data, Browse Input and Exit buttons located on the bottom
right region of the screen. It can be seen that there are five entries, which are highlighted
yellow. The data in these are already filled in and they cannot be edited. All the programs
will have some of the entries highlighted. It is possible to load the sample data by
<Clicking> Load Sample Data button. Also it is possible to load previously stored input file
by <Clicking> Browse Input button. New data has to be typed in this screen. The input in
the Control Information screen dictates the number of rows and/or columns in the tables in
all subsequent screens. It is essential that in this as well as all other screens the entries
504
including 0 (zero) be entered. Once the last entry is typed the Co-ordinates button, will
appear on the left side of the screen. <Clicking> the Co-ordinates button will open the
screen shown in Figure F.3
Figure F.1
2. CO-ORDINATES
Number of rows will be equal to the Number of Nodes and number of columns will be
equal Number of Dimensions specified in the Control Information. For programs P1, P3,
P4, P5, P6 and P7 there will be two columns and for P2 and P8 there will be three
columns. In programs P6 and P7 the co-ordinates are r and z. If there is large number of
nodes, then a vertical scroll bar will appear on the table to facilitate input of data.
505
Figure F.2
506
Figure F.3
Figure F.4
Note that it is possible to import coordinate data from a Microsoft Excel file. (Note that this
® ®
feature requires Microsoft Excel to be installed.) To import the coordinate data <Click>
® ®
the Import Co-ordinates from Excel button. It will open the window shown in Figure F.4.
Browse for the location and import the appropriate file. Typed coordinate data can also be
exported to excel for modification or reuse by <Clicking> the Save Co-ordinates to Excel
button. (Note that this feature also requires Microsoft Excel to be installed. Saving the
® ®
data to Microsoft Excel can be a time consuming process; do not interrupt the process
® ®
due to the risk of data loss from “crashing” the program.) It will open the window shown in
Figure F.4. Browse for the desired location and save the file with an appropriate file name.
As soon as the last entry is entered into the table the Element Data button will appear on
the screen. <Clicking> the Element Data button will open the screen shown in Figure F.5.
3. ELEMENT DATA
In addition to Material Set and No. of nodes per element columns, there will be number of
columns corresponding to Max Node per Elements specified in the Control Information.
For P1, P2 and P3 the Max Nodes per Elements are 2 and the pertinent column will be
highlighted. Similarly, for P4 there are 3 and for P8 there are 8. For programs P5, P6, and
P7 the Max Node per Elements can be from 3 to 9 and the pertinent column will not be
highlighted. If there is large number of elements and/or nodes in an element, then a
vertical and/or horizontal scroll bars will appear on the table to facilitate input of data.
Note that it is possible to import (export) the element data from (to) a Microsoft Excel file.
® ®
(For details see the Co-ordinates section.).

There are two options to view the Graphical Display by <Clicking> the Graphical Display
button. For programs P1, P3, P4, and P5 by not selecting the Scale to panel size option
507
with scale the 700 × 700 pixel screen in each direction. It will take
and multiply the x and y co-ordinates by Nmin. While selecting the selecting the Scale to
Panel option will scale the coordinates in x–direction by and in y–direction by . For
programs P6 and P7 r and z replace the role played by x and y. In programs P2 and P8
and multiply x, y and z co-ordinates by Nmin when Scale to panel

option is not selected. While selecting the selecting the Scale to Panel option will scale the
co-ordinates in x–direction by , in y–direction by and z–direction by .
Figure F.5
Displays from both options are shown in Figure F.6 and F.7, respectively. The screen
contains two buttons, a Print button and Save button. To print the output, <Click> the Print
button. (Note that the output will be sent to the default Microsoft Windows printer.)
® ®
<Clicking> the Save button will open a window similar to the one shown in Figure F.4.
Browse for the desired location and save the JPG file with an appropriate file name.
508
Figure F.6
Figure F.7
509
Figure F.8
As soon as the last entry is entered in the table, the Boundary Conditions and Nodal
Displacement Data button will appear on the screen as seen. <Clicking> the Boundary
Conditions and Nodal Displacement Data button will open the screen shown in Figure
F.8.
4. BOUNDARY CONDITIONS AND NODAL DISPLACEMENT DATA

The screen contains two tables, one for boundary conditions and the other for nodal
displacement data. The number of rows in the boundary conditions table is equal to the
number of supported (constrained) nodes and the number of columns is one plus the
degrees of freedom (DOF) specified in the Control Information. The number of rows in the
nodal displacement data table is equal to the number of nodes where the displacement is
specified and the number of columns is one plus the degrees of freedom specified in the
Control Information. In either table, if there is large number of nodes then a vertical scroll
bar will appear on the table to facilitate input of data.
For programs P1, P4, P5, and P6 there are two DOF while P2, P3, P7 and P8 have three
DOF. For programs P1, P2, P4, P5 and P8 except P3 the DOF refers to the displacements
u, v, and w (if applicable) in Global Co-ordinates. For P3 degrees of freedom are u, v, and
the rotation θ. For programs P6 and P7 the degrees of freedom refer to the displacements
ur,uz, and uθ.
510
As soon as both the tables are filled Material Characteristic Data button will appear on
the screen. <Clicking> the Material Characteristic Data button will open the screen
shown in Figure F.9.
Figure F.9
5. MATERIAL CHARACTERISTIC DATA

The table contains number of rows corresponding to the Number of Material Sets specified
in the Control Information. The columns are for, from left to right, assigning the material set
number, element type, (which are 1 to 8 for programs P1 to P8, respectively), and the
material data shown on the screen for each program. Each Material Set will have different
material data. If there is large number of material sets and/or material data, then a vertical
and/or horizontal scroll bars will appear on the table to facilitate input of data. Note that it is
possible to import (export) the material characteristic data from (to) a Microsoft Excel file.
® ®
(For details, see the Co-ordinates section.)

As soon as the table is filled in, the Nodal Load Data and Distributed Load button will
appear on the screen. <Clicking> the Nodal Load Data and Distributed Load button will
open the screen shown in Figure F.10.
6. NODAL LOAD DATA AND DISTRIBUTED LOAD

This screen contains two tables, one for Nodal Load Data and the other Distributed Load
Data. The number of rows in the Nodal Load Data table is equal to the Number of Loaded
Nodes and the number of columns is one plus the number of degrees of freedom specified
in the Control Information. The Distributed Load Data table if active (it is active only for
511
programs P3, P6 and P7) will have number of rows equal to the Number of Elements with
Distributed Load, and number of columns for P3 will be two and for P6 and P7 are three.
As soon as the last entry is filled in the table, the Execute button is enabled. Also enabled
at this time is the Save Input button. To store all the input data for later use or
modifications <Click> this button and the screen shown in Figure F.4 will be displayed;
browse for the desired location and save the .din file with an appropriate file name. The
program can be executed by <Clicking> the Execute button. When the execution is
finished, the Output and Result buttons will be enabled. <Click> the Output button to
view the output in a window that looks like Figure F.11.
Figure F.10
To view the complete output, scroll by using the scroll bar(s).
To print the output, <Click> the Print button. (Note that the output will be sent to the
default Microsoft Windows printer.) To save the output, <Click> the Save Output button
® ®
and the screen shown in Figure F.4 will be displayed; browse for the desired location and
save the file with an appropriate file name.
<Click> the Result button to view the results in a window that looks like Figure F.12.
To view the complete result, scroll by using the scroll bar(s).
To print the results, <Click> the Print button. (Note that the results will be sent to the
default Microsoft Windows printer.) To save the results, <Click> the Save Results button
® ®
and the screen shown in Figure F.4 will be displayed; browse for the desired location and
save the file with an appropriate file name.
512
Figure F.11
513
Figure F.12
514
Appendix G
COMPUTER PROGRAMS FOR ONE-
DIMENSIONAL AND TWO-
DIMENSIONAL PROBLEMS
G.1 COMPUTER PROGRAM FOR THE FLOW NETWORK ANALYSIS
For the pipe network problem considered in Chapter 6, Section 6.11, a computer program
in one-dimensional has been developed using the Matlab package. The detailed program
listing is given below. The program is tested with the problem mentioned in Section 6.11.1.
The data file is embedded in the program listing.
%---------------------------Listing 1 Matlab Program-------------------------------------
% Program to analyse water distribution network
% written by B.V. Rao and modified by T.I. Eldho, C.E.Dept, I.I.T.,Bombay
% H - pressure heads in meters 1 × n
% pipe_length[1:m]
% Q Discharge in the pipe [1:m]
% A(n,n) global matrix
% n number of nodes
% m number of pipes
% C consumption array [1:n]
% f friction factor
% T transmittivity array [1:m] - (12.1*D^5/(f*L))^0.5
% HF headloss in pipes [1:m]
% R right hand side array [1:n]
% Conn_nodes(m,2) shows how the pipes are connected to nodes
% Valp(npv,2) values to be prescribed before solving global matrix
% npv numbers of nodes where the values are prescribed
% p(m) temporary array
% Dia(m) diameter of pipes
Q=[];HF=[];p=[];R=[];R1=[];
H=[2, 1.91, 1.92, 1.93, 1.94, 1.95];
pipe_length=[1,1,1,1,1,1,1.5,1.5,1.5];
Dia=[0.01,0.01,0.01,0.01,0.01,0.01,0.012,0.012,0.012];
515
C=[125,-25,-25,-25,-25,-25]*1.0e-6/60;
Valp=[1,2];
Conn_nodes=[1 2;2 4;1 3;4 6;3 5;5 6;1 4;3 4;3 6];
m=9;n=6;npv=1;f=0.02; itmax=25;
A=[1:n;1:n];
for i=1:m
p(i)= sqrt(12.1*Dia(i)^5/pipe_length(i)/f);
end
it=0;s1=1.0;
% zeros(size(A));
while s1 > 1.0e-6 & it <= itmax
it=it+1;
for i=1:n
R(i)=C(i);
for j=1:n
A(i,j)=0.0;
end
end
for k=1:m
i=Conn_nodes(k,1);
j=Conn_nodes(k,2);
HF(k)=H(i)-H(j);
if abs(HF(k)) == 0, HF(k)=1.0;end
T=p(k)/sqrt(abs(HF(k)));
A(i,i)=A(i,i)+T;
A(j,j)=A(j,j)+T;
A(i,j)=A(i,j)-T;
A(j,i)=A(j,i)-T;
end
for i=1:npv
j=Valp(i,1);
c1=Valp(i,2);
for k=1:n
R(k)=R(k)-c1*A(j,k);A(k,j)=0.0;A(j,k)=0.0;
end
516
A(j,j)=1.0;R(j)=c1;
end
R1=A\R';
% R1=inv(A)*R;
% error=norm(R1-H);
% H=R1;
% To find the error norm
s2=0.0;
for i=1:n
s3=R1(i)-H(i);
s2=s2+s3*s3;
H(i)=R1(i);
end
s1=sqrt(s2);
%fprintf(‘s1=_.5f it=%5.0f\n’,s1,it);
end
fprintf(‘s1=_.5f it=%5.0f\n’,s1,it);
% The following statements compute the discharge
for k=1:m
i=Conn_nodes(k,1);
j=Conn_nodes(k,2);
HF(k)=H(i)-H(j);
if HF(k) < 0,
Q(k)=-1000*p(k)*sqrt(abs(HF(k)));
else
Q(k)=1000*p(k)*sqrt((HF(k)));
end
end
disp(‘Print Pressure head values at different nodes in met.’);
for i=1:n
fprintf(‘%5.0f _.5f\n’,i,H(i));
end
disp(‘Discharge values in different pipes in lit/s’);
for i=1:m
fprintf(‘%5.0f _.5f\n’,i,Q(i));
517
end
%--------------------program ends--------------------
G.2 COMPUTER PROGRAM FOR THE ELECTRICAL NETWORK ANALYSIS

For the electrical network problem considered in Chapter 6, Section 6.11.2, a computer
program in one-dimensional has been developed using the Matlab package. The detailed
program listing is given below. The program is tested with the problem Section 6.11.2.1.
The data file is embedded in the program listing.
%-----------------------Listing 2 Matlab Program----------------------
% Program to analyse Simple Electrical Network
% written by B.V Rao and modified by T.I. Eldho, C.E.Dept, I.I.T.,Bombay
% A(nnp,nnp) global matrix
% nnp number of nodes
% nel number of elements
% R right hand side array [1:nnp]
% Conn_nodes(nel,2) shows how the elements are connected to nodes
% Valp(npv,2) values to be prescribed before solving global matrix
% npv numbers of nodes where the values are prescribed
R=[];
Voltage=[];
current=[];
%------------------------------------------data set 1--------------------------------
resistance=[250,100,200,300,400,500];
Valp=[1,10;5,0];
Conn_nodes=[1,2;2,3;2,3;3,4;3,4;4,5];
nel=6;nnp=5;npv=2;
A=[1:5;1:5];
%------------------------------------------end of data set 1--------------------------------
for i=1:nnp
R(i)=0.0;
for j=1:nnp
A(i,j)=0.0;
end
end
for k=1:nel
i=Conn_nodes(k,1);
j=Conn_nodes(k,2);
518
g1=1.0/resistance(k);
A(i,i)=A(i,i)+g1;
A(j,j)=A(j,j)+g1;
A(i,j)=A(i,j)-g1;
A(j,i)=A(j,i)-g1;
end
for i=1:npv
j= Valp(i,1);
c1=Valp(i,2);
for k=1:nnp
R(k)=R(k)-c1*A(k,j);
A(k,j)=0.0;
A(j,k)=0.0;
end
A(j,j)=1.0;
R(j)=c1;
end
Voltage=A\R';
% The following statements compute the currents
disp(‘ Print currents in elements and Voltages’);
disp(‘Element No. currents’);
for k=1:nel
i=Conn_nodes(k,1);
j=Conn_nodes(k,2);
v1=Voltage(i);v2=Voltage(j);
c=(v1-v2)/resistance(k);
current(k)=c;
fprintf(‘%3.0f_.5f\n’,k,c);
end
disp(‘Node No. Voltage’);
for i=1:nnp
v1=Voltage(i);
fprintf(‘%3.0f_.5f\n’,i,v1);
end
%---------------------------program ends--------------------------------
519
G.3 COMPUTER PROGRAM FOR THE LAPLACE EQUATION –

HOMOGENEOUS AND NON-HOMOGENEOUS MEDIA PROBLEMS
For the Laplace equation FEM formulation considered in Chapter 7, Section 7.5, a
computer program in 2D has been developed using the FORTRAN programming
language. The detailed program listing is given below. The program can be used for
homogeneous and non-homogeneous media problems. The program is tested with the
problem given in Chapter 7, Section 7.8.
C*****************************************************************************
C Laplace1.for; F.E.M: 2D-PROGRAM - LISTING 7.5.1--ELDHO.T.I., DEPT. CIVIL, IIT
BOMBAY
C -----------------------------------------------------------------------------------------------------
C PROBLEM--POTENTIAL FLOW: POROUS MEDIA; FLOW BENEATH A WEIR-
C HOMOGENEOUS/ NONHOMOGENEOUS MEDIA
C ANALYSIS BY FINITE ELEMENT METHOD USING TRIANGUALR LINEAR
ELEMENTS
C *************************************************************************
C PROGRAM FEM-LAPLACE1
DIMENSION A(99,99),S(3,3),B(99),X(99),Y(99),VIBC(99),NOD(99,3),
& IBC(99),CK(99), D(99)
OPEN(UNIT=9,FILE=‘TAPE9.DAT’,STATUS=‘OLD’)
OPEN(UNIT=10,FILE=‘TAPE10.DAT’,STATUS=‘unknown’)
IN=9
IO=10
C NND= NO. OF NODES (TOTAL)
C NEL= NO. OF ELEMENTS
C NPR= NO.OF BOUNDARY NODES WITH PRESCRIBED HEADS
C NNODE= NO. OF NODES/ELEMENT
C CK= COEFFT. PERMEABILITY
C**********************************************************
C DATA INPUT
C**********************************************************
READ(9,9)NND,NEL,NPR,NNODE
9 FORMAT(4I10)
WRITE(10,10) NND,NEL,NPR,NNODE
10 FORMAT(// ‘ NUMBER OF NODES : ’,I10/
* ‘ NUMBER OF ELEMENTS : ’,I10/
* ‘ NUMBER OF PRI., VAL: ’,I10/
520
* ‘ NUMBER OF NODES/ELE: ’,I10///)

C
DO 20 I=1,NEL
READ(9,*) J,(NOD(J,K),K=1,NNODE),CK(I)
WRITE(10,15)J,(NOD(J,K),K=1,NNODE),CK(I)
15 FORMAT(I10,10X,3I5,5X,F5.2)
20 CONTINUE
C
DO 30 I=1,NND
READ(9,*) IR,X(IR),Y(IR)
WRITE(10,25)IR,X(IR),Y(IR)
25 FORMAT(I5,2F10.2)
30 CONTINUE
C
READ(9,*) (IBC(I),VIBC(I),I=1,NPR)
WRITE(10,35)(IBC(I),VIBC(I),I=1,NPR)
35 FORMAT(I10,F10.2)
C
C***************************************************
C FINITE ELEMENT MATRIX FORMULATION
C***************************************************
DO 60 I=1,NND
B(I)=0.0
DO 60 J=1,NND
A(I,J)=0.0
60 CONTINUE
C
C COMPUTE ELEMENT MATRICES
DO 500 K=1,NEL
K1=NOD(K,1)
K2=NOD(K,2)
K3=NOD(K,3)
X1=X(K1)
X2=X(K2)
X3=X(K3)
521
Y1=Y(K1)
Y2=Y(K2)
Y3=Y(K3)
CK1=CK(K)
A1=X2*Y3-X3*Y2
A2=X3*Y1-X1*Y3
A3=X1*Y2-X2*Y1
B1=Y2-Y3
B2=Y3-Y1
B3=Y1-Y2
C1=X3-X2
C2=X1-X3
C3=X2-X1
DELT=(A1+A2+A3)
C------EVALUATION OF S-MATRIX IE. ELEMENT MATRIX------------------------------C
S(1,1)=CK1*(B1*B1+C1*C1)/(2.*DELT)
S(1,2)=CK1*(B1*B2+C1*C2)/(2.*DELT)
S(1,3)=CK1*(B1*B3+C1*C3)/(2.0*DELT)
C-------1ST ROW OVER--------------------------------------------C
S(2,1)=S(1,2)
S(2,2)=CK1*(B2*B2+C2*C2)/(2.*DELT)
S(2,3)=CK1*(B2*B3+C2*C3)/(2.*DELT)
C-------2ND ROW OVER------------------------------------------------------C
S(3,1)=S(1,3)
S(3,2)=S(2,3)
S(3,3)=CK1*(B3*B3+C3*C3)/(2.*DELT)
C-------3RD ROW OVER----------------------------------------------------C
DO 80 IM=1,NNODE
IR=NOD(K,IM)
DO 80 IN=1,NNODE
IC=NOD(K,IN)
A(IR,IC)=A(IR,IC)+S(IM,IN)
80 CONTINUE
500 CONTINUE
C*********************************************************************
522
C-------SUBSTITUTION OF BOUNDARY CONDITIONS--------------------------------

C*********************************************************************
DO 610 I=1,NPR
N5=IBC(I)
A(N5,N5)=1.E+10
B1=VIBC(I)
B(N5)=1.E+10*B1
610 CONTINUE
C*********************************************************************
C ------------------MATRIX SOLUTION--------------------------------------------------
C*************************************************
CALL SOLVE(A,B,D,NND)
C*************************************************
C --------------------RESULT OUT PUT-----------------------------------------------
C*********************************************************************
WRITE(10,*)‘POTENTIAL VARIATION AT NODES’
WRITE(10,620)(I,B(I),I=1,NND)
620 FORMAT(I10,F10.2)
STOP
END
C*********************************************************************
SUBROUTINE SOLVE(A,B,D,N)
C SOLUTION OF SYSTEM OF EQUATIONS BY GUASS ELIMINATION TECHNIQUE
C A, ARRAY CONTAINING THE COMPLETE SYSTEM MATRIX
C B, ARRAY CONTAINING COEFFICIENTS BEFORE & VALUES OF SYSTEM
C UNKNOWNS AFTER SOLUTION.
C D, AN AUXILLARY VECTOR
C N, ACTUAL NUMBER OF UNKNOWNS
DIMENSION A(99,99),B(99),D(99)
M=N-1
DO 10 I=1,M
L=I+1
DO 10 J=L,N
IF (A(J,I)) 6,10,6
6 DO 8 K=L,N
523
8 A(J,K)=A(J,K)-A(I,K)*A(J,I)/A(I,I)
B(J)=B(J)-B(I)*A(J,I)/A(I,I)
10 CONTINUE
C COMPUTE THE LAST UNKNOWN
D(N)=B(N)/A(N,N)
C APPLY BACK-SUBSTITION TO COMPUTE THE REMAINING UNKNOWNS
DO 30 I=1,M
K=N-I
L=K+1
DO 20 J=L,N
20 B(K)=B(K)-D(J)*A(K,J)
D(K)=B(K)/A(K,K)
30 CONTINUE
DO 40 I=1,N
40 B(I)=D(I)
RETURN
END
C*********************************************************************
G.3.1 Example 1:
Here the input data and results of the Illustrative example problem given in Section 7.8.1
are described. For the problem considered, the potential variation in a homogeneous
isotropic porous media of length 100 m and width 30 m has been found. The Figures and
other details are given in Section 7.8.1.
Input Data:
524
Note: Being homogeneous media problem, Coefficient of permeability is taken as CK=1.0.
525
Results:
NUMBER OF NODES : 20
NUMBER OF ELEMENTS : 24
NUMBER OF PRI., VAL: 8
NUMBER OF NODES/ELE: 3
POTENTIAL VARIATION AT NODES IN m.
1 100.00
2 100.00
3 100.00
4 100.00
5 150.00
6 150.00
7 150.00
8 150.00
9 200.00
10 200.00
11 200.00
12 200.00
13 250.00
14 250.00
15 250.00
16 250.00
17 300.00
18 300.00
19 300.00
20 300.00
G.3.2 Example 2:
Here the input data and results of the Illustrative example problem given in Section 7.8.2
are described. For the problem considered, the potential function variation beneath a
concrete dam on a homogeneous isotropic permeable foundation has been found. Figures
and other details are given in Section 7.8.2.
Input Data (for potential variation):
526
527
528
Results:
NUMBER OF NODES : 39
NUMBER OF ELEMENTS : 44
NUMBER OF PRI., VAL: 4
NUMBER OF NODES/ELE: 3
POTENTIAL VARIATION AT NODES
1 9.39
2 9.44
3 9.56
4 9.75
5 10.00
6 9.22
7 9.28
8 9.41
9 9.65
10 10.00
11 8.59
12 8.82
13 8.83
14 9.13
15 9.16
16 8.92
529
17 9.07
18 7.54
19 7.30
20 7.40
21 7.58
22 6.11
23 6.08
24 6.33
25 6.27
26 6.20
27 6.19
28 6.00
29 6.01
30 5.62
31 5.55
32 5.45
33 5.23
34 5.00
35 5.41
530
36 5.36
37 5.28
38 5.13
39 5.00
Note: In a similar way, the input data for the stream function problem can be constructed
and solved the problem using the Computer Program Listing 7.1 and the results are
obtained.
Disclaimer
The programs contained here are intended for educational use. While every reasonable
effort has been made to ensure that the programs are accurate and error free, use of the
programs is at your own risk.
Nei-ther the authors nor publisher assume any responsibility for errors or omissions in
these materials.
These materials are provided “as is” without warranty of any kind, either express or
implied, including but not limited to, the implied warranties of merchantability or fitness for
a particular purpose.
The authors, publishers, and other parties involved in creating and delivering these
materials shall not be liable for any special, indirect, incidental, or consequential damages,
including without limitation, lost revenues or lost profits, or any other damages, which may
result from the use of these materials.
531
ACKNOWLEDGEMENTS
Authors have gained immensely through the works of several researchers in the field of
FEM and are grateful to all of them. They have lectured FEM-based courses for over a
period of thirty years to under-graduate and post-graduate students at IIT Bombay in India
and at the University of Manitoba in Canada. In addition, they have given many FEM-
based Master's level and Ph.D. level projects. Also Drs Desai and Eldho have offered
FEM-based short-term courses several times to College Teachers and Engineers at IIT
Bombay, India. These activities have helped the authors to refine many ideas and
applications of FEM for teaching. The authors are thankful to all their students, teachers
and colleagues as well as participants of the FEM courses. Special thanks are extended to
all Master's and Ph.D. Scholars whose project-related works have been extensively
referred in the book by the authors. Specifically, the first two authors wish to express their
thanks to former and present students: Dr. G. S. Ramtekkar, M. Arpita, S. Sharique, Dr. S.
M. V. Sharief, M. Meenal and A. Kulkarni and their colleagues Professor Tarun Kant and
Professor E.P. Rao. They are thankful to IIT Bombay authorities and Quality Improvement
Programme (QIP) of IIT Bombay for the partial financial support received for writing this
book.
Dr. Desai acknowledges the help and support received from his former colleagues Dr. R.
Paskaramoorthy and Dr. W. M. Karunasena during initial draft version of the book and
development of simplified finite element analysis program (sfeap). Dr. Eldho wishes to
express his sincere gratitude to the late Professor B. Vasudeva Rao, Department of Civil
Engineering, IIT Bombay, with whom he had a long-term association as a student and
later as a colleague. Some of the text materials presented in this book (especially some
parts of chapters 2, 4 and 5 and MATLAB programs) have been developed by him in
collaboration with Professor Rao. Dr. Shah would like to acknowledge his former graduate
students for making this book a reality, his colleague Dr. J. Frye for assistance with
figures, his current doctoral student Professor D. Stoyko for his help with the manual, and
computer programmers for development of the Graphical User Interface for sfeap.
A book project like this could not have been completed without the full support and
cooperation of the author's family members. Authors are especially grateful to their family
members (Dr. Desai to his wife Nilima and daughter Adveta; Dr. Eldho to his wife and
sons: Dr. Manjush, Iype and Basil; Dr. Shah to his wife Ranjan and daughters: Ketki and
Seema) for their patience and understanding while enduring the “book syndrome”.
Authors wish to express sincere gratitude to the editorial team of Pearson Education,
especially, Thomas Rajesh Mathew, Anita Yadav, Ruchi Sachdev and Gauravjeet Singh.
Finally, the authors wish to thank all those who helped directly or indirectly and made this
book possible.
532

Finite Element Method With Applications in Engineering

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Finite Element Method With Applications in Engineering

Uploaded by

Copyright:

Available Formats

Finite Element Method with Applications in Engineering

3.7 CLOSING REMARKS ............................................................................................. 43

5.4.10 Two-Dimensional serendipity elements .......................................................... 96

7.3.5 Some remarks on Triangular Elements .......................................................... 234

10.7.3 FEM in Geotechnical Engineering ................................................................ 374

F.2 EXECUTING PROGRAMS ................................................................................... 504

1.2 MATHEMATICAL MODELLING OF ENGINEERING PROBLEMS

Figure 1.1 A typical mathematical modelling procedure

1.3 TYPE OF GOVERNING EQUATIONS

Here a, b and c are functions of x and y and α is a function of x, y, h, and . Such a

1.3.1 Initial and Boundary Conditions

Natural or Neumann boundary condition

Mixed boundary condition

1.4 SOLUTION METHODOLOGIES

explored without explicitly solving it in qualitative analysis. Quantitative analysis, on the

1.4.1 Analytical Method

1.4.2 Physical Method

1.4.3 Computational Method

1.5 NUMERICAL MODELLING

and types of boundaries. Direct approach, variational principle or weighted residual

1.6 PRE-PROCESSING AND POST-PROCESSING

1.7 SCOPE OF THE BOOK

1.8 HIGHLIGHTS OF THE BOOK

1.9 HOW TO USE THE BOOK?

1.10 CLOSING REMARKS

REFERENCES AND FURTHER READING

2 Approximate Methods of Analysis

2.2 APROXIMATING METHODS

2.3 METHOD OF WEIGHTED RESIDUALS

Considering only the first two coefficients,

For F to be minimum that yields n equations to determine n values of

where are the weights.

Use boundary conditions ɸ(x = 0) = 0 and ɸ(x = 1) = 0.

Simplifying Eqs. (2.19) and (2.20), the resulting equations are

2.3.4 Galerkin's Method

Use boundary conditions ɸ(x = 0) = 0 and ɸ(x = 1) = 0.

Simplifying equations (2.30) and (2.31), resulting equations are

Solving equations (2.32) and (2.33) yields α1 = −0.14588 and α2 = −0.16278

2.4 RAYLEIGH–RITZ METHOD

Use boundary conditions ɸ(x = 0) = 0 and ɸ(x = 1) = 0.

F satisfies the Euler–Lagrangian equation of variational calculus given as

yielding the differential equation chosen. Further,

substituting all these into Eq. (2.39), , can be obtained,

and integrating separately term by term within the limits x = 0 to 1,

Minimizing Ψ with respect to undetermined coefficients α1, and α2

Solving equations (2.42) and (2.43) simultaneously, α1 = –0.1460 and α2 = –0.1628.

2.5 FURTHER NUMERICAL EXAMPLES

Limiting it to two terms,

Substituting values of in original equation, error ε is found to be

Choosing x = 0.25 and x = 0.5 as collocation points,

Simplifying and making error = 0 at x = 0.5,

Solution of Eqs. (2.50) and (2.51) yields

Method of Least Squares

This can be achieved by differentiating with respect to α1 and α2. Therefore,

Integrating and substituting the limits, it can be shown that

or ɸ″ – ɸ = x2 is obtained. Substituting Eqs. (2.46) and (2.47) into

and integrating separately term by term, following equations are obtained.

Solving, α1 = –0.05496 and α2 = –0.16279

On solving, α1 =1.136, α2 = –0.4765, α3 = 0.3404.

Solving (2.66), (2.67), and (2.68),

According to Galerkin criterion,

with boundary conditions h = 0 at x = ± a, y = ± b, using the Galerkin's method.

Substitution of Eq. (2.71) into Eq. (2.70) yields

By using Galerkin's method,