APPLIED FINITE ELEMENT ANALYSIS

APPLIED FINITE ELEMENT ANALYSIS

Second Edition

LARRY J. SEGERLIND
Agricultural Engineering Department Michigan State University

JOHN WILEY AND SONS

New York • Chichester • Brisbane . Toronto . Singapore

To my parents T. J. and Bessie and my grade school teacher Alice Brunger

Copyright© 1984, by John Wiley & Sons, Inc. All rights reserved. Published simultaneously in Canada. Reproduction or translation of any part of this work beyond that permitted by Sections 107 and 108 of the 1976 United States Copyright Act without the permission of the copyright owner is unlawful. Requests for permission or further information should be addressed to the Permissions Department, John Wiley & Sons.
Library of Congress Cataloging in Publication Data:

Segerlind, Larry J., 1937Applied finite element analysis. Bibliography: p. 411. Includes index. 1. Finite element method. I. Title. TA347.F5S43 1984 620'.001'515353 Printed in the United States of America 10 9 8 7 6 5 4 3

84-7455

Preface

The finite element method is a widely accepted numerical procedure for solving the differential equations of engineering and physics and is the computational basis of many computer-aided design systems. The teaching of the fundamentals of the finite element method is rapidly becoming a necessity in those curricula which solve problems in the general areas of structural analysis, continuum mechanics, heat transfer, and fluid flow. This is an introductory textbook covering the basic concepts of the finite element method and their application to the analysis of plane structures and twodimensional continuum problems in heat transfer, irrotational fluid flow, and elasticity. The topics can be handled by advanced senior and beginning graduate students and have been taught in a course at this level for 10 years. No prior knowledge of structural analysis or the finite element method is assumed. The major differences between this book and the first edition are the organization, the inclusion of five new chapters that introduce the analysis of plane structures, an increase in the number of homework problems, and the use of Galerkin's method in conjunction with the solution of field problems. The organization of the material is unique and makes the book more useful as a teaching tool. The book consists of four parts: basic concepts, field problems, structural and solid mechanics problems, and linear and quadratic elements. The basic concepts cover six chapters and contain information needed to study field problem and/or structural and solid mechanics applications. Once the basic concepts have been completed, either Part Two, the field problems, or Part Three, structural and solid mechanics, can be studied. These two sections of the book are independent; the students can study either part or chapters from both parts depending on their interests. Part Four, linear and quadratic elements, covers general procedures for developing the element shape functions and numerically integrating the element matrices. This part can be covered after some of the continuum applications have been completed. The organization of the book offers the instructor at least three different teaching options: (1) a general finite element course with topics selected from both field and solid mechanics applications, (2) a course emphasizing the solution of field problems, or (3) a course emphasizing the solution of solid mechanics problems. The author teaches the first type of course and covers Chapters 1 through 11, 17, 18, 23, and 24 in approximately 40 lectures.

viii

PREFACE

The chapters that introduce the analysis of plane structures are included for agricultural and mechanical engineering students who need an introduction to the matrix analysis of structures but who do not have time to take several civil engineering courses to get the material. The displacement method of structural analysis based on the principle of minimum potential energy is presented. This formulation is consistent with the finite element analysis of elasticity problems. Galerkin's method has been used to solve the field problems because it is an approach that is more readily accepted by seniors and beginning graduate students. These students have not had any variational calculus; hence, the concept of a functional is often quite mysterious to them. Galerkin's method offers two primary advantages. The inclusion of derivative boundary conditions is a straightforward procedure, and the student will find that he or she has studied a method that can be used to formulate differential equations with a first-derivative term. The disadvantage of Galerkin's method is the surfacing of the interelement requirements, which are never used and are somewhat difficult to explain away. It should be noted, however, that the interelement requirements also occur in the variational formulation when it is correctly applied. The variational method is not presented correctly in most finite element books. The number of homework problems has been increased to over 300 with most chapters having at least 10 and many having 15 or more. The problems are a mixture of numerical calculation, analytical derivations, or the evaluation of important integrals and problems requiring a computer solution. A complete solution manual is available from the publisher. The finite element method must be implemented on a digital computer; therefore, four computer programs have been included. These programs are written specifically for the beginning user and contain diagnostic checks to detect errors made by first-time users. These checks are very efficient and eliminate most ~f the student-instructor contact relative to finding data input errors. I would like to thank the many students in MMM809 who used the original handwritten pages of this book as well as several typed versions. Their questions and suggestions were an invaluable contribution and have influenced the complete organization of the book as well as specific paragraphs within the text. I would also like to thank my wife Donna for her patience during the writing of this manuscript. East Lansing, Michigan Larry J. Segerlind

Contents
PART ONE
Chapter 1

BASIC CONCEPTS
INTRODUCTION

1.1 1.2

1.3 1.4 1.5

Solution of Boundary Value Problems Integral Formulations for Numerical Solutions 1.2.1 Variational Method 1.2.2 Collocation Method 1.2.3 Subdomain Method 1.2.4 Galerkin's Method 1.2.5 Least Squares Method Potential Energy Formulations The Finite Element Method Objective and Organization Problems LINEAR ELEMENT

3 4
6

7 8 8 9 9 10 11 13 13 17
17

Chapter 2

ONE-DIMENSIONAL

2.1 2.2 2.3 2.4

Division of the Region into Elements The Linear Element A Continuous Piecewise Smooth Equation A Comment on Notation Problems

18 21 22 22 27 27 29 30 32 34 35 40 40 43 44 47 50 50

Chapter 3

A FINITE ELEMENT EXAMPLE

3.1 3.2 3.3 3.4 3.5

Chapter 4

Weighting Functions The Weighted Residual Integral Evaluation of the Integral Analysis of a Simply Supported Beam Matrix Notation Problems GALERKIN FORMULATION

ELEMENT MATRICES:

4.1 4.2 4.3 4.4 4.5

Element Matrices Direct Stiffness Method Analysis of a Simply Supported Beam Properties of the Global Stiffness Matrix General Flow of the Computations Problems

CONTENTS 5 TWO-DIMENSIONAL 5.1 5.2 5.3 5.4 ELEMENTS 51 51 56 61 64 66 68 68 70 73 73 78 80

CONTENTS Flow Around a Cylinder 10.3.1 10.3.2 Regional Aquifer Problems Chapter 11 HEAT TRANSFER 11 .1 11.2 11.3 11.4 BY CONDUCTION AND CONVECTION

xi
130 132 134 138 138 142 144 145 145 148 148 148 151 , 158 158 162 164 165 165 166 168 170 173 176

Chapter

Two-Dimensional Grids Linear Triangular Element Bilinear Rectangular Element A Continuous Piecewise Smooth Equation Problems SYSTEMS

Chapter

COORDINATE 6.1 6.2 6.3 6.4 6.5

Local Coordinate Systems Natural Coordinate Systems Rectangular Element Triangular Element: Area Coordinates Continuity Problems

11.5

The One-Dimensional Fin The Composite Wall The Two-Dimensional Fin Long Two-Dimensional Bodies 11 .4.1 Convection Boundary Condition 11.4.2 Heat Flux into the Body 11.4.3 Concluding Remarks A Computer Example Problems VIBRATIONS

PART TWO
Chapter 7

FIELD PROBLEMS
TWO-DIMENSIONAL 7.1 7.2 7.3 7.4 FIELD EQUATION

Chapter 85

12

ACOUSTICAL 12.1 12.2

87 87 89 91 94 97 100 100 101 105 107 109 111 Chapter 13

One- Dimensional Vibrations Two- Dimensional Vibrations Problems FIELD PROBLEMS

Governing Differential Equations Integral Equations for the Element Matrices Element Matrices: Triangular Element Element Matrices: Rectangular Element Problems OF NONCIRCULAR SECTIONS

AXISYMMETRIC 13.1 13.2 13.3 13.4 13.5

Chapter

TORSION 8.1 8.2 8.3 8.4 8.5

General Theory Twisting of a Square Bar Shear Stress Components Evaluation of the Twisting Torque Computer Solutions for the Square Bar Problems

The Differential Equation Axisymmetric Elements Galerkin's Method Element Matrices The Derivative Boundary Condition Problems

Chapter

14

TIME-DEPENDENT FIELD PROBLEMS: THEORETICAL CONSIDERATIONS 14.1 14.2 14.3 14.4 14.5 Galerkin's Method The Consistent Formulation The Lumped Formulation Finite Difference Solution in Time Heat Flow in a Rod Problems

177 177 178 180 183 185 188

Chapter

DERIVATIVE BOUNDARY CONDITIONS: POINT SOURCES AND SINKS 9.1 9.2 9.3 Derivative Boundary Conditions Evaluation of the Element Integrals Point Sources and Sinks Point Sources and Sinks FLOW

115 115 118 121 124 126 126 126 127 129 130 Chapter 15

TIME-DEPENDENT FIELD PROBLEMS: PRACTICAL CONSIDERATIONS 15.1 15.2 15.3 Physical Reality Numerical Oscillations Lumped or Consistent Formulation? Consistent Formulation 15.3.1 15.3.2 Lumped Formulation 15.3.3 Summary Two-Dimensional Elements 15.4.1 The Rectangular Element

190 190 191 194 194 195 196 196 196

Chapter

10

IRROTATIONAL 10.1

10.2 10.3

Flow of an Ideal Fluid 10.1.1 Streamline Formulation 10.1.2 Potential Formulation Groundwater Flow Computer Examples

15.4

xii
15.4.2 The Triangular Element 15.4.3 Summary Derivative Boundary Conditions Problems FOR TWO-DIMENSIONAL

CONTENTS 197 199 199 200

CONTENTS 21.2 21.3 The Element Stiffness Matrix The Internal Forces Problems

xiii
278 280 283 286 286 288 289 289 290 291 292 293 293 293 294 295 297 304 305 309 314 Coordinates 314 316 317 320 322

15.5 Chapter 16

COMPUTER PROGRAM FIELD PROBLEMS 16.1 16.2 16.3

Chapter 202 202 203 219

22

THEORY OF ELASTICITY 22.1 22.2 22.3 Stress, Strain, and Hooke's Law The Strain Displacement Equations The Element Matrices The Strain Energy Equation 22.3.1 The Work Terms 22.3.2 The Element Matrices 22.3.3 The Stress Components ELASTICITY

..
Chapter 17

The Equations Program Listing An Example Problem

PART THREE

STRUCTURAL

AND SOUD MECHANICS

225 Chapter 227 227 228 229 231 235 235 23

22.4

TWO-DIMENSIONAL 23.1

THE AXIAL FORCE MEMBER 17.1 17.2 17.3 17.4 17.5 The One-Dimensional Model Principle of Minimum Potential Energy The Strain Energy Equation A System of Axial Force Members Matrix Notation Problems POTENTIAL ENERGY

23.2 23.3 23.4 23.5

Plane Stress and Plane Strain Plane Stress 23.1.1 Plane Strain 23.1.2 The Displacement Equations The Element Matrices Element Stresses Discussion of a Computer Example Problems

Chapter

18

ELEMENT MATRICES; FORMULATIONS 18.1 18.2 18.3 18.4

238 238 240 242 244 245 246 246 247 250 255 261 261 263 263 265 267 271 277 277

Chapter

24

AXISYMMETRIC 24.1 24.2 24.3 24.4

ELASTICITY

The Axial Force Element A System of Axial Force Members A General Formulation Internal Forces Problems

Definitions in Cylindrical Axisymmetric Elasticity Element Matrices Surface Loads Problems

Chapter

19

THE TRUSS ELEMENT 19.1 19.2 19.3 The Structural Model The Element Matrices Analysis of a Pinned Truss Problems

Chapter

25

COMPUTER PROGRAMS AND SOLID MECHANICS 25.1 25.2 25.3 25.4

FOR STRUCTURAL

326 326 338 340 351

Program FRAME An Example Problem for FRAME Program STRESS An Example Problem for STRESS

Chapter

20

A BEAM ELEMENT 20.1 20.2 20.3 20.4 20.5 The Structural Model The Strain Energy Equation The Displacement Equation The Element Stiffness Matrix Analysis of a Statically Indeterminate Problems

PART FOUR
Chapter 26

LINEAR AND QUADRATIC

ELEMENT SHAPE FUNCTIONS 26.1 26.2 26.3 26.4 26.5

ELEMENTS

357 359 359 361 362 363 365

Beam

Chapter

21

A PLANE FRAME ELEMENT 21.1 The Structural Model

Local Node Numbers Evaluating the Shape Functions The One-Dimensional Element Triangular Elements Quadrilateral Elements

xiv
26.5.1 26.5.2 26.5.3 The Linear Quadrilateral Element The Lagrangian Element The Eight-Node Quadratic Element Problems

CONTENTS

365 366 367 368 371 371 371 372 375 376 376 377 378 380 382 383

APPLIED FINITE ELEMENT ANALYSIS

Chapter 27

ELEMENT MATRICES 27.1 Changing the Variables of Integration 27.1.1 One-Dimensional Integrals 27.1.2 Two-Dimensional Integrals Numerical Integration Techniques 27.2.1 One-Dimensional Integrals 27.2.2 Quadrilateral Regions 27.2.3 Triangular Regions An Integration Example Evaluating [8] Evaluating the Surface Integrals Problems COMPUTER PROGRAMS and {f(e)}

27.2

27.3 27.4 27.5 Chapter 28

ISOPARAMETRIC 28.1 28.2 28.3 28.4 28.5 Computer

387 387
389 390 391 405 411 413

Evaluation of [k(e)]

Curved Boundaries Solving Time- Dependent Field Problems The Computer Program ISOFLD A Computer Example

REFERENCES APPENDIX APPENPIX APPENDIX INDEX 1 2 3 MATRIX NOTATION DIFFERENTIATION MODIFYING OF MATRIX EQUATIONS

414 417 421 425

THE SYSTEM OF EQUATIONS

ANSWERS TO SELECTED PROBLEMS

PART ONE
BASIC CONCEPTS
The first six chapters cover concepts basic to all applications of the finite element method. These chapters should be covered before going to the application chapters.

Chapter

INTRODUCTION
The finite element method is a numerical procedure for obtaining solutions to many of the problems encountered in engineering analysis. It has two primary subdivisions. The first utilizes discrete elements to obtain the joint displacements and member forces of a structural framework. The second uses the continuum elements to obtain approximate solutions to heat transfer, fluid mechanics, and solid mechanics problems. The formulation using the discrete elements is referred to as the "matrix analysis of structures" and yields results identical with the classical analysis of structural frameworks. The second approach is the true finite element method. It yields approximate values of the desired parameters at specific points called nodes. A general finite element computer program, however, is capable of solving both types of problems and the name "finite element method" is often used to denote both the discrete element and the continuum element formulations. The finite element method combines several mathematical concepts to produce a system of linear or nonlinear equations. The number of equations is usually very large, anywhere from 20 to 20,000 or more and requires the computational power of the digital computer. The method has little practical value if a computer is not available. It is impossible to document the exact origin of the finite element method because the basic concepts have evolved over a period of 150 or more years. The method as we know it today is an outgrowth of several papers published in the 1950s that extended the matrix analysis of structures to continuum bodies. The space exploration ofthe 1960s provided money for basic research, which placed the method on a firm mathematical foundation and stimulated the development of multiple-purpose computer programs that implemented the method. The design of airplanes, missiles, space capsules, and the like, provided application areas. Although the origin of the method is vague, its advantages are clear. The method is easily applied to irregular-shaped objects composed of several different materials and having mixed boundary conditions. It is applicable to steady-state and timedependent problems as well as problems involving nonlinear material properties. General computer programs that are user-independent can be, and have been, developed. User-assisting programs that generate a grid from a limited number of shape-defining points are available as well as programs that analyze the results and display them in graphic form for further study. The finite element method is the computational basis of many computer-assisted design programs. The increased use of computer-assisted design makes it imperative that the practicing engineer have a knowledge of how the finite element method works.

4 1.1

BASIC CONCEPTS

INTRODUCTION

SOLUTION OF BOUNDARY VALUE PROBLEMS

The best way to solve any physical problem governed by a differential equation is to obtain the analytical solution. There are many situations, however, where the analytical solution is difficult to obtain. The region under consideration may be so irregular that it is mathematically impossible to describe the boundary. The configuration may be composed of several different materials whose regions are mathematically difficult to describe. Problems involving anisotropic materials are usually difficult to solve analytically, as are equations having nonlinear terms. A numerical method can be used to obtain an approximate solution when an analytical solution cannot be developed. All numerical solutions produce values at discrete points for one set of the independent parameters. The complete solution procedure is repeated each time these parameters change. Numerical solutions, however, are more desirable than no solution at all. The calculated values provide important information about the physical process even though they are at discrete points. There are several procedures for obtaining a numerical solution to a differential equation. The methods can be separated into three basic groupings: (1) the finite difference method, (2) the variational method, and (3) the methods that weight a residual. These methods are briefly discussed in the following paragraphs. Finite Difference Method The finite difference method approximates the derivatives in the governing differential equation using difference equations. This method is useful for solving heat transfer and fluid mechanics problems and works well for two-dimensional regions with boundaries parallel to the coordinate axes. The method, however, is rather cumbersome when regions have curved or irregular boundaries, and it is difficult to write general computer programs for the method. Variational Method The variational approach involves the integral of a function that produces a number. Each new function produces a new number. The function that produces the lowest number has the additional property of satisfying a specific differential equation. To help clarify this concept, consider the integral
(1.1)

The process can be reversed. Given a differential equation, an approxima~e solution can be obtained by substituting different trial functions int~ the appropr~ate functional. The trial function that gives the minimum value of Il IS the approximate solution. The variational method is the basis for many finite element formulations, but it has a major disadvantage: It is not applicable to any differential equation containing a first derivative term. Weighted Residual Methods The weighted residual methods also involve an integral. In these methods, an approximate solution is substituted into the di~erential ~quation. Since the approximate solution does not satisfy the equation, a residual or. er~or t~rm results. Suppose that y =h(x) is an approximate solution to (1.2).Substitution gives
D d2h~) +Q=R(x)=foO dx

(1.3)

since y that

=h(x)

does not satisfy the equation. The weighted residual methods require

Wi(x)R(x)dx=O

(1.4)

The residual R(x) is multiplied by a weighting function Wi(x), and the integral of the product is required to be zero. The number of weighting functions equals the number of unknown coefficients in the approximate solution. There are several choices for the weighting functions, and some of the more popular choices have been assigned names. Collocation Method. Impulse functions Wi(x)=J(xX;) are selected as the weighting functions. This selection is equivalent to requiring the residual to vanish at specific points. The number of points selected equals the number of undetermined coefficients in the approximate solution. Subdomain Method. Each weighting function is selected as unity, Wi (x) = 1,over a specific region. This is equivalent to requiring the integral of the residual to vanish over an interval of the region. The number of integration intervals equals the number of undetermined coefficients in the approximate solution. Galerkin's Method. Galerkin's method uses the same functions for Wi(x) that were used in the approximating equation. This approach is the basis of the finite element method for problems involving first-derivative terms. This method yields the same result as the variational method when applied to differential equations that are self-adjoint. Galerkin's method is used to develop the finite element equations for the field problems discussed in this book.

The numerical value of Il can be calculated given a specific equation Y= f(x). The calculus of variations shows, however, that the particular equation y=g(x), which yields the lowest numerical value for Il, is the solution to the differential equation D with the boundary conditions
d2y
dx2+Q=O

(1.2)

y(O) = Yo

and

y(H)=YH.

BASIC CONCEPTS

INTRODUCTION

Least Squares Method. The least squares method utilizes the residual as the weighting function and obtains a new error term defined by
Er=

The coefficient EI represents the resistance of the beam to deflection and M(x) is the bending moment equation. In this example, M(x) has the constant value Mo· An approximate equation for the beam deflection is
y(x)=A

[R(xW

dx

(1.5)

sin H

nx

(1.8)

This error is minimized with respect to the unknown coefficients in the approximate solution. The least squares method has been utilized to formulate finite element solutions, but it is not as popular as Galerkin's method and the variational approach. The variational method and the weighted residual methods each involve an integral. These methods can be grouped under the heading of integral formulations. A numerical solution based on an integral formulation is a new concept for many people; therefore, some of the more common methods are illustrated by solving a simple problem. 1.2 INTEGRAL FORMULATIONS FOR NUMERICAL SOLUTIONS The immediate objective is to illustrate how each of the integral methods discussed in the previous section can be used to obtain an approximate solution to a physical problem. The example is a simply supported beam subjected to concentrated moments at each end. The beam and its bending moment diagram are shown in Figure 1.1. The governing differential equation is
EI dx2 - M(x)=O

where A is an undetermined coefficient. This solution is an acceptable candidate because it satisfies the boundary conditions y(O) = y(H)=O and has a shape similar to the expected deflection curve. The exact solution of the differential equation is
Mox y(x)= 2EI (x- H) (1.9)

1.2.1

Variational Method

The integral for the differential equation (1.6) is n=

r Jo

[EI

y (ddx)2

+MoY

] dx

(1.10)

The value of A that makes (1.8) the best approximation to the deflection curve is the value that makes Il a minimum. To evaluate A, Il must be written as a function of A and then minimized with respect to A. Noting that -=-cosdx dy An nx

H
nx] +MoA sin H dx

d2y

(1.6)

we find that Il becomes n=

with the boundary conditions

y(O) =0

r L2 !icos H Jo rEI (An nx)2

H

and

y(H)=O

(1.7)

or (1.11 ) Minimizing Il yields

~-----------H----------~~
M(x)

and
M(x)=MO
LL_

(1.12)
~~x end moments.

The approximate solution is

y(x)=4MoH2 . nx n3EI sin H (1.13)

Figure 1.1.

A simply supported beam with concentrated

8 1.2.2 Collocation Method

BASIC CONCEPTS

INTRODUCTION

The approximate solution is

y(x)=-

The collocation method requires that the residual equation for the approximate solution be zero at as many points as there are undetermined coefficients. The residual is obtained by substituting (1.8) into (1.6). The result is
R(x) = -EI An2 tt x H2 sin H -Mo

--sin2nEI

MoH2

nx H

(1.18)

1.2.4 (1.14)

Galerkin's Method

since
d2y An2. -=---SIndx? H2 nx H

When using Galerkin's method, f W;(x)R(x) dx is evaluated using the same functions for W;(x) that were used in the approximate solution. In this example there is only one weighting function, W;(x)=sin nxl H. The residual equation is (1.14) and the integral is

There is only one undetermined coefficient; therefore, R(x) is equated to zero at one point between 0 and H. Selecting x=H/2 for convenience, we obtain
R(~)=-EI:22

An2 sin-nx[ -EI-2-sin- ttx -Mo ] dx=O H H H

Integrating yields

sin~ -Mo=O Solving gives

and
A __ MoH2 EIn2

(1.15) and the approximate solution is

(1.19)

The approximate solution is

y(x)=MoH2 nx -E 2 sinIn H

(1.16)

y(x) = -

4MoH2 nx n3 EI sin H

(1.20)

Had a point other than x=H/2 would have been obtained. 1.2.3 SubdomainMethod

been selected, a different approximate solution

This solution is identical to the solution obtained using the variational method. 1.2.5 Least SquaresMethod

The subdomain method requires that f R(x)dx=O over as many subintervals as there are undetermined coefficients. The user can choose how long to make each subinterval. In the present example there is only one unknown coefficient; thus the interval must be [0, H]. The residual equation is (1.14); thus Jo R(x)dx= Integration yields

A new error term, Er =f [R(x)Y dx, is formed when using the least squares method. Substitution of the residual equation gives
Er = Jo

[EIn2

A sin H - M 0

nx

J2 dx

Integration gives
Er_A2H(EIn2)2 2 H2

Jo

H [

-EI

An2 nx ] H2 sin H -Mo dx=O

4MoEInA H

M2H
0

The error is minimized with respect to A producing

-(2EIn) II
and
A=-

A-MoH=O

iJEr _ iJA -AH

(EIn2)2 H2

4MoEIn H

(1.21)

After solving for A, the approximate solution is

MoH2 2nEI

(1.17)

y(x)=-

4MoH2

--sinn3EI

nx H

(1.22)

10

BASIC CONCEPTS

INTRODUCTION

11

......
".' '" "'" '"
,. "" '"

__ .-----_

..

• ,. 'Subdomain method

~
QJ

<: 0

't "0
.s

•
0

,. "
;'

0.125

0.25

0.375

0.50

approached in several ways. The classical approach is to formulate the governing differential equation and obtain the analytical solution. This does not work for many problems because of difficulties in mathematically describing the structural geometry and/or the boundary conditions. A popular alternative to the classical approach is a numerical procedure based on a principle which states that the displacements at the equilibrium position occur such that the potential energy of a stable system is a minimum value. A contributing term to the potential energy is the strain energy. This is the energy stored during the deformation process. The strain energy is a volume integral involving products of the stress and strain components. For example, the strain energy in an axial force member is A=

g
"QJ

1:: !:'
QJ QJ

Jv

UxxExx

dV

(1.23)

a.. .~ocation

.,.,. """

"-

._-.
method

More will be said about the principle of minimum potential energy and strain energy in later chapters. The important concept to realize now is that the displacement analysis of structural and solid mechanics problems combines the strain energy integral with a minimization process. Computationally, the analysis of a truss or plate structure looks very similar to the variational and Galerkin approaches of the previous section. The Similarity should be apparent by the time the reader has completed this book. 1.4 THE FINITE ELEMENT METHOD

Figure 1.2. Percentage error for the five approximate solutions for the simply supported beam.

The finite element method is a numerical procedure for solving physical problems governed by a differential equation or an energy theorem. It has two characteristics that distinguish it from other numerical procedures: 1. The method utilizes an integral formulation to generate a system of algebraic equations. 2. The method uses continuous piecewise smooth functions for approximating the unknown quantity or quantities. The second characteristic distinguishes the finite element method from other numerical procedures that utilize an integral formulation. Recall the approximate solution used in the previous section, y=A sin ttxl H, This function is continuous and has an infinite number of continuous derivatives. The finite element method uses a continuous function but a function with only enough continuity in the derivatives to allow the integrals to be evaluated. For an integral formulation such as the variational method (1.10), no continuity is required in the first derivative. The integral can be evaluated when the first derivative is piecewise continuous. An equation composed of several linear segments can be used as the approximating equation. A finite element model for the beam deflection problem considered in the previous section might appear as shown in Figure 1.3. It could consist of several linear segments defined in terms of the nodal values, as shown in Figure 1.3a.

and is identical to the solutions obtained using the variational method and Galerkin's method, (1.13) and (1.20). It is not possible to state which method is the most accurate. The error depends on the approximating function and the equation being solved. Percentage error curves for the different methods are given in Figure 1.2. It appears that equation (1.22) is more accurate than either equation (1.16) or (1.18).However, it is possible to find a collocation point which produces a maximum deflection that agrees with the exact value (see Problem 1.16). The collocation points or subregions selected affect the accuracy of the approximate solutions. The important point to be gained from these examples is that the numerical solution of a differential equation can be formulated in terms of an integral. The integral formulation is a basic characteristic of the finite element method. 1.3 POTENTIAL ENERGY FORMULATIONS

The solution of solid mechanics problems, which includes the solution of two- and three-dimensional elasticity problems as well as plate and shell structures, can be

12

BASIC CONCEPTS
y

INTRODUCTION

13

r;==================~
~

~~z
1234567
(a) 4 2

_x

4. Solve the system of equations. 5. Calculate quantities of interest. These quantities are usually related to the derivative of the parameter and include the stress components, and heat flow and fluid velocities.

1.5

OBJECTIVE AND ORGANIZATION

Discontinuous slope (b)

Figure 1.3.

(a) A linear finite element model. (b) A quadratic finite element model.

The interval between each node would be considered an element, and the deflection is approximated by straight-line segments. An alternate grid could consist of three elements each defined by three node points as shown in Figure l.3b. In this case a quadratic equation is defined over each set of three points. In either case, the equations Y= f(x) or y =g(x) would not have a continuous first derivative between any pair of adjacent elements. Functions without continuous first-derivative terms can also be used with Galerkin's method. The second-derivative term, d2y/dx2, is modified using integration by parts. The finite element method can be subdivided into five basic steps. These steps are listed here and illustrated in the next two chapters. 1. Discretize the region. This includes locating and numbering the node points, as well as specifying their coordinates values. 2. Specify the approximation equation. The order of the approximation, linear or quadratic, must be specified and the equations must be written in terms of the unknown nodal values. An equation is written for each element. 3. Develop the system of equations. When using Galerkin's method, the weighting function for each unknown nodal value is defined and the weighted residual integral is evaluated. This generates one equation for each unknown nodal value. In the potential energy formulation, the potential energy of the system is written in terms of the nodal displacements and then is minimized. This gives one equation for each of the unknown displacements. .

The objective of this book is to provide a basic introduction to the finite element method as it is used to obtain solutions to heat transfer, irrotational flow and elasticity problems, and the analysis of two-dimensional structural frameworks. This book is not a comprehensive textbook on the finite element method. You should, however, be able to read and understand more advanced books and the technical literature once the material in this book has been completed. This textbook contains many example problems and problem assignments. The organization of the material has evolved over several years of teaching the material to seniors and graduate students. The book is divided into four parts: basic concepts, field problems, structural and solid mechanics problems, and numerically integrated elements. Once the basic concepts have been covered, the reader may turn to either the group of chapters covering the field problems or those chapters discussing the structural and solid mechanics problems. These two parts of the book are independent. The numerically integrated elements can be covered after either the field problems or the mechanics problems have been studied. Initial emphasis is placed on the linear elements because the matrices for these elements can be evaluated using a hand calculator. The numerically integrated elements should be covered only after the linear elements are thoroughly understood. A good knowledge of undergraduate mathematics, including some linear algebra, is all that is necessary to handle the material in the first three parts. It is assumed that the reader has a background knowledge in some of the application areas. A limited knowledge of advanced calculus is needed to understand the numerically integrated elements. A comment on the matrix notation used in this book appears in Appendix I.

PROBLEMS 1.1-1.5 Obtain an approximate displacement equation for the simply supported beam of length H and section property E1 shown in Figure Pl.l. Assume that the trial displacement equation is y(x) = A sin 1tx/ H. Com pare the deflection at the center with the theoretical value y= -5WH4/384E1. The governing differential equation is E1 dx2 d2y Wx(H - x)

14

BASIC CONCEPTS

INTRODUCTION

15

M(X)~
1.1 ll= 1.2

iI
Wx(H - x)

~-----------H------------~
JI.( _ y""x ) -

1.8 Evaluate A using the subdomain method. 1.9 Evaluate A using Galerkin's method. 1.10-1.13 Obtain an approximate displacement equation for the simply supported beam shown in Figure P1.10 using the trial solutiony(x)=A sin nx/H. Compare the deflection at the center with the theoretical value of y = PH3/48EI. The governing differential equations are
EI--

~
Figure P1.1

d2y dx2

Px -=0

dy EI dx2

P I(H-x)=O
p

-~x~H 2

Evaluate A by minimizing the integral

IH[~IG~Y

+(WX(~-X»)y

JdX

Evaluate A by requiring that the residual vanish at (a) x=H/3, and (b)x=H/2. 1.3 Evaluate A using the subdomain method. 1.4 Evaluate A using Galerkin's method. 1.5 Evaluate A using the least squares method. 1.6-1.9 Obtain an approximate displacement equation for the simply supported beam shown in Figure P1.6 using the trial solution y(x)=A sin nx/H. Compare the deflection at the center with the theoretical value y= -0.06415MoH2/EI. The governing differential equation is
EI d y _ Mox =0 dx? H
2

EI

M(X:L2<:X)
Figure P1.1 0

M(X)~
~x

~==========E=I=========~MO J)); ~
""I--,__--------H-----------~

1.10 Evaluate A by minimizing the integrals ll=

EI -y - (d )2 dx+ o 2 dx I

IH/2
0

Pxy -dx+ 2

iH
H/22

P -(H-x)ydx

Figure P1.6

1.6

Evaluate A by minimizing the integral'

I.U Evaluate A by requiring the residual to vanish at x=H/2. 1.12 Evaluate A using the subdomain method. 1.13 Determine the collocation point for which A is equal to the deflection at the center of the beam, that is, A = - PH 3/48EI. 1.14 Obtain an approximate displacement equation for the simply supported beam in Figure 1.1 using the equation nx 3nx y(x)=A sin H +B sin If in conjunction with the variational method. Hint: II must be minimized with respect to both A and B. 1.15 Do Problem 1.14 by requiring that the residual vanish at x=H/4 and

n.,
1.7

r[~1 ~Y+ M;X -lG

Evaluate A by requiring the residual to vanish at (a) x=H/2, and (b) X= 0.577H.

16
x=H/2.

BASIC CONCEPTS

1.16

1.17 1.18 1.19

Both points must be used because there are two unknown coefficients, A and B. Determine the collocation points x/H for the beam in Figure 1.1 that yields A=-MoH2/8EI (the correct value of the maximum deflection). Use y(x)=A sin 7tx/H as the approximate equation. Evaluate A using y(x)=A(x2-xH) for an approximate solution for the beam shown in Figure 1.1. Use the variational method. Evaluate A for the equation and beam in Problem 1.17 by applying the collocation method at x=H/2_ Calculate the value of Il for the beam in Figure 1.1 using (1.10) and the exact solution (1.9). Verify that the value for Il for the exact solution is less than the value for the approximate solution (1.13).

Chapter

ONE-DIMENSIONAL LINEAR ELEMENT

Our immediate objective is to discuss the division of a one-dimensional region into linear elements and to develop an element equation. The element equation is then generalized so that a continuous piecewise smooth equation can be written for the region. The linear element is used to obtain an approximate solution to
D~:~ +Q=O (2.1)

in the next chapter. This element is also used to calculate the displacements in a system of axial force members, Chapter 17. 2.1 DIVISION OF THE REGION INTO ELEMENTS

The one-dimensional region is a line segment and the division into subregions or elements is quite 'straightforward. The line segment is divided into shorter segments by using nodes (Figure 2.1). The nodes are usually numbered consecutively from left to right as are the elements. The element numbers are enclosed in parentheses to distinguish them from the node numbers. There are some rules to guide the placement of the nodes when obtaining an approximate solution to a differential equation.

(1)

(2)

(3)

(4)

5 region into elements.

Figure 2.1

Division of a one-dimensional

18

BASIC CONCEPTS

ONE-DIMENSIONAL

LINEAR ELEMENT

19

1. Place the nodes closer together in the regions where the unknown parameter changes rapidly and further apart where the unknown is relatively constant. 2. Place a node wherever there is a stepped change in the value of the coefficients D and Q of (2.1). 3. Place a node wherever the numerical value of 4> in (2.1) is needed. The first rule requires that the user have some knowledge of how the unknown parameter varies. This is where engineering experience enters the solution process. The second rule is necessary because integrals that include the parameters D and Q of (2.1) must be evaluated. The integrals are easier to ·evaluate if the coefficients do not experience a stepped change within the interval of integration. 2.2 THE LINEAR ELEMENT

to develop the pair of equations

<l>i=al <l>j=al +a2Xi +a2Xj

(2.4)

which yield al and a: as

<l>iXj-<l>jXi

x.: x,
(2.5)

x.:«,

<l>j - <l>i

Substitution of (2.5) into (2.2) and rearranging give (2.6) where Xj- Xi has been replaced by the element length L. Equation (2.6) is in a standard finite element form. The nodal values are multiplied by linear functions of x, which are called shape functions or interpolation functions. These functions are denoted by N with a subscript to indicate the node with which a specific shape function is associated. The shape functions in (2.6) are denoted by N, and N, with and Equation (2.6) can be rewritten as 4> = Ni<l>i and also as 4> = [N]{<l>} where [N]
= [N i

The one-dimensional linear element is a line segment with a length L and two nodes, one at each end (Figure 2.2). The nodes are denoted by i andj and the nodal values by <l>i and <l>j. The origin of the coordinate system is to the left of node i. The parameter 4>* varies linearly between the nodes, and the equation for 4> is
(2.2)

The coefficients

al

and

az

can be determined by using the nodal conditions at at

(2.3)

(2.7)

+ Nj<l>j

(2.8)

(2.9)

N j] is a row vector of shape functions and {<l>} = {::}

4>;

Fugure 2.2

The one-dimensional

linear element.

*The symbol rP is used throughout this text to denote a general scalar quantity. Uppercase symbols such as X, Y, «1>, and U, denote nodal values. '

is a column vector containing the element nodal values. A few comments about the shape functions are in order. Each shape function has a value of one at its own node and zero at the other node and the two shape functions sum to one. A third characteristic is that the shape functions are always polynomials of the same type as the original interpolation equation. Equation (2.2) is a linear equation and the shape functions are linear equations. If the interpolation equation had been a quadratic model defined by three nodes, the resulting shape functions would have also been quadratic equations. Another characteristic is that the derivatives of the shape functions with respect to x sum to zero. The shape functions are shown in Figure 2.3.

20

BASIC CONCEPTS

ONE-DIMENSIONAL

LINEAR ELEMENT

21

The temperature, <p,within the element is given by (2.6)

<p=(Xj:X)~;+(X~X)~j

The element data are

X;=1.5cm ~i=120°C x=4.0cm Xj=6.0em

~j=90°C
L=4.5em

Substitution yields
~<x)

<p =(6-4)

4.5

120+(4-1.5) 4.5

90

<p=103.3°C
1

The temperature gradient is the derivative of (2.6)

d<p ~r~i

tr=t':
-6.670Cjcm

(2.10)

x,
Figure 2.3

Substituting the nodal values, we obtain

Xj

The linear shape functions N; and Nj•

d<p=(90-120)= dx 4.5

ILLUSTRATIVE EXAMPLE A one-dimensional linear element has been used to approximate the temperature distribution in a fin. The solution indicates that the temperatures at nodes i and j are 120 and 90°C, respectively. Determine the temperature at a point 4 em from the origin and the temperature gradient within the element. Nodes i and j are located at 1.5 and 6 em from the origin in Figure 2.4.
<I>(x)

2.3

A CONTINUOUS PIECEWISE SMOOTH EQUATION

A continuous piecewise smooth equation for a one-dimensional region can be constructed by connecting several linear equations with the properties developed in the previous section. Each of these equations can be written as
A-.(e) _ N(e)~. 'I' I
I

+ Me)~ J

. J

(2.11)
X-Xi Xj-X;

where
N\e)
I

XJ·-x Xj-Xi

and

N(.e) J

(2.12)

The superscript (e) indicates an element quantity. All that is needed to complete the process is to insert the correct values ofi,j, and e for.each ele~~nt. The values of i andj for a corresponding e are obtained from the gnd. Node liS th.e left-hand node of an element. The element information for the grid in Figure 2.1 IS

~----~----------------~--~~X Ll';_Ji I
j

e
1 1

~6cm___J
Figure 2.4.
Nodal values for the example problem.

2 3
4

2
3 4

2 3 4 5

22

BASIC CONCEPTS

ONE-DIMENSIONAL

LINEAR ELEMENT

23

The equation for each element in Figure 2.1 is

4>(l)=NP)$l 4>(3) N~3)$3 = +N~1)$2
(a)

x
(b)
(c)

Xi
0.0 3.0 6.5 0.5 1.0

Xj
1.5
4.5 7.5 3.0 3.0

q)i

<1>j

4>(2) N~2)$2 + N~2)$3 =

+ M?)$4

(2.13)

4>(4) Ni4)$4 + N~4)$5 =

(d) (e)

0.8 3.6 7.1 1.8 2.2

60 27 63 0 60

43 33 51 -15 67

Note that N~l) and N~2) are different equations even though both involve node two. The equations for these two quantities are and Realize that each equation of (2.13) is for a single element and is not applicable outside the element. The first equation should be written as

2.2 (a-e) Evaluate d4>ldx for the corresponding element ~n P~oblem 2.1. 2.3 The shape functions for the quadratic element shown III FIgure P2.3 are
2 Ni= L2 (x-Xj)(x-Xk) Nj=4 L2(x-Xi)(X-Xk)

but the range of x is deleted in most of the finite element literature and is deleted in this book. Whenever an element equation is given, the implication is that it is valid for only a single element.

N» =2 (x- Xi)(x-

2 L

Xj)

(a) Show that these shape functions equal one at their own node and are zero at the other two nodes. Also show that the shape functions sum to one. (b) Show that the derivatives of Ni, Nj, and N; with respect to x sum to zero.

2.4

A COMMENT ON NOTATION
,p(x)

The need to denote element quantities occurs on most of the pages of this book. The following notation is used so that a superscript (e) does not have to be placed on every coefficient. 1. When brackets or parentheses have a superscript (e), that is, (G4>+ Q)(e), then every term within the brackets or parentheses should be interpreted on an element basis. 2. A quantity on the left-hand side of an equal sign with a superscript (e) implies that the quantities on the right-hand side of the sign are for a particular element. For example,
4>(e)=Ni$i+Nj$j

L_

implies that N, and N, are really Me) and Nje), and that element nodal values.

$i

and

$j

are the

PROBLEMS 2.1 The nodal coordinates Xi and Xj and the nodal values of $i and $j for several linear elements are given below. Evaluate 4> at the given value of x. The x values are in centimeters, and $i and $j are in degrees Celsius.

E~
.-

.-

~~x

Figure P2.3

2.4 The implementation of the finite element method .requi.res .the evaluation of integrals that contain the shape functions or their derivatives. Evaluate

24
(a) JXi Ni dx

BASIC CONCEPTS

ONE_DIMENSIONAL

LINEAR ELEMENT

25

(b)

XjdNidNj --dx Xi dx dx

(c)

f::

NJ dx

where
al =Vi, a3 = L2 (Vj- v;)a2=(}i, a4= L3 (Vi-Vj)+

2.5

for the linear element. The coordinate s shown in Figure P2.5 has its origin at node i and a value of L at node j. Develop the shape functions Ni(s) and Nis) starting with c/>(s)=al +a2s and solving for al and a2'
4>(s)

L (2(}i+
1

(}j)

L2 «(}i+(}j)

Develop the shape function equations for the interpolation equation

V(x) = NlVi+ N2(}i+ N3Vj+ N40j

1~----------------~_J------~s
j

1
4>j

~-----------L------------~
Figure P2.7

Figure P2.5

2.6

The coordinate ~ shown in Figure P2.6 is a natural coordinate whose origin is at the center of the element. The value of ~ at nodes i and j is 1 and - 1, respectively. Develop the shape functions Ni(~) and Ni~) starting with c/>@=al +a2~ and solving for al and ai-

2.8 The equation for c/>(x, ) in a two-dimensional rectangular element shown in y Figure P2.8 is

T
Lt j {=-l

m._

~k

t= 1

Figure P2.6

l,..,.___----j
Figure P2.8

2.7

The beam element shown in Figure P2.7 has two vertical displacements, Vi and Vj, and two rotations, 0i and OJ, defined at the end points. The displacement equation is

I~<--Ls--~~I

26
The coefficients are C1 =11>1>C3= L, (II>m-lI>d 1 C2=L(lI>j-lI>d,
s

BASIC CONCEPTS

Chapter

C4=L L (lI>j-lI>j+lI>k-lI>m)
s ,

wher~ II>j, II>j, II>b and II>mare the nodal values of ¢. Develop the shape function equations for the interpolation equation ¢(x, y)= Njll>j+ Njll>j+ Nkll>k

A FINITE ELEMENT EXAMPLE

The shape function information developed in Chapter 2 is general information that can be used in solving solid mechanics problems as well as differential equations. The objective in this chapter is to illustrate the finite element method by developing an approximate solution for the one-dimensional differential equation D dx2 +Q=O with the boundary conditions ¢(O)=¢o and (3.2)

+ N mll>m

d2¢

(3.1)

Two physical problems are embedded within (3.1): the deflection of simply supported beams when the bending moment diagram is known and heat flow through a composite wall when the surface temperatures are known. The finite element equations are obtained using Galerkin's formulation. Evaluation of the residual integral yields a nodal equation that is applied in a recursive manner to generate a system of linear equations. The nodal equation is used to solve a beam deflection problem. The reader whose immediate interest is the analysis of structural frameworks may substitute Chapters 17 and 18 for Chapters 3 and 4. Chapters 3 and 4, however, should be read before beginning Part Two of the book.

3.1

WEIGHTING FUNCTIONS
is generated
H

A system of linear equations integral*

by evaluating +Q

the weighted

residual (3.3)

r - Jo

W(x)

(dD ¢dx
2

) dx=O

using a new weighting function for each node where ¢ is unknown. Galerkin's formulation of the weighted residual method requires that the weighting functions be constructed using the shape functions N, and N]. The weighting functions in the Galerkin finite element formulation are defined as follows: The weiqhtinq function for the sth node, consists of the shape functions associated with the sth node.

w.,

"The integral has been multiplied by a negative one so that the results can be written in a more convenient form.

28

BASIC CONCEPTS

A FINITE ELEMENT EXAMPLE

29

The weig?ting function for node three of a linear grid (Figure 3.1) consists of th shape functions for node three. e X2~X~X3 X3~X~X4 Xr~X~Xs Xs~X~Xt
(3.4)

This function is occasionally called a hat function for reasons apparent in Figure 3.2b. The weighting functions for the first and last nodes are shown in Figure 3.2a and 3.2c. The respective equations are and
(3.6)

In general,

(3.5)

The weighting function for every node consists either of Ni, Nj, or a combination of the two. 3.2 THE WEIGHTED RESIDUAL INTEGRAL

After defining the weighting functions, the next step is to evaluate the residual integral (3.3). Using the sequence of nodes r, s, and t in Figure 3.2b, we find that (3.3)becomes

s, = R~e)+ R~e + 1) = Figure 3.1.

W(x)

J::[

Ns (D ~:~

+ Q)

T)

dx (3.7)

The weighting function for node three.

Xt [

N,

Xs

(d2cjJ )]<e+ 1) D-2 +Q dx=O dx

l~

(1)

(a) W(x)

because w,=0 for x-c X, and x>Xt• The integral splits into two parts because w,(x) is defined by two separate equations within the interval Xr~X~Xt. The terms R~e) and R~e + 1) represent the contributions of elements (e) and (e + 1) to the residual equation for node s. There is a problem associated with each integral in (3.7). The approximate solution does not have continuity in the first derivative dcjJfdx; therefore, the integral of d2cjJ/dx2 is not defined. This difficulty can be circumvented, however, by changing d2cjJ/dx2 into a new term. Consider the first integral in (3.7) and note that
(3.8)
X

r ~ L(e)-+-

L(e +

1).-01

Then
(3.9)

W(x)

It

(b)

Substituting into the integral gives

(p -1) p

(p -1)

~x

Jx ,
+

X. (

NsD-2
,

~L(P-1)~

(c)

Jx,

d 2 cjJ)<e) dx=dx

DN.-

dcfJ)<e)x, dx x, (3.10)

(D dNs dcfJ)<e) x d dx dx

Figure 3.2. The weighting functions for (a) the first node (b) an interior node, and (c) the last node. '

A similar set of operations applied to the first term of the second integral in (3.7) produces

31 30
BASIC CONCEPTS A FINITE ELEMENT EXAMPLE

l
XS

x' (

N.D-2

d24»(e+ dx

1)

dx=1)

DN.-

d4»(e+ dx

1)

IX'
x,

Substitution of the appropriate terms and evaluating the integrals gives

X' (

Xs

D-dx dx

a«, d4»(e+

dx

(3.11)

and

X•
r

dN.d4> D--dx=-( dxdx

D L
XS

-<I>r+<I>.)
QL

(3.16)

The first terms in each of (3.10) and (3.11) simplify because the shape functions are either zero or one at the respective nodes. The complete residual equation is

I
( dx

x,

QN.dx=T

(3.17)

. (3 16) and (3 17) are combined with the interelement contribution for EquatIOns . . element (e) to give d4»(e) D QL (3.18) R!e)=D+-(-<I>r+<I>.)2:

= _ (D d4»(e)

I
x=Xs

x=Xs

i
I

dx
X s (

Proceeding with the second integral of (3.12)

)(e) dx 4>(e+ = Ns<l>s Nt<l>t 1) +

x,

a», d4> D---N.Q dx dx

x=Xs

+(Dd4»(e+1)1 dx +
X, (

4>(e+I)= (X -x)
)(e+ 1)

T
L ' 1 L

<1>.+(x-X ) ----y;- <l>t

dN!e+ dx

(3.19)

x,

D--

s», d4>
dx dx

Thus
dx=O

-N.Q

(3.12) and

Ns

(e+O

-x __ X t_
-

I)

(3.20)

The pair of terms evaluated at x=X. establishes an interelement requirement. The residual cannot be zero until the difference between these two quantities is zero.

d4>(e+ I)

---=Evaluation of the integrals yields

dx

(-<I>s+<I>t)

(3.21)

3.3

EVALUATION OF THE INTEGRAL

Evaluation of the integrals in (3.12) yields the residual equation for an interior node. The equations for the first and last nodes can be obtained from these operations and are left as exercises. Starting with element (e)
4>(e) Nr<l>r N.<I>. = + 4>(e)=(Xs;x)

I
and

dNsd4> D D--dx=-(<I>.-<I>t) x, dx dx L

x'

(3.22) (3.23)

x'

x,

QN.dx=T

QL

R!e+I)=D- d4>\ dx x=Xs

D QL +:;-{<I>.-<I>t)- T L

(3.24)

<l>r+( x~ Xr) <1>.

(3.13)

Combining (3.18) and (3.24) gives the residual equation for node s
I d R.=(D 4>r+ )\ dx
x=Xs

Thus
N(e)- x-Xr sL ' dMe) -dx
-

d -(D 4>r)\ dx

x=Xs

(3.14)

-(~r)
-(Q Lr) 2

<l>r+[(~r)

+(~r+1)J<I>s-(~r+l)
=0

<l>t
(3.25)

and

-=-

d4>(e) 1 dx L

(-<I>r + <1>.)

(3.15)

-(Q Lr+l)
2

32

BASIC CONCEPTS

A FINITE ELEMENT EXAMPLE

33

The usual solution procedure is to generate the system of equations interelement terms. Once the equations have been solved, ( D d¢)(e dx
+
1)

without the

I _ (D d¢)(e) I
x=Xs

dx

(3.26)
M(x)

x=Xs

can be calculated. Theoretically, the value of (3.26) can be used to evaluate the quality of the grid and to indicate where the grid should be refined. A practical way to implement this idea, however, has not been developed. Equation (3.26) does contain some important information. If D(e) =D(e+ 1), then the interelement requirement reduces to ( d¢)(e dx
+
1)

J
•
Figure 3.3
(1) 1

106 N'cm

I
(3)

I _ (d¢)(e) I
x=X,

dx

(3.27)

x=Xs

• 2

(2)

• 3

• 4

(4)

5 end moments.

There must be continuity in the slopes before the residual is zero. This continuity can never be attained with a linear element unless the solution is a straight line. The value of (3.27) becomes smaller as the grid is refined, but it is never zero for all of the nodes. The interelement requirement (3.26) can be viewed as an error term similar to the one associated with finite difference approximations. The error term is not incorporated into the system of equations. It is, however, a constant reminder that the solution is only approximate. Deleting the interelement requirement from (3.25), assuming a sequential numbering of the nodes and elements, and writing all quantities in terms of s gives the nodal residual equation

A simply supported

beam subjected to concentrated

moment, M(x), is in N· em and y is the deflection, em. The boundary

conditions

are y(O) = y(800 cmj--O. Comparing (3.29) with (3.1) yields the relationships D=EI and Q=-M(x)= -106. The element data for the beam are summarized as follows (the element length must be in centimeters). Q
-106 -106 _106 -106 L 200 200 200 200

e 1 2 3 4

D 2.4(10)10 4.0(10)10 4.0(10)'0 2.4(10)10

Rs-

__ (!!_)(S L

1)

<1>s-1 +

[(!!_)(S L Ly) =0

1)

(!!_)(S)]
L

<1>s

_ (!!_)(S)
L

<1>s+1
(3.28) Since

_(Q
3.4

Ly-l)

_(Q

Q and L have constant values, (3.28) Simplifies to

-D(S-I)y._l +(D(S-I)

+ D(s»Y.-

D(s)Y.+1

ANALYSIS OF A SIMPLY SUPPORTED BEAM

Rs=

(3.30)

The general residual equation, (3.28), obtained by evaluating the weighted residual integral is now used to obtain approximate deflection values for a simply supported beam subjected to concentrated end moments. The beam in Figure 3.3 has been reinforced over the center one-half of its span through the use of steel plates that are welded to the basic section. The beam data, its bending moment diagram, and the finite element model are given in the figure. The governing differential equation for the deflection curve is d¢ EI dx2 -M(x)=O
2

where Y has been used to denote the nodal deflection values. Writing the residual equation for nodes two, three, and four gives

R2= -1.2Y1 +3.2Y2-2.0Y3+2=0 R3= -2.0Y2+4.0Y3-2.0Y4 +2=0 R4= -2.0Y3 + 3.2Y4-1.2Y5 +2=0
after the 108 multiplier has been canceled. Incorporation of the boundary ditions Y1= Y5=0 produces

(3.31) con-

(3.29)

where EI is a bending stiffness term composed of the elastic modulus of the material, E, Nzcm ', and the area moment of the cross section, I, em". The internal bending

R2= 3.2Y2-2.0Y3 =-2 R3= -2.0Y2 +4.0Y3-2.0Y4=-2 R4= -2.0Y3+3.2Y4=-2

(3.32)

34
which yields

BASIC CONCEPTS

A FINITE ELEMENT EXAMPLE

35
as (3.34)

This set of equations

is written symbolically

Y2 =

2.50 em,

Y3 =

3.0 em,

and

Y4= -2.50cm

{R}=[K]{Y}-{F}={~
where {R} is the vector of residual equations,

when solved. The solution of(3.32) for Y2, Y3, and Y4 would appear to end the example. There are many situations, however, where there is a need to calculate quantities pertaining to a particular element once the nodal values are known. A couple of items are calculated here to illustrate the procedures. The first calculation involves determining the deflection at x=3oo em. This point occurs in element two (see Figure 3.3); thus

[Kl~H2 ~~
{Fl~

o -2

(3.35)

3.2

is called the global stiffness matrix, and

y<2)=N~2)Y2

Noting that X =200 ern and X =400 em, and using the calculated we find that

=(~3_-;JY2+(~~:2J 3
(-2.50)+ -2.75 em

+ N~2)Y3

Y3 nodal values,

{=~}

(3.36)

is called the global force vector. The terms stiffness and force come from the matrix analysis of structures. The stiffness matrix, [K], is symm~tric when ~he system of equations has a potential energy formulation or a Galerkin formulation of a differential equation that is self-adjoint.

Y= ( 400-200

400 - 300)

(300 - 200) 400-200 (- 3.0)

1 = - 2(2.50+3.0)=

PROBLEMS
3.1 This node is (a) Obtain the final system of finite element equations tions of the stepped beam shown in Figure P3.1. (b) Solve the equations
y

for the nodal deflec-

For the second calculation, the slope at node one is evaluated. located in element one; therefore, using (2.10), we obtain

in (a) and calculate the deflection at x=3H/16.

dy<l)

dx =I

(- Yl + Y2)

-2.50-0 200

-0.0125

em/ern

The slope is constant within the element. The constant slope property is a major disadvantage of the linear elements. The beam deflection problem solved in this section was selected because of the governing differential equation. The reader should not leave this chapter with the idea that all beam deflection problems should be solved this way. There is a more efficient method that involves a specific beam element. This element is covered in the structural mechanics part of this book (Part Three).

M'~

Jk = r--~ +;:
~x

EI

2EI

;)Mo
~

H 2

:+:
(3)

~--l
(4)

3.5

MATRIX NOTATION

• 1

(1)

(2)

•
3

•
4

Some comments relative to the system of equations (3.32) are in order. It is both convenient and orderly to use matrix notation when working with a system of equations. One way of writing (3.32) in matrix notation is

Figure P3.1

3.2 The differential equation

-2
4

-2

-~ ]{i} - {=~}={~}
3.2 Y4 -2 0

(3.33)

d2¢1dx2 =0 is applicable to each section of the composite wall shown in Figure P3.2, where D(e) is the thermal conductivity. Calculate the nodal temperature values within the wall and evaluate the heat flow through each material. The heat flow is given by q = - D(e) d¢ldx.
D(e)

A unit of surface area is assumed.

36

BASIC CONCEPT:

A FINITE ELEMENT EXAMPLE

37 Q
(a)

4>(0) 0
1

4>(2) 3.0 2.0 0.5 1.5 -2.0

D = 0.0035

emW_oe

(b) (e)
(d)

-3 -5

4 6

2 2
-1

20°C

-15°C

(e)

1.3 em

__J k-s__J •• 12
(1) (2)

1--2.5

<f>

•• 3

(3)
4

Figure P3.2

1 •
1

3.3

Perform the calculations in Problem 3.2 for the configuration shown ill Figure P3.3.

t05-.L5~05~051
3 Figure P3.4

(1)

• 2

(2)

(3)

• 4

(4)

• 5

D = 1 emW_oe

D= 0.2

ern-

Woe

3.5 Evaluate the residual equation for node one using the we ig ti~ghf(unc~i)on h) shown in Figure 3.2a. Note that the answer is the same as (3. 2 4 Wit e + = (1),s=1,andt=2. . .. h 3.6 Evaluate the residual equation for node p using the weighting f~nc(ho) (s o~n) n in Figure 3.2c. Note that the answer is the same as (3.18) with e = p - , r=p-l, and s=p. 3.7 Evaluate the contribution of

to the Galerkin residual equation R, when Q varies linearly over an element

WQdX

••
12

(1)

(2)

(3)

Q(x)

Figure P3.3

3.4

Start with (3.28) and develop and solve the system of finite element equations for an approximate solution to the differential equation d2cjJ/dx2 + Q =0 using the value for Q and the boundary conditions given in the following table. Divide the interval [0,2] into four elements, each with a length of 0.5 em. The nodes and elements are numbered as shown in Figure P3.4.

~'
Qr
(e) (e + L_--~~--~---.--~~1)

~x

Figure P3.7

38
(Figure P3.7). The equation for Q(x) in each element is

BASIC CONCEPTS:

A FINITE ELEMENT EXAMPLE 12000 N

39

Q(e) = Me)Qr + Me)Qs Q(e+l) =N~e+ l)Qs+ Nle+ l)Qt

where Q" Q., and Qt are the nodal values of Q. Note that (X a- X b)3 X; - 3X;Xb+ 3XaX~ - X~. The residual equation for a uniform grid and a linear variation of Q(x) between nodes is given by

~I
M(X)f

!
2

k-3m~~+E~----9m------~
(1) (2) (3) (4)

3.8

Rs= -D(s-l)y._l

+ [D(S-l)+D(S)]Y.-D(S)Y.+l
L 0 for the beam shown in

_L(Qs-l+4Qs+Qs+d
6

Use this equation to obtain the nodal displacements Figure P3.8. The governing differential equation is EI dx2 and M(x) 2(lOlO)N'

v.~
~N.cm

... x

d2¢

Figure P3.9

M(x)=O Each element is 200 em long; EI =

crrr'.

is given

in the figure.

":JOOO N

i
3

~ ~
(1)

4m (2)
2

JE
(3)

~
4m (4)
4

~I
5

M(x)

Figure P3.8

3.9

Use the residual equation in Problem 3.8 to obtain the nodal displacements for the beam shown in Figure P3.9. The governing differential equation is d2¢ D dx2 and M(x) is given 2(lOlO)N· cm '. in the figure.
-

M(x)=O Each element is 300 cm long; EI =

ELEMENT MATRICES:

GALERKIN

FORMULATION

41

Chapter

W1(x)

lK~
~ (2)

ELEMENT MATRICES: GALERKIN FORMULATION

The computer implementation of the finite element method parallels the computer solution of structural frameworks. The matrix analysis of structures places the emphasis on the element. The system of equations is constructed by calculating the element's contribution and placing the values into their correct positions within the.final system of equations. The final set of equations emerges after all of the elements have been considered. The objective in this chapter is to determine the element contributions to the final system of equations and show where these contributions are located. Matrix notation is used, and an element stiffness matrix and an element force vector are defined for the differential equation analyzed in the previous chapter. There are three points to keep in mind as the element matrices are developed. First, the residual equations are always arranged in numerical sequence, that is, RI> R2, ... ,Rp_1> Rp where there are p nodal values. Second, the nodal values <1>1> <1>2> •.. , <l>p are arranged sequentially within an equation. Third, an equation is developed for each node. The boundary conditions are incorporated after all of the equations have been developed. 4.1 ELEMENT MATRICES

• 3

(3)

• 4

(4)

:> x

Figure 4.1

The weighting functions for a four-element grid.

four. Its contribution to equation three is R~3)=-

The discussion starts by defining a column vector {R}. Each component of {R} represents a residual equation. The vector is Rl R2 {R}= Rp-1 Rp where R{J is the residual equation for node p. The residual equation is further subdivided into element contributions. For example, Rbe) is the contribution of element (e) to the residual equation for node p. The weighting functions for a four-element grid are shown in Figure 4.1. Analyzing these functions from an element point of view rather than an equation point of view shows that element three contributes nonzero values to equations three and
(4.1)

IX'

N~3)(Dd2~

XJ

dx

+Q)dX

(4.2)

while its contribution to equation four is

Ri3) = -

f:"

Ni3) (D ~:~ +Q )dX

(4.3)

Element three contributes nothing to the other equations because every weighting function except W3 and W4 are zero in element three. The results for element three can be generalized. Given an arbitrary element with nodes i and j (Figure 4.2~ we find that it contributes R!e)=_

J:l

Ni(X)(D~:~

+Q)dX

(4.4)

42

BASIC CONCEPT~

ELEMENT MATRICES:

GALERKIN

FORMULATION

43

{I(e)}={W)}={ Iy)
j

-D-

D~~'x=x'l d¢/
dx =».

(4.10)

hich is the element contribution to the interelement requirement. Equation (4.10) is deleted from the discussion except when derivative boundary conditions are specified at node one or p for the reason given relative to (3.26) and (3.27). The element matrices are the important results and they are

[k(e)]=~[

•
Figure 4.2.

1 -1

-IJ 1

(4.11)

and (4.12) The element nodal values, program for coefficient is The vector stiffness matrix is the matrix that multiplies the column vector of {<l>(e)}.Equations (4.11) and (4.12) are useful because they are easy to computer evaluation; also, it can be quickly determined where each located in the final system of equations. {R} represents a system of equations that symbolically is

The weighting functions for element (e).

to equation

i and (4.5)

to equation j.

{R}
in Chapter 3. The first integral,

= [K]{

<l>} {F} -

= {O}

(4.13)

Rle), is equivalent to R!e+1), (3.24), with s=i and t=j whereas RJe) is the same as R!e), (3.18), with r=; and s=]. Using (3.18) and (3.24) gives
Ri =D(e) (e)

The integrals in (4.4) and (4.5) were evaluated

d¢1
dx
x=X,

D +-(<l>i-<l>j)--

QL

2
QL

(4.6)· (4.7)

Rj =-DEquations

d¢1
dx
x=X]

D +-(-<l>i+<l>j)--

Equation (4.8) states that the coefficients in the first row of [k(e)] and {f(e)} are located in row i of [K] and {F} because row i is the row associated with R~e). Similarly, the coefficients in the second row of [k(e)] and {f(e)} are located in row j of [K] and {F} because this row is associated with Rje). The coefficients of [k(e)] are located in columns i and j of [K] because the coefficients in the first column multiply <l>i nd those in the second column multiply <l>j. a

(4.6) and (4.7) can be written as

4.2
-IJ{<l>i}_ 1 <l>j QL{I} 21 (4.8)

DIRECT STIFFNESS METHOD

R~e)} { Rje)
or

{lie)} e)

+I

D[

-1

The direct stiffness method is the name given to the procedure for incorporating the element matrices into the final system of equations. The method is simple and straightforward. The numerical values of i andj for a specific element are written over the columns of [k(e)] and along the side of [k(e)] and {pel}, that is, j

(4.9) where {R(e)} is the contribution of element (e) to the final system of equations .. This contribution consists of an element stiffness matrix [k(e)] and an element force vector {fIe)}. The other vectors are

k12 k22
The direct stiffness procedure matrices

i j by using the hypothetical

(4.14) set of

{<l>(e)}

={:J

is illustrated

which is the column vector of nodal values and

44 for a linear element between nodes 2 and 3 (i=2,j=3). andj gives

3

!
BASIC CONCE~

ELEMENT MATRICES:

GALERKIN

FORMULATION

45

Using these values of

l
'

6J 2 7 3'
and the location of the coefficients in [K] and {F} are 4 adds 6 adds 5 adds 7 adds 8 adds 9 adds to to to to to to K22 K23 K32 K33 F2 F3

M(J
(1) 2

106 N'cm

I
(3)
(4) 4

(2)

~
5

Figure 4.3

A simply supported beam subjected to concentrated end moments.

The word add is emphasized because there could be contributions to K22, K23• K32, K33, F 2, and F 3 from other elements which have not been considered. 'I The direct stiffness method is easily incorporated into a computer program1
GSM for [I<] ESM for [k(e)] GF for {F} EF for {f(e)} If we use the variable names

The element data in tabular form are

QL
2 -108 _108

1
e
1 2 3 4 1 2 3 4 j 2 3 4 5

L
1.2(108) 2.0(108) 2.0(108) 1.2(108)

and assume that the numerical values of i and j are contained in the array NS~ the computer implementation of the direct stiffness method is
DO 10 I

_108 _108

The substitution of D/L into (4.11) and QL/2 into (4.12) gives the element matrices

1,2

10

II ~ NS (I) GF(II) = GF(II) + EF(I) DO 10 J = 1,2 JJ = NS (J) GSM(II,JJ) = GSM(II ,JJ)

2 [k(1)] = 108[
+ ESM(I ,J)

1.2 -1.2 2

-1.2J 1 1.2 2' 3 -2J 2 4 -2J 2 3 4' 5 -1.2J 4 1.2 5' 2 3'

{f(1)} = _108

{a ~

It is assumed that NS has a dimension of at least two and that the element has two nodes. The direct stiffness procedure shows how to utilize the element matrices. The method is further illustrated in the following section where the beam deflection problem of Chapter 3 is reworked.
4.3 ANALYSIS OF A SIMPLY SUPPORTED BEAM

[k(2)] = 108[

-2

2 3

{f(2)} = -108

{a ~ {a ! {a ~

[k(3)] = 108 [

2 -2 4

{f(3)} = -108

The direct stiffness method is now illustrated by reworking the beam deflection problem considered in Chapter 3. The beam and the element model in Figure 3.3 are presented in Figure 4.3 for convenience. The element stiffness matrix [k(e)] is given by (4.11) and {pel} is given by (4.12).

[k(4)] = 108~ 1.2 1.2

{f(4)} = -108

The global matrices [K] and [F] are first initialized with zeros and then built by adding the coefficients of the element matrices an element at a time. Adding

46 the values of element one gives

BASIC CONCEPT~

! ELEMENT MATRICES: GALERKIN FORMULATION

47

1.2 [K]
= 108

-1.2 0 0 0

-1.2 1.2
0 0 0

0 0 0 0 0

0 0 0 0 0

0 0 0 0 0

{F}

-108

1 0 0 0

The nodal values Y1 and Ys can be eliminated from the equations in (4.16) because they have known values; each is zero. The terms -1.2 Y1 and -1.2 Ys vanish. If either or both Y1 and Y2 were nonzero values, then - 1.2Y1 and - 1.2Ys are evaluated and added to the two coefficient in the respective row. Deleting the rerms involving Y1 and Ys yields the system of three equations R2 =3.2Y2 -2Y3 +2=0 R3=-2Y2+4Y3-2Y4 +2=0 R4= -2Y3 +3.2Y4 +2=0 These equations are identical to (3.32).

Adding element two gives

(4.17)

[K]

1.2 -1.2
= 108

0 0 0

-1.2 3.2 -2
0 0

-2 2
0 0 0

0 0 0 0 0

0 0 0 0 0

{F}

2
=

-108

1 0 0 1

4.4

PROPERTIES OF THE GLOBAL STIFFNESS MATRIX

Adding element three gives

1.2 -1.2 [K]=108

0 0 0

-1.2 3.2 -2
0 0

-2
4

0 0

1 0 The addition of element four finishes the summation through the elements giving the system of equations 0 0 1.2 -1.2 0 0 0
8

-2

-2 2

0 0 0 0 0

{F}=-108

2 2

-1.2 3.2 -2
0 0

-2
4

0 0

-2
0

-2 3.2
-1.2

0 0 0

-1.2 1.2

Y1 Y2 Y3 Y4 Ys

-1

-2 -2 -2
-1

0 0 0 0 0

(4.15)

after 10 has been canceled. The final result is a 5 x 5 stiffness matrix [K] and a 5 x 1 column vector {F}. The size of the system of equations differs from the three equations in (3.32)ecause b the boundary conditions have not been incorporated. To incorporate the boundary conditions, note that Y1 and Ys are known quantities, Y1 = Ys =0. Equations one and five in (4.15) are eliminated because equations are not written for the nodes whose values are already known. Elimination of these two equations produces three equations containing the five nodal values. These equations are R2= -1.2Y1 +3.2Y2-2Y3 R3= -2Y2 +4Y3 -2Y4 R4= -2Y3 +3.2Y4-1.2Ys +2=0 +2=0 +2=0 (4.16)

The stiffness matrix, [K], is always symmetric and positive definite for structural problems and for governing differential equations that are selfadjoint. The diagonal coefficients, Kit, are always positive and relatively large when compared to the off-diagonal values in the same row. Finite element equations are usually solved using Gaussian elimination or efficient modifications of Gaussian elimination (Conte and deBoor, 1980)because the system of equations is not always diagonally dominant; that is, Kii can be less than the sum of the off-diagonal coefficients of row i. This occurs in some structural problems. The relatively large diagonal coefficient allows Gaussian elimination to be performed without interchanging rows (pivoting). This is an important fact because only the nonzero coefficients need to be stored within the computer. The symmetry property is also important because it eliminates the need to store the coefficients below the main diagonal. The symmetry and positive definite properties are a result of the mathematical formulation. Another important property possessed by [K] is related to the element grid and the node numbers. The global matrix [K] is banded. A banded matrix has the characteristic that all of the nonzero coefficients are located relatively close to the main diagonal and all of the coefficients beyond the bandwidth are zero. This is illustrated schematically in Figure 4.4 where the bandwidth is shown by the dashed diagonal line. The C's denote nonzero terms. It is permissible to have zero coefficients within the bandwidth. The bandwidth of [K] is related to the numbering of the nodes. The bandwidth of a one-dimensional grid of the linear elements whose nodes are numbered in succession from left to right is two. The matrix consists of the main diagonal and a diagonal on each side of it. All of the other coefficients are zero. This fact can be observed by looking at (4.15). The reason for the banded property is obtained by studying a system of equations as it is constructed. Each individual equation is associated with a node; that is, the third equation of a system is the residual equation for the third node. The nonzero coefficients in the third equation occur in the columns corresponding to the node numbers of the elements that touch node three. Consider the one-

48 I---Bandwidth-----.I C C C 0 C C C C C C C C 0 C C C C C 0 C <, C C 0"" C 0 C 0"" C 0 0 0"" C 0 0 0 (j"

<, <, <,

BASIC CONCEPTI

ELEMENT MATRICES:

GALERKIN

FORMULATION

49

C C 0 C C C C 0 C

,,0 <, C"" C C C C C C C

<,

0 0 0 0 0 0 0 C"""O C <, 0 C C 0 C C C C C C 0 C C C 0 C C
<, <,

here N BW is the bandwidth, and BW(e) is the difference between the largest and :maUest node numbers for an element. The largest value of Bw(e) is used in the calculation. The way to minimize the bandwidth is to number the nodes such that Bw(e) is as small as possible in each element. This is done by keeping the node numbers of each element as close as possible.

~
Input the number of nodes and number of elements Modify [K] and {F) to incorporate known nodal values

Figure 4.4

A banded matrix.

dimensional grid in Figure 4.5a. Elements two and three touch node three so that columns two, three, and four will contain nonzero coefficients. Columns one and five contain zero values because elements one and four do not touch node three. A more general situation occurs in two-dimensional grids. Consider the four triangles touching node 12, Figure 4.5b. Nonzero coefficients OCcurin columns 6, 10, 12, 14, and 21. The bandwidth is one plus the greatest distance between the diagonal coefficient and the last nonzero coefficient in the row. All rows must be considered in this calculation. The general equation for calculating the bandwidth of [K] for a finite element grid is NBW =max [Bw(e)] + 1 (4.18)
e

I
Initialize [K] and
{F) with zero values

I
Solve the system of equations for the noda I va lues

I
~
DO on the number of elements

I
~
DO on the number
of elements

I
Gather element data: node numbers, nodal values of x and Q, element value for D

I
Recall or input element data

(1) 2

(2) 3

(3)

(4) 4 5

I
Calculate [k(e)] and {fie)} using (4.11) and (4.12)

I
Retrieve <l>i and <l>j from the solution vector

(a)

',10....... ".. ......

" -,

"

I
Add [k(e)] to [K] and {fie)} to {F) using the direct stiffness procedure of Section 4.2

I
Calculate gradient value using (2.10)

,
/
I

I
Output gradient value

I
End of loop

~21
I I /

"-

I
End of loop

" -,

(b)

"

Figure 4.5. Node numbering in (a) a one-dimensional region, (b) a segment of a two-dimensional region.

Figure 4.6

I
END

50
4.5 GENERAL FLOW OF THE COMPUTATIONS

BASIC CONCEPTI

The general flow of the computations using the direct stiffness procedure is showi in the flow diagram in Figure 4.6. The computations start with the informatioj that defines the problem. These data include the number of nodes, the number OJ elements, the nodal coordinates, and the element values of D and Q. The global matrices [K] and {F} are initialized with zero values before starting to assemble the system of equations. The direct stiffness construction of the global matrices is done within a DO loop over the number of elements. The first step is to collect the element data, which include the element node numbers and the values of D and Q. The length is calculated from the nodal coordinates. Once the element data are available, the element matrices [k(e)] and {fIe)} are calculated. This step is followed by the direct stiffness procedure during which the coefficients of [k(e)] and {f(e)} are added to the correct locations of [K] and {F}. The global matrices [K] and {F}, which exist when the direct stiffness procedure is completed, must be modified to include the known nodal conditions. This modification produces a system of equations that is solved using a direct approach such as Gaussian elimination. Completion of the latter step produces a set of nodal values that is one of the desired results. In some situations, the nodal values are all that is desired and the problem is finished. Whenever quantities related to the element gradient values are needed, another DO loop on the elements is necessary. The first step of this loop is to recall the element data from memory. Once the element data are available, the element nodal values of ¢ are selected from the solution vector and the gradient value is calculated. The flow chart in Figure 4.6 gives the general flow of the computations. It is not, however, a flow diagram for an actual computer program. Good computer programs incorporate techniques that minimize memory requirements and make the program both computationally efficient and transferable. PROBLEMS 4.1-4.4 Develop the system of equations for the following problems using the element matrix and direct stiffness concepts discussed in this chapter. Modify the system of equations to incorporate the boundary conditions and solve for the unknown nodal values. 4.1 Problem 3.1. 4.2 Problem 3.2. 4.3 Problem 3.3. 4.4(a-e) Problem 3.4(a-e). 4.5 Solve the system of equations (4.15) when the left end of the beam has settled 1em, that is, Y1= -1 and Ys =0.

Chapter

TWO- DI M ENSIONAL ELEMENTS

A primary advantage of the finite element method is the ease with which it can be generalized to solve two-dimensional problems composed of several different materials and having irregular boundaries. Many general-purpose finite element programs are available for solving two-dimensional problems. All of these programs use triangular and rectangular elements or generalizations of these elements. The discussion of two-dimensional problems begins by considering the linear triangular and bilinear rectangular elements. The shape functions and pertinent local coordinate systems are discussed in this chapter. Some convenient natural coordinate systems are discussed in the next chapter. These elements are then used to solve heat transfer, irrotational fluid flow, and solid mechanics problems. Generalizations of these elements are considered in Part Four of this book. ·5.1 TWO-DIMENSIONAL GRIDS

The linear triangular element (Figure 5.1a) has straight sides and a node at each corner. The interpolation equation for a scalar quantity is (5.1) which is a complete linear polynomial because it contains a constant term and alI of the possible linear terms, namely, x and y. As a result, the triangular element can take on any orientation and satisfy the continuity requirements involving adjacent elements. The bilinear rectangular element (Figure 5.1b) has straight sides and a node at each corner. The interpolation equation for a scalar quantity is
(5.2)

This equation contains only one of the three possible second-order terms, xy. The rectangle cannot be arbitrarily oriented because the x2 and y2 terms are not present. The sides of the rectangle must remain parallel to the xy-coordinate system. A grid of rectangular elements is easy to construct. AlI of the elements in a row paralIel to the x-axis must be the same height. All of the elements in a column paralIel to the y-axis must be the same width. It should be clear that the rectangular element is best suited for square or rectangular regions. Both triangular and

52
y

BASIC CONCEPTS

TWO-DIMENSIONAL

ELEMENTS

53

~
(a)

~x

(a)

m.-------------~k

._

.. j (b)

~--------------------------~x
(b)

Figure

5.2

Regions divided into triangular elements.

Figure 5.1

The linear triangular element and the bilinear rectangular element.

rectangular elements should be used in irregular regions. The triangular elements' are used to model the irregular boundary. ,: The division of a region into triangular elements is most easily accomplishedl by first dividing it into large quadrilateral and triangular subregions. Each of these.l subregions is then divided into triangles. A triangular subregion is most easily: divided into elements by specifying the same number of nodes along each side and', then connecting the appropriate nodes by straight lines and placing nodes at the! intersection points. The traingular region shown in Figure S.2a has been divided] into nine elements after placing four nodes on a side. There is no reason why the! nodes have to be equally spaced along a side. A variation in the spacing allows the! size of the elements to be changed. There are (n - 1)2 triangular elements in 1:\1 triangular region, where n is the number of nodes on a side. 11 When the triangular region has curved sides, the boundary elements model thel curvature using straight-line segments. The division of a curved triangular regioIlI into linear triangular elements is shown in Figure S.2b. The dashed line is th~~ original shape and the solid lines denote the elements.
.1

The quadrilateral subregion is easily divided into triangular elements by connecting the nodes on opposite sides using line segments (Figure S.3a). The interior nodes are located at the intersection points. The interior quadrilaterals are divided into triangular elements by inserting the shortest diagonal (Figure S.3b). Division using the shortest diagonal is preferable because elements close to an equilateral shape produce more accurate results than long narrow triangles. The number of nodes along adjacent sides of a quadrilateral subregion do not have to be the same, but the number of nodes on opposite sides should be equal unless the grid is to be refined (or enlarged). The spacing between boundary nodes can be varied to produce elements of difference sizes. There are 2(n-l)(m-l) triangular elements in a quadrilateral, where nand m are the number of nodes on a pair of adjacent sides. The nodes on the boundary between subregions must be identical in number and must have the same relative position. This property is necessary to insure the continuity of cjJ across an element boundary. The application of the concepts relative to the discretization of a region are illustrated in Figure S.4. The node spacing has been varied along the edges of the quadrilateral in order to have smaller elements in the vicinity of the curved boundary.

54

BASIC CONCE1
I ',.

oIt----:4.----....,--"
I\
I'

"

,I \I

__

I 'I ......- _ -;- -I ,I

I, ~ -,. I /I I I

_I-

\I

"

",

I" \ 'I

"

1 I

\1

I" I ,.

,.

\,

I'

(a)

6J CZJ
Desirable Undesirable (b)

",'"
I I ....." .... "'.
, ...'"

//

r- - - - - ---- "-J,<-- -I,..."

I/",.

.... I I I I

//

,.

,,\

\' \ ,;, /

// \

---

-- /

-/If- -------"1
\

//

,
\

Figure 5.5

Using an expansion region to change the element size.

BW(l) = 13

2
(1)

Figure 5.3
....

Division of a quadrilateral subregion into triangular elements

6 14

...........
...

,,

.........

13
-,
...

------~

.... ,

"

"
"

I',

I~-----"",----"
,
' 'I

..

'

r
I

',

>.

--,

----

,.",'"

",'"

-... " ::.."-----~.:J... , -_

J',
I

I ~

\ --_

',

12

11
(a)

"1

"

",,;'

BW(l)=4

1 (1)

13

Figure 5.4

Division of a region into subregions and then into triangular elements: 2 5

14

A regular mesh with all elements the same size and shape is not necessary because there usually are regions in which the nodal variable is relatively constant. Larget elements can be used in these regions. The ability to vary the element size is all important advantage of the triangular element. The easiest way to make a transi· tion in element size is to employ a quadrilateral region that has an unequal numbei of nodes on two opposite sides. A good combination is to place two nodes on one side for every three nodes on the opposite side. Such a region is shown in Figure 5.5.

15

6
(b)

Figure 5.6

Two sets of node numbers resulting in different band widths.

56

BASIC CONCEPTS

TWO- DI M ENSIONAL

ELEM ENTS

57

Assigning the node numbers would be a trivial operation if the numbering did not influence the bandwidth of the system of equations. The bandwidth, NBW, as discussed in Chapter 4, is given by NBW=max
e

with the nodal conditions

<P =<I>j

at at at

X=Xj,Y= x=Xj,Y= x=Xk,Y=

Y; lj Yk the system of equations

[Bw(e)]

+1

(5.3)
node numbers by numbering assigning node in Figure 5.6a values of Bw(e) Substitution

<p =<I>j <p = <l>k

of these conditions

where Bw(e) is the difference between the largest and smallest in an element. The minimization of N BW in simple regions can be accomplished across the dimension of the body with the fewest nodes when numbers. Two different numbering schemes for a set of nodes are shown and b. The values of BW(I) are 13 and 4, respectively. The largest are 13 and 5. The respective bandwidths are 14 and 6.

into (5.4) produces <l>j=(X1 +(XZXj+(X3Y; <l>j=(X1 +(XZXj+(X3lj <l>k=(X1 +(XZXk+(X3lk

(5.5)

which yields (Xl = 21A [(X (Xz =(X3=1 2A 1 2A

j

lk - X k lj)<J>j + (X k Y; - Xj lk)<J>j Y;)<J>j+(Y;-lj)<J>k] X k)<J>j+ (Xj-

+ (Xj

lj - X j Y;)<J>k]

5.2

LINEAR TRIANGULAR ELEMENT

The linear triangular element shown in Figure 5.7 has straight sides and three nodes, one at each corner. A consistent labeling of the nodes is a necessity and the labeling in this book proceeds counterclockwise from node i, which is arbitrarily specified. The nodal values of <P are <l>j, <l>j, and <l>k whereas the nodal coordinates are (Xj, Y;), (Xj, lj), and (Xb lk). The interpolation polynomial is

[(lj-lk)<J>j+(Yk[(Xk - Xj)<J>j + (Xj-

Xj)<J>k]

where the determinant

(5.4)

x,

x,

Y; lj =2A
(5.6)

Yk
an equation for (5.7)

and A is the area of the triangle. Substituting for (Xj, (Xz, and (X3 in (5.4) and rearranging produces cp in terms of three shape functions and <l>j, <l>j, and <l>k that is

where Nj=2A 1 [aj+bjx+cjY]

(5.8)

1 Ni=x: [aj+bjx+cjY] 2A
y

(5.9)

1 Nv=:': [ak+bkx+CkY] 2A and

(5.1 0)

x j (Xi'

Yjl

aj=Xj aj=Xk ak=Xj

Yk-Xk Y;-Xj Yj-

lj, lk,

bj= lj-lk,
b j= Yk- Y;, bk= Y;-lj,

and and and

Cj=Xk-Xj Cj=Xj-Xk Ck=Xj-Xj

Figure 5.7

Parameters for the linear triangular element.

x, Y;,

58

BASIC CONCEPTS

TWO-DIMENSIONAL

ELEMENTS

59
are

The scalar quantity cp is related to the nodal values by a set of shape functions that are linear in x and y. This means that the gradients acp/ax and acp/ay are constant within the element. For example,

(5.9), and (5.10). The coefficients for the shape function equations ai=Xi Y,.- Xk Y;=4(5)-2(0.5)= aj=Xk Y;- Xi Yk =2(0)-0(5)=0 ak= Xi Y;- Xj Y;=0(0.5)-4(0)=0 bi= Y;- Yk=0.5-5= -4.5 bj= Y,.- Y;=5-0=5
bk= Y;- Y;=0-0.5
=

19

acp_ aNi <I> aNj <I> aNk ax i+ ax r+ ax <l>k

-ax
--

(5.11)

but

aNp ax

--

bp

2A

/3=i,j, k

Therefore,

(5.12)
Since b., bj, and b, are constants (they are fixed once the nodal coordinates are specified) and <l>j, l>j, < and <l>k independent of the space coordinates, the derivaare tive has a constant value. A constant gradient within any element means that, many small elements have to be used to accurately approximate a rapid change in cp.

Ci=Xk-Xj=2-4= Cj=Xi- Xk=0-2=-2 Ck= X j - Xi = 4 - 0 = 4

whereas

-0.5 -2

2A = 1 =20-1
Substitution

Xi Xj Xk =19

Y;

Y; = 1 1 Yk

0 4 2

0 0.5 5

of the coefficients

into the shape function

equations

gives

ILLUSTRATIVE

EXAMPLE

Evaluate the element shape functions and calculate the value of the pressure at point A in Figure 5.8 if the nodal values are <l>j=40 Nzcm ', <l>j=34 N/cm2, and' <l>k=46N/cm2• Point A is located at (2,1.5). The pressure cp is given by (5.7), and the shape functions are defined by (5.8),

Ni=

19-4.5x-2y 19

5x-2y Nj=-19Nk= Note that Ni+Nj+Nk=1.

-0.5x+4y 19
for the pressure is
k

The equation

_(19-4.5.X-2Y) ¢19

<1>,+

. (5X-2Y) 19

<1>,+

. (-0.5X+4Y)<I> 19

The value of ¢ at point A, (2, 1.5), is

cp= (~)40
</>= 34 ,
I

+ (~)34 + (~)46

cp=39.4 N/cm2
The shape functions defined for the triangular element satisfy the properties discussed relative to the one-dimensional shape functions. Each shape function has a value of one at its own node and is zero at the other two. The three functions also sum to one. There are two other important properties whose proofs are left as exercises. First, a shape function varies linearly along the sides between its node and the other two nodes, that is, N, varies linearly along sides ij and ik. A shape function is zero along the side opposite its node; that is, N, is zero along side jk.

I I

I I

I I

I
(0.0) i

Figure 5.8

Parameters for the example problem.

60

BASIC CONCEPTS

TWO-DIMENSIONAL

ELEMENTS

61

A resul.t of the first property is that ¢ varies linearly along each of the three sides., ~no~her lmp?rtant characteristic of ¢ is that any line of constant ¢ is a straight line mtersectmg tw~ sides of the element (unless all nodes have the same value). These two properties make it easy to locate contour lines as illustrated in the following example. ILLUSTRA TIVE EXAMPLE Dete~mine the I.ocation of the 42 N/cm2 contour line for the triangular element used m the previous illustrative example. The pressure contour for 42 Nzcm ' intersects sides ik and jk. Simple ratios are used to obtain the coordinate values because the pressure varies linearly along, each side. For side jk .
46-42 2-x x=2.67

5.3

BILINEAR RECTANGULAR ELEMENT

The bilinear rectangular element has a length 2b and a height 2a. The nodes are labeled i, j, k, and m with node i always at the lower left-hand corner. The element and the important coordinate systems are shown in Figure 5.10. The interpolation equation (5.2) is written in terms of local coordinates sand t. There are at least three choices with
(5.13)

46- 34 =2-4 and 46-42 _ 5-y 46-34 5-0.5 Similar ratios for side ik yield x=jcm

or
em

2-x

12

-2

or

y=3.5 cm

being the most useful. The other choices would replace the st term by either S2 or t', Equation (5.13) is used because ¢ is linear in s along any line of constant t and linear in t along any line of constant s. Because of these properties, the element is often said to be bilinear. Equation (5.13) is written relative to a local coordinate system, whose origin is at node i because the shape functions are easier to evaluate in this reference frame. Another popular coordinate system is qr, which has its origin located at the center of the element (Figure 5.10). The coefficients C C 2, C 3, and C4 in (5.13) are obtained by using the nodal values of ¢ and the nodal coordinates (in the st system) to generate four equations. These equations are <l>i=C1 <l>j=C1 + (2b)C2
J,

<l>k

and

= C 1 + (2b)C 2 + (2a)C 3 + (4ab)C4

<l>m=C1+ (2a)C3

The contour line is shown in Figure 5.9.

I I
1

I
I
I

cpo = 34
J

I I
I I I
I

.~

/1

,_.j-

I~Y

I
k

x~

Figure 5.9

The 42 N/cm2 contour line,

Figure 5.10.

Parameters for the bilinear rectangular element.

62
Solving gives

BASIC CONCEPn

TWO-DIMENSIONAL

ELEMENTS

63

C1=<1>i CZ=2b
1

(<1>j-<1>;) (5.15J

1 C 3 = 2a (<1>m <1>;) -

C4= 4ab (<1>i-<1>j+<1>k-<1>m)

Substitution where of (5.15) into (5.13) and rearranging gives

The shape functions defined by (5.19) are useful because they lead to a natural coordinate system that allows the rectangle to be deformed into a general quadrilateral. A contour line in a rectangular element is generally curved. The intersection of the contour line with the edges can be obtained using linear interpolation. The easiest method for obtaining a third point is to set s or t to zero in the shape function equations and solve (5.16) for the other coordinate value. This procedure is illustrated in the following example.

ILLUSTRATIVE EXAMPLE (5.16)

Determine three points on the 50 e contour line for the rectangular element shown in Figure 5.11. The nodal values are <1>i =42°e, <1>j 54°e, <1>k 56°e, and = =
0

¢ = Ni<1>i+ Nj<1>j+ Nk<1>k+ Nm<1>m Ni=(lNj=;b

N _ st

;b)(I(1- 2~)

2~)

c1>m=46°C.
The lengths of the sides are

2b=Xj-Xi=8-5=3 2a = Ym - Y; = 5 - 3 = 2 (5.17)
Substituting these values into (5.17) gives the shape functions

k- 4ab

Nm=;a

(1-

Ni=(1;b)

})(10

D,

Nj=}(l-~)

The shape functions for the bilinear rectangular element have properties similar to those possessed by the triangular element. Each shape function varies linearly along the edges between its node and the two adjacent nodes. For example, N, varies linearly along sides ij and im. Each shape function is also zero along the sides its node does not touch, that is, N, is zero along sides jk and km. The linear variation of ¢ along an edge of the rectangular element and an edge of the triangular element means that these two elements are compatible and can be used adjacent to one another. The transformation equations between the qr and st coordinate and systems are

Nm=~(1-0
Inspection reveals that the 50 e contour line intersects the sides ij and km; therefore, we need to assume values of t and calculate values of s. Along side ij. t =0 and
'I'
A.

= (1 - 3 <1> 3 <1> 50 = J
I

.1') +.1'

Substituting

for <1>i and <1>j nd solving gives s = 2.0. a

Along side km, t = 2a = 2 and

s=b+q
Substitution

¢=~<1>k+(1Substituting

~)<1>m=50

t=a+r

(5.18)

of (5.18) into (5.17) gives the shape functions in terms of q and r

for <1>knd <1>m a and solving gives s = 1.2.

y

Ni=~
Nj

(1-

t)(1-

~)
1- ~)

= ~ ( 1+

t)(
t)(1

I
I

(5. 5)Dk

(8, 5)

N k = ~( 1 +

Nm=~(1-

t)( +~) +~)

1

(5.19)
I

(5, 3)

}_s (8,3)

L-------------------------~x
Figure 5.11. Nodal coordinates for the example problem.

64

BASIC CONCEPTS

TWO-DIMENSIONAL

ELEMENTS 8 7

65

t
I

(12,2)

(4)

-------.._--~ _s
(164, 1) (2,0)

(2)

2 ._-------(1)

.. 5

Figure 5.12.

The 50 contour line.

0

To obtain the third point, assume that t =a -I,=~(1'f'2 Substituting

= 1; then
Figure 5.13.

* _---------~4 A four-element grid with node numbers.

.s')<I>.+.s'<I>.+.s'<I>k+~(l3 '6J6 23m

.s')<I> =50 The interpolation equation for element one is (5.20)

the nodal values gives

6' (-42

+ 54+ 56 -46)+2

(42+46)=50

¢(l)=NP)<I>l + N~1)<I>4+ N ~1)<I>5 N~1)<I>2 +

Solving yields s = 1.64. The st coordinates of the three points are (from top to bottom) (1.2, 2), (1.64, 1) and (2, 0). The xy coordinates of these points are (6.2, 5), (6.64, 4) and (7, 3). A straight line from (6.2, 5) to (7,3) passes through the point (6.60, 4); therefore, the contour line is not straight (Figure 5.12).

Note that the element node numbers are no longer consecutive. This is the usual case with two-dimensional elements. The shape functions in (5.17) are a function of the global coordinates only in the sense that 2b=Xj-Xi=X4-XI and 2a= Ym-

Yi= Y2 - Y1
(5.21)

5.4

A CONTINUOUS PIECEWISE SMOOTH EQUATION

The interpolation

equation

for element four is

The element equation for ¢ defined by either (5.7) or (5.16) can be used for any triangular or rectangular element by specifying the numerical values 'of i, j, and k or i, j, k, and m. Any node of a triangular element may be node i. An asterisk is used to distinguish it from the other nodes. Node i of the rectangular element is always at the origin of the st coordinate system. The element nodal data for the four-element grid in Figure 5.13 is

¢(4) = N~4)<I>8+ N~4)<I>3+ N~4)<I>7

The shape functions in (5.21) are a function of the global coordinates and the specification of i, j, and k immediately indicates which coordinates to use. Consider, for example, N~4). Using (5.8) gives 1 N~4) =- (a~4) + b~4)X+ C~4)y)

2A

e 1 2
3 4

k 5

where a~4)=X
3

1 2
3 8

2
3

Y7-X7
3

Y3

5
6

6
7

b~4)= Y3 - Y7 C~4)=X7-X sincej=3 andk=7. The area, A, is that of element four.

66 PROBLEMS
5.1 5.2

BASIC CONCEPTS

TWO-DIMENSIONAL

ELEMENTS

67

5.3

Verify that N, for the triangular element is equal to one at node i and equal to zero at nodes j and k. Verify that N, for the triangular element in Figure 5.1a is zero everywhere along side jk. Hint: Write an equation of the form y =d + mx for side jk and substitute for yin (5.8). Verify that the shape functions for the triangular element sum to one, that is, Ni+Nj+Nk=l. Comment on the behavior of the following summations: (i) ai+aj+ak' (ii) bi+bj+bk. (iii) Ci+Cj+Ck'

5.12-5.16 The nodal values for five different rectangular elements are summarized below. Each column of values is associated with an element and a specific problem. (a) Calculate (b) Determine (c) Evaluate the value of 4> at the coordinates three sets of of point B.

84>/ox

and

xy 84>/8y

coordinates at point B.

for the specified contour line.

Element Quantity 5.12 0.31 0.18 0.38 0.25 115 85 76 105 0.35 0.22 90

Problem Number 5.13 0.25 0.18 0.31 0.25 140 115 104 124 0.26 0.22 130 5.14 0.25 0.06 0.31 0.13 158 130 125 150 0.28 0.10 154 5.15 0.31 0.13 0.38 0.19 125 92 86 116 0.34 0.15 100 5.16 0.40 0.00 0.44 0.03 76 54 60 80 0.42 0.01 70

Xi

Verify that the shape functions for the rectangular element given by (5.17) sum to one. Also check those given by (5.19). 5.5 Determine the requirement that the shape functions must satisfy in order to correctly model the condition that 4> equals a constant within an element. 5.6 Verify that a line of constant 4> in a rectangular element is in general not a straight line. Under what conditions will it be a straight line? 5.7-5.11 The nodal values for five different triangular elements are summarized below. Each column of values is associated with an element and a specific problem. (a) (b) the (c) Calculate the value of 4> at the coordinates of point A. Determine the xy coordinates where the specified contour element boundaries. Evaluate o4>/ox and o4>/oy within the element. line intersects

5.4

Illj Illk Illm Point B

Xj Ym Illi

li

x
y Contour line

Element Quantity 5.7 0.13 0.01 0.25 0.06 0.13 0.13 190 160 185 0.20 0.06 170

Problem Number 5.8 0.31 0.06 0.38 0.09 0.31 0.13 130 94 125 0.36 0.09 110 5.9 0.13 0.13 0.25 0.13 0.19 0.19 185 151 160 0.18 0.13 170 5.10 0.13 0.00 0.25 0.00 0.25 0.07 194 160 158 0.20 0.03 180 5.11 0.44 0.25 0.50 0.25 0.50 0.38 43 60 52 0.47 0.30 55

Xi Xj Xk

li

}j Y,. Illj Illk Point A

Illi

x
y Contour line

COORDINATE SYSTEMS

69

Chapter

6
SYSTEMS
(a)

COORDINATE

All finite element solutions require the evaluation of integrals. 'Some of these are easily evaluated while others are very difficult. Many are impossible to evaluate analytically so that numerical techniques are employed. The difficulties associated with evaluating an integral can often be decreased by changing the variables of integration. This involves writing the integral in a new coordinate system. The objective of this chapter is to discuss several coordinate systems that can be used to eliminate some of the difficulties associated with finite element integrals. Local and natural coordinate systems are discussed in this chapter. These systems are discussed relative to the one-dimensional linear element, and then the two-dimensional triangular and rectangular elements. and

(b)

Figure 6.1

Local coordinate systems for the one-dimensional

element.

6.1

LOCAL COORDINATE SYSTEMS

in Chapter and

The linear shape functions developed X·-x Ni(x)=_J_ L

2,
NAx)=--'

X-Xi Nj(S)=-L-=

Xi+S-Xi L

=I

(6.4)

x-x
L

(6.1)

are for an element in which the origin of the coordinate system is to the left of node i. These are general equations valid for all linear elements regardless of their location. !he d~sadvantage of these shape functions shows up when evaluating integrals mvolvmg products of the shape functions such as

Note that each shape function equals one at its own node and is zero at the other node. The two sum to one as did the pair in (6.1). The shape functions for a coordinate system located at the center of the ele~ent are obtained from (6.1) by replacing x by x=Xi+(L/2)+q. The shape functions relative to this origin are Ni(q)=G -

Ni(x)Nj(x)

dx

or

J:

Ni (X) dx

i)

and

(6.5)

(6.2)

Integrals similar to these occur in the consideration of both field problems and solid mechanics problems. The integrations in (6.2) are simplified by developing new shape functions defined relative to a coordinate system whose origin is located on the element. This type of system is called a local coordinate system. The two most common local coordinate systems for the one-dimensional element have the origin located at node i or at the center of the element (Figure 6.1). The shape functions for a coordinate system located at node i are obtained from (6.1) by replacing x by x = Xi + s. This substitution produces Ni(s)= Xj-x L
= Xj-(Xi+S)

where the coordinate variable q ranges from - L/2 to L/2. The shape functions, (6.3) and (6.4), as well as the pair in (6.5), are u.seful only if a change in the integration variables is performed. The change of vanable formula from integral calculus (Olmstead, f(x)dx=

1961) is

[2

f(9(p){d(~~»JdP

(6.6) relating x and p,

where p is the new coordinate that is, x = g(p). Interpretation

variable and g(p) is the equation

of (6.6) relative to the coordinate

systems in Figure 6.1 goes as

1- ~
L

(6.3)

follows. For the coordinate

s, where x = Xi

+S

70

BASIC CONCEPTS

COORDINATE

SYSTEMS

71

J J
Xj

Xj

f(x)dx=

Is2
SI

d(X+s)

Xi

ds

ds =

L h(s)ds
0

(6.7)

f=-l

where h(s) is f(x) written in terms of s. The limits of integration were obtained substituting Xi and Xj for x in x=Xi+s and solving for s. F or the coordinate q, where x = Xi + L/2 + q f(x)dx=

by

iq2 r(q) d(X·+L/2+q)

I

Xi

ql

dq

dq

fLIZ -LIZ

l ~~--J
(a)

f= 1

r(q)dq

(6.8)

where r(q) is f(x) written in terms of q. The usefulness of (6.7) and (6.8) comes when integrals such as

are evaluated.

Using the coordinate

variable s, we obtain iL()Z 1-~ 0 L ds=- L 3

Figure 6.2

il

r..S-------;L=-__ J _
(b)

J J
Xj

Xj

NiZ(x)dx=

iL N/(s)ds=
0

Xi

Using the q coordinate,

we obtain fLIZ -LIZ from LIZ NiZ(q)dq =f -LIZ (~_

Natural coordinate systems for the one-dimensional

element.

NiZ(x)dx=

Xi

i)Z
L

dq=~

The result, L/3, is obtained

The advantage of the coordinate variable ~ is the - 1 to + 1 limits of integration. Most computer programs use numerical integration techniques to evaluate the element matrices. A numerical integration scheme used in finite element programs is the Gauss-Legendre method (Conte and deBoor, 1980), which has the sampling points and weighting coefficients defined on a-I, + 1 interval. Another interesting natural coordinate system consists of a pair of length ratios, Figure 6.2b. If s is the distance from node i, then {I and {z are defined as the ratios and
{z=-

only after a rather complicated

expression

is recognized

as being L3.

6.2

NATURAL COORDINATE SYSTEMS

(6.11 )

This pair of coordinates

is not independent

because (6.12)

The local coordinate systems sand q can be converted to natural coordinate systems. A natural coordinate system is a local system that permits the specification of a point within the element by a dimensionless number whose absolute magnitude never exceeds unity. Start with the q coordinate in Figure 6.1 and form the ratio q/(L/2)=2q/L=~. The coordinate ~ varies from - 1 to + 1 (Figure 6.2a). The shape functions in (6.5) can be written in terms of ~ by replacing q by q = ~L/2. The new shape functions are and The change of variables in the integration f -LIZ r(q) dq L IZ yields d~ (6.9)

{I +{z=1

The most important characteristic of (6.11) and (6.12) is that {I and {z are identical to the shape functions defined by (6.3) and (6.4). The usefulness of these coordinates is associated with the evaluation of integrals of the type

N~(s)NJ(s)ds

(6.13)

f~2 g(~) d(~L/2) d~

~I

="2

Lfl
-I

g(~) d~

(6.10)

which involve the product of shape functions. The length ratio coordinates result in a simple formula for evaluating an integral similar to (6.13). The change of variable rule and the relationships Ni(s)={J, Nj(S)={2, s=L{z, and ds/d{ 2 =L give (6.14)

where g(~) is r(q) written in terms of ~.

72
The integral on the right-hand side of (6.14) can be changed to L (l-e2)ae~

BASIC CONCEPTS.i

COORDINATE SYSTEMS

73

Another example is (6.15)

«.
r(z)r(w) r(z+w)

Jo

N?(s)N](s)ds=L

Jo elelde2=L(3+~~1)!=60

3'2'

L for the one-

using (6.12). The integral in (6.15) is of the same form as

The coordinate systems, shape functions, and limits of integration dimensional linear element are summarized in Table 6.1. (6.16)

1
o

t 1(1Z -

t)w-l dt

6.3

RECTANGULAR ELEMENT

where r(n+ 1)=n! (Abramowitz L

Jo

and Stegun, 1964). Thus

eaeb de =L r(a+ l)r(b+ 1) 12 2 r(a + b + 1+ 1) alb! =L~~~(a+b+ 1)! (6.17)

Natural coordinate systems can be defined for two-dimensional elements; they have the same advantage as observed for the one-dimensional formulation. They are more convenient for both analytical and numerical integration. The natural coordinate system for the rectangular element is shown in Figure 6.3. It is located at the center of the element and the coordinates are the length ratios

~=9__

Equation (6.17) is useful because it states that a rather complicated integral can be evaluated using an equation which involves only the length of the element and the powers involved in the product. Evaluation of a pair of integrals illustrates the usefulness of (6.17). Starting with

and

r 1]=-

(6.18) in (5.19) are easily

where q and r are the local coordinates. The shape functions converted to the natural coordinate system. The results are

N, =t(1- ~)(1 -1]),

Nk=i(l It should be clear that ~ and +~)(1 +1]),
I]

Nj=t(l

+~)(1-1]) (6.19)

Nm=t(l-~)(1 +1])

range between plus and minus one, that is, and

(6.12) gives

1]=1

Table 6.1

Coordinate Systems One-Dimensional Element Type of System Coordinate Variable

and Limits of Integration

for the

Shape Functions

Limits of Integration

e= -1

~----------+-~f
f=l
1]=-1

Global Local Local Natural Natural

x
s q

X'-x Ni=--y-' N=1-.5_

I

Nj=x-Xi Nj=s

Xi,
0, L

Xj
L L Figure 6.3. A natural

coordinate

system

for the rectangular

element.

L' L'

N=(!_i) 2
I

N=(!+i) L }2 J Nj=-(I +~) 2 Nj=[2

2'
-1, 0,

6.4

TRIANGULAR ELEMENT: AREA COORDINATES

~
[2

J Ni=-(J-O, 2 Ni=[j,

A natural coordinate system for the triangular element is obtained by defining the three length ratios Lh L2, and L3 shown in Figure 6.4a. Each coordinate is the ratio of a perpendicular distance from one side, s, to the altitude, h, of that

74
k

BASIC CONCEPTS!

COORDINATE SYSTEMS

75

(a)

Figure 6.5.

A triangle divided into the areas corresponding to the area coordinates.

is A and is given by
B.

A=~

bh

whereas the area of the shaded triangle (B, j, k) is

(b)

Al='2
k

bs

(6.20)

Forming the ratio A l/A yields (6.21) The area coordinate L, is the ratio of the shaded area in Figure 6.5 to the total area. Similar equations can be written for L2 and L3 giving and Since A 1 + A 2 + A 3 = A, (6.22)

(c)

(6.23) An equation relating the three coordinates was expected because the coordinates are not independent. The location of a point can be specified using two of the coordinates. Equation (6.21) can be reworked into another form. Multiplying the top and bottom by two gives (6.24)

Figure 6.4.

The three area coordinates for a triangular element.

same side. This is illustrated in Figure 6.4b. Each coordinate is a length ratio that varies between zero and one. The lines of constant L, are shown in Figure 6.4c. Each of these lines is parallel to the side from which L, is measured. The coordinates l-i, L2, and L3 are called area coordinates because their values give the ratio of the area of a subtriangular region to the area of the complete triangle. Consider point B as shown in Figure 6.5. The area of the complete triangle

76
Using the determinant expansion for 2A
1

BASIC CONCEPT~

COORDINATE SYSTEMS
y

77

produces

x
2Al or

=1
1

(6.25) where x and yare (6.24) yields the coordinates

1

of B in Figure 6.5. Substituting

(6.25) into

LJ =2A [(Xj}kEquation

Xk}j)+ (lj-}k)x+(Xk-

XJy]

(6.26)

(6.26) is identical to (5.8); thus Ll=Ni (6.27)

A similar analysis for L2 and LJ shows that and (6.28)

Figure 6.6. The area coordinates for a point on the edge of a triangle.

The area coordinates for the linear triangular element are identical to the shape functions, and the two sets of quantities can be interchanged. The advantage of using the area coordinate system is the existence of an integration equation that simplifies the evaluation of area integrals (Eisenberg and Malvern, 1973). This integral equation is related to (6.17) and is

coordinate L3 is zero and L, is the ratio of the shaded area to the total area. Define the coordinate variable s, which is parallel to side ij and measured from node i. If the coordinate of point B is s, and the length of the side is b, then 2h(b-s) 2 2 The area coordinate L2 is L2=S

i L

b a!b!c! L~L2L3 dA =( b 2),2A a+ +c+ .

(6.29) product (6.30)

t., =-2A-=-2-bh-

2Al

b+s

The use of (6.29) can be illustrated

by evaluating

the shape function

-=1-b b

(6.31)

Ni(x, y)NAx, y)dA

over the area of a triangle. The area integral is NiNjdA

(6.32)

LiL1L~ dA 1!1 to! 2A

=-

2A 4!

(1+1+0+2)1

c.,-------c-~c-c

=-

The area coordinates L, and L2 reduce to the one-dimensional shape functions Ni(s) and Nj(s) defined by (6.3) and (6.4). Using the one-dimensional natural coordinates, eland e 2, defined by (6.11), the area coordinates become and The relationships for the other two sides are L2 =e L3 =e
1

12

side i-+ j

(6.33)

The area coordinates L, and L2 can be substituted for N, and Nj, respectively. Since Nk was not in the product, L3 is included to the zero power. Zero factorial is defined as one. The incorporation of derivative boundary conditions or surface loads into a finite element analysis requires the evaluation of an integral along the edge of an element. These integrals are easy to evaluate once it is known how the area coordinates behave on an edge. Consider point B on the side ij (Figure 6.6). The

and and

L3 =e 2 L, =e
2

side j-+k side k-« i

(6.34) (6.35)

The importance of the relationships in (6.33), (6.34), and (6.35) is that any integral over the edge of a triangular element can be replaced by a line integral written in

78
terms of s or

BASIC CONCE~

COORDINATE SYSTEMS

79

t 2,

that is,

Jrf(L1.L2,L3)dr=

g(s)ds=L

h(t2)dt2

(6.36J of a twO'

and evaluated using the factorial formula dimensional element is denoted by F.

(6.17). The boundary

ILLUSTRATIVE

EXAMPLE element.

Evaluate f r [NY dr over side ik of a linear triangular The integral is

since the linear triangular shape functions and the area coordinates Along side ik, l-s =t 1. L2 =0, and L3 =t 2; thus

are equivalent,

Figure 6.7.

A two-element grid.

using (6.35) and then (6.17).

exercise. Continuity of 4> along a common boundary of two arbitrarily oriented triangular elements is more complicated and is considered here. Consider two adjacent elements (Figure 6.7) with the coordinate system originating at node one. The nodal values are <1>1.<1>2,<1>3, and <1>4' The equations for 4> are

6.5

CONTINUITY

The function for approximating 4>(x, y) consists of a set of continuous piecewise smooth equations, each defined over a single element. The need to integrate this piecewise smooth function places a requirement on the order of continuity between elements. The integral

4>(1) = Njl)<I>1 +N~I)<I>3+Nil)<I>4 4>(2) = NF)<I>1 + N~2)<I>2 N~2)<I>3 +

(6.37)

The properties of the shape functions indicate that N~2)=Nil)=0 along the common boundary. Recalling the equality between the shape functions and the area coordinates, (6.27) and (6.28), allows (6.37) to be written as

dn4> -d ndx ox

4>(1) = Ljl)<I>1 + Li1)<1>3 4>(2) = Li2)<I> 1 + L~2)<I>3

Remember that the subscripts on the area coordinates numbers. Since L~I)=Li2)=0, (6.38) can be reworked into

(6.38) are not related to the node

is defined only if 4> has continuity of order (n -1) (Olmstead, 1961). This ensures that only finite jump discontinuities exist in the nth derivative. This requiremenl means that the first derivative of the approximating function must be continuous between elements if the integral contains second-derivative terms, n = 2. All of the integrals in this book, except the beam element, contain first-derivative terms Therefore, 4> must be continuous between elements, but its derivatives do not have to be continuous. Continuity in the derivative is required for the beam element. Continuity of 4> in the one-dimensional element is assured, since two adjacent elements have a common node. Continuity in 4> along a common boundary between two rectangular elements is relatively easy to prove and is left as an

4>(1) = Ljl)<I>1 +(1-Li1»<1>3 4>(2) = Li2)<I> 1 + (1 - Lj2»<I> 3

(6.33)

using (6.23). The proof is completed when it is shown that Lil) = Li2). A point on the common boundary is shown in Figure 6.8 with the areas associated with Li1) and Li2) shaded. Defining the distance from point B to node three as c

80
y

BASIC CONCEPT~

COORDINATE SYSTEMS

81

Figure P6.1

6.2

Do Problem 6.1 for the natural coordinate

(Figure P6.2).

~t
Figure 6.8. The area coordinates

L;')

and

L;2)

along a common boundary.

and the length of side 1-3 as b,

2ch(1)
(I)

I:
6.3

~I
L

~I
P6.2

LI

2A~I) -2-_ =2A(1)=2bh(I)-b

Figure
C

2
and
2ch(2) LI (2)_ 2A~2) _-2-_ c _ -2A(2)-2bh(2)-b-L1
(I)

Do Problem 6.1 for the natural coordinates

eland

(Figure P6.3).

2 The proof is complete.

Figure P6.3

PROBLEMS
6.1 The shape functions for the three node one-dimensional quadratic element (Figure P6.1) relative to the local coordinate s follow. Write these shape, functions in terms of the local coordinate q.

6.4

A local coordinate-system r has its origin at the one-third point of a linear element (Figure P6.4). Develop the shape functions in terms of r :

Ni=(1-

~)(1-f),
Nk=

Nj= ~

(1- f)

~I

13

2S(s_~)2

~I:~-----L------~~~I
Figure P6.4

82 (a) Starting with the x-coordinate system. (b) Starting with the s-coordinate system. (c) Starting with the q-coordinate system. 6.5

BASIC CONCEPli

COORDINATE SYSTEMS

83

6.11

Verify that ¢(1)=¢(2) along the common boundary between the trian~ular and rectangular elements in Figure P6.1l. Hint: Use the st-coordmate system for the rectangular element.
4

The integration of d¢/dx occurs in many finite element applications. Whel N; and N, are written in terms of a new coordinate variable, d¢/dx mus be evaluated using the chain rule, that is, for the local coordinate s,
d¢ ds

dsb dx
dx ds

and

d¢

d¢ -_-dx ds dx/ds

(1)

3
y

Show that the following relationships hold for the one-dimensional co ordinate systems studied in this chapter.
d¢ ds d¢ dx ' dq dx

1
I
~x

and

Figure P6.11

6.6

The quadratic shape functions given in Problem 6.1 are 6.12 when written in terms of the natural coordinates eland equations for the shape functions and (6.17),evaluate

2.

Using these

6.13

Show that the shape functions for the rectangular element, (6.19), reduce to the linear shape functions, (6.9),along side ij. The three shape functions along one edge of a quadratic triangular element are given in Figure P6.13. Show that these equations reduce to th~ sha~e functions for the one-dimensional quadratic element that are given m Problem 6.6.
y

6.7 6.S 6.9

Verify that L2 = N} for the linear triangular element. Verify that any line of constant ¢ in a triangular element is a straight line Hint: Investigate the value of ¢ along the side of a triangle when two node have the same value. Evaluate the following integrals using (6.29).
(a) L N; Nk dA (c) L(N;2+Nj)dA (b) L N;NjNk dA (d) L(NjNk+NJdA

N, = Lj (2L1 Nj Nk

1)

4L1L2 L2 (2L2 - 1)

6.10

Verify that ¢(1)=¢(2) along the common boundary between the two reo tangular elements shown in Figure P6.10.
2
4

6
L-

~x Figure P6.13

(1)

(2)

L.
y

Figure P6.10

PART TWO FIELD PROBLEMS

Chapters 7 through 16 cover the finite element solution of steady state and time-dependent field problems. Specific application areas include heat transfer, irrotational flow, and acoustical vibrations. Thereader whose interest is in structural applications can go directly to Part Three, which covers structural and solid mechanics applications.

Chapter TWO- DI M ENSIONAL

FIELD EQUATION
The implementation of the finite element method can be subdivided into three broad steps: (1) establishing the element interpolation properties, (2) evaluating the element matrices, and (3) solving an actual problem. We have discussed the properties of a pair of two-dimensional elements in Chapters 5 and 6. Specific applica tion areas are considered in the next several chapters. Our immediate objective is centered on the discussion of the element matrices associated with the twodimensional field equation: D'; ox2 +Dy oy2 -G<jJ+Q=O

02<jJ

02<jJ

(7.1)

We start by briefly discussing the governing differential equations for several physical problems embedded within (7.1). The emphasis in the remainder of the chapter is on the derivation of the integral equations for the element matrices and the evaluation of these equations for the linear triangular and bilinear rectangular elements.

7.1 GOVERNING DIFFERENTIAL EQUATIONS

The general field equation, (7.1), has many important applications in the physical sciences. A few of these applications are discussed in this section. Since our objective is to establish the usefulness of (7.1), a discussion of the boundary conditions is deferred to the application chapters that follow. The first application area is the torsion of noncircular sections. The governing differential equation is

1 02<jJ 1 02<jJ --+--+2e=0 9 ox2 9 oy2

(7.2)

where 9 is the shear modulus of the material and is the angle of twist. Equation (7.2) is obtained from (7.1) by noting that Dx=Dy=l/g, G=O and Q=2e. The variable <jJ is a stress function, and the shear stresses within the shaft are related the derivatives of <jJ with respect to x and y. Several fluid mechanics problems are embedded within (7.1). The streamline and potential formulations for an ideal irrotational fluid are governed by

02<jJ 02<jJ ox2 + oy2 =0

(7.3)

88
and

FIELD PROBLE

TWO_DIMENSIONAL

FIELD EQUATION

89
(7.9)

~+~=O
ux

021jJ 021jJ oy

02W 02W 4n2 w-O h ox2 +h oy2 +gT2 -

(1,

respectively. The streamlines, 1jJ,are perpendicular to the constant potential li ¢, and the velocity components are related to the derivatives of either </J or IjJ wl respect to x and y. Each of (7.3) and (7.4) are obtained from (7.1) using Dx=Dy';" and G=Q=O. ' The flow of water within the earth is governed by equations that are embedd within (7.1). The seepage of water under a dam or retaining wall and within confined aquifer is governed by

is the water depth at the quiescent state, w is the wave height above the where h level 9 is the gravitational constant, and T IS t h e peno 0 f OSCI'11' tiion . iod . nt a qUlesce , . A fluid vibrating within closed volume IS governed by

02 P 02 P w2 -+~+2 p=O ox2 oy C

(7.10)

where P is the pressure excess above ambient pressure, ~ is the wave frequency, and c is the wave velocity in the medium. The waves descnbed by (7.10) would not be a function of the z-direction. , . Equations (7.2) through (7.10) describe nine different physical problems that are contained within the general differential equation (7.1). It should be clear t~at a discussion of applying the finite element method to solve (7.1) is worthy of our time.

02¢ 02¢ Dx~+Dy~=O

ox uy where D'; and D; are the permeabilities of the earth material and ¢ represents t piezometric head. The water level around a well during the pumping process' governed by

02¢ 02¢ Dx ox2 +Dy+ oy2 +Q=O

where Q is a point sink term. The other coefficients are the same as defined f , (7.5), and the aquifer is assumed to be confined. There are two heat transfer equations embedded within (7.1). The heat transf from a two-dimensional fin to the surrounding fluid by convection is governed b

7.2 INTEGRAL EQUATIONS FOR THE ELEMENT MATRICES

Our immediate objective is to derive the integral equations that define the element matrices for the group of problems embedded in (7.1). The element contribution to the system of equations __ {R(e)} is given by

r [N]T JA

(D 02¢ x ox2

+D a ¢ Y oy2

G¢

+ Q)

dA

(7.11)

02T 02T 2h 2hTf Dx;)z+Dv;;z - ~ T+--=O

ox ' uy t t The coefficients D'; and D; represent the thermal conductivities in the x and directions, respectively: h is the convection coefficient; t is the thickness of th' fin; Tf is the ambient temperature of the surrounding fluid; and T is the temperatu " of the fin. Relating (7.7) to (7.1), we find that G =2h/t and Q = 2hTf/t. The fin of (7.7) is assumed to be thin and the heat loss from the edges is neglecte .' When the body is very long in the z-direction and the temperature is a function 0 only the x- and y-coordinate directions, the heat transfer is governed by

where [N] is the row vector containing the element shape functions. Since the interpolation function, ¢(x, y), does not have continuous derivatives between the elements , the second-derivative terms in (7.11) must be replaced by firstderivative terms. The second-derivative terms in (7.11) can be replaced by applying rule for differentiation. Consider the quantity the product

O !__([NY ¢) ox ox
Differentiation gives

(7.12)

02T 02T Dx~+Dy~=O

ox
UY2

(7.8

where Dx, Dy, and T represent the same variables as given for (7.7). Heat transfer by convection is related to (7.8) through the boundary conditions. When G in (7.1) is a negative coefficient and Q equals zero, the differential equation is called the Helmholtz equation. A negative G leads to the sol ution of an eigenvalue problem. A couple of physical problems governed by the Helmholtz' equation are the Seiche motion of water and acoustic vibrations. Seiche motion, which describes the standing waves on a bounded shallow body of water, is governed by

!__([NyO¢)=[Ny02</J ox ox
Rearranging and substituting

ox2

+ o[NY o¢
ox 2

ox

(7.13)

for [NY 02 </J/ox in (7.11) produces

- /. [NY

o. ox2

02¢

dA

= -- JA o, 'ox [NY oX

0(

O¢')

dA

+ JA o, --a;- ox

o[Nya</J

dA

(7.14)

The first integral on the right-hand side of (7.14) can be replaced by an integral around the boundary using Green's theorem (Olmstead, 1961). Application of the

90
theorem yields

FIELD PROBlE.

TWO_DIMENSIONAL

FIELD EQUATION

91

(k(e)] = JA
(7.1 wh~re 0 is th~ angle to t~e outward normal and tuting (7.15) mto (7.14) gives the final relationship

r (D x ~[ox NY

o[ N] + D o[ NY o[ N]) dA + G[N]T[N] ox y oy oy JA

dA

(7.23)

r is the

element boundary. SubS1 for the second-derivative term I

and (7.24) The variable ¢ in (7.22) was not replaced because the quantity D; occurs in the derivative boundary

- JA o, [NY r
produces

~2~

ox

dA = -

Jr

r o, [NY

o¢ cos 0 st:+ o, o[NY aX JA oX

o¢ dA

ox

(7.11

A similar set of operations

starting with

o¢ o¢ . ax cosO+Dy oy smO
conditions and is considered

(7.25) in detail in Chapter by defining (7.26)

r -1 Dy[NY
A

8"2dA=Y

02¢

ir

~([NYO¢) oy oy Dy[NY o¢ -sinOdr+ oy

9.
The first integral in (7.23) can be written more compactly

JA

r o, o[NYo¢ 8y dA oy

(7.1 and the gradient vector

[D]=[;

Substitution

of (7.16) and (7.17) into (7.11) gives

{R(e)} =

-1
t
JA

([NY

(Dx ~: cos O+Dy ~~ sin oX o¢ +Dy o[NY o¢)dA ox oy oy

0) dr
(7.18

+ +

r (Dx o[NY
G[NY¢

{g,i

da-

~1{t~~l if}
[BY =[O[NY oX O[NY] oy

($("] ~[B1W"]

(7.27)

Q[NY dA

Eq~ation (7.18~ is close to the desired form. It can be put in a final form by substl tuting for ¢ using the relationship,

The firs; row of {gv} is the derivative of [N] with respect to x; the second row is the derivative of [N] with respect to y. The transpose of [B] contains two columns and is given by (7.28)

¢(e) = [N]{ <I>(e)}

Substitution and rearranging gives

(7.19

If we use (7.26), (7.27), and (7.28), it is easy to verify that

{R(e)}=-1

[NY(Dx~:

cosO+Dy ~~ sinO )dr O~~] + o, o[~Y O~~]) ) dA {<I>(e)}

+(

JA

[BY[D][B]dA=

JA

r (DxO[NYO[N]+DyO[NYO[N])dA oy oX ax oy

(7.29)

The stiffness matrix for field problems is usually written as (7.30) (7.20: and the individual integrals are denoted in this book by

(Dx o[~y

+(t G[NY[N]dA
which has the general form where

) {<I>(e)}_ Q[NY dA

[k~)] and [k~)], where

(7.31)

[k(e)] = [k~)] + [k~)]

(7.21]

{R(e)} = {](e)} + [k(e)]{<I>(e)} {fIe)} _

7.3
(7.22)

ELEMENT MATRICES: TRIANGULAR ELEMENTS

Our objective for the rest of this chapter is to evaluate the element matrices for the two-dimensional elements discussed in Chapter 5.

92
The scalar quantity cjJ is defined over a triangular region by

FIELD PROBl

TWO_DIMENSIONAL

FIELD EQUATION

93

where

NiNj Ni=2A Nj=2A Nk=2A

1

(ai+bix+CiY) (aj+bjx+ Cjy)

N2J
NjNk LIL2
L~ (7.37)

1 1

(ak+bkx+CkY)
5. The gradient vector fI

L2L3
since Ni=Lb Nj=L2, and Nk=L3 for the linear triangle. formula (6.29) to evaluate each integral yields 1 Using the factorial

a~d the a, b, .and this element IS

coefficients were defined in Chapter

(gv}

= ONi OY

[aN, ox

e»,
OY bj
Cj

ONj ox

e», (
oY

aN'] ox

2
f <l>(e)}

(7.38)

(7.3

or
f (gvf 1

The element stiffness matrix for the triangular element is the sum of (7.36) and (7.38) when G is nonzero. The element force vector also involves the shape functions, and the evaluation of (7.24) is quite similar to the evaluation of [k~)J. Substitution gives (7.39)

=- 1 [bi 2A c,

bkJ {<l>(e)} = [B]{ <l>(e)} Ck

assuming that Q is constant (6.29) produces

within the element. Using the integration

equation

[k~)] =
or

f [BY[D][B]
A

dA = [BY[D][B]

f dA
A

(7.40)

[k~)] = [BY[D] [B]A

Expanding the matrix product yields

ILLUSTRA TIVE EXAMPLE

Heat transfer from a two-dimensional fin is governed by (7.7). Calculate the element matrices for the element in Figure 5.8 when D'; =Dy =0.5 W [cui-C, h = 0.01 W/cm2-C, t=0.5 em, and Tf=10C. The element stiffness matrix [k(e)] is given by (7.30), where [k~)] and [k~)] are defined by (7.36) and (7.38), respectively. The band C coefficients for (7.36) were calculated in the example associated with Figure 5.8. These values are

(7.36j
I
I

IS

. The secon~ i~tegral of (7.30) involves the shape functions. If we assume that constant within the element, this integral becomes .1

d !

11

bi= -4.5,
Ci=

-2,

bj=5, ci= -2,

I~

94
The area was also calculated and is 9.5 cm '. The parameters matrices in (7.36) and (7.38) are 4A

FIELD PROBLEMl

TWO_DIMENSIONAL

FIELD EQUATION

95

multiplying

the

o, o,

a rectangular element defined relative to an st-coordinate system can be sum~arized as follows. Since the st-coordinate system is parallel to the xy-coordinate system and a unit length in either s or t is the same as a unit length of x or y,

= 4A = 4(9.5) = .0132
2(0.01)9.5 _ 03 - 12(0.5) -0. 17 coefficients into (7.36] 4
4

0.5

j(x, y) dx dy=

j(s, r) ds dt

(7.41) The chain rule gives (7.42)

GA _ 2hA 12 - 12t Substitution of the band and (7.38) gives [k(e)] as 0.0132 20.3 -22.5 2.25 1 2 or

Equally important

is the relationship oNp ox

between the derivatives. and oNp oy oNp ot

c coefficients and the calculated

as

oNp

- 22.5 25 - 2.50

2.25 ] [ -2.50 +0.0132 0.250

4 4 -8

-8
16

-8]

The shape functions (5.17) are s Nj=2b st - 4ab

-8

~]
=

The evaluation of [k(e)] and {f(e)} is illustrated by considering in each case. The easiest integrations are associated with {j(e)}, -0.213 0.446 -0.107 -0.0442] -0.107 0.278

[k(e)]

0.384 0.213

-0.0442

{f('))~

L
b
=

Q[NydA~

r r l~lld)
Q a 4~b dt ds A 4
0

a specific integral which is

(7.44)

The element force vector thus

{J(e)} is given by (7.40). In this example, Q = 2hcP fit;

Considering the third coefficient gives

QA _ 2hcPfA _ 2(0.01)(9.5)(10) _ 2 33t 3(0.5) -1. 7

and

r: f:b
Jo

n, dt ds=
8ab
0

f:b r:

2b -st212a ds= J2b -ds=as

2b

(7.45)

{J(e)} = l.27} 1.27

{ 1.27

The other three integrals yield the same result and

7.4

ELEMENT MATRICES: RECTANGULAR ELEMENT

{f(,))~Q:

(7.46)

Evaluation of the element matrices for the rectangular element cannot be performed as quickly as the integrations in the previous section. Each coefficient involves integrating a polynomial over an area. The integrals can be evaluated using the shape functions given by either of (5.17) or (5.19). We will use (5.17) because of the similarity between the st- and xy-coordinate systems. The shape functions in (5.17) were developed relative to the st-coordinate system; this presents a minor problem. All of the integrals are defined relative to the xy system. In particular, the gradient matrix [B] has coefficients related to the derivative of the shape functions with respect to x or y. The change of variables equation for double integrals is discussed in detail in Chapter 27 with regard to quadratic elements. The application of this equation

The integral associated

with

[k~)] is

NiNk NjNk N~ NkNm

(7.47)

96
Selecting the

FIELD PROBLEM!

TWO.DIMENSIONAL

FIELD EQUATION

97
is the sum of

Nf

term, we have

Jo Jo

I" [2a (_!!__)2 dt ds =

4ab

Jo Jo

[2b [2a ~ 4ab 16a2b2 dt ds =9=9

The element stiffness matrix [k(e)] for the rectangular element (7.49) and (7.55). The element force vector is given by (7.46).

(7.48 PROBLEMS

The complete set of coefficients is

[k(e)J
G

= GA

36

f;
1 2

2
4

7.1

2
4

Evaluate one of the following integrals rectangular element.

(a)

related to the force vector for the

2 1

11
of the shape functions.

(7.49;

QNi dA

(b)

QNj dA

T~e evalua~ion of [k~)J gradient matnx [BJ is

involves the derivatives

The

7.2

Evaluate element.

one of the following integrals related to

[k~)] for the rectangular

[BJ = ox'
ONi oy

e»,
Ox oNj oy

ONk ox ONk oy

mOl
ox oNm oy

(7.50]

7.3

Evaluate the integrals for one of the following coefficients in rectangular element. (a) Row 1, column 3. (b) Row 2, column 4. (c) Row 4, column 4.

[k~)] for the

U~ing th~ r~lationships given in (7.42) allows DIfferentIatIOn of the shape functions gives

[BJ to be written in terms of sand t.

-t
S

[BJ=_l

[-(2a-t) 4ab -(2b-s)

(2a-t) -s

(2b-s)

(7.51) after the

7.4

The .co~ffi~ient in t~e first row and first column of [k~)J is available multiphcations [BJ [DJ[BJ have been performed. This coefficient is 16a2b2 and the associated 2b 12a 1o which integrates
0

Evaluate the coefficients in (7.24) when Q varies linearly over a triangular element, that is, Q(x, y) is given by Q(x, y)=NiQi+ NjQj+ NkQk

o,

(2

a-t

)2

+ 16a2b2 (2b-s)2
2b 2a
0 0

o,

7.5 (7.52)

Evaluate the coefficients in (7.24) when element, that is, Q(s, t) is given by

Q varies linearly over a rectangular

integral is D 16 ;b2(2a-t)2dtds+1 a

Q(s, t)=NiQi+NjQj+NkQk+NmQm

7.6 D; 2 (2b-s)2dtds 16a b (7.53)

to Dxa Dyb 3b +3;; (7.54)

Write the transformation equations that apply between the xy- and stcoordinate systems of the rectangular element. Use these equations and the chain rule equations

e», es, ox es, oy -=--+-os ox os oy os

The complete result for [k~)J is

en, oNpox oNpoy -=--+--ot ox ot

oy ot

[k~)l~

D""[- ~
6b -1

-2 2

-1

1
2 -2

-2

-}D~ [ :
6a-1 2 -2

1
2 -2

-1
-2 2

-1

-1

=]
(7.55)

7.7

to verify the relationships given in (7.42). Write the transformation equations that apply between the xy- and qrcoordinate systems. Use these equations and chain rule equations similar to those in Problem 7.6 to show that
--

en,
ox

oNp oq

and

en, -oy

oNp or

98
7.8 7.9

TWO_DIMENSIONAL

FIELD EQUATION

99

Develop a relationship for [B] similar to (7.51) that applies in the q coordinate system. Calculate [keel] and {pel} for a linear triangular element when (a) Dx =Dy = 1, G =4, Q = 5, and the coordinates are those given in Proble 5.7. (b) Dx=Dy=O.25, Problem 5.8.
G=6, Q=3,

- JA [N]

02ct> dA oxoy (act> cos oy

=~
2

JA

r (O[NY

aX

act>+ o[NY oy oy

act» dA ax

_ Jr [NY r
7.13

e + act> sin e) dr ax
oX act>+ o[NY oy oy oct» dA ax

and the coordinates are those given i'

Evaluate

7.10

(c) Dx=Dy=2, G=6, Q=6, and the coordinates are those given in Problem 5.9. Calculate [keel] and {fee)} for a bilinear rectangular element when (a) Dx =Dy = 1, G = 12, Q = 5, and the coordinates are those given in Problem; 5.12. (b) Dx=Dy=O.5, G=10, Q=40, and the coordinates are those given in Problem 5.13. . (c) Dx=Dy=2, G=2, Q =6, and the coordinates are those given in Problem, 5.14. Given the definitions ct>(e) [N]{ <1>(e)}and = dct>(e)= deN] dx dx start with {<1>(e)}= [B]{<1>(e)}

~ r (O[NY
2

JA

given that ct> = [N]{ <1>(el} and [N] consists of the shape functions for the linear triangular element.

7.11

and obtain the general form {R(e)} where

D ~~IX;Xi

= {I(e)} + [k(e)]{ <1>(e)}- {fee)}

[keel] and

dct>1 -Ddx
X;Xj
j

D[ BY[ B] dx

{feel} =

J:j. Q[ NY

dx

7.12

for the one-dimensional linear element. The mixed-derivative term 02ct>/OX oy occurs in some two-dimensional differential equations. Show that

TORSION OF NONCIRCULAR

SECTIONS

101

Chapter

8
SECTIONS

--6
z

TORSION OF NONCIRCULAR

The element matrices for the linear triangular and bilinear rectangular elements were evaluated in the previous chapter. Application of this information to obtain a1 numerical solution of a realistic problem is discussed in this chapter. We will! calculate the shear stresses in a square steel bar subjected to a twisting torque, The torsion of noncircular sections has been selected as the initial application area: because it has the simplest of the possible boundary conditions; ¢ is zero on the', boundary. ' 8.1 GENERAL THEORY

There are two theories for calculating the shear stresses in a solid noncircular1 shaft subjected to torsion. St. Venant developed one theory, and Prandtl proposed' the other. Both theories are discussed by Fung (1965). Prandtl's theory is used In' this chapter. ! The shear stress components in a noncircular shaft subjected to a twisting1 moment T about the z axis (Figure 8.1a) can be calculated using l
LZX

(b)

a¢ =ay

and differential equation is (8.2)l

I

Figure 8.1. (a) The shear stress components in a noncircular section subjected to a torque loading. (b) The if> surface and related shear stress components. The stress function represents a surface covering the cross section of the shaft (Figure 8.lb). The torque is proportional to the volume under the surface while the shear stresses are related to the surface gradients in the x and y coordinate directions. Equation (8.2) is usually written as

where ¢(x, y) is a stress function. The governing

g ox

1 a2¢ 1 a2¢ --::;~+--+20=0 2

g ay2

with

¢ =0

(8.3)1

a¢ a¢
2 2

on the boundary. The physical parameters in (8.2) are the shear modulus, (Nzcm "), and the angle of twist per unit length, 0, (rad/cm).l Prandtl's formulation does not have the applied torque, T, (N . em) in the govern-j ing equation. Instead, T is calculated using

g,.j

-a x 2 +-a.2 +2g0=0 y
when the shaft is composed of a single material. This equation is obtained
(7.1) by noting that Ir, =Dy= I, G =0, and

(8.5)
from

Q =2g0.

T=2
once ¢(x, y) is known.

¢dA

(8.4)

8.2

TWISTING

OF A SQUARE BAR

The square shaft (Figure 8.2a) is used to illustrate the evaluation and the assemblage of the element matrices into a set of linear equations. This shaft has four axes of

102

FIELD PROBLENij

TORSION OF NONCIRCULAR

SECTIONS

103

The matrices for the triangular element are given by (7.36) and (7.40) while those for the rectangle are given by (7.46) and (7.55). The matrices become
CjCj

The shaft is 1 em square.

g
IJ

= = 0.01 degree/em
8(106) N/em2

1 [b; [k)))] = 4A b.b, bjbk

c,

(8.6)

= 0.0001745 rad/em

and
(8.7)

(a)

for the triangular

6

element and

(3) 5 (2) (1)

2

-t _l

[k"'] ~~
and

0.25 em

r=~

-1
4

l-l

-2 -1
4

-1 -2

-11
-2 -1
4

(8.8)

-1

0.25em

(8.9)

for the rectangular

1--025
>~ (b)

element. Equation

(8.8) also incorporates

the fact that element

0.25--1

two is square, 2a = 2b. We shall now evaluate (8.6) for elements one and three. The element area is -b and 4A(1)=~. The band C coefficients are

Figure 8.2

Element subdividion for the torsion of a square shaft.

W)= Y2 - Y4= -0.25,

b~l)= Y4 - Yl =0.25, b'1)= Yl- Y2=0,
Substituting these values into (8.6) gives -1 1

C\l)= X

4-

X 2 =0

symmetry; therefore, only one-eighth of the cross section needs to be analyzed. This fractional portion is divided into three elements (Figure 8.2b). Three elements are not sufficient to obtain an accurate answer, but they are enough to illustrate the calculations. The calculations have three significant digits of accuracy. The element node numbers are e
1
j

C~l)=Xl -X4= -0.25 C~1)=X2-Xl =0.25

0] 8 [0
~ + 16
~

o
-1

k
4 5 6

m after noting
4

o ±-h. Adding
that 1/4A(1)=8

-:J

(8.10)

2
3

2
4

2 3 5

and the bjbj,

CjCrtype products

are either zero or

the two matrices yields

-1
and the same dimensions; 2 (8.11)

Elements one and three have the same orientation therefore, their matrices are identical.

-1

104

FIELD PROBLEMS

TORSION OF NONCIRCULAR

SECTIONS

105

The element force vector, {f(1)}, is readily obtained once the 2g0 parameter been evaluated. Substituting the values given in Figure 8.2 produces 2g& =2(8)(106)(0.01) Substitution of 2790 and A(
1)

has

;0) = 2790

<PI = 217

= l"2 into (8.7) gives

29.1

{f(1)}T=[29.1

29.1]={f(3)?

(8.12)

The stiffness matrix for element two is given by (8.8). The force vector, (8.9), is [j(2)}
T

= [43.6

43.6

43.6

43.6]

(8.13)

since A(2) =~ and 2gB is 2790. The element matrices are summarized below. The node numbers indicate th~ rows and columns of [K] and {F] to which the individual coefficients add.

[k"'] ~

[k'''] ~

I[_: Ir-~
1 0 2 -2 -1

2 -1 2

-1
3 -1 4 -1 -2

-n
5 -2 -1 4 -1 6

{f(1)] = 29.1

tl} I
2 29.1 4
r /(2)1
_

(8.14)

-Il
4 -2 -1 4

[ 43.6l2 43.6 3 5 4

<P6

=0

'

1-143.61 43.6

(8.15)

Figure B.3.

The rj; surface for the torsion problem.

[k'''] ~~ [

-l
0 -4 4 -2 -4

5 -1 2 -1

-:J
0 -2 -1 -4 10 -3 3 -3 0 0 0 0

{f(3)} = 29.l

tl} 4
5 29.1 6 stiffness procedure

(8.16) and

H
Solution yields

-3
10

-4

-~]{::} ={~} -{~~!}

10 <1>4 611 0 <1>2 159, = and

Adding the element contributions using the direct multiplying through by six gives the system of equations 3 -3 0 0 0 0 -3 10 -1 -4 -2 0 -2 -1 0

10 0 0

175
436 262 436 175

-1

The nodal values <1>3, and <1>6re on the external boundary and each is zero; <1>5, a therefore, equations three, five, and six are eliminated. Columns three, five, and six must be incorporated into {F}. Since <1>3 =<1>5 =<1>6 =0, they contribute nothing to {F} and the modified system of equations is

l~f

<1>2

1 01
0 0 (8.17) 0 0

61f o[

The ¢ surface for this set of nodal values is shown in Figure 8.3. The determination of the nodal values is a major part of the solution, but there is usually a set of element quantities that must be calculated once the nodal values are known. The shear stress components and the twisting torque are of interest in the present example. The evaluation of these quantities is discussed in the next two sections.

8.3

SHEAR STRESS COMPONENTS

because the shear stress

The gradients of the nodal parameter, ¢, are important components are related to these gradients by
Tzx=-

o¢

oy

and

!zy=

--

o¢

ox

(8.18)

106
The gradient vector for the triangular

FIELD PROBLEM

TORSION OF NONCIRCULAR

SECTIONS Positive shear stress components

107

element is given by (7.34) as

{gv}
The gradient

1 =_ 2A

[b.

Tzy

c,

(8.19
6

vector for the rectangular

element is given by (7.51) as

{gv} = 4ab -(2b-s)

1 [-(2a-t)

(2a- t) -s

(8.20
y

568

The area as well as the band C coefficients are the same for both triangula elements and they were evaluated prior to (8.10). Using these values and the calculated values for <1>[, <1>2, and <1>4 gives

{gv(l)}
and

=-

16[-1
4

1
fI------

68_j 636
..... 2~-----

-1

OJ {217} ={-232} 159 -136

1 125

... 3

i~

Figure 8.4

The shear stress values for the square shaft.

!zx

(1)

=-=

o<jJ

oy

-136 N/ em 2 N / em 2
in a similar manner givin

and and The shear stress values calculated above are shown in Figure 8.4 along with the values of !zx and !Zy at the center of the rectangular element. The shear stress values calculated for each triangular element are constant within the element and are usually assumed to be the values at the center of the element. There are at least two ways to improve the stress values obtained for this example. First, a larger number of elements can be used. As the element size decreases, the existence of a constant value within the element becomes more realistic. An alternative approach is to use elements with more nodes and a quadratic or cubic interpolating polynomial. Differentiation will then yield gradients that are a function of the location within the element.

!zy

(1)

o<jJ = - -=232

ox

The stress components

for element three are calculated !~~)=O and

!W=500 Nzcm?

The gradient values are not constant within the rectangular element. Thus w can calculate the shear stress components at node three, which is where the larges value of r., occurs. The local coordinates of node three are s=2b and t=O; als 2a=2b=0.25. The gradient vector is 0

I.gv

(2)_16[ -1 -4 0

-1

~]{~J
-~36}

8.4

EVALUATION OF THE TWISTING TORQUE

is the twisting

Another quantity of interest in the analysis of a torsion member torque T defined in (8.4). This integral is equivalent to T The integral

or

{gV(2)}

= [-4
0

4 -4

~J ~ ={
125

r}

e=

21

<jJ(e) dA

(8.21)

A(e)

108
is
JA(e)

FIELD PROBL

TORSION OF NONCIRCULAR

SECTIONS

109

4>(e) dA =

i (<1>i+ <1>j <I>k) + 3

the actual applied torque is

!zy=

(Tact)
Teal Tzy

for the triangular

element and

JA(e)

4>(e) dA =~4 (<1>i+ <1>j <1>k+ <1>m) +

=(250) (636)=909 175

N/cm2

(8.26)

for the rectangular element The torque, T, is the sum of the element contributions T= The element contributions are +<1>2 +<1>4) + 159 + 125)= 10.4 +<1>5 +<1>6)=2.60 T(1)+ T(2)+ T(3)

or

8.5

COMPUTER SOLUTIONS FOR THE SQUARE BAR

T(1)=2(i2)(~)(<1>1 =is(217 T(3)=ls(<1>4 The torque is T=

T(2) = 2(1o)(j:)(<1>2 <1>3 <1>5 <1>4) 8.88 + + + =

10.4+8.88 +2.60=21.9

N· em

The three-element grid discussed in the previous sections i~ the s~al1est number of ts that can be used to solve the torsion problem involving a square bar. I eemenaccurate values for 0 and the shear stresses are b taine d usi M using gnld s ith a ore mber of linear elements or by using elements with a higher level of interlarger nu . d ] thi ti n Three computer solutions for the square bar are summanze m IS poI 10 . a f h I' . I · The first grid consists of SO elements and is a mixture 0 t e mear tnang e sect Ion. h (F' 8 6) and the bilinear rectangular elements (Figure 8.5). !he ot er two igure '. are mixtures of the quadratic triangular and quadnlateral elements. One gnd has three elements, the other seven. . . The maximum shear stress and the maximum value of 4> for the vanous gnds are summarized in Table 8.1. The 50-element grid is a definite improvement over

":1

(8.

This torque acts on one-eighth of the cross section. Thus the torque acting on square bar is 8(21.9)= 175 N· ern. I t takes a torque of 175 N· em to produce a twist of 1 in a l-cm square st shaft that is 100 em long. The accuracy of this result is questionable, howey because of the coarseness of the grid. In fact, our answer is 11 percent below t theoretical value* of 196 N . em. The calculated value of the torque seldom corresponds to the actual appJi value because we have guessed at the angle of twist, O. The correct values of Tzx, Tz and 0 are obtained by scaling the calculated value by the ratio T.etual/Tcaleulated' F··· example, suppose that the torque applied to the shaft in Figure 8.2 was 250 N . The true angle of twist is
0

or
0tTue

250) = ( 175 (0.01)=0.0143

rad/cm
Tzy

The shear stress values are scaled in the same way. The largest value of

fo
<Pmax = 208.6

*The relationship between the applied torque and the angle of twist for a square of dimension 20 given by T=0.1406g0(2a)4 (Timoshenko and Goodier, 1970, equation 170, p. 313). For our exampl 20=1 and T=0.1406g0=196N·cm.

49 nodes, 50 elements

Figure B.S.

A 50-element grid for the square shaft.

110

FIELD PROBLE~
J

TORSION OF NONCIRCULAR

SECTIONS

111
of a

Table 8.1
Square Bar Grid Figure 8.2b Figure 8.5 Figure 8.6a Figure 8.6b Theoretical

Computer Solutions

Torque 175 192 196 196 196

<lImax 217 212 206 206 207

'tmax

640 893 928 937 945

4>max = 206.2 14 nodes, 3 elements

grid is within 6 percent of its theoretical value. If we are interested in derivativerelated quantities, then we need a fine grid of linear elements or several quadratic elements. The maximum shear stress for the 50-element grid is 893 Njcm ', which is 5.5 percent below the theoretical maximum. It took 50 elements to obtain the same accuracy in a derivative quantity that three elements produced in the nodal values. This ratio will not be the same for all problems, but it should be kept in mind. PROBLEMS 8.1 Solve the torsion problem for a square shaft using the four triangular elements shown in Figure P8.1. Set up the equations, solve for the nodal values of </1, and calculate the maximum shear stress in element three. Use 2g() =2790.

4>max = 206.2 24 nodes, 7 elements

~5+
elements to solve the square shaft
L...___ _

t 0.25em
0.25 em

Figure 8.6.
problem.

Two grids using quadratic

3i
_j

0.5em

the 3-element grid of Figure 8.2h. The grids with the quadratic elements, howeve give more accurate results than the grid with 50 linear elements. Table 8.1 indicates why the higher-order elements are included in many CODl mercial computer programs: Greater accuracy can be obtained with fewer el ements. Fewer elements means less work preparing the element data. A comparlso of the .esults for the quadratic grids indicates that additional elements are nd needed. Table 8.1 also emphasizes a fact that should be kept in mind when constructin the grids for a finite element problem. A rather coarse grid gives good results] we are only interested in the nodal values. Note that <l>max for the three-elemen

~IE~------lem--------~>~1

Figure P8.1

8.2 Solve the torsion problem for a square shaft using the four rectangular
elements shown in Figure P8.2. Set up the equations, solve for the nodal values of </1, and calculate the maximum shear stress at node three. Use 2g() = 2790.

112
7 8 9

FIELD PROB~

TORSION OF NONCIRCULAR

SECTIONS
y

113

(3) 4*
(1)

(4) 5• (2) 2

r.0.25+0.25'_

31
_j
9

6+

t0.25em
0.25em

3em

II
1
2b

0.5em

4>=-

G8a b (a2 + b2)

22

(~+~-1)
a2

b2

2G8a2b
Tm"=(a2+b2)

at

y=rxb

P8.14
1--0.5

Figure P8.2

8.3

Do Problem

8.2 using the four-element

7 (3) 8

grid shown in Figure P8.3.

~----~-----a----~~

(4) 5• (2) 3 6

4•
1

(1) • 2• k02~03-~

-t

0.3em

J
IkE--------1em------

1
0.2 0.5em

Figure P8.3 Use the nodal coordinates and ¢ values for the triangular elemer given in Problems 5.7-5.11 and calculate the values of T zx and T zy "or t 11 element. Also calculate the element contribution to the twisting torque T 8.9-8.13 Use the nodal coordinates and ¢ values for the rectangular elerner given in Problems 5.12 to 5.16 and calculate the values of Tzx and Tzy at t center of the element. Also calculate the element contribution to the twisth torque T'". 8.14-8.15 The twisting of the shapes shown in Figure P8.14 and P8.15 ha analytical solutions. Obtain finite element solutions for these problems usli the computer program TDFIELD (Chapter 16) and a reasonable numb of elements (20 to 50). Compare the maximum values of ¢ and the she stress for the finite e!ement solution with the analytical values. Use 2ge = 27S

1
[

4>=-G8

1 [ 2(x

2 +y)2

1 2a(x

3 -3xy)- 2

2 na 2J

Tm.,

= G~a

at the middle of each side

8.4-8.8

Figure P8.15

8.16-8.19 Obtain finite element solutions to the torsion problem for the cross
sections shown in Figures P8.16 to P8.19 using the computer program TDFIELD (Chapter 16). Use a minimum of 30 elements. Use 2ge=2790.

I _~I}
t____,_

kIE~------6em--------~>~1

Figure P8.16

-3

2~

_j_
1<
7cm
Figure P8.17

i -i
4cm 3cm

3cm

Chapter

DERIVATIVE BOUNDARY CONDITIONS: POINT SOURCES AND SINKS

The torsion of noncircular sections is a unique two-dimensional problem because the value of ¢(x, y) is specified around the entire boundary. Most physical problems have a mixture of boundary conditions. The values of ¢ are specified on part of the boundary, and values related to the derivatives a¢/ax and a¢/ay are specified on other parts of the boundary. This latter situation occurs in both fluid flow and heat transfer problems. We cannot discuss either of these topics until we have studied how to handle derivative boundary conditions. Another phenomenon that occurs in fluid flow and heat transfer problems is the point source or sink. In this situation, Q is concentrated at a point. Physical examples of a sink and a source include the pumping of water from an aquifer and heat generated by electrical lines embedded within a material. The objective in this chapter is to discuss the concepts related to derivative boundary conditions and sources and sinks before we begin the study of irrotational flow and heat transfer.

-+ + --f-12cm 11

1
11

___l
16cm
Figure P8.18

12

9.1 DERIVATIVE BOUNDARY CONDITIONS

The two types of boundary conditions for two-dimensional field problems are shown schematically in Figure 9.1. Over part of the boundary, call it I'l s ¢ is specified. A boundary condition of the type

r l
4cm

a¢ a¢ DXaxcosO+DYaySinO=-M¢b+S
is specified on the rest of the boundary, I'2. When Dx=Dy. (9.1) reduces to

(9.1)

Dx a¢ an
x

= -M¢b+S

(9.2)

where a¢/an is the derivative normal to the boundary. In each case, ¢b represents the value of ¢ on I'2 and is unknown. Equations (9.1) and (9.2) simplify to
D;

a¢ a¢ . ax cce G-i-D, ay sm(}=O

(9.3)

and (9.4)
....I:

Figure P8.19

116
<f> known on is

FIELD PROBLE~

DERIVATIVE BOUNDARY

CONDITIONS;POINT

SOURCES AND SINKS

117

r1

where <l>b is given by the element equation

<I>(e) [N] {<l>(e)} =

Substituting (9.9) into (9.8) produces

(9.9)

{Ib~)} =
which can be separated r2
Dx-cosO+Dy-sinO=-M¢b+S

Jrbe

[NY(M[N]{

<l>(e)} S) st: -

(9.10)

{Ib~}
o¢
on I",

=( r

into

Jrbe

M[NY[N]

dr)

{<l>(e)}-

JrbC

S[NY

dr

(9.11)

o¢ OX

oy

Figure 9.1.

The two types of boundary conditions for field problems.

The boundary condition produces two components. One component adds to [k(e)] because it multiplies {<l>(e)}.The other adds to {J(e)}. The integr~ls, ~owever, are different from those previously encountered because they are line integrals instead of area integrals. The two components in {Il~)}in (9.11) can be defined as

when M =S=O. This situation occurs on insulated or impermeable boundaries; on axes of symmetry. The situation where either M or S, but not both, are zero; also possible. : The inclusion of the derivative boundary condition into the finite eleme analysis of field problems is done using the interelement vector {I(e)} given 1 (7.22). This vector is l

{Il:)} = [k~]{ <l>(e)} {J~e)} where

(9.12)

[k~J =
and

Jrbe

M[NY[N]

dr

(9.13)

{I(e)} = -

Ir [NY

(Dx ~~ cos 8+Dy ~; sin

8)

dr

(9j

{J~e)}
Before we proceed

where the integral is around the boundary of the element in a counterclockwt direction. The integral in (9.5) is the sum of three integrals (one for each side) when int grating around a triangular element and the sum of four integrals for the rectan~ lar element. We shall separate {I(e)} into two components

JrbC

S[NY

dr

(9.14) it should be

with the evaluation

-D

of the element integrals,

Yay

i!!E.

=M<f> -8
b

fI(e)) l
where

1-

- {I(e)} be

+ {I(e)) i

(9~

(9.:
and rbc is the side of the element over which the boundary condition is specific e The vector )} contains the integrals of (9.5), which occur on the element sidj that do not have a boundary condition specified on them. These integrals lead 1 the interelement requirements that must be satisfied before Galerkin's residual.i zero. Using our definition for {Ib~)} and substituting the relationship in (9.1) gives

{n

D'!.P. =M<f>b-8 x ax

Dx

i!!E. ay

=M<f>b -S

-D

x ax

a<l>=M<f>b-S

-D x i!!E. =M<f>b-S ax

{Ib~)} =

r [NY(M<I>b-S)dr Jrbe

D a<l>=M<f>b-S
Yay

Figure 9.2.

The derivative boundary conditions for a hollow rectangular region.

118

FIELD PROBLE~

noted that the boundary condition (9.1) takes on different forms as we traver~ .. i the outside and inside of a rectangular region with a hole. The various forms f<>1

DERIVATIVE BOUNDARY

CONDITIONS:

POINT SOURCES AND SINKS

119

_ (Dx ~fos e + o, ~; sin e) c

The evaluation of (9.14) for the triangular element gives results that are very similar to those in (9.17) and (9.18). The results are

(9.1~

are given in Figure 9.2. The sign associated with M and S must be determined relative to the information contained in Figure 9.2. This determination is illustrated in the application chapters that follow. Remember that is the angle from th4 x-axis to the outward normal. '

{fI"}~sHl s~"m.
{fJe)}
=

and

s~"m
Ij,

(9.19)

for sides ij, jk, and ik, respectively. The quantities Lij' Ljk' and Lik are the le.ngths of the respective sides. They are not the area coordinates. The area coordmates have numerical subscripts. The first result of (9.19) is obtained . ... as follows. Given Side

9.2

EVALUATION

OF THE ELEMENT

INTEGRALS

The integrals in (9.13) and (9.14) are valid for any two-dimensional element an~ can be evaluated once the element shape functions are known. We shall evaluat, these for the linear triangular and bilinear rectangular elements that were studi~ in Chapter 5. We shall start with (9.14) and the rectangular element because this combination is the easiest to evaluate. l Assuming that S is specified over side ij and that the element has a unit thickness gives

l tbe S[NY

.u: =Lij

In S {Ni} ~~
1

de 2

(9.20)

Since

is zero along side ij,

{fJe)} =Lij
because the shape functions

1 {N
S
~j

i}
de 2 =Lij
and

r 1 {e I}
Jo
S e~ de 2
(9.21)

N, and N, reduce to

(9.16)
'j

Ni=L1=e1

Nj=L2=e2

(9.22)

where the shape functions are those defined by (5.19) for the qr-coordinate system. Note, however, that Nk=Nm=O along side ij. Substituting for the nonzero shapd functions and noting that r = _ a,

along side ij. This fact was discussed in Chapter 6. The integr~tion is along a Ii~e; thus we can use the factorial formula (6.17), and the result m (9.19) follows Immediately. . .I h The integrals associated with [k~] are evaluated in a manner identica to t. ose just discussed. The major difference is that there are more terms to consider. The integral in (9.13) expands into

{fje,} ~

r.~ }dq~ S~)j m r:~:;

SLkmt~}l' 2
1 and

(9.m

[k~]=
The quantity S is multiplied by the length of side ij, Lij' which is 2b and divided equally between the two nodes on side i]. .' There are three other evaluations for the surface integral, one for each of the remaining three sides. It is left as an exercise to show that the other results fOJ"I for the rectangular

Ni2 N·N· [be M [ N~NJ ,k NiNm

u,«,
NJ NJ.Nk NjNm

u.»,
NjNk N~ N kN m

n.«;
u,»;
N;,

NiNml

dr

(9.23)

element. If we assume that M is specified over side ij, then

{n~}~

.
I

Nk=Nm=O

and (9.23) becomes

[kWl~
Evaluation

(9.18)

r.

N2 [ N~:Vj

o
Jb(b-q)2
-b --2-

o o o

(9.24)

o
_ a gives
(9.25)

of the individual

coefficients after noting r =

for sides jk, km, and im, respectively. If S is specified on more than one side of ~ element, the values for {fJe)} for the appropriate sides are added together.

N, dq=

-b

4b

dq_

_2b_Lij

3_ 3

120

FIELD PROBLEM

DERIVATIVE BOUNDARY

CONDITIONS:

POINT SOURCES AND SINKS

121

(9.2t and

9.3

POINT

SOURCES AND SINKS

fb

N~d
J

-b

=fb

_b

(b+q)2d
4b2

q_ 3 _ 3

_2b_Lij

(9.21

Using these results, we have 1

[k~]

= MLijl~

o o
0

o o o o
0
2

iJ

(9.28

There are three other results for results are

[k~], one for each of the other sides. TheS4

[kl1l~MLjr 6

0 0

1 0 0 0

0 0
0 2

[kl1l~ M6 0 L'-l~
0 and

0 0 0 0 0 0

f]

~]

An important physical situation is the concept of a point source or sink. A source or sink is said to exist whenever Q occurs over a very small area. Examples of line sources include steam and/or hot water pipes within the earth and conducting electrical wires embedded within a product. In each case, the cross-sectional area of the pipe or conductor is very small compared with the surrounding media. Point sinks occur in groundwater problems: They are pumps removing water from an aquifer. Sources and sinks occur often enough in the real world to warrant our attention. Our discussion is structured around the two-dimensional element, but the procedure can be quickly modified to handle the one- or three-dimensional element. Consider the triangular element in Figure 9.3 with a source Q* located at (Xo, Yo)· Since the source is located at a point, Q is no longer constant throughout the element but is a function of x and y. Using unit impulse functions, 6(x-Xo) and 6(y _ Yo) (Kaplan, 1962), we can write (9.35) The integral

(9.29

{jIP} =
(9.30; becomes

Q[NY dA

(9.36)

[kl1l~ M

L;·l~
0
1 6

0
0

(9.37)

0
0

where L jk, Lkm, and Lim are the lengths of the respective sides. The evaluation of (9.13) for the triangular element leads to

~J

(9.31]
y

[kl1l~MLf

2 0 0 2 1

~]
rJ

(9.32)

[kl1l~ MLj'[~
6 and

(9.33)

[kl1l~ ML;'[~
6

0 0

~]

(9.34)

~---------

~x

Figure 9.3.

An element with a point source or sink.

122

FIELD PROBLE~

DERIVATIVE BOUNDARY

CONDITIONS:

POINT SOURCES AND SINKS

123

The integral of a quantity multiplied by an impulse function, however, is equ~ to the quantity evaluated at X 0 and Yo. Therefore, "

Substituting x=X 0=5 and Y= Yo =2 produces N,

= /3

{fJe)} =Q*

Ni(X 0, Nj(Xo, yo) Nk(XO, yo)

YO)}
(9.3~

[28 - 4(5)- 2] =

6 13

Nj=

/3[6 +5 - 3(2)] = /3
to nodes i, j, and k by the fractions

Nk=-fJ[ -21 +3(5)+4(2)]=A The value of Q* is allotted respectively. Therefore,

The proportion of Q* allocated to each node is based on the relative values ~ Ni, N], and N; evaluated using the coordinates of the point source. Since til shape functions sum to one at every point within the element, we are not allocati~ more than Q*.
(

fJ,

5 13

and

A,

52 {fJe») =13 ~ = {24} 2~

{6}

ILLUSTRATIVE

EXAMPLE The best location for a source or sink is at a node. This location changes the result given in (9.38). If we assume that the source is at node j (Figure 9.5), then Ni=Nk=O and

A line source Q* =52 W/cm is located at (5, 2) in the element shown in Figure 9.~ Determine the amount of Q* allocated to each node. The values of the a, b, and c constants are ai=28, bi=-4, Ci= -1, The shape function equations aj=6, as= -21

bj=l,
Cj= -3,

bk=3
ck=4

(fJ"H'

{I}

(9.39)

can be written after recalling that ai+aj+ak=2A

= 13

The equations

are Ni=ti[28-4xNj=-fJ[6+x-3y] y]

The magnitude of Q*, however, must be modified when the source (sink) is shared by more than one element. The magnitude of the source is divided among the elements joining at the node. The source is allocated according to the ratio of the angle in the element to 360. The correct equation for element (e) in Figure 9.5 is
r rle)1 UQ J

= rxQ* {~,}
360

N« =-fJ[ - 21 + 3x+4y]
y

o
I

(9.40)

f\----

_J.I

\ \ \ ____ /
/

/,~

_.I..~~

(3,3)

~... /"\
\ \

_ j

Figure 9.4.

A point source in a triangular element.

Figure 9.5.

A point source at a node.

124

FIELD PROBLE~

DERIVATIVE BOUNDARY

CONDITIONS:

POINT SOURCES AND KINKS

125

There is no need to evaluate o: for the various elements. When the equations al assembled using the direct stiffness method, the element contributions at thi node add to Q*. An easier procedure for implementing a node source is to add th value of Q* to the row of {F} corresponding to the node number. A source i positive whereas the sink has a negative sign.

(d) Side jk of a triangular (e) Side ik of a triangular

element. element.

9.4-9.8 Evaluate for a point source, Q* =40 W/cm, located at point A for the corresponding triangular element in Problems 5.7-5.11. 9.9-9.13 Evaluate for a point source, Q*=40 W/cm, located at point B for the corresponding rectangular element in Problems 5.12 to 5.16.

UJe)}

UJe)}

PROBLEMS
9.1 The boundary condition around the outside of a rectangular region foil OWl Determine the magnitude and sign for M and S on each of the sides; th sides are labeled as shown in Figure P9.1. (a) D';

o¢

ox =6¢b-

3,

side 2 side 1 side 4 side 3

Side 3

(b) D, oy (c) Dx

o¢

= -4¢b+6,

ox =5¢b+2,
o¢

o¢

(d) D, oy =8¢b-4,

Side

41

L___....___

_I

Sid, 2

Side 1

Figure P9.1

9.2

Evaluate

the integral in (9.14) for: element. element. element. element. element. element. element. element.

(a) Side jk of a rectangular (b) Side km of a rectangular (c) Side im of a rectangular (d) Side jk of a triangular (e) Side ik of a triangular 9.3 (a) Side jk of a rectangular (b) Side km of a rectangular (c) Side im of a rectangular

Evaluate the integral in (9.13) for

IRROTATIONAL

FLOW

127

Chapter

10
FLOW

Vx = 5 em/sec 12 em

IRROTATIONAL

Figure 10.1.

Flow around a cylinder

1/1= 30
<lob

The irrotational flow of an ideal fluid has been studied extensively because of t information that can be obtained about flow around corners, over weirs, throu constructions, and about airfoils. Ideal irrotational flow is an approximation. assumes that no friction occurs between the fluid and a surface (ideal) and that n rotation or distortion of the fluid particles occurs during the movement (irro tional). . T~e flo~ of water through the earth can also be closely approximated by ass mg.lrrotatlOna! flow. The analysis of groundwater flow is an important aspect regional planning because many regions within the country depend wholly or part on groundwater for their water supply. The flow of water through and und dams as well as into underground drains are some other important areas that be studied using the theory.

'!!I! =0
iJx

ax

=0

Figure 10.2.

Boundary conditions for streamline flow around a cylinder.

10.1

FLOW OF AN IDEAL FLUID

The two-dimensional flow of an ideal fluid can be formulated in terms of a stre fun~tion i/J or a velocity potential function <p. Lines of constant i/J are perpendicul to lines of constant <p and the governing differential equations are identical. Th equations are and

(10.1

respectively. The boundary conditions for i/J and <p, however, are not the sam and this difference leads to different calculated values. Both of these formulatio are discussed in this section.

10.1.1

Streamline Formulation

Li~es of consta?t i/J .are called streamlines. The volume flow rate, Qij, between an, pair of streamlines IS equal to the difference in their values,:

The assumption of an ideal fluid implies that the motion of the fluid does not penetrate into the surrounding body or separate from the surface of the body and leave empty spaces. These conditions imply that the component of the fluid velocity normal to the surface is equal to the velocity of the surface in the same direction (Duncan et aI., 1970). The above implies no flow perpendicular to a fixed boundary and, therefore, no velocity perpendicular to the boundary. Fixed boundaries as well as an axis of symmetry parallel to the flow are streamlines because there is no fluid velocity perpendicular to them. The boundary conditions for streamline flow are discussed relative to the problem illustrated in Figure 10.1. On the left boundary, Vx is a uniform 5 cm/sec. Since Vy=O, it is concluded from (10.3) that oi/J/ox=O on this edge. The same boundary condition applies to the right-hand vertical side because it is an axis of symmetry and the streamlines must be symmetrical about this edge. The horizontal axis of symmetry and the cylinder boundary form a streamline as well as the upper boundary. A zero value is assigned to the lower streamline. The upper streamline can have any nonzero value; an appropriate value is 30 because the flow rate is 30 cm3/sec for a unit thickness (one-half of the total flow). The four boundary conditions are shown in Figure 10.2. 10.1.2 Potential Formulation in the potential Vx=formulation Vy=are related to <p by (10.4)

Qij=i/Ji-i/Jj
There is no flow perpendicular to a streamline. obtained from the calculated values of i/J using and The velocity components

(10.2j

at1
j
(10,1

The velocity components

v. - - iN yox

o<p

ox

and

o<p

oy

128

FIELD PROBlE.

IRROTATIONAl

FLOW

129

Relating these to the flow around the cylinder (Figure 10.1) gives o</J/ox = 5 on tl left-hand edge and ocp/oy =0 along the horizontal axis of symmetry and the upp boundary. Since the velocity normal to the cylinder is zero, ocp/on =0 along ~ cylinder. We draw upon our knowledge that potential lines are perpendicular streamlines to establish the last boundary condition. The right-hand edge must be potential line because it is an axis of symmetry and all the streamlines are pc pendicular to it. It is assigned an arbitrary value of 50. The value does not influeol the results, since the velocity components are related to the gradient values. 11 boundary conditions are shown in Figure 10.3. The numerical value of the flux boundary condition along the left edge is dete mined by comparing the actual boundary condition with the theoretical bounda1 condition. The actual boundary condition is

side of the shape until a uniform velocity is again attained (Figure 10.4). At this point, constant potential lines are perpendicular to the flow and cp can be given an arbitrary value.

10.2 GROUNDWATER FLOW

There are two important groundwater The first is the seepage of groundwater problems governed by the field equation. under dams. The governing equation is

Dx ox2 +Dy

02cp

02cp oy2

=0

(10.7)

ocp =5
Ox
and the theoretical equation (see Figure 9.2) is

(10;

-Dx-=S

ocp
Ox

or

-=-S

acp
Ox

(10.1

when D'; = 1. Equating the second of (l0.6) with (l0.5) gives S = - 5. The boundary condition cp = 50 on the right edge of Figure 10.3 is a specil case. When the body is irregular in shape, the grid must include the downstreai
iJ.P. =0
ay

where D'; and D, are the coefficients of permeability (m/day) and cp is the piezometric head, in meters, measured from the bottom of a confined aquifer. The boundary conditions generally consist of known values of cp beneath the water, and a zero seepage condition on the other boundaries (Figure 10.5). A impermeable vertical wall beneath the dam is modeled by using a narrow gap for the wall. The finite element method automatically enforces the impermeable boundary condition, ocp/on =0, on each side of this gap, when no other conditions are specified. The second groundwater problem is the calculation of the drawdown at a well that is removing water from an aquifer. The governing equation for a confined aquifer is D';

a2cp 02cp ox2 +Dy oy2 +Q=O

(l0.8)

<1>= 50

=5
a¢

an
a¢

=0

where Dx, Dy, and cp are the same as defined for (10.7). The Q term in (l0.8) represents a point sink, the well, and should be evaluated using the concepts discussed in Chapter 9. The best results occur when the well is located at a node. The boundary conditions associated with (l0.8) consist of known values on all

ay

=0
for potential flow around a CYlind~
<I>=b

Figure

10.3.

Boundary

conditions
a¢

GO
- =0 ay
'!:!!_ =0 ay

<I>

= Constant

Figure 10.4.

Boundary conditions for potential flow around an irregular shap~

Figure 10.5.

Boundary conditions for groundwater seepage under a dam.

130

FIELD PROBLE~

IRROTATIONAL

FLOW

131

or a part of the boundary and/or the seepage of water into the aquifer along t~ boundary. Seepage is described by the derivative boundary condition,~

-(Dx

~!

cosO+Dy ~~ sinO)=s
are calculated using Darcy's law

(10.~

The fluid velocity components

'I
!.

The length of 8.25 em from the cylinder represents a guess as to where a uniform locity exists on the upstream side of the cylinder. yeThe nodal coordinates and calculated values of cP and t/J are given in Table 10.1. Contour lines for various values of cP and t/J are shown in Figure 10-7. The cP =0 contour line is not vertical so that a uniform velocity, Vx, does not exist along the left edge.

(10.10]

Coordinate Data and Results Table 10.1 for Irrotational Flow Around a Cylinder Node

X
0.0 0.0 0.0 0.0 0.0 2.7 2.7 2.7 2.7 2.7 4.5 4.5 4.5 4.5 4.5 6.0 6.0 6.0 6.0 6.0 6.9 7.2 7.8 6.9 7.5 7.2 8.25 8.42 8.91 10.64 10.0 8.7 8.4 10.5 10.5 10.5 10.5

y 0.0 1.5 3.0 4.5 6.0 0.0 1.5 3.0 4.5 6.0 0.0 1.5 3.0 4.5 6.0 0.0 1.5 3.0 4.5 6.0 0.0 1.2 2.7 3.6 4.5 6.0 0.0 0.86 1.59 2.08 3.3 4.5 6.0 2.25 3.6 4.8 6.0

rf> -10.33 -10.30 -10.24 -10.19 -10.16 4.04 4.10 4.26 4.41 4.47 10.6 10.8 10.3 10.7 10.8 110.3 110.8 21.0 21.7 22.0 22.9 24.8 30.1 26.1 30.2 28.8 26.5 28.3 33.3 41.1 38.5 37.6 36.1 50.0 50.0 50.0 50.0

i/J
0.0 7.43 14.9 22.5 30.0 0.0 7.32 110.8 22.3 30.0 0.0 7.01 14.3 22.0 30.0 0.0 6.25 10.5 21.5 30.0 0.0 0.0 10.66 15.9 20.8 30.0 0.0 0.0 0.0 0.0 10.3 110.9 30.0 0.0 10.3 21.5 30.0

10.3

COMPUTER EXAMPLES

The computer solutions for two problems using the steady-state field program TDFI ELD given in Chapter 16 are discussed here. The first problem involvel ideal flow around a cylinder as shown in Figure 10.1. The second involves th4 pumping of water from a confined aquifer.

10.3.1

Flow Around ACyl inder

A grid consisting of rectangular and triangular elements for the configuration it1 Figure 10.2 is shown in Figure 10.6. The grid consists of 38 elements and 37 nodes.

6em 3~----------~-------+----~~~

~--------------8.25 Figure 10.6.

em -------------~

Finite element grid for irrotational flow around a cylinder.

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37

36

Figure 10.7. a cylinder.

cJ>=10 cJ>=25 :1 Streamlined and constant potential lines for irrotational flow arourJ4

10.3.2

Reg ional Aqu ifer

Figure 10.9.

Finite element grid for the regional aquifer.

A small regional aquifer (Figure 10.8) has a single pump removing water. The; upper and lower edges are impermeable, and the left and right edges are far enough' from the pump so that a constant head of 200 m is maintained. The pump has a' 3 capacity of 1500 m /day. The permeabilities, D'; and D; are 15 m/day. The grid used to solve the problem is shown in Figure 10.9. The nodal values and, contour lines for ¢ are given in Figure 10.10. The nodal values of X and Y as well: as the calculated values of ¢ are given in Table 10.2. The maximum drawdown

200

179.8
I

I I I

169,3 ·165.1

168.7 168.7·
I

3000 m

I< I'E~---

3000 m-----i)o""l1
Impermeable

179.4

I /
I

179.5

o=
200

180 •

/(/>=170
I

:=200m 2000~
pump·T

177 9,'. '1165.7

I

160 cJ>= "'1'5'-7-\ / 9, \.160.8

...... ... _

.168,3
I

,,
I

,
200

.~200m

-1
2000 m

/,'
" /

"

cJ> 150 =
I , -159.8 // "

," cJ>=180
I .183.5
I

,
I

: 1643.':.

I'
,

\
\

"
-,

124.9
~I

200

:.178
I
I I

\\ • -,..2¥ ·-~:,·/·162
\166.1

/1

.1'69.9
I

I I

1500 m

L~~~~~~~~~~~
~

t-</-E;;-.....-----Figure 10.8.

Impermeable

5000 m------~)o;;..l

_1.200

,
\.179.6
I
I

,,

160.4

.184.3

200

,, ,
\ \ \

• 166.4
/
/ /

/ ·171.7 172.5
/

,,
I
I I

\ 180,2

169 /

185.5

200

A pump in a regional aquifer.

Figure 10.10.

Nodal values and contour lines for the regional aquifer.

136 134
Table 10.2 Coordinate Data and Results for the Regional Aquifer Problem Node I 2 3 4 5 6 7 8 9 10 II 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 X 0 0 0 0 1000 1000 1000 1000 1000 1000 1500 1500 1500 2000 2000 2000 2000 2000 2000 2000 2500 2500 2500 3000 3000 3000 3000 3000 3000 4000 4000 4000 4000 5000 5000 5000 Y 0 1000 2000 3000 0 500 1000 2000 2500 3000 1000 1500 2000 0 500 1000 1500 2000 2500 3000 1000 1500 2000 0 500 1000 2000 2500 3000 0 1000 1500 2500 0 1000 2000 rf> 200.0 200.0 200.0 200.0 180.2 179.6 178.0 177.9 179.4 179.8 166.1 164.3 165.7 169.0 166.4 160.4 124.9 159.7 165.1 169.3 162.0 159.8 160.8 172.5 171.7 169.9 168.3 168.7 168.7 185.5 184.3 183.5 179.5 200.0 200.0 200.0 15 em 40 em/sec
FIELD PROBLe4 IRROTATIONAL FLOW

10.1

Determine and plot the location of several streamlines and equal potential lines for the flow of an ideal fluid about the elliptical shape shown in Figure PI0.I.

L
Figure P10.1

30'Cm

10.2

Do Problem 10.1 for the rectangular shape shown in Figure PI0.2.

12 em

50 em/sec

1_
Figure P10.2

24 em

10.3

Do Problem

ml

for the junction region shown in Figure PIO.3.

32 em

occurs at the pump as expected, but this occurs only because the pump is at a node. The greatest drawdown occurs at the node closest to the pump when the pump is not located at a node.
PROBLEMS

407nVsee\

---JFigure P10.3

12 em

__l

Use the computer program TDFIELD discussed in Chapter 16 to analyze thei following problems. I

136
10.4 Do Problem 10.1 for the triangular

FIELD PROBL~

IRROTATIONAL

FLOW

137

region shown in Figure PIO.4.

1m

15 em 50 em/see

1~.•..••...... .....•...•.....•...•...•...•.•....•.....•...•.•.. .....•..•.•......•.•.•... · ·.· . .•.•. . ·•.•.•. ~J

Jem 30em m

L20<m-J

Impermeable

Figure P10.4 10.5 Determine the drawdown at the pump in an infinite media when the pu is removing 2500 m 3/day, D'; = D, = 20 m/day and ¢ = 300 m at a dista far removed from the pump. Two pumps are located in the confined aquifer shown in Figure PIO. Calculate the drawdown at each pump and plot some of the contour lin The pumping rates are 600 m 3/day for pump I and 400 m 3/day for pump The permeabilities are D'; =Dy = 30 m/day. The region is impermeable two sides, and ¢ is maintained at 100 m on the other two sides.
Impermeable

Figure P10.7 10.8 Repeat Problem 10.7 for the region shown in Figure PIO.8. The permeabilities are Dx = 20 m/day and D; = 15 m/day. The lower layer is impermeable. An impermeable wall 30 m deep has been placed on the upstream side of the dam.

10.6

Pump 2 Q = -400 m3/day (2000, 2000)

.:l00m

D, : D, : 30 mid"

Pump 1 Q: -600m' /d,.

1 J
3000m

30m

Figure P10.8

~~~=~~10~0~m~~~~~~~(3~0~0~O~,5~0~0~)~~~~~~~

I
10.7

Impermeable

~<~-------------5000m--------------~~~

Figure P10.6 Determine the location and plot some of the equal potential lines for t~1 groundwater seepage problem shown in Figure PlO.7. The bottom layer" impermeable. The permeabilities are Dx=Dy=20 m/day. .

HEATTRANSFER BY CONDUCTION

AND CONVECTION

139

Chapter

11

The governing differential equation d2¢

(11.1) has the general form (11.4) to the Galerkin

D--G¢+Q=O dx2
where D=kA, G=hP, and Q=hP¢f. residual equation {R(e)} is The element contribution

HEAT TRANSFER BY CONDUCTION AND CONVECTION

One of the first applications of the finite element method to nonstructural probl was in the area of heat transfer by conduction and convection (Visser, 1965). solution of heat transfer problems using the finite element method is particul popular with individuals responsible for analyzing thermal stress problems.' solution of the heat transfer problem becomes input to the stress analysis probl and the same grid can be used to solve both problems. We shall discuss the solution of four different heat transfer problems in t chapter. Two of these problems involve heat transfer from a fin. The third is analysis of a composite wall, and the fourth is the classical two-dimensio problem with convection boundary conditions. 11.1 THE ONE-DIMENSIONAL equation FIN for steady-state heat transfer from a
0

{R(e)} = _

=-

r [NY (D d2~ -G¢+Q) dx Jx; dx r (d ¢ ) r Jx; [NY D dx2+Q dx+ Jx; G[NY¢dx
Xi Xi 2 Xi

(11.5)

The first integral in (11.5) is the one we studied in Chapters 3 and 4 and is equivalent to {I(e)}+ [k(e)]{<D}_{f(el}. The second integral in (11.5) is a new quantity that needs to be evaluated. Recalling that ¢(e) = [N]{ <D(e)} and substituting this into the integral, we find that

f:

G[NY dX=(J:i

G[NY[N]dX)

{<D(e)}

(11.6)

Since the integral multiplies we define

{<D(e)}, it is a part of the element stiffness matrix. If

[k~)]=
then

J:i
,

G[NY[N]dx

(11.7)

{R(e)}

= {I(e)}

+ ([k~)] + [kl;)]){ <D(e)} - {fJe)}

(11.8)

The governing differential dimensional fin is

where [k~)] is given by (4.11) and {fJe)} by (4.12). The integral in (11. 7) is most easily evaluated in either the s or f b f 2 coordinate systems discussed in Chapter 6. It is left as an exercise for the reader to show that

(11., ., where k is the thermal conductivity, h is the convection coefficient, A is the cro sectional area, P is the distance around the fin, and ¢ is the temperature. temperature has a single value for all points on the cross section for a particu value of x. The boundary conditions associated with (11.1) are usually a specifi temperature at x =0 ¢(O)=¢o and convection heat loss at the free end +k A dx =hA(¢b-¢f)

~J

(11.9) into the

The derivative boundary condition defined by (11.3) is incorporated formulation using the interelement vector, {I(e)}. This vector is, (4.10),

(11.
and can be split into

"f

l(en = {

d¢ -D- I dx x=X]

D ~~

IX=X;J

(11.10)

d¢

atx=H

(11.

where ¢b is the temperature at the end of the fin and is not known prior to solution of the problem. The convection coefficient in (11.3) mayor may not the same as the one in (11.1).

{I(e)}={D~~lx=x;}+

{_Dd~1 } dx x=Xi

(11.11)

140
which is equivalent to

HEAT TRANSFER BY CONDUCTION

AND CONVECTION

141

{I(e)} = {Ile)} + {Ibe)}

where {Ile)} is the interelement requirement and {I~e)} is associated with boundary condition. The nonzero term in {I~e)}, however, is the left-hand si (11.3). Thus

~C_~_
f-oElc(c---~(1) 8em'---~~

lem

l~ I

(11.
since <Ph is the same as I1>j. Equation (11.12) is equivalent to

r--

4em

---1

{Ib )}
where
[(e)]

=[~
0

(2)
2 3

(3)
4

(4)
5

h~J{:J-{hA°<pJ { l1>(e)} - {f~e)}

= [kW]

(11,1
.~

1
J

Figure 11.1.

A rectangular fin.

!~

kM

[0

hA'

0J

r ds(e)}

hA<pf

} for

where the third matrix in [k(e)] and the second vector in {fIe)} are applicable for element four. The values of the various parameters are

only

(11~1 {I(e)} in
(1..1·· ..·

kA=3(4)=6
L 66°C 2

The complete and is

residual equation

is obtained

by substituting

W °C 0.333 W W 0C 20 W W

.••

hPL = 0.1(10)2

{R(e) = {I)e)} + ([k~)] + [k~)] + [kW]){ l1>(e)} - {fJe)} - {f~e)}

Neglecting the interelement requirement

(1L
.~

{Ile)} gives

hA =0.1(4)=0.400 hPL<p f = 0.1(10)(20)(2)

2 2

(lq The contribution of [kW] to [k(e)] occurs only for the last element of the fin ~ only when h is nonzero for the end of the fin. For example, [kW] is zero if the.,. of the fin is insulated. I' The element quantities ILLUSTRATIVE EXAMPLE
,~ [ (e)]

hA<Pf=0.1(4)(20)=8
are -5.667J 6.666

=[

-5.667

6.666

'

(e)}

={20}

20

Calculate the temperature distribution in a one-dimensional fin with the Phyi ... · properties given in Figure 11.1. The fin is rectangular in shape, and is 8 em 1o, 4 em wide, and 1 em thick. Assume that convection heat loss occurs from , end of the fin. . The fin is modeled matrices are by four elements each with a length of 2 em. The eleme

for elements one, two, and three and

[ (e)]

=[

-5.667

6.666

-5.667J 7.066

'

r _{20} d (enf- 28

-IJ
1
r l

hPL[2
6

and

h~J
hA<Pf
0 }

for element four. The assembly of the element matrices using the direct stiffness procedure produces the system of equations 6.666 - 5.667 0 0 0 - 5.667 13.33 - 5.667 0 0 0 - 5.667 13.33 - 5.667 0 0 0 -5.667 13.33 - 5.667 0 0 0 - 5.667 7.006 11>1 11>2 11>3 11>4 11>5 20 40 40 40 28

pe)}=hPL<Pf{I}+{
2

142

FIELD PROBLEJ

HEAT TRANSFER

BY CONDUCTION

AND CONVECTION

143

The temperature at node one is known, o!>, ~ 80'C; thus the first equation m~ be deleted and the others modified. The new system of equations iSl 13.33 -5.667 -5.667 13.33 -5.667

ILLUSTRATIVE

EXAMPLE

o
-5.667 13.33 -5.667 - ~.667 7.066

Determine the temperature distribution through the composite wall shown in Figure 11.2 when convection heat loss occurs on the left surface. Assume a unit area. The parameters for element one are kA = 0.2(1) =01 W L 2 . °C hA =0.1(1)=0.1 hA¢f=0.1(1)(-5)=

r
obtained

o o

]{<1>211493} :: = :~ <1>5 28

The nodal values are

{<1>}T =[80
which compares very favorably

53.9

39.9

32.8

30.3]

0C
-0.5 W

with the theoretical 54.3 40.2

values of 30.6] and

{<1>th}T =[80
using equation

33.2

2-47 in (Kreith, 1973).

~ 'I; ~

[k(1)]
The parameter

=[

0.2 -0.1

-O.1J 0.1 '

11.2

THE COMPOSITE WALLi

differential equation

for element two is kA = 0.06(1 LO.Ol W L 6 °C

The governing

for heat transfer through a composite wall ~ kA

d2¢
-2

dx<

=0

(11.17~

~
and

where either ¢ is known at one or both surfaces or convection heat loss occu~ from one or both surfaces. The convection boundary conditions are] kA ~~ =hA(¢b-¢f) and atx=H The element stiffness matrix is given by [k(e)]=k:[ = [k~)] _~ at x=O (11.18

0.01 [ (2)]=[ k -0.01

-O.OlJ 0.01 gives

1
~
'J I

Assembly of the element matrices 0.20 [ -0.10 -0.10 0.21 -0.01

using the direct stiffness procedure

<~

(11.19~

o -0.01 ]{<1>1} {-0.5} <1>2= 0

0.01 <1>3 0

-~J+[h~i~J+[~ h~Jj
(11.21
<Pf= -5°C

+ [k~.J + [k~J

where the second matrix, [k~.J, results from the convection boundary condition at node i and the third matrix, [k~J, results from the convection boundary co dition at node j and is the same as given in (11.14). The element force vector is I Ute)} ={hAi¢f

}+{

k = 0.06 cmWoc

0} hAj¢f

where the first vector comes from (11.18) and the second vector from (11.19 The quantities, Ute)}, [K~.J, and [K~j] are neglected when temperature' specified at both surfaces.

•• 12
Figure 11.2.

(1)

(2)

• 3
wall.

A composite

144 Since <1>3 = 20°C, equation The final system is

FIELD PROBl,

HEAT TRANSFER BY CONDUCTION

AND CONVECTION

145

three is deleted and the second equation - 0.1 {<I> 0.21 <I>2

is modifi~

0.20 [ - 0.1 0 and the nodal values are

OJ

I}

1
,1
:1
;1

and a¢ a¢ kxt - cos 8+kyt -a sin 8=0

ax y

={ -

0.50} 0.20

(11.24 )

{<I>} = [ - 2.66

-0.31

20J

along the outer edge. The latter condition is the insulated boundary condition. The heat transfer in the fin is a two-dimensional problem because the fin is too thin to develop a temperature gradient in the z-direction. The governing equation (11.22) has a form identical to (7.1) where Dx=tkx, Dy=tky, G=2h, Q=2h¢f (11.25)

The calculated values are also the theoretical values because the solution of t~ differential equation consists of straight-line segments. 11.3 THE TWO-DIMENSIONAL FIN

The element matrices developed in Chapter 7 are applicable without modification. The element matrices for an element in a fin were calculated in the Illustrative Example in Section 7.3 and should be reviewed at this time. 11.4 LONG TWO-DIMENSIONAL BODIES

A two-dimensional fin is a thin piece of metal attached to a hot water or st~ pipe as shown in Figure 11.3. Heat is transferred from the pipe to the fin by co~ duction and from the fin to the surrounding media (usually air) by convectloi Convection heat loss occurs from both faces of the fin and the edge. The surf a.·. area of the edge, however, is small compared to the area of the two faces and '. convection losses from the edge can be neglected..! The governing differential equation for the two-dimensional fin is

1
~ ~

Another form of two-dimensional heat transfer is that which occurs in long bodies of constant cross section subjected to the same boundary conditions along the entire length of the body. In this situation, the temperature gradient in the zdirection is zero and the governing differential equation is

(11.~ where k; and k; are the thermal conductivities in the x- and y-directions, res~ tively, t is the thickness of the fin, h is the convection coefficient, and ¢f is t~ temperature of the surrounding fluid. The boundary conditions are along the pipe boundary

a2¢ a2¢ k; ax2 +ky ay2

+ Q =0

(11.26)

(l1.2l,

where k, and ky are the thermal conductivities and Q is an internal heat source or sink. The internal source or sink must exist along the entire z-direction for the heat transfer to be two-dimensional. A two-dimensional problem exists because the temperature can vary in the xy plane. The differential equation (11.26) is embedded within (7.1). The parameters for (7.1)are Q=Q 01.27)

","" '~'\

"' ....

,~

,... -,\
I

\''')

...............

'd>(I') SP:cified

and the element matrices developed in Chapter 7 are applicable. The item that makes heat transfer in a long body different from that in a fin is the boundary conditions. The possible boundary conditions are shown in Figure 11.4 and consist of prescribed temperature values, convection heat transfer, and surface heat fluxes. The latter two conditions were treated in general terms in Chapter 9. We shall now look at these conditions relative to heat transfer and assume that k, =k ; =k, which changes the derivative boundary condition (9.1) to (11.28 )

k,1 Figure 11.3.

;t cos 0+ k,1 ~

,;00 ~ 0

A two-dimensional

fin and boundary conditions.

11.4.1

Convect ion Boundary Cond it ion

The objective here is to determine the parameters that comprise M and S in (11.28) as well as the signs on these two quantities for the convection boundary

146
Prescribed temperatures

FIELD PROB~

HEATTRANSFER BY CONDUCTION

AND CONVECTION

147

Surface heat flux

(a)

Convection heat transfer

Figure 11.4.

Types of boundary conditions

for the classical heat transfer probl

condition. We shall consider two situations: heat leaving the body and entering the body. Consider the convection condition shown in Figure 11.5a, where heat is lea the body. If heat is leaving, then the gradient o¢/ox must be negative and heat conducted to the surface is

The heat conducted

to the surface must equal that leaving the surface; theref -kA On =hA(¢b-¢f)

o¢

(11:-,
(b)

Rewriting (11.30) in the form of (11.28) gives k-= and we conclude that

Figure 11.5.

Convection

heat transfer at a surface.

o¢ on

-h¢b+h¢f and

d
M =h and S=h¢f (l1.~ the bO~ " .:jj We again conclude that and

(11.35)

now consider the convection situation when heat is entering (Figure 11.5b). The gradient o¢/on is positive; therefore,

'!'Ie

o¢ qc=kA on
A negative sign. is not needed because the gradient to the heat commg to the surface gives is positive. Equating

(11.36)

(l1.3j,

The conclusion of the analysis is that M = hand S = h¢f regardless of whether heat is coming into or leaving a body during the process of convection heat transfer. These results hold for exterior surfaces and interior surfaces such as those in a chimney. The above analysis assumes that a positive outward normal, n, is always directed away from the surface.

148
11.4.2 Heat Flux into The Body

FIELD PROBLE~
l

HEAT TRANSFER

BY CONDUCTION

AND CONVECTION

149

The situation where heat is applied to a portion of the boundary is sho~ schematically in Figure 11.6. The heat must be conducted away from the boundal'l Thus qc=kA

/Heatin.g

cables,

•
Region to be analyzed

•
6

t- --1
4cm

o¢

on

(11.31
Insulation

The heat applied to the surface is q*A and equating heat flows gives kA Equating (11.38) and (11.28) yields M=O and S=q*

o¢ -=q*A
On

Figure 11.7.

Heating cables within a conducting media.

(11.31

(IUS

where q* is the heat flux per unit area. If heat were being removed from the bo~ then S= -q*. The conclusion of this analysis is that a heat flux on the boundary does havel sign associated with it. Heat is positive if it is moving into the body and negath if it is being removed. The coefficient M is zero when a heat flux occurs.

exposed surface (Figure 11.7). The cables are 4 em on center and 2 em below the surface. The slab rests on a thick layer of insulation and the heat loss from the bottom can be neglected. The conductives are k; =ky =0.0180 Wjcrn - °C and the surface convection coefficient is h =0.00340 Wjcm2 - "C. The latter corresponds to about a 30-35 krn/hr wind velocity. The objective is to determine the surface
2 (1) 6 (5) 11 (9) 16 (13) 21 (17) 26 (21) 31 (25) 27 (22) 32 (26) 37 30) 17 (14) 22 (18) 28 (23) 33 (27) 38 7 (6) 12 (10) 18 (15) 23 (19) 29 (24) 34 (28) 39 13 (11) 19 (16) 24 (20) 30 25 (2) 8 (7) 14 (12) 20

3
(3)

4
(4) 9 (8)

5
10

15

35

Figure 11.6.

A heat flux applied to a surface.

36

11.4.3

Conclud ing Remarks

~~ 1

W~
2 33) (36) 45 (38) 48

40

43

Convection heat transfer and a flux boundary condition are incorporated int~ finite element analysis using the element matrices developed in Chapter 9. ' the case of convection heat transfer, M =h and S=h¢ f whereas M =0 and S= ±. for the flux condition. Heat into the body is considered positive for both q* and 11.5 A COMPUTER EXAMPLE

(35) 44

46

(37) 47

We close this chapter with a computer example. A grid of heating cables have em bedded in a thin concrete slab for the purpose of melting the snow on

49

Figure 11.8.

The finite element grid for the heat transfer example.

150
temperature is -5°C.

FIELD PROBlE

HEAT TRANSFER

BY CONDUCTION

AND CONVECTION

151

when the cables produce 0.050 W fcm of heat and the air temperat

Table 11.1 Node X

Coordinate Data and Results for the Heat Transfer Example

y

The repeated symmetry in the problem reduces the region to be analyzed to 0 which is 2 em wide and 6 em deep (Figure 11.7). The grid for this region (Figui 11.8) consists of 38 elements arranged so that node 21 is located at a heating cab Larger elements are used in the lower portion of the grid because previous e, perience indicates that the temperature probably does not vary a great deal the, Elements (1), (2), (3), and (4) have the convection boundary condition. The valu~ of M and S in each case are .' M=h=0.OO340 and S=h¢ f=(0.OO340X - 5)= -0.0170.

rP
2.4 2.4 2.4 2.3 2.3 3.1 3.1 3.0 3.0 3.0 3.9 3.9 3.7 3.6 3.6 4.9 4.8 4.3 4.2 4.1 7.3 5.3 4.8 4.6 4.5

Node

rP
5.6 5.5 5.0 4.9 4.8 5.3 5.3 5.1 5.0 5.0 5.2 5.2 5.1 5.1 5.1 5.2 5.1 5.1 5.1 5.1 5.1 5.1 5.1 5.1

2.4

2.4

2.4 3°e

2.3

2.3

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25

0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0

6.0 6.0 6.0 6.0 6.0 5.5 5.5 5.5 5.5 5.0 5.0 5.0 5.0 5.0 4.5 4.5 4.5 4.5 4.5 4.0 4.0 4.0 4.0 4.0

s.s

26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49

0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 0.5 1.0 1.5 2.0 0.0 1.0 2.0 0.0 1.0 2.0 0.0 1.0 2.0

3.5 3.5 3.5 3.5 3.5 3.0 3.0 3.0 3.0 3.0 2.5 2.5 2.5 2.5 2.5 2.0 2.0 2.0 1.0 1.0 1.0 0.0 0.0 0.0

The nodal coordinates and calculated temperature values (to two significant digits) are given in Table 11.1. Some contour lines and nodal temperature values are shown in Figure 11.9. The top surface is close to a uniform value of 2.4°C. The lower part of the region is a uniform 5.1 °C justifying the use of large elements.

PROBLEMS 11.1 11.2

Verify that [k~)] as defined by (11.7) is equal to (11.9). Evaluate [k~] and u~e)} for the derivative boundary condition Dd<P =hA(<Pb-<Pf) dx
5.1 5.1

at

x=O

11.3

Determine the temperature distribution grid shown in Figure Pl1.3.

(1) 2 (2)

for the fin in Figure 11.1 using the

(4)
4

(3) 3

Figure 11.9.

Temperature contour lines and some nodal values for the heat transfer example.
Figure P11.3

3cm---1

152
11.4

FIELD PROBLEfIl

HEAT TRANSFER

BY CONDUCTION

AND CONVECTION

153

Determine the temperature distribution in the circular fin shown in Fig~ PIIA. Include the convection heat loss from the end of the fin.
k

=2

w'e' ern-

= 0.2 em2.'e'

50oc~f-------_____.1

OT
5
I

11.7 Calculate the surface temperature at x = 4 in Problem 11.6 when the surface temperature on the left side is specified at 11.8 Calculate the surface temperature at x=O in Problem 11.6 when the surface temperature on the right-hand side is specified at 30°C. 11.9-11.12 Evaluate [k(e)] and {f(e)} for the triangular elements shown in Figures Pl1.9 through Pl1.12. The conductivities arekx=ky=2W;oC-cm

soc.

I...
(1)

6cm~.~

Q* = 5 '!!_ em

(2)

(3)

(4)

Figure P11.4

11.5

Determine the temperature distribution in the circular fin of Problem nl using the three-element grid shown in Figure Pl1.S. Include the convectlq heat loss from the end of the fin.
(1) (2) (3)

L.
y

1(10,1) q,f= 20'e

Figure P11.9

4
Q*= 1O~ em

Figure P11.S
T

11.6

Calculate the surface temperatures for the wall shown in Figure Pl1.~ Convection heat transfer occurs on both surfaces. Assume a unit of surfac area.
~,~

L.
y

(4,2)

Figure P11.10

h = 1.5 em· ~

'e

h=0.05

~e em·
q,f= 15'C
y

L.
• (1) •

(-2, -2)

Figure P11.6

Figure P11.11

, 154
FIELD PROBL~

'i .,

-:
/
J

(5,1)

Figure P11.12

and h =0.2 W/cm2_ e. The heat sources, Q* in Problems 11.9 and 11., are line sources. . .. 11.13-11.14 Use the computer program TDFIELD, Chapter 16, to analyze Oi of the two-dimensional fins shown in Figures P1Ll3 and P1Ll4. Tl physical parameters arekx=ky=3 W/cm_oC, h=O.OlO W/cm2_oC,and <PI: 20°e. Each fin is 0.3 em thick. The pipe temperature is 8S°e. 11.15-11.19 Use the computer program TDFIELD, Chapter 16, to analyze O! of the long two-dimensional bodies shown in Figures PILlS to P1U The physical parameters and the boundary conditions are defined in d respective figures.
0

Figure P11.14

1<

~20+-60---\ 5
Insulated

10 1.5em 5 q, = 50°C
kx

q, = 200°C

= ky =

3 cm_OC

80em

1<

10em Figure P11.13

>1

160em Figure P11.15

em·'f'c <Pr= 20'e h=1.5

h = 0.3 cm2.,C <Pr=

ro-c

All exterior surfaces

kx = ky=

L,
y

2 em~C

Boundary of the inner cylinder is at 140 C Diameter of the inner cylinder, 2 em Diameter of the outer cylinder, 8 em

Figure P11.16 Figure P11.18

Insulated

= 03

'f'c' crrr-

<Pr= 25ce

--;r--

4cm

t
~
foo!<E----------30 cm

l
<p = 60ce

50cm

----~~~I
W --em-DC

Material!

kx=k}'=O.! 2

Material

kx=k,.=4

W --em-DC

~~J
'--

<p

IO'C

50

Insulate~1 100 cm---------;~~·

Figure P11.17

Figure P11.19

ACOUSTICAL VIBRATIONS

159 2

Chapter

12
VIBRATIONS
field equatloi

The element stiffness matrix for the first term, d cjJ/dx . . WI. .b (411) The contribution of the - GcjJ term to the element stiffness matnx isglven Y . . . f (125)· is given by (11.9). The complete element stiffness matnx or . IS -·thD- 1 an d G -

W2/C2

ACOUSTICAL

[k(e)]=~[

_~

-~J-:~;e ~J
.. III

(12.6)

The element force vector, {le)}, consists of zeros, since there is no source term in 12.5) and the dcjJ/dx=O boundary conditions (both ends) do not generate any

A zero value for Q and a negative value for G in the steady-state (7.1), yields the Helmholtz equation

nonzero terms III . . The problem under consid~ration is the closed pipe element model consists of two elements. Substituting multiplying through by H/2 gives

. {f(e)}

Th fi . Figure 12:1. e nite H/2 for L III (12.6) and

02cjJ 02cjJ D; ox2 + D, oy2 + GcjJ =0

(12.1
where

[k(e)]=[_~

-~J-ze ~J

(12.7)

Physical problems governed by (12.1) include the wave motion for shallow bod~ of water and acoustical vibrations in closed rooms or compartments. The solution of the Helmholtz equation results in the need to solve an eigenvalu problem because the boundary conditions are such that the global force vecto {F}, is zero. The global system of equations has the form [K] {<I>} = {O}. ' The solution of the Helmholtz equation is the topic of discussion for this chapte Acoustical vibrations is the subject matter area used to illustrate the solutio! technique. We start by considering the one-dimensional problem because ~ calculations can be done by hand. The natural frequencies and vibration mode for a two-dimensional problem conclude the chapter.

(12.8)

(12.9)

or (12.10)

12.1

ONE-DIMENSIONAL

VIBRATIONS

The governing differential equation for the pressure field associated with acoustlcs vibrations in a two-dimensional room with rigid boundaries is

(12.2
where cjJ is the change in pressure from some ambient value, w is the wave fl'Ci quency, and c is the wave velocity in the media. The condition to be satisfied 01 each boundary is

[KG] is positive definite whereas Eigenvalue the?~y states that all of the eigenvalues, Zi; that satisfy (12.9) are distinct, real, and positive numbers and the corresponding eigenvectors {tl>}; are independent. The eigenvalues, Zi, are the values of Z that make the determinant of (12.9)

[KD] is semidefinite (it has a zero determinant).

Both matrices, [KD] and [KG]'

are symmetric;

c£<E=o dx

l_

~=o

ocjJ=0

on

(12.3

~-----------H----------~~

The one-dimensional

analog to (12.2) is

d2cjJ w2 -d 2+2:cjJ=0 x c
with dcjJ/dx=O at each end. Equation (12.4) has the general form

(12.4

(1)

(2)

d2cjJ D--GcjJ=O dx2

(12.;'

Figure 12.1. A two-element grid for a one-dimensional pipe.

160 zero. Combining the two matrices gives (1-2Z) [ The determinant -(1 +Z) - (1 + Z) (2-4Z) -(1 +Z)

FIELD PROBlE,

ACOUSTICAL VIBRATIONS

161 frequencies, nne

The theoretical

values for the natural

w., are given by

(12.19)

(12.1

w·=H

is

The calculated values for w, are obtained by substituting the roots Z b Z 2, and Z 3 given by (12.13) into (12.8) and solving for w n- The calculated values of IV,

(12.1
which has the roots and (12.l~
I

IV1=0,

3.464e
W2=~'

and values of and

IV3=~

6.928e

compare fairly well with the theoretical

There is an eigenvector, {<1> , associated with each root in (12.13). It is impossib L to uniquely determine the three components of {<1>} because the set of equatioi i is homogeneous. The usual procedure is to assign an arbitrary value to one con ponent and solve for the remaining components in terms of the assigned value.' The eigenvector {<1>}1 is determined by substituting Z1 =0 into (12.11). n resulting equations are

IV1-, -0

W2 =--H

3.142e

considering the grid consisted of only two elements. The theoretical mode shapes have the general form P=cos(nnx/H). The theoretical mode shapes and the calculated eigenvectors are shown in Figure 12.2. The theoretical and calculated shapes for the first mode coincide.
"'(xl
---Calculated --- - - - Theoretical

<1>1- <1>2 =0 -<1>1 +2<1>2-<1>3=0 -<1>2+<1>3=0

The first equation says that <1>1 =<1>2 while the third equation thus <1>1 =<1>2 =<1>3 and the eigenvector is

(121;
states that <1>2 =cJ)1
,I

._

._ First mode

~~x

{<1>}f = [1
when <1>1 1 is used as the arbitrary value. = Substituting Z 2 into (12.11) gives

1]

(12.1!

",(x)

=i

3 - - <1>2 2 (12.1~ - - <1>2

2 The first and third equations Using <1>1 1 as the arbitrary = state that <1>2 =0 while the second yields <1>3 -cJ)1 = value, we obtain 3

",(x)

{<1>H =[1

o
-1

-1]
is

(12.11
Third mode

It is left for the reader to show that the third eigenvector

{<1>H =[1

1]

(l2J~

Figure 12.2. Pipe.

The theoretical and calculated node shapes for the one-dimensional

162 12.2 TWO-DIMENSIONAL VIBRATIONS

FIELD PROBLEIl

i
,

ACOUSTICAL VIBRATIONS

163
for (12.20) are 1, -1, 1, -1, -0.5, compare -1, 1, 1, -1, -1, -0.5, reasonably 1] 0] 0] 0] -0.5, 1] values (12.24)

The five eigenvalues and eigenvectors Zl::::0, Z2::::0.03O, Z3::::0.120, Z4::::0.150, Zs::::0.450,
2 ::::W2/C ,

The two-dimensional acoustical vibration equation and its boundary conditia were given in (12.2) and (12.3). The element matrices for this equation are th1 given by (7.36) and (7.38) for the triangular element, and (7.49) and (7.55) for ~ rectangular element. A rectangular room, 20 m by 10 m, is divided into four triangular elem~ (Figure 12.3). Defining Z = W2/C2, we find that the global system of equations i$
(

{<I>H=[I, {<I>}I [1, = {<I>H =[1, {<I>}r=[I, I{<I>H -0.5, =I which are

1, -1,

1,

([KD]-Z
where

[KG]){<I>} ={O}
-3 0 3 10 -10 25 0 25 100 50 50 50 50 50 200 3 -10 -10 -10 -10 40

(12J
'J

The calculated eigenvalues of Z

well with the theoretical

10 [KD] =g
1 -3 -10 100 -10 3 0 -3

3 10 0 -3

Z2 =0.0247, (12j
and Z5 =0.395 Each of the mode shapes can be illustrated graphically as was done for the onedimensional case. The mode shapes corresponding to {<I>} and {<I>} are shown in 2 4 Figure 12.4. (12.2
</>(x,y) Dashed lines are in the xy plane.

10 -10
0 25 100 25 50

25 100 25 0 50

[KG] =(;

25 0 25 50

and

{<I>}T =<1>1
Hand

<1>2

<1>3

<1>4

<1>5]

(12.2:
3

computation of the values of Z that makes the determinant 4 zero is unreasonable; a computer program should be use Discussions of direct and iterative methods of evaluating eigenvalues are given I many textbooks. Bathe and Wilson (1976) present a comprehensive discussion c eigenvalue calculations relative to finite element problems.

[KD]-Z[KG])

4~--31

Mode shape {<I>12 </>(x, y)

eo::::-------_~2J
Figure 12.3. A four-element gird for a two-dimensional room. Figure 12.4.

Mode shape {<I> 14

The second and fourth mode shapes for a two-dimensional

room.

164
PROBLEMS 12.1 12.2 12.3 12.4 12.5 12.6 12.7

FIELD PROBLE~

12.8

Calculate the eigenvalues and eigenvectors for the one-dimensional problej in Figure 12.1 using one element of length H.l Calculate the eigenvalues and eigenvectors for the one-dimensional problel in Figure 12.1 using three equal length elements. j Solve the vibration problem in Figure 12.1 using several elements and! computer program to evaluate the eigenvalues and eigenvectors. . Verify the mode shape {<I>h given in (12.18). Verify the mode shape {<I>L given in (12.24). Verify the mode shape {<I>h given in (12.24). The acoustic field in an automotive compartment is an important applies tion area for acoustical vibrations. Construct and analyze a grid simila to that given by Shuku and Ishihara (1973) using linear triangular 3.Ili 1 bilinear rectangular elements. The buckling load of a simply supported column (Figure PI2.8) is governs by the differential equation

Chapter 13 AXISYMMETRIC PROBLEMS

FIELD

d2y dx2

+ EI y=O

There is a group of three-dimensional field problems that can be solved using twodimensional elements. These problems possess symmetry about an axis of rotation and are known as axisymmetric problems. The boundary conditions as well as as the region geometry must be independent of the circumferential direction. The Galerkin formulation and the element equations are similar to those for two-dimensional field problems, but they do differ in some significant ways. The derivation and results differ enough from those in Chapters 7 and 9 to warrant the special consideration given in this chapter.

where P is the buckling load and EI is a section property. The boundar conditions are y(O) = y(H)=O. Evaluate the eigenvalues and eigenvector for the four-element grid shown in the figure. Compare the calculate critical loads with the theoretical values given by Pcr=n2n2EI/H2.

13.1

DIFFERENTIAL

EQUATION coordinate system (r, (J, z) is

The field equation

in a cylindrical

02¢ Dr o¢ De 02¢ 02¢ Dr-;:;-z+--;)+--y ::l(J2 +Dz~+Q=O Or r ur ru uz

An axisymmetric problem is independent of(J; thus (13.1) reduces to

(13.1 )

02¢ Dr o¢ Dr ;;-z+-;;-+Dz ur r ur
(1)

02¢ ;;-z+Q=O uz

(13.2)

(2)

(3)

(4)

which can also be written as

~[Dr~(r r or
Figure P12.8 12.9 Calculate the critical buckling loads for the column in Figure P12.8 usi~ (a) two equal-length elements and (b) three equal-length elements. . on a part of the boundary,

O¢)J+Dz or

~2~

oz

+Q=O

(13.3)

assuming that Dr is a constant. The boundary conditions associated

with (13.3) are (13.4 )

¢(r) = specified values call it

r band
z

Dr

o¢

a,: cos (J+D

o¢ OZ sin(J=-M¢b+S

(13.5)

on the rest of the boundary, r 2' Both of these conditions must be independent of the circumferential direction. Equation 13.5 is similar to (9.1) and the numerical values for M and S are determined similar to the examples in Chapters 10 and 11.

166
13.2 AXISYMMETRIC ELEMENTS

FIELD PROBLENI

AXISYMMETRIC

FIELD PROBLEMS

167
shape functions in the new coordinate system are
(13.6)

The variable ¢ and the triangular

The axisymmetric element is obtained by rotating a two-dimensional elem~ about the z-axis to obtain a torus. The idea is illustrated with the triangular elemCl1 in Figure 13.1. _ j~ A single triangular element in the r - z plane is shown in Figure 13.2. element is identical to the triangular element discussed in Chapter 5 except that ~ coordinate variables are rand z instead of x and y. The element shape functi~ are identical to those given in (5.7) through (5.9) with x and y replaced by rand j

where 1 Ni=2A (ai+bir+ciz) 1 N J.=- (a·+b·r+c·z) J 2A J J (13.7) where ai=RjZkaj=RkZiak=RiZjRkZj, RiZk, RjZi' bi=Zj-Zk> bj=Zk-Zj, bk=Zi-Zj, ci=Rkcj=Rick=Rj-Ri R,

TI.I

...---------

..........

R,

A single rectangular element in the r-z plane is shown in Figure 13.3. This element is similar to the one III Figure 5.10 and was discussed in Section 5.3. The shape functions, however, are different from those given in Section 5.3 because they must be referenced relative to the origin of the r-z coordinate system. All of the rectangular shape functions in Chapter 5 were written relative to local coordinate systems. Noting that r=Ri+s
Figure 13.1. z The axisymmetric triangular element.
z

and

(13.8)

j --j
Z
Figure 13.2. The axisymmetric triangular element in the r-z plane. Figure 13.3.

ZiT

r
The axisymmetric rectangular element in the r-z plane.

168
rearranging into

FIELD PROBL.

AXISYMMETRIC

FIELD PROBLEMS

169

Substitution

of (13.13) and (13.15) into (13.11) produces

s=r-Ri

and

t=Z-Zi

(1~ shape functions in the ~

{R(e)}=

and substituting for sand t in (5.17) gives the rectangular coordinate system

Jv

r (Dr o[NY Or

o<jJ Dz o[NY o<jJ-[NYQ)dV + Or oz oz oz

(13.16) using

N·=I

1 4ab} (R--r)(Z 1

-z) -z)

_ Jv (Dr~([NYro<jJ)+Dzj_([NyO<jJ))dV r r or Or oz
The second volume integral can be transformed Gauss's theorem (Olmstead, 1961). The result is

into a surface integral

N·=-(r-R-)(Z } 4ab Nk=

1 4ab 1

(r-R;)(z-Z;)

Ir (~r
Ii (13.1~
,~

([NYr

~~) cos + Dz[NY

~~ sin

0) dr

(13.17)

which Simplifies to (13.18) The complete residual integral is

Nm= 4ab (Rj-r)(z-Z;)

13.3

GALERKIN'S METHOD
integral for an axisymmetric field problem is the vol~

The weighted residual integral

i
f,i

(13.19) Since <jJ(e) [N]{ <I>(e)}, = placed by

(13.~
The derivative terms must be transformed into lower-order forms using the prod~ rule for differentiation and Gauss's theorem. The term involving D, reduces iIi: manner identical to (7.12). The product rule for differentiation gives

o<jJ/or and o<jJ/oz in the first integral of (13.19) can be reand (13.20)

giving

(13.1;
Rearranging gives

{R(e)}=(

Jv

r (Dr o[NY Or
dV

O[N]+Dz o[NY Or oz

O[N])dV) oz

f<l>(e))

(13.1:
The first term in (13.11) is replaced once it is determined that (13.1' Rearranging gives

- Iv Q[NY - Jr [NY

(Dr ~~ cosO+Dz

~~

sinO)dr

(13.21 )

[NY

(!~ O<jJ))=!~ Or ([NYr (r r Or Or r

The first integral in (13.21) multiplies {<I>(e)}; thus it is the element stiffness ~atrix. The integral containing Q becomes {f(e)}, while the surface integral is the IOterelement requirement for interior element boundaries and the derivative boundary conditions for element boundaries on I" 2. The general form of {R(e)} is (13.22)

O<jJ)_ o[NY o<jJ Or Or Or

(13Ji

170
where
{I(e)

FIELD PROBLEM!

AXISYMMETRIC

FIELD PROBLEMS

171

=Jv

[k(e)]=
and

r (Dr o[NY

Jr

r [NY
Or

(Dr 0<jJ cos or O[N]+Dz or

e + D,
o[NY oz

0<jJ sin

oz

e) dr

The volume of an area revolved about the z-axis is V =2nrA, where r is the radial distance to the centroid of the area. The element stiffness matrix is

(13.23 (13.2~
This matrix product

[k(e)] =2nrA[BY[D][B]
is easily evaluated to give

(13.31 )

O[N])dV oz

UJe)}

= Iv Q[NY dV

(13.2~

[k(e)] =4:,4 b.b, bibk

2 -s», -

[bf

bibj b~ J bjbk

b.b; ] b b;

bf

+-4A'

2ttr z [ -D

C,C· J c.c,

cf

c.c,
Cj

c.c, ] CjCk

(13.32)

CjCk

The terms in the integral for [k(e)] are similar to those in (7.20), and (13.2~ can be written in the compact form 'f

The radial distance to the centroid of a triangular

element is

[k(e)] =
where

Iv [BY[D][

_ Ri+Rj+Rk r 3
The element force vector UJe)} is

(13.33)

B] dV

(13.~

(13.34)
(13.~ since dV=2nrdA. The shape functions in (13.34) can be replaced ordinates, and the radial distance r can be written as by area co-

and

r=NiR+NjRj+NkRk
and the integral for UJe)} is

=LIR+L2Rj+L3Rk

(13.35)

13.4

ELEMENT MATRICES

UJe)}=2nQ

The immediate objective is to evaluate the volume integrals that give [k(e)] Ill! {pel}. The contribution of the derivative boundary condition to these integl'lll is discussed in the next section. The discussion is limited to the triangular eleme! because the integrals are relatively easy to evaluate. The integrals for the rectangt lar element are usually evaluated using numerical techniques. . The coefficients in [B] are obtained by differentiating the shape functio! relative to rand z, This produces a matrix identical to the one in (7.34)

,~

Ll(Ll Ri+ L2Rj+ L3Rd} L2(LIR+L2Rj+L3Rk) { L3(L1R+ L2Rj+ L3Rk)

products

dA

Evaluating the integrals of the area coordinate 1

using (6.29) produces

(13.36)

[B] =_1 [bi 2A c.

Each coefficient in [B] is a constant. constant coefficients'!l Since

b,
Cj

bkJ Ck

(13.2

[D], as given by (13.28), also consistSQ

ILLUSTRATIVE EXAMPLE

n
j(

A uniform Q within the element is not distributed equally among the nodes as occurred with the two-dimensional element. Each node receives an amount related to its radial distance from the origin.

[k(e)] =

Iv [BY[

D][ B] dV V

= [BY[ D][ B]

Iv dV

= [BY[D][B]

(13.~

An aXisymmetric triangular element with the nodal coordinates 13.4 has a uniform heat generation of Q = 3 W/cm3. Calculate WhenDr=Dz= 1.5 W/cm_°c.

given in Figure

[k(e)] and UJe)}

172
z

FIELD

PROB"

AXISYMMETRIC FIELD PROBLEMS

173

i>.
(22,10)

~15),

Since Q is a constant

within the element, the total heat generated

is

Iv Q dV =Q Iv dV =2nrAQ=1296n
The components of UJe)} equally among the nodes. 13.5

W is not distributed

Figure 13.4.

The triangular element for the example problem.

sum to this value, but the quantity

By using (13.7), the determinant b;= -4, c;=-2,

equation bj=5,

(5.6) and (13.33) gives bk=-1 THE DERIVATIVE BOUNDARY CONDITION The element matrices (13.32) and (13.36) for [k(e)] and UJe)} , respectively, are valid for interior elements and boundary elements when ¢(r, z) is specified on the boundary. When the derivative boundary condition (13.5) is specified, there are additional contributions to [k(e)] and Ute)}. These contributions are considered in this section. The element contributions due to the derivative boundary condition come from the interelement vector (13.23) after substituting the relationship in (13.5). Assuming that rbe is the surface of the element with the boundary condition 2n or

r=
and

_ 22+26+24

3
2A =18 em?

24cm

The multiplier

coefficients are 4T 4T2nQA 12 2nr Dr 2nr

o,

2n(24)(1.5) 2(18) 4.5 n

{Ib~)} = -

Jrbc

[NY( - M¢b+S)

dr

(13.37)

2n(3)(9) 12

The element stiffness matrix is

[k(e)] = 2n - 2~
or

16

-20 25 -5

-:]+2' [ :
1 -8 -32 58 -26 -26 -8 34

4 4 -8

-8

-8]
16

[lb~)}=r

lbc

M[NY¢bdr-r

lbc
{<l>(e)} -

S[NYdr

(13.38)

The value of ¢ on the boundary,

¢b, is given by ¢(e)=[N]{<l>(e)}; therefore,

[lb~)l =(

lbc

M[NY[N]

dr)

lbc

S[NY

.u:

(13.39)

[k(e)] = n [ - 4D 32
The element force vector is

-8]

The first integral contributes to [k(e)], since it multiplies {<l>(e»); the second is part of [fie);. Both integrals in (13.39) are surface integrals and are evaluated similarly to those discussed in Chapter 1O. The integrals in (13.39) are (13.40)

{fo"} ~4.5. [:

m~J

and (13.41)

174
The integral in (13.41) is the easiest to evaluate giving

FIELD PROBlE

..

AXISYMMETRIC

FIELD PROBLEMS

175

Substituting (13.44) for r and using the factorial formula (6.17) yields [k~J .

u~e)} =s

Jrbc
2

Ni} dr =Ljk n, {N
k

Jo

s {LI} L2
L
3

Znr dt

(13.4

= 2n%L

jk

[~

assuming

that the integration L3 =f

is along side jk. The area coordinates

reduce

LI =0, L2 =f band

along this side [{se~(}6'34)] and the integral

J beCO'.'.•.
(13.41

II

(3Rj: Rk) (Rj+Rd

.: Rd] (Rj+3Rd

(13.49)

The other two results for (13.40) are (3Ri+Rj) [k~J=2n~Lij [ (Ri~Rj) (13.50)

(f!"») ~ 2.SL j.

L ~:;
l

«.
R +l ,R.
J

Theradialdistanceto:::::; :n~~;j::::::;~
since Ni=O. Substitution for r in (13.43) produces {f~e)}=2nSLjk

(13.1
i'~

2 ML. [(3Ri+Rd [k~J= n ,k 0 12 (Rj+Rk) for sides ij and ik, respectively. ILLUSTRATIVE EXAMPLE

o
o

(13.51)

{fl(fIR~+f2Rd}df2 o e 2(f I Rj+f 2Rd

(13.~ )~

Evaluate [k~J and {fde)} for side ij of the element shown in Figure M=4 and S=3. The pertinent quantities are and Lij=.J(26 The multiplying coefficients are 2nMLij 12 2(3.14)(4)(4.12)=862 12 . 2(3.14)(3)(4.12) 6 29 1. - 22)2 + (11-10)2 =4.12 em

13.4 when

The integrals of ti, fIt 2, and f~ are evaluated

using (6.17). The final result is; f~ Rk} (13.~

{fd )}

= 2n~Ljk {2Rj:

Rj+2Rk The other two results for {fde)} are

(13.4~

r
for sides ij and ik, respectively. The surface integral (13.40) is evaluated jk, (13.40) becomes in a similar manner. Considering : sidi

2nSLij 6 while

3Ri+ Rj=3(22)+26=92 R;+ Rj=22+26=48 R, + 3Rj=22 R, + 2Rj=22

r
The element quantities

+ 3(26)= + 2(26)=

100 74

2Ri+ Rj=2(22)+26=70 are 92 [k~J = 8.62 [ 4~ 48 100 0]

~

=2nMLjk

t1[0 ~

[793 = 41~

414 862

~]

176
and

FIELDPROBl"

Ud
Note that the diagonal same.

e )}

po 1
J

j903\

= 12.9l ~4 = 19~5

r
Ud
e)}

Chapter
are not.

14

values in

[k~] as well as the values in

PROBLEMS
Evaluate (13.40) for side jk of a triangular element. Evaluate (13.40) for side km of a rectangular element. Evaluate (13.41) for side ki of a triangular element. Evaluate (13.41) for side km of a rectangular element. Evaluate (13.25) for a rectangular element using the defined by (13.10). 13.6 Verify that (13.46) and (13.49) behave similarly to (9.19) and (9.33) w nodes j and k are on a vertical side, that is, R, = Ri: 13.7-13.11 The nodal coordinate and coefficient values for five axisymm elements follow. Evaluate [k~)] and {fJe)} for one of the elements. Ass that 13.1 13.2 13.3 13.4 13.5

TIME-DEPENDENT FIELD PROBLEMS: THEORETICAL CONSIDERATIONS

The field problems considered in the previous chapters were steady-state problems. An equally important set of physical problems are those that are time-dependent. The finite element solution of time-dependent problems is introduced in this chapter with the primary emphasis placed on the theoretical aspects. Some practical aspects of solving time-dependent problems are discussed in Chapter 15. The discussion in each chapter is build around the one-dimensional equation

o.s»;

Problem Number Element Quantity Dx

D~'I'+Q=Je~ 2

a2A.

ox

aA.
at

(14.1 ) plots

13.7 2 20 2 4 3 0 0 4

13.8

13.9 2 40
8

13.10 4 50 20 20 16 20 24 24

13.11 5 60 6
10

Zj

Ri Rj Rk Zi

3 30
10

because the numerical calculations are relatively easy and three-dimensional of ¢-x-t can be constructed. The two-dimensional time-dependent equation is

14 14
10

12
10

Zk
13.12 13.13 13.14 13.15

12 16

6
10

10 12
8

02¢ 02¢ ,o¢ Dx~o 2+Dy---;;;---y-G¢+Q=A:;x oy ut

(14.2)

The only new term in each of(14.1) and (14.2) is Jeo¢/ot. We shall see that it behaves the same in each dimension so that a discussion of (14.1) is adequate.

14

14

Use the element data for Problem 13.7 and the values M=6 and e Evaluate [k~] and )} along side ij. Use the element data for Problem 13.8 and the values M =10 and s:=:l e Evaluate [k~] and )} along side ik. ' Use the element data for Problem 13.10 and the values M =6 and S:=:-: Evaluate [k~] and {f~e)} along side jk. The governing differential equation for a radial symmetric problem is

Ud

s~J

14.1 GALERKIN METHOD

The general procedure for analyzing (14.1) or (14.2) is to evaluate the Galerkin residual integral with respect to the space coordinate(s) for a fixed instant of time. This yields a system of differential equations that are solved to obtain the variation of ¢ with time. By rearranging (14.1) into (14.3) the residual integral is

Ud

02¢ Dro¢ Dr~o 2 +--0 +Q=O r rr

Obtain the general equations for

[k~)] and

UJe)}

by evaluating the residual integl1

{R(e)} = _

r [NY JA

2 (Dr a ¢ or2

+ Dr ?__~+
r Or

Q) dA

{R(e)}=_

Xj

Xi

[WY (02¢ 2 D~+Q-JeaX

a¢) ot

dx

(14.4 )

178
where [WY into contains the Galerkin weighting functions. Equation

FIELD PROBlE",

TIME-DEPENDENT

FIELD PROBLEMS:

THEORETICAL

CONSIDERATIONS

179

14.4 separatt 1
.~

element can be written as (14.9)

Of

(14'1
or

{R(e)]f=( fR(e)]s + (fR(e)]f o A

(14.f

,"
If we also use the linear shape functions for the weighting [WY = [NY, the residual integral is {R\\)}
coefficients,

(14.10) that is,

Defining [WY =[NY in the first integral of (14.5).yields the same integrals tb) were analyzed in Chapter 3 and ~

(14.i
The only new term is {R~)}

J:i [WY

(~ ~~) dx

(14.'

J:i =( J:i

A[NY[N]{<I>(e)} A[NY[N]

dx

dX)

{<I>(e)}

(14.11)
(14.12)

Equation 14.8 has two different solutions. One solution is called the conslstei formulation and the other is referred to as the lumped formulation. We sha consider each of these solutions starting with the consistent formulation.;

= [c(e)]{$(e)}
A similar analysis for (14.2) yields

14.2

THE CONSISTENT FORMULATION

(14.13) where [N] contains the triangular or rectangular shape functions. When [c(e)] is combined with the other element matrices and summed over all the elements using the direct stiffness procedure, the final result is a system of firstorder differential equations given by

A set of weighting functions and the variation of o¢/ot with respect to x must b defined before the integral in (14.8) can be evaluated. The parameter o¢/od shown schematically in Figure 14.1. One equation that defines the variation stan that the time derivative varies linearly between the nodal values. If we denote tb nodal values of o¢lot by $b $2,... ,$p, the linear variation of o¢/ot within a
q,

[C]{$}

+ [K]{<I>} -{F} ={O}

(14.14)

where the boundary conditions in {I} have been incorporat~d into and the interelement requirements discarded. The vector {<I>} is
f

[K] and {F}

(14.15)

$1 T = [0<1>
J

in

0<1> 2

ot

..•

0<1> in

pJ

The new matrix [C], is usually called the capacitance matrix. The integrals that define [c(e)] are the same as those associated with the G term in the steady-state form of the differential equation. We immediately conclude that

[c(e)]=
for the one-dimensional

A~G

linear element,
1

~J

(14.16)

2
Figure 14.1. The 8lj>/8t as a function of x.

(14.17)

180
for the two-dimensional triangular element, and

FIELD PROBLE~

TIME-DEPENDENT

FIELD PROBLEMS:

THEORETICAL

CONSIDERATIONS

181

The variation functions 2

4

of

a4>/at within an element can be written explicitly using step

(14.19)

2
4

JJ
x

(14.1.

i'

where

for the two-dimensional rectangular element [see (11.9), (7.38), and (7.49)]. The formulation discussed in this section is called the consistent fOrmUlatii" because the linear variation of a<p/at with respect to within an element is identi ,. ' or consistent with the linear variation assumed for 4>(x).The same set of weighti functions, namely [NY, is used in both of the integrals in (14.5). .1 14.3 THE LUMPED FORMULATION
I

(14.20)

and s is measured from node i. The quantities multiplying be thought of as new shape functions, Nt and st. where

$i

and

$j

in (14.19) can

An alternate approach to defining the variation in a4>/at with respect to x is tC assume that it is constant between the midpoints of adjacent elements. This conceQ is shown schematically in Figure 14.2a for a single element and in Figure 14.~ for a grid of five elements.

Nt=1-h(S-~)
(14.21) Upon using this concept, (14.19) becomes

a4>_N*a$i+N*a$j=[N*r<D(en at at J at
I

(14.22)

Returning to the residual integral,

{Riel}

=
Using the same functions in

(a)

[Riel} =(
The element capacitance

at

act>

{Riel}, and using (14.22) gives x a4> A[WY ~dx x; at

J

},[WY[N*]{$(el}

ds for

(14.23)

[WY

that are used in the relationship

a4>/at gives
(14.24)

},[N*Y[N*]

dS) {$(el}=[c(el]{$(el}

matrix is defined by

[c(el] =
which is easily evaluated
2 Figure 14.2. (b) for a grid, The step variation of 3
(b)

},[N*Y[N*]

ds

(14.25)

because
I )

N*N*=N*N*=1 J
I

and

The element capacitance

matrix is

a </1/ at

as a function of x (a) within an elemen1

[c(el] = A~ [~

~J

(14.26)

182

FIELD PROBLEMS:

TIME-DEPENDENT

FIELD PROBLEMS:

THEORETICAL

CONSIDERATIONS

183

for the one-dimensional linear element. It is a diagonal matrix, which means that [C] is also diagonal. Since the formulation produces a diagonal matrix, it i$ usually referred to as a lumped formulation. The generalization of the step function variation for two-dimensional problems is accomplished by defining functions similar to those shown in Figure 14.3. The' shaded region for Nt' has corners at node i, at the midpoints of the sides that touch node i and at the center of the element. Defining the functions in this manner produces N: N] products whose value is zero. The resulting diagonal matrices are

for the triangular

element and 0 0 0 0 1 0 0

['(')l~': [~

~J

(14.28)

o o
Nt(x. y)

O~]

(14.27)

14.4 FINITE DIFFERENCE SOLUTION IN TIME

The finite element solution of time-dependent field problems produces a system of linear first-order differential equations in the time domain. These equations must be solved before the variation of ¢ in space and time is known. There are several procedures for numerically solving the equations given by (14.14). We shall concentrate on the methods associated with the solution of heat transfer problems that use a finite difference approximation in the time domain to generate a numerical solution. Given a function ¢(t) and the interval [a, b], we can use the mean value theorem for differentiation to develop an equation for ¢(t). The mean value theorem states that there is a value of t, call it ~, such that ¢(b)- ¢(a)=(bd¢ a) - (~) dt (14.29)

//
(a)

Rearranging

gives d¢ (~)=¢(b)-¢(a) dt I'1t (14.30)

m
Nt(x. y)
y

where I'1t=(b- a) is the length of the interval. The location of ~ in (14.29) and (14.30) is not known. This aspect will be dealt with in a few moments. The value of ¢ at t =a, ¢(a), can be approximated as shown in Figure 14.4. ¢(a)=¢(~)Rearranging gives d¢ ¢(~)=¢(a)+ (~- a) dt (~-a)

dt (~)

d¢

(14.31)

(~)

(14.32)

and after substituting

(14.30) ¢(¢)=¢(a)+ ¢(b)- ¢(a) (¢ -a) I'1t (14.33)

x
k (b)

If we define the ratio

elements. (a) ~

e as
(14.34)

Figure 14.3. The step-function vanation for two-dimensional triangular element. (b) A rectangular element.