Plates and Shells: Theory and Computation

- 4D9 Dr Fehmi Cirak (fc286@)

Office: Inglis building mezzanine level (INO 31)

Outline -1!

This part of the module consists of seven lectures and will focus on finite
elements for beams, plates and shells. More specifically, we will consider
!

Review of elasticity equations in strong and weak form

Beam models and their finite element discretisation

!
!

Page 2

Euler-Bernoulli beam
Timoshenko beam

Plate models and their finite element discretisation

Shells as an assembly of plate and membrane finite elements

Introduction to geometrically exact shell finite elements

Dynamics

F. Cirak

Outline -2!

There will be opportunities to gain hands-on experience with the

implementation of finite elements using MATLAB
!
!

Page 3

One hour lab session on implementation of beam finite elements (will be not marked)
Coursework on implementation of plate finite elements and dynamics

F. Cirak

Why Learn Plate and Shell FEs?

!

Beam, plate and shell FE are available in almost all finite element software
packages
!

The intelligent use of this software and correct interpretation of output requires basic
understanding of the underlying theories

FEM is able to solve problems on geometrically complicated domains

!

Analytic methods introduced in the first part of the module are only suitable for computing plates
and shells with regular geometries, like disks, cylinders, spheres etc.
Many shell structures consist of free form surfaces and/or have a complex topology
!

Page 4

Computational methods are the only tool for designing such shell structures

FEM is able to solve problems involving large deformations, non-linear

material models and/or dynamics
FEM is very cost effective and fast compared to experimentation

F. Cirak

Literature
!

JN Reddy, An introduction to the finite element method, McGraw-Hill (2006)

TJR Hughes, The finite element method, linear static and dynamic finite element
analysis, Prentice-Hall (1987)

K-J Bathe, Finite element procedures, Prentice Hall (1996)

J Fish, T Belytschko, A first course on finite elements, John Wiley & Sons (2007)

3D7 - Finite element methods - handouts

Page 5

F. Cirak

Examples of Shell Structures -1!

Civil engineering

Masonry shell structure (Eladio Dieste)

!

Mechanical engineering and aeronautics

Ship hull (sheet metal and frame)

Page 6

Concrete roof structure (Pier Luigi Nervi)

Fuselage (sheet metal and frame)

F. Cirak

Examples of Shell Structures -2!

Consumer products

Nature

Crusteceans
Page 7

Ficus elastica leaf

Red blood cells

F. Cirak

Representative Finite Element Computations

Wrinkling of an inflated party balloon

Virtual crash test (BMW)

buckling of carbon nanotubes

Sheet metal stamping (Abaqus)
Page 8

F. Cirak

Shell-Fluid Coupled Airbag Inflation -1-

0.49 m

0.86 m

6m
0.8

0.025 m

0.74 m

0.123 m
Shell mesh: 10176 elements

Page 9

Fluid mesh: 48x48x62 cells

F. Cirak

Shell-Fluid Coupled Airbag Inflation -2-

Page 10

F. Cirak

Detonation Driven Fracture -1Fractured tubes (Al 6061-T6)

Modeling and simulation challenges

!
!

Page 11

Ductile mixed mode fracture

Fluid-shell interaction

F. Cirak

Detonation Driven Fracture -2-

Page 12

F. Cirak

Roadmap for the Derivation of FEM

!

As introduced in 3D7, there are two distinct ingredients that are combined
to arrive at the discrete system of FE equations
!
!

In the derivation of the weak form for beams, plates and shells the
following approach will be pursued
1)
2)
3)

Page 13

The weak form

A mesh and the corresponding shape functions

Assume how a beam, plate or shell deforms across its thickness

Introduce the assumed deformations into the weak form of three-dimensional elasticity
Integrate the resulting three-dimensional elasticity equations along the thickness direction
analytically

F. Cirak

Elasticity Theory -1!

Consider a body in its undeformed (reference) configuration

Kinematic equations; defined based on displacements of an infinitesimal

volume element)
!

Page 14

The body deforms due to loading and the material points move by a displacement

Axial strains

F. Cirak

Elasticity Theory -2!

Page 15

Shear components

Stresses
!

Normal stress components

Shear stress component

Shear stresses are symmetric

F. Cirak

Elasticity Theory -3!

Equilibrium equations (determined from equilibrium of an infinitesimal

volume element)
!

Equilibrium in x-direction

Equilibrium in y-direction

Equilibrium in z-direction

Page 16

are the components of the external loading vector (e.g., gravity)

F. Cirak

Elasticity Theory -4!

Hookes law (linear elastic material equations)

Page 17

With the material constants Youngs modulus and Poissons ratio

F. Cirak

Index Notation -1!

Page 18

The notation used on the previous slides is rather clumsy and leads to very
long expressions
Matrices and vectors can also be expressed in index notation, e.g.

Summation convention: a repeated index implies summation over 1,2,3, e.g.

A comma denotes differentiation

F. Cirak

Index Notation -2!

Page 19

Kronecker delta

Examples:

F. Cirak

Elasticity Theory in Index Notation -1!

Kinematic equations

Equilibrium equations
!

Page 20

Note that these are six equations

Note that these are three equations

Linear elastic material equations

Inverse relationship

Instead of the Youngs modulus and Poissons ratio the Lame constants can be used

F. Cirak

Weak Form of Equilibrium Equations -1!

The equilibrium, kinematic and material equations can be combined into

three coupled second order partial differential equations

Next the equilibrium equations in weak form are considered in preparation

for finite elements
!

Page 21

In structural analysis the weak form is also known as the principle of virtual displacements
To simplify the derivations we assume that the boundaries of the domain are fixed (built-in, zero
displacements)
The weak form is constructed by multiplying the equilibrium equations with test functions vi which
are zero at fixed boundaries but otherwise arbitrary

F. Cirak

Weak Form of Equilibrium Equations -1!

Aside: divergence theorem

!

Consider a vector field

The divergence theorem states

and its divergence

Using the divergence theorem equation (1) reduces to

Page 22

Further make use of integration by parts

which leads to the principle of virtual displacements

F. Cirak

Weak Form of Equilibrium Equations -2!

The integral on the right hand side is the external virtual work performed by the external forces due to virtual
displacements

Note that the material equations have not been used in the preceding derivation.
Hence, the principle of virtual work is independent of material (valid for elastic, plastic,
)
The internal virtual work can also be written with virtual strains so that the principle of
virtual work reads

Page 23

The integral on the left hand side is the internal virtual work performed by the internal stresses due to virtual
displacements

Try to prove
F. Cirak

Finite Element Formulation for Beams

- Handout 2 Dr Fehmi Cirak (fc286@)

Completed Version

Review of Euler-Bernoulli Beam

Physical beam model

midline

Page 25

Beam domain in three-dimensions

Midline, also called the neutral axis, has the coordinate

Key assumptions: beam axis is in its unloaded configuration straight
Loads are normal to the beam axis

F Cirak

Kinematics of Euler-Bernoulli Beam -1

Assumed displacements during loading

deformed
configuration
reference
configuration

Kinematic assumption: Material points on the normal to the midline remain on the normal during
the deformation

Slope of midline:

The kinematic assumption determines the axial displacement of the material points across
thickness

Page 26

Note this is valid only for small deflections, else

F Cirak

Kinematics of Euler-Bernoulli Beam -2

Introducing the displacements into the strain equations of threedimensional elasticity leads to

Axial strains

Page 27

Axial strains vary linearly across thickness

All other strain components are zero

Shear strain in the

Through-the-thickness strain (no stretching of the midline normal during deformation)

No deformations in

and

planes so that the corresponding strains are zero

F Cirak

Weak Form of Euler-Bernoulli Beam

The beam strains introduced into the internal virtual work expression
of three-dimensional elasticity

with the standard definition of bending moment:

External virtual work

Weak work of beam equation

Page 28

Boundary terms only present if force/moment boundary conditions present

F Cirak

Stress-Strain Law

The only non-zero stress component is given by Hookes law

This leads to the usual relationship between the moment and curvature

Weak form work as will be used for FE discretization

Page 29

with the second moment of area

EI assumed to be constant

F Cirak

Finite Element Method

Beam is represented as a (disjoint) collection of finite elements

On each element displacements and the test function are interpolated using
shape functions and the corresponding nodal values

Number of nodes per element

Shape function of node K

Nodal values of displacements

Nodal values of test functions

To obtain the FE equations the preceding interpolation equations are

introduced into the weak form

Page 30

Note that the integrals in the weak form depend on the second order derivatives of u3 and v
F Cirak

Aside: Smoothness of Functions

A function f: is of class Ck=Ck() if its derivatives of order j, where

0 j k, exist and are continuous functions

For example, a C0 function is simply a continuous function

For example, a C function is a function with all the derivatives continuous

C1-continuous function

differentiation

C0-continuous function

Page 31

The shape functions for the Euler-Bernoulli beam have to be C1-continuous

so that their second order derivatives in the weak form can be integrated
F Cirak

Hermite Interpolation -1

To achieve C1-smoothness Hermite shape functions can be used

Hermite shape functions for an element of length

Shape functions of node 1

Page 32

with

F Cirak

Hermite Interpolation -2

Shape functions of Node 2

Page 33

with

F Cirak

Element Stiffness Matrix

Page 34

According to Hermite interpolation the degrees of freedom for each element are the
displacements
and slopes
at the two nodes

Interpolation of the displacements

Test functions are interpolated in the same way like displacements

Introducing the displacement and test functions interpolations into weak form gives the element stiffness matris

F Cirak

Element Load Vector

Load vector computation analogous to the stiffness matrix derivation

The global stiffness matrix and the global load vector are obtained by assembling the
individual element contributions

Page 35

The assembly procedure is identical to usual finite elements

Global stiffness matrix

Global load vector

All nodal displacements and rotations

F Cirak

Stiffness Matrix of Euler-Bernoulli Beam

Page 36

Element stiffness matrix of an element with length le

F Cirak

Kinematics of Timoshenko Beam -1

Assumed displacements during loading

deformed
configuration
reference
configuration

Kinematic assumption: a plane section originally normal to the centroid remains plane, but in
addition also shear deformations occur

Rotation angle of the normal:

Angle of shearing:
Slope of midline:

The kinematic assumption determines the axial displacement of the material points across
thickness

Page 37

Note that this is only valid for small rotations, else

F Cirak

Kinematics of Timoshenko Beam -2

Introducing the displacements into the strain equations of threedimensional elasticity leads to

Axial strain

Shear strain

Page 38

Axial strain varies linearly across thickness

Shear strain is constant across thickness

All the other strain components are zero

F Cirak

Weak Form of Timoshenko Beam

The beam strains introduced into the internal virtual work expression
of three-dimensional elasticity give

Hookess law

Introducing the expressions for strain and Hookes law into the weak form gives

Page 39

virtual displacements and rotations:

shear correction factor necessary because across thickness shear stresses are parabolic
according to elasticity theory but constant according to Timoshenko beam theory

shear correction factor for a rectangular cross section

shear modulus

External virtual work similar to Euler-Bernoulli beam

F Cirak

Euler-Bernoulli vs. Timoshenko -1

Comparison of the displacements of a cantilever beam analytically

computed with the Euler-Bernoulli and Timoshenko beam theories

Page 40

Bernoulli beam

Governing equation:

Boundary conditions:

Timoshenko beam

Governing equations:

Boundary conditions:

F Cirak

Euler-Bernoulli vs. Timoshenko -2

Maximum tip deflection computed by integrating the differential equations

Bernoulli beam

Timoshenko beam

Ratio

Page 41

For slender beams (L/t > 20) both theories give the same result
For stocky beams (Lt < 10) Timoshenko beam is physically more realistic because it includes the shear
deformations
F Cirak

Finite Element Discretization

Page 42

The weak form essentially contains

test functions

C0 interpolation appears to be sufficient, e.g. linear interpolation

Interpolation of displacements and rotation angle

and the corresponding

F Cirak

Element Stiffness Matrix

Shear angle

Curvature

Page 43

Test functions are interpolated in the same way like displacements and rotations
Introducing the interpolations into the weak form leads to the element stiffness matrices

Shear component of the stiffness matrix

Bending component of the stiffness matrix

F Cirak

Review: Numerical Integration

Gaussian Quadrature

Page 44

The locations of the quadrature points and weights are determined for maximum accuracy

nint=1

nint=2

nint=3

Note that polynomials with order (2nint-1) or less are exactly integrated

The element domain is usually different from [-1,+1) and an isoparametric

mapping can be used

F Cirak

Stiffness Matrix of the Timoshenko Beam -1

Necessary number of quadrature points for linear shape functions

Page 45

Bending stiffness: one integration point sufficient because

Shear stiffness: two integration points necessary because

is constant
is linear

Element bending stiffness matrix of an element with length le and one integration
point

Element shear stiffness matrix of an element with length le and two integration points

F Cirak

Limitations of the Timoshenko Beam FE

Recap: Degrees of freedom for the Timoshenko beam

Physics dictates that for t0 (so-called Euler-Bernoulli limit) the shear angle
has to go to zero (
)

If linear shape functions are used for u3 and

Adding a constant and a linear function will never give zero!

Hence, since the shear strains cannot be arbitrarily small everywhere, an erroneous shear strain
energy will be included in the energy balance

Page 46

In practice, the computed finite element displacements will be much smaller than the exact solution

F Cirak

Shear Locking: Example -1

Displacements of a cantilever beam

Influence of the beam thickness on the normalized tip displacement

TWO integration points

Thin beam

Thick beam
# elem.

2 point

# elem.

2 point

0.0416

0.0002

0.445

0.0008

0.762

0.0003

0.927

0.0013

from TJR Hughes, The finite element method.

Page 47

F Cirak

Stiffness Matrix of the Timoshenko Beam -2

The beam element with only linear shape functions appears not to be ideal for very
thin beams
The problem is caused by non-matching u3 and interpolation

How can we fix this problem?

Page 48

For very thin beams it is not possible to reproduce

Lets try with using only one integration point for integrating the element shear stiffness matrix
Element shear stiffness matrix of an element with length le and one integration points

F Cirak

Shear Locking: Example -2

Displacements of a cantilever beam

Influence of the beam thickness on the normalized displacement

ONE integration point

Thin beam

Thick beam
# elem.

1 point

# elem.

1 point

0.762

0.750

0.940

0.938

0.985

0.984

0.996

0.996

from TJR Hughes, The finite element method.

Page 49

F Cirak

Reduced Integration Beam Elements

Page 50

If the displacements and rotations are interpolated with the same shape
functions, there is tendency to lock (too stiff numerical behavior)
Reduced integration is the most basic engineering approach to resolve
this problem

Shape function
order

Linear

Quadratic

Cubic

Quadrature rule

One-point

Two-point

Three-point

Mathematically more rigorous approaches: Mixed variational principles

based e.g. on the Hellinger-Reissner functional

F Cirak

Finite Element Formulation for Plates

- Handout 3 Dr Fehmi Cirak (fc286@)

Completed Version

Definitions

A plate is a three dimensional solid body with

one of the plate dimensions much smaller than the other two
zero curvature of the plate mid-surface in the reference configuration
loading that causes bending deformation
mid-surface
or mid-plane

A shell is a three dimensional solid body with

Page 52

one of the shell dimensions much smaller than the other two
non-zero curvature of the shell mid-surface in the current configuration
loading that causes bending and stretching deformation

F Cirak

Membrane versus Bending Response

For a plate membrane and bending response are decoupled

loading in the plane of the mid-surface

(membrane response active)

For most practical problems membrane and bending response can be investigated independently
and later superposed
Membrane response can be investigated using the two-dimensional finite elements introduced in
3D7
Bending response can be investigated using the plate finite elements introduced in this handout

For plate problems involving large deflections membrane and bending

response are coupled

Page 53

loading orthogonal to the mid-surface

(bending response active)

For example, the stamping of a flat sheet metal into a complicated shape can only be simulated
using shell elements
F Cirak

Overview of Plate Theories

In analogy to beams there are several different plate theories

thick

thin

very thin

Lengt / thickness

~5 to ~10

~10 to ~100

> ~100

physical
characteristics

transverse shear
deformations

negligible transverse
shear deformations

geometrically nonlinear

The extension of the Euler-Bernoulli beam theory to plates is the Kirchhoff plate theory

The extension of Timoshenko beam theory to plates is the Reissner-Mindlin plate theory

Suitable for thick and thin plates

As discussed for beams the related finite elements have problems if applied to thin problems

In very thin plates deflections always large

Page 54

Suitable only for thin plates

Geometrically nonlinear plate theory crucial (such as the one introduced for buckling of plates)

F Cirak

Kinematics of Kirchhoff Plate -1

Assumed displacements during loading

deformed

reference

undeformed and deformed geometries

along one of the coordinate axis

Kinematic assumption: Material points which lie on the mid-surface normal remain on the midsurface normal during the deformation
Kinematic equations

In-plane displacements

Page 55

In this equation and in following all Greek indices take only values 1 or 2
It is assumed that rotations are small

Out-of-plane displacements

F Cirak

Kinematics of Kirchhoff Plate -2

Page 56

Introducing the displacements into the strain equations of three-dimensional

elasticity leads to

Axial strains and in-plane shear strain

All other strain components are zero

Out-of-plane shear

Through-the-thickness strain (no stretching of the mid-surface normal during deformation)

F Cirak

Weak Form of Kirchhoff Plate -1

The plate strains introduced into the internal virtual work expression of
three-dimensional elasticity

Note that the summation convention is used (summation over repeated indices)

Definition of bending moments

External virtual work

Distributed surface load

Weak form of Kirchhoff Plate

Page 57

For other type of external loadings see TJR Hughes book

Boundary terms only present if force/moment boundary conditions present

F Cirak

Weak Form of Kirchhoff Plate -2

Moment and curvature matrices

Constitutive equation (Hookes law)

Plane stress assumption for thin plates

must be used

Hookes law for three-dimensional elasticity (with Lam constants)

Page 58

Both matrices are symmetric

Through-the-thickness strain can be determined using plane stress assumption

Introducing the determined through-the-thickness strain

Hookes law for plane stress

back into the Hookes law yields the

F Cirak

Weak Form of Kirchhoff Plate -3

Integration over the plate thickness leads to

Page 59

Note the change to Youngs modulus and Poissons ratio

The two sets of material constants are related by

F Cirak

Finite Element Discretization

The problem domain is partitioned into a collection of pre-selected finite

elements (either triangular or quadrilateral)
On each element displacements and test functions are interpolated using
shape functions and the corresponding nodal values

Shape functions

Nodal values

To obtain the FE equations the preceding interpolation equations are

introduced into the weak form

Page 60

Similar to Euler-Bernoulli Beam the internal virtual work depends on the second order derivatives
of the deflection
and virtual deflection
C1-continuous smooth shape functions are necessary in order to render the internal virtual work
computable
F Cirak

Review: Isoparametric Shape Functions -1

In finite element analysis of two and three dimensional problems the

isoparametric concept is particularly useful
Isoparametric mapping of a four-node quadrilateral

parent element
physical element

Shape functions are defined on the parent (or master) element

Page 61

Each element on the mesh has exactly the same shape functions

Shape functions are used for interpolating the element coordinates and deflections

F Cirak

Review: Isoparametric Shape Functions -2

In the computation of field variable derivatives the Jacobian of the mapping has to be
considered

Page 62

The Jacobian is computed using the coordinate interpolation equation

F Cirak

Shape Functions in Two Dimensions -1

In 3D7 shape functions were derived in a more or less ad hoc way

Shape functions can be systematically developed with the help of the Pascals
triangle (which contains the terms of polynomials, also called monomials, of various
degrees)

Page 63

Triangular elements

Three-node triangle linear interpolation

Six-node triangle quadratic interpolation

Quadrilateral elements

Four-node quadrilateral bi-linear interpolation

Nine-node quadrilateral bi-quadratic interpolation

Pascals triangle
(with constants a, b, c, d, )

It is for the convergence of the finite element method important to use only complete polynomials up to a certain
desired polynomial order
F Cirak

Shape Functions in Two Dimensions -2

The constants a, b, c, d, e, in the polynomial expansions can be

expressed in dependence of the nodal values

For example in case of a a four-node quadrilateral element

As mentioned the plate internal virtual work depends on the second

derivatives of deflections and test functions so that C1-continuous smooth
shape functions are necessary

Page 64

with the shape functions

It is not possible to use the shape functions shown above

F Cirak

Early Smooth Shape Functions -1

For the Euler-Bernoulli beam the Hermite interpolation was used which has the nodal
deflections and slopes as degrees-of-freedom
The equivalent 2D element is the Adini-Clough quadrilateral (1961)

Degrees-of-freedom are the nodal deflections and slopes

Interpolation with a polynomial with 12 (=3x4) constants

monomials

Page 65

Surprisingly this element does not produce C1- continuous smooth interpolation
(explanation on next page)

F Cirak

Early Smooth Shape Functions -2

Consider an edge between two Adini-Clough elements

Page 66

For simplicity the considered boundary is assumed to be along the

The deflections and slopes along the edge are

so that there are 8 unknown constants in these equations

axis in both elements

If the interpolation is smooth, the deflection and the slopes in both elements along the edge have
to agree
It is not possible to uniquely define a smooth interpolation between the two elements because
there are only 6 nodal values available for the edge (displacements and slopes of the two nodes).
There are however 8 unknown constants which control the smoothness between the two
elements.

Elements that violate continuity conditions are known as nonconforming

elements. The Adini-Clough element is a nonconforming element. Despite
this deficiency the element is known to give good results
F Cirak

Early Smooth Shape Functions -3

Bogner-Fox-Schmidt quadrilateral (1966)

Page 67

This element is conforming because there are now

8 parameters on a edge between two elements in order to
generate a C1-continuous function

Problems

Degrees-of-freedom are the nodal deflections, first derivatives and second mixed derivatives

monomials

Physical meaning of cross derivatives not clear

At boundaries it is not clear how to prescribe the cross derivatives
The stiffness matrix is very large (16x16)

Due to these problems such elements are not widely used in present day
commercial software
F Cirak

New Developments in Smooth Interpolation

Recently, research on finite elements has been reinvigorated by the use of

smooth surface representation techniques from computer graphics and
geometric design

Smooth surfaces are crucial for computer graphics, gaming and geometric design

Fifa 07, computer game

Page 68

F Cirak

Splines - Piecewise Polynomial Curves

Splines are piecewise polynomial curves for smooth interpolation

For example, consider cubic spline shape functions

cubic
polynomial
cubic
polynomial

Page 69

cubic
polynomial
cubic
polynomial

Each cubic spline is composed out of four cubic polynomials; neighboring curve segments are C2
continuously connected (i.e., continuous up to second order derivatives)
An interpolation constructed out of cubic spline shape functions is C2 continuous

F Cirak

Tensor Product B-Spline Surfaces -1

A b-spline surface can be constructed as the tensor-product of b-spline

curves

two
dimensional

one
dimensional

Tensor product b-spline surfaces are only possible over regular meshes
A presently active area of research are the b-spline like surfaces over irregular meshes

The new approaches developed will most likely be available in next generation finite element software

irregular mesh
Page 70

one
dimensional

spline like surface

generated on
irregular mesh
F Cirak

Finite Element Formulation for Plates

- Handout 4 Dr Fehmi Cirak (fc286@)

Completed Version

Kinematics of Reissner-Mindlin Plate -1

The extension of Timoshenko beam theory to plates is the Reissner-Mindlin

plate theory
In Reissner-Mindlin plate theory the out-of-plane shear deformations are
non-zero (in contrast to Kirchhoff plate theory)
Almost all commercial codes (Abaqus, LS-Dyna, Ansys, ) use ReissnerMindlin type plate finite elements
Assumed displacements during loading

deformed

reference
undeformed and deformed geometries
along one of the coordinate axis

Page 72

Kinematic assumption: a plane section originally normal to the mid-surface remains plane, but in
addition also shear deformations occur
F Cirak

Kinematics of Reissner-Mindlin Plate -2

Kinematic equations

In plane-displacements

Page 73

In this equation and in following all Greek indices only take values 1 or 2
It is assumed that rotations are small
Rotation angle of normal:
Angle of shearing:
Slope of midsurface:

Out-of-plane displacements

The independent variables of the Reissner-Mindlin plate theory are the

rotation angle
and mid-surface displacement
Introducing the displacements into the strain equation of three-dimensional
elasticity leads to the strains of the plate

F Cirak

Weak Form of Reissner-Mindlin Plate -1

Axial strains and in-plane shear:

Out-of-plane shear:

Through-the-thickness strain:

The plate strains introduced into the internal virtual work of threedimensional elasticity give the internal virtual work of the plate

Page 74

Note that always

with virtual displacements and rotations:

F Cirak

Weak Form of Reissner-Mindlin Plate -2-

Definition of bending moments:

Definition of shear forces:

External virtual work

Distributed surface load

Weak form of Reissner-Mindlin plate

Page 75

As usual summation convention applies

F Cirak

Weak Form of Reissner-Mindlin Plate -3

Constitutive equations

For bending moments (same as Kirchhoff plate)

For shear forces

Page 76

Note that the curvature

and the shear angle

is a function of rotation angle

is a function of rotation angle

and the mid-surface displacement

F Cirak

Finite Element Discretization -1

The independent variables in the weak form are

corresponding test functions

Weak form contains only

Page 77

and the

so that C0-interpolation is sufficient

Usual (Lagrange) shape functions such as used in 3D7 can be used

Nodal values of variables:

Nodal values test functions:

Interpolation equations introduced into the kinematic equations yield

four-node isoparametric
element

F Cirak

Finite Element Discretization -2-

Constitutive equations in matrix notation

Bending moments:

Shear forces:

Element stiffness matrix of a four-node quadrilateral element

Page 78

Bending stiffness (12x12 matrix)

F Cirak

Finite Element Discretization -3

Shear stiffness (12x12 matrix)

Element stiffness matrix

The integrals are evaluated with numerical integration. If too few integration points are used,
element stiffness matrix will be rank deficient.

The global stiffness matrix and global load vector are obtained by
assembling the individual element stiffness matrices and load
vectors

Page 79

The necessary number of integration points for the bilinear element are 2x2 Gauss points

The assembly procedure is identical to usual finite elements

F Cirak

Shear Locking Problem -1

As discussed for the Bernoulli and Timoshenko beams with increasing plate
slenderness physics dictates that shear deformations have to vanish
(i.e.,
)

Reissner-Mindlin plate and Timoshenko beam finite elements have problems to approximate
deformation states with zero shear deformations (shear locking problem)

1D example: Cantilever beam with applied tip moment

Page 80

Bending moment and curvature constant along the beam

Shear force and hence shear angle zero along the beam
Displacements quadratic along the beam

Discretized with one two-node Timoshenko beam element

F Cirak

Shear Locking Problem -2

Deflection interpolation:

Rotation interpolation:

Shear angle:

For the shear angle to be zero along the beam, the displacements and rotations have to be zero. Hence, a
shear strain in the beam can only be reached when there are no deformations!
Similarly, enforcing

at two integration points leads to zero

displacements and rotations!

However, enforcing

only at one integration point (midpoint of the

beam) leads to non-zero displacements

In the following several techniques will be introduced to circumvent the

shear locking problem

Page 81

Use of higher-order elements

Uniform and selective reduced integration
Discrete Kirchhoff elements
Assumed strain elements
F Cirak

Constraint Ratio (Hughes et al.) -1

Constraint ratio is a semi-heuristic number for estimating an elements

tendency to shear lock

Continuous problem

Number of equilibrium equations: 3 (two for bending moments + one for shear force)
Number of shear strain constraints in the thin limit: 2

Constraint ratio:

With four-node quadrilateral finite elements discretized problem

Number of degrees of freedom per element on a very large mesh is ~3

Number of constraints per element for 2x2 integration per element is 8

Number of constraints per element for one integration point per element is 2

Page 82

Constraint ratio:

Constraint ratio:

F Cirak

Constraint Ratio (Hughes et al.) -2

Constraint ratio for a 9 node element

Number of degrees of freedom per element on a very large mesh is ~ 4x3 =12
Number of constraints per element for 3x3 integration is 18

Constraint ratio:

Constraint ratio for a 16 node element

Number of degrees of freedom per element on a very large mesh is ~ 9x3=27

Number of constraints per element for 4x4 integration is 32

Constraint ratio:

Page 83

As indicated by the constraint ratio higher-order elements are less likely to

exhibit shear locking

F Cirak

Uniform And Selective Reduced Integration -1

The easiest approach to avoid shear locking in thin plates is to use

some form of reduced integration

In selective reduced integration the bending term is integrated with the normal rule and
the shear term with a lower-order rule

Uniform reduced integrated elements have usually rank deficiency

(i.e. there are internal mechanisms; deformations which do not need
energy)

Page 84

In uniform reduced integration the bending and shear terms are integrated with the
same rule, which is lower than the normal

The global stiffness matrix is not invertible

Not useful for practical applications

F Cirak

Uniform And Selective Reduced Integration -2-

Shape functions

Bilinear

Biquadratic

Bicubic

Uniform reduced
integration

1x1

2x2

3x3

Selective reduced
integration

1x1 shear
2x2 bending

2x2 shear
3x3 bending

3x3 shear
4x4 bending

Page 85

Shear refers to the integration of the element shear stiffness matrix

Bending refers to the integration of the element bending stiffness matrix
F Cirak

Discrete Kirchhoff Elements

Page 86

The principal approach is best illustrated with a Timoshenko beam

The displacements and rotations are approximated with quadratic shape
functions

The inner variables are eliminated by enforcing zero shear stress at the two
gauss points

Back inserting
into the interpolation equations leads to a beam
element with 4 nodal parameters
F Cirak

Assumed Strain Elements

It is assumed that the out-of-plane shear strains at edge centres are of

higher quality (similar to the midpoint of a beam)

First, the shear angle at the edge centres

is computed using the displacement and rotation nodal values

Subsequently, the shear angles from the edge centres are interpolated back

Page 87

Note that the shape functions

are special edge shape functions

These reinterpolated shear angles are introduced into the weak form and are for element stiffness
matrix computation used
F Cirak

Finite Element Formulation for Shells

- Handout 5 Dr Fehmi Cirak (fc286@)

Completed Version

Overview of Shell Finite Elements

There are three different approaches for deriving shell finite

elements

Flat shell elements

Degenerated shell elements

Page 89

The geometry of a shell is approximated with flat finite elements

Flat shell elements are obtained by combining plate elements with plate stress elements

Elements are derived by degenerating a three dimensional solid finite element into a shell
surface element

F Cirak

Flat Shell Finite Elements

Example: Discretization of a cylindrical shell with flat shell finite elements

Cylindrical shell

Fine mesh

Note that due to symmetry only one eight of the shell is discretized

The quality of the surface approximation improves if more and more flat elements are used
Flat shell finite elements are derived by superposition of plate finite elements with plane stress
finite elements

Page 90

Coarse mesh

As plate finite elements usually Reissner-Mindlin plate elements are used

As plane stress elements the finite elements derived in 3D7 are used
Overall approach equivalent to deriving frame finite elements by superposition of beam and truss finite
elements
F Cirak

Four-Noded Flat Shell Element -1

First the degrees of freedom of a plate and plane-stress finite element in a

local element-aligned coordinate system are considered
Plate element

Page 91

The local base vectors

element

Plane stress element

Flat shell element

are in the plane of the element and

is orthogonal to the

The plate element has three degrees of freedom per node (one out-of-plane displacement and two
rotations)
The plane stress element has two degrees of freedom per node node (two in plane displacements)
The resulting flat shell element has five degrees of freedom per node

F Cirak

Four-Noded Flat Shell Element -2

Stiffness matrix of the plate in the local coordinate system:

Stiffness matrix of the plane stress element in the local coordinate system:

Stiffness matrix of the flat shell element in the local coordinate system

Stiffness matrix of the flat shell element can be augmented to include the rotations
figure on previous page)

Page 92

(see

Stiffness components corresponding to

are zero because neither the plate nor the plane
stress element has corresponding stiffness components

F Cirak

Four-Noded Flat Shell Element -3

Transformation of the element stiffness matrix from the local to the global
coordinate system

Discrete element equilibrium equation in the local coordinate system

Transformation of vectors from the local to the global coordinate system

Page 93

Nodal displacements and rotations of element

Element force vector

Rotation matrix (or also known as the direction cosine matrix)

Note that for all rotation matrices

Transformation of element stiffness matrix from the local to global coordinate system

Discrete element equilibrium equation in the global coordinate system

F Cirak

Four-Noded Flat Shell Element -4

The global stiffness matrix for the shell structure is constructred by

transforming each element matrix into the global coordinate system prior to
assembly
The global force vector of the shell structure is constructed by transforming
each element force vector into the global coordinate system prior to
assembly
Remember that there was no stiffness associated with the local rotation
degrees of freedom
. Therefore, the global stiffness matrix will be rank
deficient if all elements are coplanar.

It is possible to add some small stiffness for element stiffness components corresponding to
in order to make global stiffness matrix invertible

Add small stiffness in order to make

stiffness matrix invertible

Page 94

F Cirak

Degenerated Shell Elements -1

First a three-dimensional solid element and the corresponding parent

element are considered (isoparametric mapping)

parent element

Page 95

solid element

In the following it is assumed that the solid element has on its top and bottom surfaces nine nodes
so that the total number of nodes is eighteen

The derivations can easily be generalised to arbitrary number of nodes

Coordinates of the nodes on the top surface are

Coordinates of the nodes on the bottom surface are

F Cirak

Degenerated Shell Elements -2

There are nine isoparametric shape functions for interpolating the top and bottom surfaces

with the natural coordinates

Note that these shape functions are identical to the ones for two dimensional elasticity

The geometry of the solid element can be interpolate with

Definitions

Shell mid-surface node

Shell director (or fibre) at node

Page 96

The shell director is a unit vector and is approximately orthogonal to the mid-surface
F Cirak

Degenerated Shell Elements -3

Using the previous definitions the solid element geometry can be

interpolated with

The displacements of the solid element are assumed to be

The first component is the mid-surface displacement and the second component is the director
displacement

Note that the deformed mid-surface nodal coordinates can be computed with
director with

The director displacement

stretch

Page 97

with the solid element thickness

has to be constructed so that the director

and the deformed nodal

can rotate but not

This was one of the of the Reissner-Mindlin theory assumptions

F Cirak

Degenerated Shell Elements -4

The director displacements are expressed in terms of rotations at the nodes

To accomplish this a local orthonormal coordinate system

is constructed at each node

The relationship between the director displacements and the two rotation angles in the local
coordinate system is

Page 98

The definition of the orthonormal coordinate system in not unique.

In a finite element implementation it is necessary to store at each
node the established coordinate base vectors
.

Rotations
the right-hand rule

are defined as positive with

It is assumed that the rotation angles are small

so that the director length does not change
F Cirak

Degenerated Shell Elements -5

Displacement of the shell element in dependence of the mid-surface

displacements and director rotations

Introducing the displacements into the strain equation of three-dimensional

elasticity leads to the strains of the shell element

Page 99

The element has nine nodes

There are five unknowns per node (three mid-surface displacements and two director rotations)
This assumption about the possible displacements is equivalent to the Reissner-Mindlin
assumption

In computing the displacement derivatives the chain rule needs to be used

F Cirak

Degenerated Shell Elements -6

Page 100

The Jacobian is computed from the geometry interpolation

The shell strains introduced into the internal virtual work of three-dimensional
elasticity give the internal virtual work of the shell
For shear locking similar techniques such as developed for the Reissner-Mindlin plate
need to be considered

F Cirak

Dynamics of Beams, Plates and Shells

- Handout 6 Dr Fehmi Cirak (fc286@)

Completed Version

Strong and Weak Form of Elastodynamics

Equilibrium equations for elastodynamics (strong form)

Displacement initial condition (at time=0)

Velocity initial condition (at time=0)

The weak form of the equilibrium equations for elastodynamics is known as

the dAlemberts principle

It can be obtained by the standard procedure: Multiply the strong form with a test function,
integrate by parts and apply the divergence theorem

Page 102

Density
Acceleration vector
Stress matrix
Distributed body force vector

Virtual displacement vector

Build-in boundaries are assumed so that no boundary integrals are present in the above equation
F. Cirak

FE Discretization of Elastodynamics -1

The weak form over one typical finite element element

Approximation of displacements, accelerations and virtual displacements

Page 103

are the nodal displacements, accelerations and virtual displacements, respectively

Shape functions are the same as in statics
Nodal variables are now a function of time!

Element mass matrix is computed by introducing the approximations into kinetic virtual work

Element stiffness matrix and the load vector are the same as for the static case
F. Cirak

FE Discretization of Elastodynamics -2

Global semi-discrete equation of motion after assembly of element matrices

Global mass matrix

Global stiffness matrix
Global external force vector
Initial conditions

Equation called semi-discrete because it is discretized in space but still continuous in time

Global semi-discrete equation of motion with viscous damping

Rayleigh damping (widely used in structural engineering)

Page 104

Damping matrix
Damping proportional to velocity

and

are two scalar structure properties which are determined from experiments

F. Cirak

Dynamics of Timoshenko Beams

Assumed deformations during dynamic loading

deformed
configuration
reference
configuration

Page 105

Kinematic assumption: a plane section originally normal to the centroid remains plane,
but in addition also shear deformations occur

Key kinematic relations for statics:

Corresponding relations for dynamics:

F. Cirak

Kinetic Virtual Work for Timoshenko Beam

Introducing the beam accelerations and test functions into the

kinetic virtual work of elastodynamics gives

Virtual axial displacements

Virtual deflections and rotations

Kinetic virtual work due to rotation

Page 106

Rotationary inertia (very small for thin beams)

Kinetic virtual work due to deflection

F. Cirak

Finite Element Discretization - Mass Matrix

Interpolation with shape functions (e.g. linear shape functions)

Consistent mass matrix is computed by introducing the interpolations into the kinetic
virtual work

For practical computations lumped mass matrix sufficient (the sum of the elements of each row of the consistent
mass matrix is used as the diagonal element)

Page 107

The components of the mass matrix are simply the total element mass and rotational inertia divided by two

F. Cirak

Most Basic Time Integration Scheme -1

Semi-discrete equation of motion

Time integration

Central difference formula for the velocity

Central difference formula for the acceleration

Discrete equilibrium at t=tn

Page 108

Assume displacements, velocities and accelerations are known

for ttn

Substituting the computed acceleration into the central difference formula for acceleration yields the
displacements at t=tn+1

F. Cirak

Most Basic Time Integration Scheme -2

These equations are repeatedly used in order to march in time and to obtain solutions
at times t=tn+2, tn+3,
Provided that the mass matrix is diagonal displacements and velocities are computed without
inverting any matrices. Such a scheme is called explicit.

Matrix inversion is usually the most time consuming part of finite element analysis
In most real world applications, explicit time integration schemes are used

Explicit time integration is very easy to implement. The disadvantage is

conditional stability. If the time step exceeds a critical value the solution will
grow unboundedly

Critical time step size

Characteristic element size

Wave speed

Longitudinal wave speed in solids (material property)

exact solution

Page 109

F. Cirak

Advanced Time Integration Schemes -1

It is instructive to consider the time integration of the semi-discrete

heat equation before attempting the time integration for elasticity

Page 110

Temperature vector and its time derivative

Heat capacity matrix
Heat conductivity matrix
Heat supply vector
Initial conditions

- family of time integration schemes

with

Combining both gives

Introducing into the semidiscrete heat equation gives an equation for determining

F. Cirak

Advanced Time Integration Schemes -2

Common names for the resulting methods

forward differences; forward Euler
trapezoidal rule; midpoint rule; Crank-Nicholson
backward differences; backward Euler

Explicit vs. implicit methods and their stability

The method is explicit and conditionally stable for

The method is implicit and stable for

Time step size not restricted. However, for large time steps less accurate.

Implementation in a predictor-corrector form

Simplifies the computer implementation. Does not change the basic method.
Compute a predictor
with known solution

Introducing into the semidiscrete heat equation gives an equation for determining

Page 111

Time step size restricted

F. Cirak

Newmarks Scheme -1

Newmark family of time integration schemes are widely used in

structural dynamics

Assume that the displacements, velocities, and accelerations are known for ttn
Displacements and velocities are approximated with

Implementation in a-form (according to Hughes)

Page 112

with two scalar parameters

The two scalar parameters determine the accuracy of the scheme
Unconditionally stable and undamped for

Simplifies the computer implementation. Does not change the basic method.
Compute predictor velocities and displacements

F. Cirak

Newmarks Scheme -2

Displacement and velocity approximation using predictor values

Introducing the displacement approximation into the semidiscrete equation gives an equation for
computing the new accelerations at time t=tn+1

Page 113

The new displacements and velocities are computed from the displacement and velocity approximation
equations

F. Cirak

Elastodynamics - Motivation

Page 90

Fehmi Cirak

Elastodynamics -1!

The discrete elastodynamics equations can be derived from either

Hamiltonian, Lagrangian or principle of virtual work for (dAlemberts
principle)
(build-in boundaries)
!

with initial conditions

Discretization with finite elements

Element mass matrix

element mass matrix

Page 91

The stiffness matrix and the load vector are the same as for the static case
Fehmi Cirak

Elastodynamics -2!

Semi-discrete equation of motion

Mass matrix
Stiffness matrix
External force vector
Initial conditions

Semi-discrete because it is discretized in space but continuous in time

!
!
!

Viscous damping

Rayleigh damping

Page 92

Fehmi Cirak

Timoshenko Beam - Virtual Kinetic Work

deformed
configuration
reference
configuration
!

Kinetic virtual work

Rotationary inertia (very small for thin beams)

Page 93

Fehmi Cirak

Timoshenko Beam - Mass Matrix

!

Discretization with linear shape functions

Lumped mass matrix (lumping by row-sum technique)

In practice the rotational contribution can mostly be neglected

For the equivalent Reissner-Mindlin plate, the components of the mass matrix are
simply the total element mass divided by four

Page 94

Fehmi Cirak

Explicit Time Integration -1!

Semi-discrete equation of motion

Discretization in time (or integration in time)

!

Page 95

Assume displacements, velocities, and accelerations

for t!tn are known

Central difference formula for the velocity

Central difference formula for the acceleration

Discrete equilibrium at t=tn

Displacements at t=tn+1 follow from these equations as

Fehmi Cirak

Explicit Time Integration -2!

Provided that the mass matrix is diagonal the update of displacements and
velocities can be accomplished without solving any equations
Explicit time integration is very easy to implement. The disadvantage is
conditional stability. If the time step exceeds a critical value the solution will
grow unboundedly
!

Critical time step size

Longitudinal wave speed in solids

exact solution

Page 96

Fehmi Cirak

Semi-discrete Heat Equation -1!

Semi-discrete heat equation

!
!
!
!

Page 97

is the temperature vector and

is the heat capacity matrix
is the heat conductivity matrix
is the heat supply vector

its time derivative

Initial conditions

Family of time integrators

Fehmi Cirak

Semi-discrete Heat Equation -2!

Common names for the resulting methods

forward differences; forward Euler
trapezoidal rule; midpoint rule; Crank-Nicholson
backward differences; backward Euler

!
!
!

Explicit vs. implicit methods

!
!

For
For

the method is explicit

the method is implicit

Implementation: Predictor-corrector form

substituting in
known

Page 98

Fehmi Cirak

The Newmark Method -1!

For elastodynamics most widely used family of time integration schemes

Assume that the displacements, velocities, and accelerations for t!tn are
known

Implementation: a-form (according to Hughes)

!

Page 99

Unconditionally stable and undamped for

Compute predictors

Fehmi Cirak

The Newmark Method -2-

Page 100

To compute the accelerations at n+1 following equation needs to be solved

Fehmi Cirak

Coursework
Plates and Shells: Analysis and Computation (4D9)
Dr Fehmi Cirak
Deadline: The deadline for the report and software is 11 March 2010, 5pm
Estimated time to complete: 10 hours
Introductory lab session: 3 March 2010, 11-12
Form of submission: A typed report of at least three pages and a working MATLAB
implementation. The preferred form of submission is via email to fc286@. Make sure that
you submit the entire CEAKIT directory on your computer and not just the functions
you implemented.

Problem description
The objective of this coursework is to implement a finite element code for static and dynamic analysis of plate structures. There is a MATLAB finite element library CEAKIT
(Computational Engineering Analysis Kit) to be used for this coursework, which can be
obtained from:
http://www-g.eng.cam.ac.uk/csml/teaching/4d9/CEAKIT.tar.gz
The dowloaded file CEAKIT.tar.gz can be unpacked with tar -xzf CEAKIT.tar.gz,
which will create the CEAKIT directory containing several *.m files. Running the driver
function Ceakit in MATLAB will plot a deflected plate.

Tasks to be completed
1. Study the Ceakit.m driver and describe with few sentences the purpose of each
function called.
2. Implement a function which computes the load vector for uniform pressure loading.
3. Figure 1 shows the geometry of a plate to be analysed. The boundaries of the plate
are simply supported and the plate thickness is t = 0.2. The Youngs modulus
of the material is E = 35000 and the Poissons ratio is = 0.3. The plate is
loaded by uniform pressure loading of p = 0.003. Study the convergence of the
maximal displacements for fully and selectively reduced integrated finite elements.
The meshes to be used should have 4 4, 8 8, 16 16 and 32 32 elements.
4. Implement a function for computing the element mass matrix of a plate finite element and extend Ceakit.m for assembling the global mass matrix.
5. Implement the implicit Newmark time integration scheme.
1

Coursework

4D9

F Cirak

15.0

10.0

10.0

Figure 1:
6. The dynamics of the plate in Fig. 1 due to sudden uniform loading is to be studied.
The mass density of the material is = 2000. Apply a sudden uniform loading of
p = 0.003 and plot the evolution of the maximal displacements over time.
Note, it is not sufficient just to submit a working MATLAB implementation.
It is important that you submit a report, which addresses item by item each
point of the previous list. Do not forget to include the requested plots and
the implemented equations.