Lecture Note

THE FINITE ELEMENT METHOD

2014

IS 00:32:09 0*T409:00 2008

J: .... J.

Taein Yeo
School of Mechanical Engineering
University of Ulsan
Korea
Copyright ©2014 by Taein Yeo
All right reserved

2
Table of Contents

1. Introduction to FEM
1.1 Overall View of FEM for Structural Analysis
1.2 One Dimensional Finite Element Analysis
1.3 Two Dimensional Truss Analysis

2. Finite Element Formulation by Force Equilibrium

2.1 The Concept of Shape Function
2.2 Spring Element Revisited
2.3 Euler Beam Element

3. Approximate Methods
3.1 Review of Calculus of Variation
3.2 Approximate Methods

4. Continuum 2D Elements by Advanced Approach

4.1 Shape Functions in Physical Coordinates
4.2 Linear Quadrilateral Element
4.3 Shape Functions in Parametric Coordinates
4.4 Change of Variable for the Integral
4.5 Numerical Integration
4.6 Linear Triangular Element
4.7 Thermal Stresses and Vibration Problems
4.8 Higher Order Elements

5. Imposition of Constraints
5.1 Lagrange Multiplier Method
5.2 Imposition of Constraints onto a System of Equations

6. Axisymmetric and 3D Elements

6.1 Axisymmetric QUAD Element
6.2 Axisymmetric Triangular Element
6.3 3D Elements

7. Beam Elements

3
7.1 Euler Beam Element
7.2 Timoshenko Beam Element

8. Plate and Shell Elements

8.1 Kirchhoff Plate Element
8.2 Reissner-Mindlin Plate Element
8.3 Decomposition of Stiffness Matrix
8.4 Shell Element

9. Modeling and Miscellaneous Topics

10. Conduction Heat Transfer Analysis

10.1 Heat Conduction Equation
10.2 Weak Form
10.3 Finite Element Formulation

11. Nonlinear Finite Element Analysis

11.1 Solution of Nonlinear Equation
11.2 Sources of Nonlinearity
11.3 Finite Element Solution of Nonlinear System
11.4 Calculation of Jacobian Matrix
11.5 Finite Element Formulation

12. Time Integration

12.1 Hyperbolic Problems.
12.2 Parabolic Problems

13. Mixed Formulations

4
1. Introduction to FEM
1.1 Overall View of FEM for Structural Analysis

1) Problem to be analyzed

Fu F,«
fixed

F29

2) Mesh generation & finite element model

© •••

'

IP¬

3) Global & local node numbers S’

\4 © >2
1

2 r 3

17 r ®
33 4 ©

2
2
25 i< ©
3

Global node numbers 2

Local node numbers

4
4) Nodal displacement vectors
Nodal displacement vector of element #1 & #2 :
(1) (2) d\y
du du (i)
(i) (2)
dAy 1
4y d\y
U?
dix
4 2
a) (2) Jt t
d'iy IV ©
3

5) Unique displacement at common nodes

(1) (2) 0) (2)
Displacement components at global node #14 : dAx = dXx = dX4x, d4y = = dx4y

6) Nodal force vectors

Nodal force vector of element #1 & #2 :
(i) (2)

fu fu
(1) (2)
fry f\y At « \<D
• W(2> = \
(1) (2)
/k/4x
4 2
[Ay Ay t t
i\ ®
3
t

7) Construction of element stiffness matrix by F.E.M

Element shape &
Material properties
Element equation : [*](i) {d}(/) = {/}(;)
I4L
ML

5
 t
8) Force equilibrium at common nodes
4
t
Non unique nodal force vector
0) (2) (3) (4) (1) (2) (3) (4)

f\x * fix *fix * fu-> fiy * fly * fly * f\y •15

t ®
Externally applied forces: FUx, FUy l J (if & (2)
tSA-r
tX hx 1T fy 3
<Force equilibrium at nodes > 2 (i)

(1) (2) (3) (4)

uy\ 1 \Ju
0)
2
At node #14: f4x+ flx+f3x+ flx,= FUx 4
<fx 2
tX t
(1) (2) (3)
f*y+ f'y +Ay+ fy ~
(4)
i ©
3
(2) (4) (2) (4) t
At node #15: f3x+ f3x~ÿÿ fy'ÿ'fy=® 2~*

-
9) Global stiffness matrix, nodal displacement vectors and nodal force vectors

Global nodal displacement vector: {<i} = {ÿ}(1) +{ÿ}(2) -1 (net)

Global nodal force vector: {F} = +{/}(2) *ÿ {/}(nel)

J
r * • \l
\
W:

--
*
1
*
Global equation: jX]{r/} ={Fj !
* *•"
V.....4T
Assembly of global stiffness matrix: [F] = [&](]) + [&](2) b[ÿ](nel)

10) Boundary conditions

Prescribed displacements: jdj

,
/ £4
N.
\ F
t\b
X
Applied forces: |FJ

MMJW
[[*..] 1 F„

J*«] M \Fn
\

6
11) Calculation of nodal displacement by a solver
- Iteration method
- Elimination method

12) Deformed shape

/

1
&
%
13) Stress and strain calculation, etc.

Strain: —
{t/} > |w(x)} > } —
Stress: {cr| = [£)]{£ÿ}

Reaction forces: IF} =

14) General remarks

a) In finite element method, unknown variables are nodal force and displacement vectors defined only on
the nodes. Therefore, nodes should be placed on the points where forces are applied or displacements
are restricted. Distributed forces should be converted to equivalent nodal force vectors.
b) Due to the continuity of displacement, the neighboring elements sharing a node must have an unique
(1) (2) (i) (2)
displacement vector at the node and hence, d4x = dXx = dl4x, d4y = dXy = dUy .
c) However, nodal force vectors at a common node of the neighboring elements are not the same;
(1) (2) (3) (4) (1) (2) (3) (4)

f4x * fix *fix * fix’ fy * fly * fly * fly

Instead, their vector sum is equal to the external force applied on the node and hence, for example, force
equilibrium at node #14 gives
(1) (2) (3) (4) (1) (2) (3) (4)
At node #14, fx+fix+fx+f\x,~FUx, fy+fy +fy+fy ~

7
d) The number of displacement components at a node is called the nodal degrees of freedom(d.o.f.). The
total number of nodal displacement components of an element is called d.o.f. of the element. For example,
d.o.f. at a node of a 2D 3-node triangular element is 2 while d.o.f. of the element is 6.
e) For each nodal displacement component there should be a corresponding nodal force component. For
example, if there exists a rotational displacement at a node as for a beam element, so does a moment. The
number of total nodal displacement components of the system should be the same as the one of total
nodal force components.

8
1.2 One Dimensional Finite Element Anaylsis
A one dimensional system composed of two spring elements is shown in the figure with global node
numbers.

m
rr d3 = 1mm
P2=3N
i 1 *, 3 2

k = kx — k2 = 3 N / mm
1) Element equation

k = AE/L
rAAAA 2
M f
Figure shows a typical spring element with forces and displacements applied on the nodes 1, 2. It is
customary to designate them beginning from left to right. For the systematic symbolic notations, the
forces and displacements are taken to be positive when they are directed in the positive coordinate
directions.

In order to find an element equation that relates nodal forces with nodal displacements for this simple
spring element, the Hooke’s law can be used. Since there are two components in displacement and force
vector, the relation, i.e., the element equation will take the matrix form

fi Ki Ki I d\
A 21 .

a) Element type
Since there are two nodes in the element, linear spring element is to be chosen.

b) Force and displacement relationships

k
T +
i
W\Ai2 T

From elementary solid mechanics, using Hooke’s law we can directly write
—
T k(d2 ~dx).

9
c) Element stiffness matrix
We need to write the equation in terms of nodal force vector, noted fx, f2 instead of T.
A =-T, —fx = k(d2 — dx) at node #1,

f,=T, f2=k(d2~dy ) at node #2

The two equations give the element equation in matrix notation

A k -k\\dx
A rk k J [d2
d) Element equations using global node numbers
Firstly, the element equation is written for each element using the displacement and force vector defined
in the figure. The nodal displacement vectors are represented with global node number considering the
uniqueness of nodal displacements.

® f«) f(2) (2) f(T)

•AAAA-*
i 3
Wd2
(i)
/, kx kx dx
For element#!, (i)
/a —kx kx d2
(2)

For element #2,

/a
(2) (2)

Element stiffness matrix . which depends on geometry and material, relates displacement vector

{ j}(.) to nodal force vector {/}(/) such that {<ÿ}(() = {/}(,) holds. For a simple element
such as 1-D truss element, [ÿ](() can he obtained directly as in this example, but for more complicated
elements, approximation methods should be employed by assuming the solution to take a pre-specified
function form, the simplest one being the polynomial function.

Secondly, in anticipation of the convenience in the assembly, the nodal displacement components in the
element equation for the element #2 should be placed in the increasing order. To this end, firstly we
exchange the orders of the equation, which means the row rotation of the force vector components as well
as the stiffness matrix. To maintain the symmetry of the stiffness matrix, additional column rotation is
carried out for the stiffness matrix, which should be accompanied by the row rotation of the displacement

10
vector. We, then, finally have for the element #2
(2)

A Jc2 -k2 \d2

(2)

/a
_-k2 *2JU
2) Assembly
In order to get global stiffness matrix [K] , assembly of all the element stiffness matrices is carried out.
Modifying the nodal displacement vector so that they include all the nodal displacement in the system, we
write
(i)

kx 0 4 f
For element #1, 0 0 0 o
0)
-kx 0 4 A
'ooo
o
4 (2)
For element #2, 0 k2 -k2 A
(2)
0 -k2 k2 4 [ÿ3
with nodal displacement vector|d} = \dx d2 d3ÿ .
Adding the two equations gives
0)

kj 0-kx f dx f
(2)
0 k2 -k2 A = A
(1) (2)
-k\ —k2 kx + k2 [d3
/3 /3
+

Externally applied force Ft on the node are shown below. Note the global node number was used in the
figure.

0 0 F2
A
3 2

From the equilibrium equation of a nodal point /', the sum of the forces on the node must be equal to the
(1) (2) (1) (2)
externally applied force Ft on the node, and hence fx= F{, f2—F2, f3+f3 = F3 . Finally, the

11
global equation becomes
WM-«
where the symmetric stiffness matrix, the nodal force vector are defined, respectively, by
kj 0 -k\
[£] = o k2 -k2
-kÿ -k2 k\ + k2
F, nr
Using the numerical values given, the global equation of the system becomes
'3 0 -Site Fx
0 3-3 d2\ = \F2
-3 -3 6 4 .Fj

3) Boundary conditions applied

If {of} and |jp} are considered as the unknowns, the global equations have multiple solutions as the

number of unknowns, displacement or force components, is greater than the equation number.

In fact, we find Det1\K§ = 0 and, hence, [AT] is singular ! Mathematically, for a system of an
equations [/l]{x} = {&} in which a singular square matrix [A] and a column vector are given,
the solution jjcj of the equation does not exist or is not unique depending on j&j . The situation,

however, is somewhat different in our occasion where the unknowns are on the both side of the equation.
By taking the appropriate force and displacement B.C’s into consideration, a unique solution can be
found. Physically this corresponds to removing rigid body motion of the system. Without the application
of the boundary conditions,

At all the nodes of the body, either force or displacement boundary conditions should be prescribed.
Applying the boundary conditions for the present example, we write
'3 0 -3] f 0 ] f Fx
0 3-3 4H3
-3 -3 6 i F;
The unknowns are on the both side of the equation and their number is the same as the equation’s.
Unique solution, therefore, does exist even with the singular matrix !

4) Solution of the global equation

To adopt the partitioning method, it is required to group the unknowns separately from the knowns in

12
terms of displacement and force components. To do this, the row rotation should be done on the
stiffness matrix as well as the column rotations to maintain the symmetry of the stiffness matrix. In the
event, the global equation will end up with
IMi [*Llfpn JWl
[Mi MJlMdlHi
where \K\.. is the submatrix, and {r/}2 andÿFÿ are the unknowns, jflf and are the known

boundary conditions.

From the 2nd equation, we have

or

and from the 1st equation, we have the reaction forces

In our case, rearranging the equation we have

'3-30 0 Fx
-3 6 -3 1
0-3 3 d2 3
the solutions of which are = —3, F3 = 0, d2 = 2

5) Calculation of stresses
Stress in element #2 are
<7 = Es _E(d2x-dix)
4
When using a commercial FEM software to analysis the physical system shown, we determine a
mathematical model where three rods with uniform area are chosen. The associated finite element model
is displayed with three line elements with boundary conditions. The displacement solution is piecewise
linear whereas the stress is constant in an element. FEM analysis can predicts quantities the exactly at a
certain location in an element when the analysis is done appropriately.

13
 p p
> X
Finite Element Model
Mathematical Model

— CT* — FEM /
/

Exact J Exact
/

i
I

i 4 4

6) Interpretation of the results

Since the finite element analysis is done using numerical numbers by a computer, the input and output
units should be taken care of by the user. For example, A[mm2], L\m\, E[N/m2\, E[iV] are
used in the ID analysis, the displacement would be in the unit of fj,m

14
PROBLEMS
Problem 1
a) Determine the displacement at the node where load P is applied for the system shown by finite
element method. (Don’t use a commercial FEM program; use only a calculator or a computer program
coded by yourself to deal with the matrices).
b) Using NASTRAN or ABAQUS or any other FEM softwares available, solve the problem again. You
should show the FEM model, results of the deformed shape, output file with the solution clearly marked.
Note: Do not attach the whole output file; just one page showing the solution is enough.
c) Discuss about your numerical results.

P
(D K
Fu
KAA/V*!
Qk3 m
kx = 1000 N / m, k2 = 600TV / m, Jt3 = 400JV / m , = 2000N I m, P = 10N

Problem 2
a) Determine the displacement at the node where load P is applied for the system shown by finite
element method. (Don’t use a commercial FEM program; use only a calculator or a computer program
coded by yourself to deal with the matrices).
b) Using NASTRAN or ABAQUS or any other FEM softwares available, solve the problem again. You
should show the FEM model, results of the deformed shape, output file with the solution clearly marked.
Note: Do not attach the whole output file; just one page showing the solution is enough.
c) Discuss about your numerical results.
= 1000 N / m, k2 = 600 N / m, k2 = 400 N / m , k4 = 1200N / m , P = 1(W , dA = -8mm
P

Fu
1
LKAAA/V\A/Vl
CD *3 d<

15
1.3 Two Dimensional Truss Analysis
1.3.1 Coordinate Transformation of a Vector Y
y
A vector V is to be described using two coordinate systems,
—
for example, the global X Y coordinate system and
—
a local X y coordinate system, which have angle 6
between them. The relation between the components
v
X

vx
vx,vy in the local coordinates and , vY in the global j *A e x
coordinates can be found by the equation

V = VXl + VyJ
= vx (i cos 0 - jsin 0) + vY (i sin 0 + jcos 0)
= (vÿ cos0 + Vy sin 0)i + (-vÿ. sin 0 + vY cos 0) j
= + v,j
From the last two equations, we have the relations for the components

vx = vx cos 6 + vY sin G , vy= —vx sin 6 + vY cos 6

Form equation, the coordinate transformation of a vector is given by in matrix form

c 5 Iv*
vy -S C [Vy
where c and s notes for COS 6 and sin# , respectively. With the help of the relation, coordinate
transformation of two vectors V, , V2 can be written directly

vi* c s 0 0 vi*
vix -sc 0 0 %
V2, 0 0 c s V2X
v2y 0 0 -sc v2Y

1.3.2 Coordinate Transformation of a Stiffness Matrix

Consider a truss element in two dimensional space placed with an angle 6 from the global X -axis as
shown the figure.
iY
2\ d
y k ’ f2y
2y

&M- X

0
>
X

16
The axial rigidity is given by k = AE / L . The element equation developed in the previous article was
in reference to a coordinate system of which x -axis is aligned with the element axis. To make use of it,
—
we place a local X y coordinate system which are aligned with the element axis. With reference to the
local coordinate system, the element equation is rewritten as

fu
-k k j|rf2x A
which is the force-displace relation of thex components only. But the problem at hand is two

dimensional, the displacement vector at a node i has two components dix , diy with a local

—
x y coordinate system. If we include y components in the equation, we have

k 0 -k OlK fu
0 0 0 0 dXy fy
'
-k 0 k 0 d2x fu
ooo OJ[A fir
or in the matrix form,
fcfal-fr,}
where ] is the local element stiffness matrix defined by

'k 0 -it O'

0 0 0 0
M- -k 0 k 0
0 0 0 0

d<y and {/}-[/u Ay /2,T are the nodal

displacement and force vector, respectively with reference to the local coordinates.

From equation, transformation from the global components to local components for the displacements can
be written as

du c s 0 0 dxx
d\y -s c 0 0 d\Y
A 0 0 c s dix
d2y 0 0 -sc d2r
or, symbolically

17
 W-P1W.1
where {dG } = \dxx dXY d2X d2Ylf are the nodal displacement vector with reference to the

global coordinates. \T] is the coordinate transformation matrix defined by

c s 0 0

m= -s
0
c 0 0
0 c s
0 0 -sc

Similarly, for the force vector we write

where {fG} = \f\x f\y fix fiY ]? are the nodal force vector with reference to the global

coordinates. Plugging the representations for displacement and force in the local element equation, we
have

wmi-pii/.}-

Premultiplying the equation by the inverse of[!T] , we have

prwmw/.}

The inverse of the coordinate transformation matrix can be obtained easily since it is the orthonormal

matrix. In general, an orthonomal matrix [g] is defined to have the relation

1, ifi = j
{Q)]{Q)r 0, ifi*j

where {Q}t representss a column vector of [<2] = [{£?}i {Q}I * •]•

For orthonomal matrix [Q] , it can be proved the inverse is obtained by its transpose, hence

[QX'=M-

18
We note the coordinate transformation matrix [!T] is a orthonormal matrix and, therefore, have the

relation \r] 1
= [jÿ , which enable us write the force-displacement relation in terms of the global

components. From equation, the global equation is written as

fclKM/c}
where the global element stiffness matrix is

c2 cs -c2 -cs
s2 -cs -s2
PbHirwPl-PTMPI-* sym. c2 cs
s2

1.3.3 2D Truss Analysis

To show how to analyze a truss structure, consider

the system shown in the figure. The displacement
at node 2 of is to calculated for P
kx=k2=k = AN I mm .
= 9N and
a

1
Y

©
K
X 2

For element # 1, the stiffness matrix and the

corresponding nodal displacement vector are
>
0 -k O'
"
k d\x X
dlY
w«,,= 0
-k
0 0
0 k 0
!-H„= d2X
0 0 0 0 d2Y
We assume a local coordinate system for element # 2 is positioned so that the angle becomes 6 = 120°

and hence, c2 = 1 / 4, cs = —y/3 / 4, s2 = 3 / 4 . From equation, the stiffness matrix and the

corresponding nodal displacement vector of element # 2 are

19
 1 -V3 -1 V3
V3
a, =7 ->/3
-1
3
V3 1 -V3
-3

>/3 -3 -V3 3
Note we used the global node number for the nodal displacements in the eqations. The assembly of the
element stiffness matrix gives the global stiffness matrix

'4 0 -4 0 0 0
0 0 0 0 0

W-7
5
—s/3 3
-1
yfe
VJ
-3
1 -S
3
along with the nodal vectors

M=Kr d« dix dn d,x d,rf ,

{*•}=[*;, Flr F„ F„ F„ F„f

The boundary conditions are written in the global coordinate system as follows;

-force:
—
- displacement : dlx = dlY d3X = d3Y —
F2X = Psm30° = P /2, F2Y = Pcos30° = y/3P/2
0

These are imposed on the global equation = {F} and the resulting matrix equation can be

solved using the matrix portioning method. When the numerical values are used for the portioned
matrix euation, we have the system of linear equations for the unknown displacement

5 -A/TI (d2x 9 1
y/3 3 \d2r. w
the solution of the equation is given by

1
d2X 2

d2Y
and the finally the forces are

20
 2
Flx -6} \F3X 2

Fir .OJ’ Ur 2

Exercise 1
What are the units of the solutions in the above example? What would the numerical values of the
solutions and their units be if a different unit, for example k = 4000 N / m , is used for the spring
constants while the force unit remains unchanged ?

Exercise 2
Redo the same analysis of the example by placing the local coordinate system for the element #2 so that
the x-axis is directed from node 3 to node 2.

PROBLEMS

Problem 1
a) For a 2-Dimensional truss structure shown, find the member forces and the deflection at the center B
by using finite element method. (Don’t use a commercial FEM program). Hint: For the sake of
simplicity, make use of the symmetry of the system.
b) Using NASTRAN or ABAQUS or any other FEM software available, solve the problem again. You
should show the FEM model, results of the deformed shape, stresses in the members, output data file with
the solution clearly marked. Note : you can submit the just page of the output file with answer in it, i.e.,
do not include the whole output in you report.
c) Solve the problem again to get the member forces by means of the joint or section method from statics.
d) Make some discussions for your results above.

A = 300mm2
30° B 30V E = 200GPa
2000mm 2000 mm

F = 500 N

Problem 2
a) For a 2-Dimensional truss structure shown, find the member forces and the deflection A between the

21
point a,b using finite element method. (Don’t use a commercial FEM program). Hint: For the sake of
simplicity, make use of the symmetry of the system.
b) Using NASTRAN or ABAQUS or any other FEM software available, solve the problem again. You
should show the FEM model, results of the deformed shape, stresses in the members, output data file with
the solution clearly marked. Note : you can submit the just page of the output file with answer in it, i.e.,
do not include the whole output in you report.
c) Solve the problem again to get the member forces by means of the joint or section method from statics.
d) Make some discussions for your results above.

A = 300mm2
E = 200GPa
a
F = 50KN
60°
b
F = 50 KN
V
L = 2000mm

Problem 3
a) For the truss shown, write the stiffness matrix \kx ]4 4
of element #1 with nodal d.o.f.

{ÿ|= \dyx dXY d2x d2 Y using coordinate transformation.

b) Assemble the stiffness matrices [&,] and [&2] to fmd the global stiffness matrix [Aÿ]6x6 with

nodal d.o.f. |i/| dx x dx y d2 x d2 y dx |

y

c) Apply the boundary conditions and obtain displacement d2Y for which matrix partitioning method

must be used.

F = 20

Y
60 h [2*.-5 (2) 3

t k =10 Q
i
w,
m

22
2. Finite Element Formulation by Force Equilibrium
2.1 The Concept of Shape Function
When the nodal displacements dx ,d2 of a 2-node bar element are obtained, the displacement variation
inside the element can be assumed on a reasonable sense to vary linearly as shown in the figure. In other
words, the displacement variation u(x) can be obtained by employing the method of interpolation with
the set of the known quantities of in the table.
Table 1
d2
X *1 *2 ir u(x)
u d2 <*2 1 2
L
x
We, therefore, assume a linear interpolating function for u(x) of the form
u = ax + b
Solving for a, b and putting them back in the interpolating equation, we get

u=—-
w,
*1"*2
—
u2- x + U\X2 ~U2X\ or
x2-xl
ll
_ x2 -X

*2"*.
w.+
X, -X
xl-x2
U2
The equation can be written as
u = Nx (x)u, + N2 (X)U2
where
. x, - X
.r ,
— . x, - X
(x) = - and N2=—-
x2-x,
T

x,-x2
Nt (x) is called interpolating function or shape function and is a linear function for the 2-node element as
shown in the figure. In fact, Af (x) ' can be easily written using Lagrange polynomial and they have
the following characteristics

Characteristics of shape Function

i)i>,(x)=i
i=i

2)NXXj) =
1
0
if
if
i~j
i*j
L
",
N(x)

X]
. N(x)

X
3) Nj(x) is function of X

When X, = 0, x2 = L , the equation in matrix form becoms

!]0 or u .[», »j(jj

23
where Nl(x) = l—x/ L and N2 -x!L.
Exercise 1 Determine the axial displacement u and stress <7 of the 2-node bar element at the
distance of L/3 from the node 1 when d{ = 2d2 = 40/im , L = 20 mm .

2.2 Spring Element Revisited

Consider a 2-node spring element shown. Now, the stiffness matrix is to be derived using the concept of
the shape function.
x
d, d2
A. r
r.•AAAAÿ >
fx ft
In other words, By applying the force equilibrium at each node, we can write the element equation, in
which the nodal displacements are the unknowns.
1) Displacement interpolation
We first assume the displacement variation in the element by means of the interpolating function. The
nodal displacements dx , d2 are temporarily regarded to be known quantities.

2) Strain-displacement
-HM* - MS
From the assumed displacement variation, strain variation can be obtained.

_du _d2-dx or
S~~dx~ L
£
4H <1
3) Stress-strain
Stress variation is given by Hooke’s law

a = Es = —E
L
(d2-dl) or <7
-fi" <;}
In this case, the stress is constant through the element.

4) Internal and External Force Equilibrium

At a node, the external forces should be equilibrated by the internal forces. For 2D and 3D element, the

24
internal forces on a node due to internal stresses are difficult to find. The best approximate relation can
be obtained by applying Ritz or Galerkin method, which will be discussed later. For this very simple ID
spring element, the internal force at a node#l is obtained simply by multiplying the stress <T, at the

node by area A . Hence, from equation the internal nodal force at node #1 is given by

flf.=Aal-k[-1 l]jd24 with k = AE / L

This must be balanced by the external force fx such that

fx + Acrx =0, k(d2-dx) = -fx,

Similarly, at node #2, the nodal force f2 must be balanced by the internal stress cr2 such that

f2-Acr2 =0, k(d2-dx) = f2

Finally, writing the equations in matrix form, we have the element equation

k ~k]\dx f
-* *JU A

2.3 Euler Beam Elements

2.3.1 Sign Conventions
The force-displacement relation for a beam is given by
EIv" = M, EIvm = V, EIvm' = q
and relations between internal shear force, internal moment, external distributed force are
V = dM / dx, q = dV/dx,
where moment M , shear force V , traction q and the deflection v are defined in the figure.

v
>M(
vM
X
* +v

fTTirm +<?«

25
2.3.2 Derivation of Beam Elements by Direct Method
In finite element method, the external forces are defined only at the nodes. Between the nodes, therefore,
there exist no distributed forces, i.e., q = 0 and the differential equation demand the deflection v(x)
be varied as a 3rd order polynomial. In the mean time, the 3rd order polynomial can be determined
uniquely when 4 displacements are specified at the nodes. If there is a discontinuity in curvature at the
boundary nodes as shown in the figure, the calculated results would be wrong, since beam bending is
governed by the rotation.

The most appropriate choice would be displacementÿ and rotation J3 at a node, which means two

node 2D beam element has four degrees of freedom. With choice of the rotation as an additional degree
of freedom, the continuity of the cross sectional rotation at the node is achieved as well as the translation
between the neighboring elements. In contrast, if we choose the translation dy at all the nodes of a 4
node-element as d.o.f, we lost the continuity of the rotation at the boundary nodes.

©
i Hr
r
To write v(x) in terms of the displacement dy and rotation f3 , we, therefore, begin with a polynomial
with 4 coefficients;
v(x) = a,x3 + a2x2 + a3x + aA
Applying v(0) = dly, v '(()) = /?,, v(L) = d2y, V\L) = J32, we have

where
M=K A d2> is the nodal displacement vector and

[iV] = [JV, N2 N3 TV4 ] is the shape function defined by

TV, (x) = (2x3 - 3x2Z- + Z3 )/ Z3

TV2 (X) = (x3Z - 2x2!2 + xZ3 )/ L3
N3(X) = (-2X3 +3X2L)/L3

TV4 (X) = {X3L- X2L2 )/ L3

Applying the formula from the elementary solid mechanics, it follows that

26
 = V(0) = £A"(0) = - j(l2d„ + 6LA -
A, + )
m, = -A/(0) = -£A = -~(6W]l +4/?/?l - 6M2, + 22=/?,)

/„ =-F(£) = -£/v’(2) =-j(-12<f„ -61/?, +12d;, -6LA)

m2 = U(L) = Eh-(L) = ~((,LdXlÿ2l:P, - + 4 ///?,)

In matrix notation, we can write the beam element equation in matrix form

MM ={/)
where the element stiffness matrix of beam element is given by
6L -12 6L
12
El 41) -6L 2L2
[*]-•£ 12 -6Z
57m
4Z2
and {/}=[4 Ay m2J is the nodal force vector.

Exercise 1
When a beam element is positioned with an angle 6 from the x-axis, the stiffness matrix with reference
to the global coordinates can be written in the form [&G] = \Tÿ [A,][7"] Write the appropriate
•

coordinate transformation matrix \T] and the stiffness matrix [&, ] in the local coordinates.

L y
\f2xAx
x m2,p2
\e AY’AY
f\x>duXÿnmvP
flY’d\Y X

2.3.3 Distributed Loads on Beam Elements

For a uniformly distributed load on a beam element, the equivalent concentrated loads can be found from
reversing the directions of the reaction forces from the fixed-fixed support as in the figure.

27
 n
m
4o <hk
nuiTT* &
L
m aJL
12 12

M
2 2

As an example, in case two beam elements are used the distributed force can be replaced by the force
system as shown in the figure. However, the deformed shape of the beam from finite element analysis is
not expected to coincide with exact solution, although the reaction forces are the same.

%(a+b) %b
iinnnnni 2 2

7 %b2
» 12
§5
1 1:
a b
a b

Exercise 2 Find the nodal force vector for a beam element with linearly distributed loads as shown.

q0 =10 KN/m x

2w
4 2« 4
2.3.4 2D Beam Element with Axial Flexibility
We are going to derive the element equation of the 2D 2-node Euler beam element with the nodal degree
of freedom as shown.
We first construct the stiffness matrix of a two dimensional 2-node beam element in the local
X — y coordinates that has an axial flexibility EA , hence, with the nodal degrees of freedom

dx,dy,p. y

fxx’
mx,px i m2,P2
9
fix’ d2x
x

L
f\yAy fly Ay
By use of the appropriate coordinate transformation, we can get the stiffness matrix relating the 6 nodal

28
displacement components to 6 nodal forces.
2.3.5 General 3D Beam Element
m2y’Ply
\ m2x’Plx
"hy’Ply
flyAÿ 2
y *rf2**u
\ ”hx’Plx X ffizAz
Y fly’dij\
1 %Ax 2 t
m2z’P2z
yrfxY AzAz
X
t
mi Z’Plz
ZA
In case of general 3D Euler beam element, d.o.f of a node is six. With the nodal displacement and force
vector

{di}—\d\x dXy du PXx PXy PXz d2x d2y d2z P2x P2y P2i\
{f,}=\_Ax Ay Ax mix mX y ™lz fix fly fix ™2x ™2y ™lzj >

the element equation is given by

WWW/}

where the information required for the construction of [£, ] would be

a) length L and the angles CC,P,y with the axes, or coordinates x,,x2 of the end nodes .
b) axial flexibility EA , torsional flexibility GJ , two bending flexibilities El

Coordinate transformation of vector components in 3D

a\\ a\i an VJT

v> = a2\ a22 a23 vr or {vz}=[«]{vG}
V, «31 a!2 «33 VZ
where the component of
global /-axis.
[a] is au — COS 6U and 0U is the angle between the local i -axis and the

The stiffness matrix with reference to the global coordinates can be written in the form

M-PTMP1
29
with the corresponding nodal force and displacement vector

{da}-\d\x d\Y dxz fixx fiXY (ixz d2x d2Y d2Z (32X fi2Y P2Z\
{fG}=[fx fir f\z m\X m\Y miz fix fir fiz m2X m2Y mizf
Exercise 3 Write the appropriate coordinate transformation matrix [T] and the stiffness matrix \k, ]
for a general 3D beam element

2.3.6 Use of Different Types of Elements

A structure, composed of two members, is subject to a force in the middle of a member as shown. It is
required to find the deflection at B. Although the problem can be solved by using beam elements only,
the best finite element model would be composed of 2 beam elements and one truss element.

Exercise 4
Find the global stiffness matrix for the system shown with the minimum
number of elements. The members should be able to experience the axial
IT. 7*>
'B
deflection as well as bending.
§
§
§
PROBLEMS

Problem 1
For a statically indeterminate system shown, find the reaction forces at x=0, 500mm and the displacement
at x=250, 1000, 1500 mm. The beam has the cross section of bxh = 20mm x 50mm and
E = 200GPa.
a) Use an analytical method to solve the problem. (Any method will do such as energy method,
integration method or superposition method.)
b) Use the finite element method with 2 Euler beam elements to solve the problem. (Don’t use a
commercial FEM program; use calculator only).
c) Use a commercial FEM software to solve the problem 1) with the distributed forces and 2) then with
the equivalent concentrated forces.
d) Compare the displacements from the methods with some comments.
w = 20KN / m
x
Ullllll
SN
»
500 mm 1000mm

30
Problem 2
For a beam with El = 9.2 x 106 Nm2 , find the reaction at the supports and maximum deflection using
1) the equivalent forces and 2) distributed forces.
If there are differences between the results from the two methods, write your reasoning about them.

q=20KN/m
P=2000KN

1 L=2000mm
L L

Problem 3
A structure, composed of two members with the same cross sectional shape, is subject to a force at the

middle of member 1 as shown. El = 16 x 104 Nm2 , AE = 200 x 106 N

a) Find the deflection 8X , Sy at node B with an elementary solid mechanics.

b) For finite element analysis, choose the appropriate elements and write the element stiffness matrices in
the global coordinates.
F = 10KN
c) Assemble them to construct the global stiffness matrix.
i A I B
d) Show the global equation with boundary conditions applied.
The number of the unknowns must equal to the equation number.
e) Solve the equation to get the deflection 8x,8y at node B
1
1
Z, = 100mm

f) Find the deflection

FEM software.
8x,8y at node B using a commercial

g) Make some discussions about the results.

1
II
L2 = 200mm

1
1 C

Problem 4

a) For a cantilever beam shown, find the displacement 8y, 8Z and the distortional rotation (f>x at the

tip using a commercial FEM program with the beam offset, b = 50mm, h = 100mm, L = 1000mm ,

31
E = 200GPa , P = IKN.
b) Compare the numerical result with analytical solution (If analytical solution is not available, you can
do the finite element analysis again without using the offset method for the purpose of comparison.).

*.y
p ty
£ 1 z_ b \p
L
1h

32
3. Approximation Methods
3.1 Review of Calculus of Variation
3.1.1 The History of Variational Problem
Brachistochrone is Greek for "shortest time". The brachistochrone problem is to find the curve that will
yield the shortest possible time for the decent of an object. This problem was originally posed as a
challenge to other mathematicians by John Bernoulli in 1696.

y
D A :Start

h
x
Q

/ C :Finish

Mathmatical formulation for the problem leads to minimize time

n > fSfc
y/2g h yjh-y
where /, h are the horizontal distance and the height, respectively. The resulting differential equation
is called Euler Equation

dx = -
yjc-h + y
The solution of the differential equation is the curve of shortest time which happens to be a cycloid. If
we take one-half of a cycloid and turn it upside-down we get the brachistochrone. Notice that it is
vertical at the start to get up lots of initial speed and then flattens out at the end. This path has another
interesting property, namely if the ride could somehow be started from rest at any point B on the curve,
the ride would last the same length of time as if it started from point A.

3.1.2 Differential Equation

A function whose derivatives are continuous up to m-th order is said to have Cm continuity. For
example, the function y — f (x) defined by

y= f\ =ai(x-b)2, x<b ax > a2 > 0

f2 = a2(x-b)2, x>b’
has C1 continuity.

33
 y

fir 3,(X-Z>)2
\ f2ÿ2(x-b)2 f2=2a2(x-b)
b X X x

y
yf[~- 2ax(x-b)
fl=2a2
x f> |
0 /2"=0 X
infinite

Mathematical formulation of a physical system can be given in differential forms:

a) Governing differential equation

where <f) is the state variable, L2m is a differential operator with up to (2m)th order
b) Boundary conditions
on i = 1,2,...,2m
where Bt is a differential operator. The order of Bt is from 0-th up to (2m-l)th. When it is less or equal
to (m-l)th order, it is called geometric(essential, Dirichlet) B.C’s. When it is greater or equal to m-th
order, it is called force(natural, Neumann) B.C’s.
c) Initial conditions
For example, 3-D elasticity problems can be described by Navier’s equation
d2u
fN2u + (A + //)V(V •u) + b = p
dt2
where displacement filed U is differentiable with up to second order in space and time. An example of

the geometric B.C. is u = 0 on . When traction force t is given, the B.C. is t = fs(du)n on

rCT , respectively.

Example 1. Beam bending problem

Governing equation is

d2 d2v
dx2' dx2 =
~ El f, 0 <x<L

On a boundary, either V or V(= EIvxxx ) and either v' or M(= EIv") should be prescribed.
The possible combinations of the boundary conditions on a boundary are, therefore, following;

34
 v and v'; v and M ; v' and V ; V and M
On the contrary, V and V can not be prescribed on a boundary.

Example 2. Rectangular plate

Governing equation is
DV*w = p (0<x<a,0<y<b)
and B.C.’s are; y
On a boundary, for example X = a , either 2L
z Mx A
My Mxy/o
dv2 W (l-v)w
w or VX\=D dx
+— xyy

y
and wx or Mx(= D(wxx +VWyy)ÿj must be prescribed.

3.1.3 Calculus of Variation

Let 3 be a set of functions f(x) satisfying certain conditions(for example, existence of continuous
first derivative at all points in an interval a < x<b, satisfaction of essential boundary conditions, etc.).
Then any quantity II which takes on a specific numerical value corresponding to each function /(x)
in 3 is said to be a functional on set 3 . Examples are

n= jV) dx, n = {{*V(*)/M-[/«f} dx

Calculus of variation deals with determination of the function y = f(x) that makes the maxima or
minima(extreme, stationary points) of the functional II .

If the function y = /(x) in equation is used to calculate the functional I = J F(x, y, y', y”) dx
* , it

would give a definite value. Hence, determining y(x) that make / stationary value is termed C1
variational problem. However, if the same function is used to calculate the functional

I=ÿF(x,y,y',y\y’)dx
it can not be calculated. In this case y = f(x) should be C2 and determination of >*(x) that make
/ stationary value is termed C2 variational problem.

Example 1. In static problem, the variatioal problem can be posed as

35
 Maximize or Minimize / =
£ F(x, u, u')dv with geometric B.C’s.

The posed problem is, therefore, C° variational problem while u should be of C1 continuity in the
differential representation.

Consider a functional that has up to m-th order derivatives of a function in it. In order for the functional to
have a definite value, the function must be up to (m-l)th order continuous and determination of the
]
function is called C'" variational problem.

To find the function form which makes the stationary value of functional FI , we take a variation of the
function j'(x) about its solution. We define the variation 6y as an arbitrary small but sufficiently
smooth function that is zero at the essential boundary conditions. Then the variation of the
functional <5TI corresponding to 8>’ , should be zero for all possible 8y , because the variation is taken
about its extreme state.

Consider the general form of C° variational problem with I defined by

1= [2F(x,y,y')dx
in which /should be minimized for all possible variations in y, which means Sy and Sy' are
independent each other. Taking variation and putting it to zero, we have

bI = d[2F(x,y,y')dx
= [26F(x,y,y')dx
Since we can treat X, y,y 1 being independent, chain rule gives
OF . DF _ dF s ,
SF =
dx — — —
<Jx + Sy+ Sy
dy dy'
Variation takes place only on the function itself not on X and Sx = 0 . Since the variation and
differentiation are independent, it follows that

The equation becomes

Jx> L dy dy
which by integrating by parts, leads to

36
 x2

SI<
*, dF d dF
dy dx\dy — dF
- 8ydc+ -dy
8y
— xi
=0 \/Sy

or

i
dF d dF
dy dx\dy —
5ydx+
dF(x2)
dy
f-Sy(x2)~
dF(xx)
dy
7~dy(x,) = 0 VSy

dF d dF
Euler Equation : - =0 , Governing Equation in Physical System
dy dx\dy'
Boundary Condition :
Natural B.C’s Essential B.C’s

At* = JC,, —=
dF
- 0 or
dy'
5y = 0

and
dF
AtX = X2, -= 0 or 5y = 0
dy'

Exercise 1. Euler equation of a C° variational problem

Obtain the Euler equation and boundary conditions relevant to the determination of extremals of the

integral I= - yy'+ y} dx

3.1.4 Variational Statements For Lumped Systems

Consider a lumped mass supported by springs and subject to a force f(t) . An mathematical
description of the system is given by a differential equation
mx + kx = f

Alternatively, the system can be described by a variational statement. With the force f given, we
assume the force field is conservative and construct a potential energy V which satisfies f —
= VV .
Then, the potential energy V can be obtained by

V= -ÿi-du K 1(0
1

When the force is a function of time only, i.e., f(0 , m H

*3
equation gives the potential energy V . :
\
f(0
% 37
 V = -fu

Kinetic energy is given by

u(o,n n m
Strain energy is given by

u(t)+Su(t)\
where A( = A, (u) and ® represents
product of 2nd order tensor k and vector (1st order tensor) u.
i
I t
Then, using Hamilton’s principle, we can define the total potential

n = {T -(U +vj\dt
which should be minimized or maximized, i.e., stationary, for the real equilibrium path u . Therefore,
in the application of Hamilton’s principle, the functional to be stationary happens to be potential II .

The following examples shows the differential equations and variational statement are equivalent to each
other.

Example 3. Newton’s 2nd law of motion

Consider a mass m subject to a constant force f . Assume
also the position X of the mass is known at beginning and final time of the motion, hence the essential
boundary(initial) conditions are X = X, at /, and X = X2 at t2.

Since U = 0 and Hamilton’s principle gives the total potential

n =ÿ{T-V}dt with T= and E= -J[f-Jx

which should be minimized.

Since the variation Sx take place around the correct path x of the motion, the external force f(t)
remains constant when we take variation of V . That is, variation of the potential energy V = f •X —
becomes —
8V = f •Sx .

When the external force is dependent on both space and time, i.e., f (x,f) , we still prove
SV = VV -Sx = f -Sx . — When using Hamilton’s principle for a system with space dependent

38
external force, we can use
of variation
—
V = f-u with the premise of taking f(x,/) to be constant in the course

Taking variation and setting it to zero, we must have

m = s£(T-V)dt = 0
Since variation has nothing to do with differential or integration, we can write

m= [\8T-8V)dt = 0
Variation of the kinetic energy becomes
1 1
ST = -mSx-x+—m x-Sx = mx-Sx
2 2
The equation becomes

Sn= |f2[mx-ÿ(<ÿx) + f -Sxÿit = j£2[-mx + f]-£xt#+|mx-£x||2

In the equation, Sx was replaced by d(5x)/ dt , since variation and differentiation are independent
of each other. We require all the candidate X satisfy the initial and final real value at t = tx,t2 which
means <?x = 0 at t = tx,t2 and hence,

<5TI= |2[-wx + f]-£>X£fr = 0 \/5x

so that

[-wx + f]-ÿx = 0 V£x

From the equation, we finally obtain the Newton’s second law

mx = f

Therefore, Hamilton’s principle, which happen to be a variational formulation of the problem, is the
another representation of the Newton’s 2nd law.

Example 4. Elongation of a linear spring

Consider a spring with a force F(jt) . Assume F(t) is
u x
F
changed to the final value F2 from zero in quasi-static condition k
and the corresponding extension also reach the final value x2 .
The solid curve represents the real path taken by the spring and
x(t)+Sx(t)
the dotted curve a disturbed path by an arbitrary, but small
virtual variation. Note the variation is zero at the initial and i
//\X(t)
i

7*
the final time since X is prescribed.
Strain and potential energy are given by
1
U = - kx(tf and V --W = —Fx(t)

where W is the work done by external forces F . The variational statement is:

Minimize or Maximize n(o =

£{u + v}dt = |2{u - wjdt

Since <5TI =
£{kxSx - F8x}dt = F}Sxdt = 0 V8x(t)

we have
kx = F

This example shows that Hamilton’s principle or a variational method leads to force equilibrium, i.e.,
the Newton’s 2nd law.

It should be noted that the formulation has nothing to do with the principle of conservation of energy.
Although the strain energy U is identical in both principles, the potential energy V in variational
principle and the external work W in principle of conservation of energy are not the same in their
magnitude. When the potential energy V is constructed in the variational principle, the external force
F is assumed constant. In contrast, in the construction of external work W for energy principle, the
external force F is assumed to vary in the quasi-static state. For example, if the energy principle is
1
applied to the one dimensional spring of Example 2, work done by external force is given by W = Fx .—
By the conservation of energy principle, then, it follows U = W which leads to kx = F .

As can be seen in the examples, the variational principle concerns the equilibrium state of a physical
phenomenon, which will be manifested later in article 3.1.4.3 and 3.1.4.4. It, therefore, can be applied to
a non-conservative system, i.e., plasticity problem. This is why the finite element method, developed on
the basis of the variational principle, can used to find the solution of any kinds of problem of which the
physical phenomenon can be described by a equation.

In this static problem, the integrand of n has no time derivatves and there is no need to integrate in
time as in example 3. For the purpose of simplicity, therefore, the functional can be constructed without
the time integration. Variation, however, is considered to occur in time space. For example, for the
single spring problem of the example, we can write the functional

40
 1
n = U + V = -kx2-Fx
2
Then,
Sn = kxSx-fFSx = (kx-F)Sx = 0 VSx or kx = F

This concept can be extended to general 3D static elastic problems. Form Hamilton’s principle,
n would be in the form of double integration, i.e., space and time integration with field variable
u = u(x,t) . The integrand, however, has no time derivatives and there is no need to integrate in time.
We construct the functional II(u(x,/)) which is volume integration only but still, variation is considered
to occur both in time and space, i.e., 5u(x,f) . In fact this method leads to a convenience in static
system and its validity will be clear to be true later when the variational formulation is derived directly
from the differential equation.

For a lumped system with multiple degrees of freedom or discretized system in a static equilibrium can be
described by a variational statement:

Minimize II = ~\dÿ {/} V \Sd} with the appropriate constraints

The corresponding equilibrium equation can be written in the matrix equation

In summary, we have shown that the differential or equilibrium equation relating force and displacement
of a system can be replaced by stationary condition of a functional. In this case, the appropriate
functional / is found to be the total potential II of the system by Hamilton’s principle.

The dynamic system shown in the figure can be considered to be a static problem when we treat the
inertia force as one of the external forces. The external force is, therefore, consisted of the actual
external force f and Coriolis force fb and hence, F = f + fb with fb = —m\i . Potential energy
is

F = -(f+fb)-u
The Hamilton’s principle reduces to minimizing n = U+ V.

Exercise 2
Write the potential II in matrix notation, and by minimizing II , show it leads to

41
 [-M+MM-I/1

3.1.5 Variational Statement for a Continuous System

Consider a deformable body in space subject to a conservative force.
The potential energy and the kinetic energy u (f,x)
are written as
f
V= -[[idndv
P,[C]
Tÿ\[p\ÿdv
x
The strain energy is z
u
where

Using Hamilton’s principle, we can get the functional form for the variational formulation as

n= {T -{U + V)}dt
and the mathematical description for the system is complete with 81 = 0. Since u(x,t ) depends on
space and time, the variation 8u(x,t) takes place in space and time domain. If we note the kinetic
energy terms corresponds to the inertia force term of the differential equation, the time dependent
problem be regarded as a static problem by treating the inertia force as a body force. In this case,
variation 5 means variation in space only and, hence the most general variational statement can take
the form;

Maximize or Minimize / =
£ F(x, u, u',-)dv with geometric B.C’s,

Example 5. Wave propagation in the rod u(x,t)

The problem is C° variational problem.
To avoid the complexity of solving a problem in time
and space, i.e.,vibration of a continuous system, we will m x
c L
p

treat the inertia term as a body force b by D’Alambert principle. Strain energy is given by

42
 [
J2
X-*zdV=Jfi2 dx
‘
<ix:

and the variational statement can be put as follows;

Minimize II for all possible u(x) with «(0) = 0 with the definition of

n= £T[l£)dx~£uidx~u(L)P
To minimize fi , we take the variation of IT and set it to zero as follows:

= dx-s£uAbdx-Su(L)P
A dx {dx) h SuAb dx-Su(L)P
£

4 —
--
d2u du
EA
dx
V>u dx + \EA — du -Su(L)P
dx o
d2u du du
EA-—j + Ab 1ducbc-Su(0) EA— + Su(L) EA P
dx dx j=0 dx x=L

--
=0 VSu
Each term should be zero. As u is prescribed at X = 0, 5u(0) = 0 . However, as u is not
du
prescribed at x = L, Su(L) 0 and hence, EA
dx
P = 0 should be true, meaning that this
x=L
should be given as a boundary condition.

Differential statement is as follows;

Euler Equation : E —d2u

dx2
+ 6 = 0, 0<x<L

--
B.C’s :
At x = 0, u = 0
du
At x- L, EA P=0
dx

We should note both ways of description require the geometric boundary conditions to be provided, while
in variational statement, the natural boundary conditions need not to be prescribed as a boundary
conditions as the natural boundary conditions is included in the main(goveming) equation. We also note
the 2nd order differential equation becomes a C° variational problem.

43
 Q2y
Using D’Alambert principle, let b = p — , then we have wave propagation Equation

d2u _ p d2u
~dÿ~~E~dt2

Example 6 Beam bending problem

The problem is to minimize the total potential energy function of a cantilever beam.

y
q(x)

TTTTTTTTTZ)M
X

\p
Strain energy due to bending moment is

where q is the force per unit length.

Variational statement is:

Minimize
n(v) = U + V = U-W=

II for u{x) , where 1/(0) = 0 and v'(0) = 0.

")2 dx - ££ qv cbc + v(L)P + v \L)M
J
Taking variation, we have

|£/v"£(v")/&-|ÿ q 8v dx+Sv(L)P+Sv'(L)M j = 0
<5TI(v) = \/Sv,Sv

where Sv,Sv' are any arbitrary, small variations about the equilibrium state or the solution. Since,
geometric B.C., v(0) = 0 and v'(0) = 0 is given, it follows that the variation should
satisfy <5v(0) = 0 5v '(0) = 0 . The first term on the right is integrated by parts two times as follows

(First Term )
ALEIv"d dx2
=| dx

= [E/v"(ÿv)-]oL-f(E/v") dx

44
 = [EIv"Jv'j -[(EIv")'Svfo + (EIv")"Sv dx£
Combining,

0= £{(£/v”)--?}«v +[<Sv(0)(£/v-)']-)-ÿv(I)j-/'-[(£/v-)']i_ij
+£v,(0)[-.E7v"]i=0 +SSV\L)[-M+[EIV"]ÿL) VSV,SV(0)

Each term must be zero. The first term gives the governing differential equation.

At X = 0, £v(0) = 0 and Sv'(0) = 0 in the second and the fourth term. [(ÿv")'X-0 311(1

[EIv "]ÿ o therefore, need not to be zero and remains as the unknown shear force and moments.

At X = L , since natural B.C., ——F =0 and —M + [EJV"]ÿL=0 is specified in

the third and fifth term, Sv(L) and 8v\L) need not to be zero. This means v(Z-) and
v'(Z) should not be specified. They should be unknowns. We, therefore, have the differential
equation as follows;

Euler Equation : (F/v")" = q, 0 <x<L

boundary conditions :
At x = 0, v(0) = 0 and v'(0) = 0

At x = L, -F-pF/v")'ÿÿ and -M + [EIV"]X=L = 0

Here, the 4th order differential equation can be rewritten as a C1 variational problem.

Example 7. Heat transfer equation

By Fourier law, heat flux is given by q = —KWT .

Heat conduction equation is given by KV2T = pet or V2T = at where a= —pc is the

diffusivity. If heat generation per unit length is denoted b , a suitable functional for this problem is
given by

n
-fHSH Tb dx-T(0,t)q0
*C
x Tw

»
45
§n = 5Tb dx-5T(0, t)q0
L
d2T ST
~fl K—5T + b\5Tdx+K—5T -q05T(0)
dx o
=0
d2T
Euler Equation : K-—ÿ- + b = 0

--
B.C’s :
Natural B.C’s Essential B.C’s
dT
At x = 0, K qÿ =Q or T is prescribed.
dx

At x = L,
dx
o or T is prescribed

dT d2T dT
,
Let b = —pc-, then transient heat conduction equation K
dt ——
dx2 = pc-
dt

3.1.6 Determination of functional form from differential equations

So far we assume the functional form of the variational statement is known in advance. In fact, in case of
elasticity problem, the Hamilton’s principle gives the appropriate functional. For other types of problem,
however, there may exist no such principles. Reverse process is, however, always possible; for any well
posed problem, i.e., differential equation with appropriate boundary conditions, we can derive an
equivalent variational problem. Consider the rod problem in the example.

Differential formulation or strong form u(x)

P
EAÿ+b
dx2 =
0 0 <x<L
x L
u \x=0 = 0
du
EA—
dx x-L
=P

Multiplying the governing equation by Su and integrating over the body, we have the weak form

46
 udx = 0 V Su

du
u\lx=0 =0 and EA —
dx x=L
=P
The variation must satisfy Su(0) = 0.

Integrating by parts, the equation becomes

td[— \eA—
* [dx] dx
dx= fbSudx+EA—Sul1.10]
* dx
By applying the boundary conditions, we have the following weak form

= £bSudx+PSu\ V du

u x=0 =0

In terms of stress and strain, the description is found to be the principle of the virtual work;

|AcrSsdx= £bSu dx+PSu\ Su

u\ x=0 = 0.

From the equation, we have

81
* =<s[f"*A+i>"L*]
Therefore, the variational formulation is;

Minimize FI with the geometric boundary condition u\x_0 = 0 or find u so that (5TI = 0 for all

Su with Su\x=fj = 0 .

47
In general, mathematical formulation of a physical system that is described with the differential operator

1
Llm can be put in the Cm variational form. In case of time and space dependent continuum

problem, natural boundary conditions can be included in the integral, but the geometric and initial
conditions are not. The latter two are taken care of when we choose a function set F for the field
variable.

3.1.7 Principle of Virtual work

When Hamilton’s principle is applied to a static problem, we have
n =-(u-w)
Variation of II about equilibrium must be zero or <5TI = 0 gives
su = sw
which is the principle of virtual displacements and it can be put in the following statement;
If a general structure which is in equilibrium with its applied forces is subjected to a set of small,
compatible virtual displacements, the virtual work done by the external forces is equal to the virtual strain
energy of internal stresses. (The principle has nothing to do with conservation of energy. It can be
applied even to a non-conservative system, e.g., plastic deformation.)

lP' t
P3
y
P2

z xl
The system shown has concentrated forces P, , traction force t , the body force per unit volume fb . It
also has a fixed boundary condition.

],{&}’{a}dV = Y.{ÿ}T{P},*i{Su}T{l}ds+[{Su}T {ft}dV

l

where {P}j is the force vector at xr

The principle can be derived by taking the similar procedure as in section (3.1.4.3) beginning with 3D
force equilibrium equation.

48
Variational formulation corresponding to equation is :
Minimize

n= \ l (ÿ }r w x{“(x<
e dv -
i
ip}, - l Mr w*- L wr {f»}dv
•

with B.C’s.

PROBLEMS

Problem 1
It is required to determine the continuously differentiable function _y(x) which minimizes the integral

/= (1 + y’2)dx , and satisfies the end conditions >>(0) = 0,ÿ(1) = 1 . Obtain the relevant Euler

equation, and show that the stationary function is y = X

Problem 2. Minimize the functional of the brachistochrone problem to get the Euler equation.

Problem 3. Starting from the equilibrium equation for 3D, derive the principle of the virtual
displacements. Hint: Use the Galerkin’s approach

Problem 4.
Determine the natural boundary conditions associated with the determination of extremal of the
functional

I=l(xy*-yy\y-)dx
For a junctional I = (\ + y’1 )dx - 6y(f) with a geometric boundary condition y(0j 1 , — jind the

differential equation and natural boundary condition at X =1.

Problem 5.
Minimize the total potential energy function El of a cantilever beam subject to a concentrated force P,
moment M and body force b to find the governing differential equation and relevant boundary
conditions.

49
 LEl( J2vY
n(v) = c/-r= J[ 2 U2 (ix- |'L6VJX + V(I)P+

m t b(x) M

Problem 6.
-x
I
I
Minimize

to prove

a!fj+fi au ~aji

50
3.2 Approximation Methods
Minimize the error or residual by which the differential equation deviates from the equilibrium.
L2mW = r
with a constraint equation
on S', i = \,2,...2m
where Llm is the linear, symmetric, positive definite differential operator with up to (2m)th order and
(f){x), r(x) are the state variable and forcing function, respectively.

3.2.1 The Weighted Residual Method(WRM)

Assume a trial function of the form

*~±ajfj
>1
where fj is linearly independent base function, 2m times differentiable and must satisfy all B.C’s. So
does (/) , since it is the linear combination of fj. aÿs are the coefficients to be determined in
approximation method.

The trial function can take the from

where f0 satisfies the nonhomogeneous boundary conditions and fj satisfy the homogeneous
boundary conditions.

Defining residual or error as R=r — L2m[0 ] , three methods are introduced here depending on the
definition of error and the sense by which the error is minimized.

1) Petrov-Galerkin Method
In this method a set of linear equations are obtained by the condition

\DgjRdD = 0, 7 = 1, 2,..,n
which implies the residual R is orthogonal to a set of functions gj . With arbitrary choice of test
function gj the equations can result in an nonsymmetric matrix. When the base functions are chosen
for gj , i.e., gj = fj, it is called Bubnov-Galerkin method or simply Galerkin method and it results a
symmetric matrix.

2) Least Squares Method

The best approximation is sought in the sense of least squares, namely,

—[
da.j
R2dD = 0, y = l, 2,...,n

51
Using equation gives

\oRL2m[fj]dD = 0, j = l,2,...,n
This method is identical to Petrov-Galerkin Mathod with Llm[fj] as weighting functions and results a

symmetric matrix.

3) Collocation Method

The constants are chosen so that the residual vanish at n discrete points, namely,

R(Xj) = 0, j = 1,2,..., n
which corresponds to

r ~ L2 m E
M
°jfj\ = °» J' = n

and results a nonsymmetric matrix results.

Example 1. Cable problem

As an application of the methods, consider a cable problem in which a string is fixed at the points (0,0)
and (/, h) and is subject to a transverse loading of
p(x) = —q0x / / . Find the static deflection y I y
4o /T{x)
h
The governing equation is given by

where
T°i?+P=°
T0=T cos 6 = const, and B.C.' s are;
T(x) x
l
at x = 0 , y = 0 and
at x = l , y = h .

The equation has the same form as steady state heat conduction equation with heat source p . For the
sake of simplicity, consider the case when l = h = 1, !T0 = 1. The differential equation is then reduced
to
d dy
-f -?o* = 0
dx\dx
and the B.C.' s are:
at x = 0, y = 0 and

52
 at x = l, y = 1
Exact solution of the problem is —
y = x + q0x(x2 1) / 6 .

To find an approximate solution, we can choose the base functions

fo(x) = x, f1(x) = x(x-1), /w(x) = x"(x-l)
—
In this example, we take a trial function with one coefficient: </> = x + a,x(x 1)

1) Collocation method
d2y
-q0x = 2al-q0x = 0
dx2
=
Take the point at X 1/2 asa collocation point, then

a.= —4 , y»x + -x(x-l)

4
Take the point at JC = 1/ 4 asa collocation point, then
0o
8’ y&x+-ÿx(x-l)
Hence, the result is different depending on the choice of collocation points.

2) Bubnov-Galerkin Method

dx = 0

= ,y»x+ÿ-x(x-l)
y
4 4

2) Petrov -Galerkin Method

cbt = 0

The choice of gt is arbitrary, for example,

1 0o
Choose g, = x , then a,=— q0, y ~ x + -ÿ-x(x-l)

Choose gj = x2 , then al=—q0, >p«x+-ÿ-x(x-l)

8 8

Example 2. Bar with variable cross section

As an example of a system with geometric and forced boundary conditions, consider a bar with cross
sectional area of A(x), fixed at one end and subject to a concentrated force at the other end.
Determine the axial displacement for a tapered bar with a concentrated force P without a body force
where A
——
4 x and L = 2 .

53
 p

1
The problem can be described by the governing differential equation
d du
AE— =0
dx dx

with the geometric B.C.’s «(0) =0 and the forced B.C’s AEdu / dx\x_L = P.

The exact solution is u(x) = -P ln(l — x/4) with deflection at the tip u(2) = 0.6932P .
Since the trial function must satisfy the forced boundary condition at the tip, for example, we begin with
u = a]x+a2x2
, for which we have
du
AE—
dx x=t
= a, + 4a2 4 = P; a{ = -Aa2 + —P
Therefore, a suitable trial function for WRM method is

x= + a2(x2- 4x)

1) Collocation

d
[AE%

---
dx

Choosing X = 1 as a collocation point, it follows that

P PX
«2=“ 16’ u = 2
P, 2
(x - 4x)s .
16
Bar elongation at the tip x = 2 is u = 0.750 P .

2)Bubnov-Galerkin

|fxRdx -|(x2 -4xÿ-4a2 -4a2x--ÿ j<ix = 0

a2 =1.146xl0"3i>

54
3.2.2 Ritz Method
In this method, the problem is described in the Cm~x variaional form and the solution <f> is

approximated using the trial function <f) = a ,f, The trial function needs to satisfy the geometric
7=1
boundary conditions only and be m-times differentiable. Note that it does not need to satisfy the forced
boundary conditions, which are satisfied approximately when the functional is minimized because they
are included in II .

When trial function is substituted in the functional, the functional is written in terms of the unknown
coefficients, i.e., Tl(ax, a2, ...) . The variation of the functional <31(ÿ) is associated with variation
of the function 5(j) , which in turn is possible only through the variation of the coefficients Sai . We
then write

<3TI = —
dax
8ax +—Sa2 +...+— 5an =0
da2 dan
VSa,
or
dn
- = 0, i = 1,2,..., n
da,

Example 3.
In this example, we will apply Ritz method to the cable problem with l = h = 1, 7ÿ = 1. To obtain the
variational formulation, we multiply the differential equation by Sy and integrate it over the problem
domain.

d2y
-ÿ-q0x\Sydx = 0
In order to include the boundary conditions, we integrate it by parts to get

5yÿ =0

The problem is given the geometric boundary conditions. The variation of these geometric boundary
terms, Sy is assumed to vanish at the boundaries.

j£(y,5y'+q0xSy)dx = 0 \/8y, £y(0) = <?y(l) =

The equation is finally in the variational formulation

1
n= -y’2+q0xy , m = 0\/Sy, Sy(0) = Sy(Y) = 0
The base functions satisfying the geometric boundary conditions are
fo(.x) = x, fj(x) = x(x-1),

55
Hence, the trial function is chosen as
);&(/) = x + x(x-l){ax + a2x+... + anxn ')
When n = 1 , the functional becomes
n= f(±{l + (2*-1)«! }2 +q0xal{x + x(x-1)} jcfr
By the condition

or <5n = 0 y8ai
dax
we have al=q0/ 4 and the approximated solution is

0(x) = x + Vo
4

When n = 2 , the similar procedure gives al=q0/ 6, a2 — q0 / 6.

3.2.3 Ritz Method and Galerkin Method

In the cable problem, the result from the Bubnov-Galerkin method is identical to the one from the Ritz
method. The situation is generally true for problems where there exist only geometric boundary
conditions. In this case, the trial functions are identical in the two methods. In addition it can be proved
the approximation equations are equivalent to each other. In fact, the condition STl = 0 in Ritz
method can be written as
sn = 8ÿ(\ya+q0xy)dx= |(y'8y'+qoxSy)dx = -ÿSy(y"-q0x)dx = 0 VSy
where integration by parts was used and the resulting boundary terms vanish since the geometric
boundary conditions are given. In approximation, we choose y « X + a,x(x 1) , then
Sy = 8ax x(x -1) = fx 8ax
—
and the above equation becomes

fx{y"-q(,x)dx = 8a,i =0
or

[fx(y"-q0x)dx=o
which is the identical equation if Bubnov-Galerkin method would have been adopted

Therefore the two methods are virtually identical when there exist the geometric boundary conditions.
When force boundary conditions are included in the problem, however, they are different. In this case
the trial functions don’t have to be the same for Ritz method and Galerkin method. For example, if trial
function of 2nd order polynomial is to be used in Ritz method for the rod problem, we can use

U
— axx + a2x2 and minimize fl to determine ax, a2. In this case the results would be also

different.

56
In summary, in Ritz method the trial function needs not to satisfy the natural boundary conditions because
the functional includes them in it, while WRM requires the trial function to satisfy all the boundary
conditions. In addition Ritz method allows the trial functions need to be differentiable up to less order
compared to WRM. As result, Ritz method has a larger functional space for trial functions and allows
the user to choose the trial functions more easily over WRM.

To overcome the difficulties in choosing the trial functions, the Bubnov-Galerkin method is used in
practice to allow the use of the same trial functions as in Ritz method. This is possible by integrating by
parts the approximation equation and hence, including the essential boundary conditions in it. This
procedure would result the equation corresponding to the weak form we have seen earlier. The Bubnov-
Galerkin method in this modified form gives the same approximation as the Ritz method. Finite element
formulation can also be done by this method. In FEM, this type of Bubnov-Galerkin’s Method is called
just Galerkin’s method.

Conclusion : In Galerkin method, the natural boundary conditions are included into the integral by
integration by parts and the same base function ft is used for the test function gt . The solution from
the Galerkin method is the same as the Ritz method which is the theory behind the FEM. Hence an
alternative approach to finite element formulation is the Galerkin’s method.

Method Trial
exactly _
function must satisfy Trial function must Approximation is done for
have__
Weighted Both Essential and Natural continuity GE. only
Residual B.C’s
Ritz Essential B.C.’s only Cm 1 continuity Both GE and Natural
B.C’s ._

Exercise 1. Determine the approximations using 2nd order polynomial function by Ritz and WRM for
the rod problem with A(x) = AQ + Aÿx. The exact solution is
u
— —A In I —AA x
P
+1

3.2.4 Formulation of FEM

l)Direct Approach
This approach can be applied to find the stiffness matrix of 2 node spring element, 2 node beam element,
3 node triangular element.
2)Rayleigh-Ritz Method
The method is called variational method. In static problem, it is minimum potential energy theory.
3)GaIerkin’s Method

57
This is the WRM with natural boundary condition included in the residual approximation equation. In
fact, the same equation can be obtained by taking variation of the functional. In mathematical term, it is
the same as weak form. In static problem, it corresponds to principle of virtual work.

3.2.5 Ritz Method and FEM

In Ritz method, the solution function is approximated by a linear combination of n functions with
Cm_I continuity over the whole domain. The solution function is optimized for the n ai 's{i \,ri),
the coefficients of the base functions ft's.
—
In FEM, the problem domain is subdivided into n discrete local elements. The solution function is
approximated by a linear combination of n base functions. The i th base function is usually polynomial
differentiable up to m-th order in the element I and zero out the element. In addition the trial function
must be Cm_1 continuous over the whole domain. This is possible when the base functions of
neighboring elements share the unique value up to the m-1 th derivatives at the boundaries. The
uniqueness at the boundary can be accomplished when the same order of polynomials are used for the
neighboring elements as well as the unique value up to m-1 th derivatives at the related nodes. The
common values at the nodes are the unknowns to be determined when Ritz method is applied and global
element equation is obtained.

An alternative approach is to apply the Ritz method to each discrete local domain. At the nodes of each
element the derivatives up to m-1 th order is assumed to be unkowns. A polynomial trial function is
defined in terms of the nodal values in the element. For a Cm~' variational problem, the number of
nodal values are m-l(i.e., oth,1st, .. m-lth). If the number of the nodes in an element is nel , the number
of the unkowns in Ritz will be fle,xm. When Ritz method is applied, element equation is obtained.
To ensure the element connectivity, the element equations are assembled by using the common nodal
values for the neighboring elements,

3.2.6 FDM v.s. FEM

Either differential or variational formulation can be used in FDM. The derivatives in the governing
equatin are replaced by finite difference approximation for which only 0th order state variables at nodes
are used. Therefore, the nodal values are in FDM 0th order state variables while in FEM the unknowns
are state variables up to m-1 th order at nodes. When the derivatives in FDM is a piecewise constant
function between the nodes while in fern or WRM a higher order function can be adopted.

Example 4. Determine the axial displacement for a tapered bar with a concentrated force and a body
—
force where A = 2c (x / L)c , fb = const. Use Ritz method.
The exact solution is

„w=ÿx+WLln(1.JL)+Wizÿln(1.JL)
Ec Ec 2L Ec 2L

58
 - fbu)dx - Pu(L) P
x
2
7fbL2+6PL 3(PL-fhL2)
— —
L
X
u w at + a2
x
L 13Ec ’ a2 13Ec

When fb = 0 , the exact solution is

-PL
»(*) = Ind-ÿ)
Ec 2L
while approximate solution is

,2
3PL lx
u«
13Ec L
+
X
L —
<End deflection>
n= 2 3 4
Approx w(Z-)
0.6923 —
PL
Ec
PL
0.6931—
Ec
PL
0.6931—
Ec
Exact u(L) PL
0.6932—
Ec

Example 5 Find the natural frequency of the system shown

mrÿ P,E,A,L k m

Governing equation for the rod is

—I
dx
EA—
dx
- pA —d2u =
7
dt 2
0, 0<x<Z,; t> 0

Assuming u(x,t) = u0(x)e,al , we have

—dxd EA —du0
dx
-\+pAco2uQ= 0, 0<x<Z.

When this equation is compared with the equation

d2u
Eÿj + b = 0
dx2

59
 of static rod problem of example 1, the second terms is regarded as the body force b = pAa)2u0 .

Variational statement weak form) becomes

-bu0 dx+ÿu0(L)2
Taking variation while keeping b constant, we have

sn = J ({MÿSÿybSuAdK+kuo(.L)Suo<.L)=0 V«oW
2
As a trial function satisfying the geometric B.C., u0 = atx‘ is chosen to give
i=i

||EA(ax + 2a2x)(Sax + 2x8a2 ) - pAco1(axx + a2x2 )(xSax + x2Sa2 ) j dx \fSax,Sa2

2
+k(axL + a2L \LSax + L25a2 ) = 0
Rearranging, we have

£ [EA(ax + 2a2x) - pAco2(axx + a2x2)x}dx +k(axL + a2E)L j<5a,

\/Sax,Sa2
+ 2a2x)2x - pAco2 (a{x+a2x2 )x2 j dx +k(a] L + a2l})lfÿSa2 =0
The equation gives us an eigenvalue problem

EAL- EA£ £ÿ+u?k-0

EAL2- £ÿ+tt.L+{iÿ.£ÿ+WU=„
When 1= 1, —pL2 =
E
kL
1 , -= 1 , the equation becoms in matrix form
EA

\2 2lw =d>2
«2
The eigenvalues of the equation are (Ox = 2.038, C02 = 6.026 . The first natural frequency is quite
close to the exact solution COx = 2.029 , while the second one is somewhat different from the exact
solution a>2 =4.913.

Weighted Residual Method(Bubnov-Galerkin)

2
Again as a trial function satisfying the geometric B.C., we can choose u0 = , but satisfying the
i= i

forced boundary condition gives 2c, + 3C2L = 0 and hence, we finally have

60
 u0=-ÿ-(3Lx-2x2) = ai(SLx-2x2)
0= £ MEAa,ÿÿ + pAo)2aJx)dx or jj<y2 -y ja, = 0
cox =4.167

PROBLEMS

Problem 1
What are the differences between fem and fdm?
What are the differences between fem and Ritz?
What are the differences between Ritz and Galerkin?

Problem 2
For a cantilever with varying 2nd moment of inertia I(x) , subject to a force P at the tip,
approximation method is to be used with polynomials of up to 5th order. Write trial function V for
a)WRM b)Ritz method.

v \p
I(x),L v«
X

Problem 3
For the cable problem, choose <j) = ax x3 and

1. Collocation Method
2. Galerkin’s Method in original form
3. Ritz Method

to find approximate solution and plot (j) - X graphs together with the exact solution and discuss their
advantages.

Problem 4
Eigenvalue problem for membrane with radius a fixed at its edge.
1 d du
r
r dr dr
— = Xu : X is an eigenvalue

The governing equation is the 2nd order diff. equation with boundary conditions of up to 1st order.
1) When the membrane is fixed at its edge, Oth order u(a) = 0

61
 8u
2) When the membrane is free to deflect transversely, 1st order —
dr
=0
For the first case, the exact solution for the first eigenvalue is \ — 5.779
a2 ‘

. nr . Znr
Candidate base functions are /,=cos —
2a
, J2 = cos-
2a
which satisfy the essential boundary

5.832
conditions. Show that the Bubnov-Galerkin method gives \ = a2 when n=l, and 2ÿ — 5.792
a2
when n=2.

Problem 5
- (ax3 / 6) + (R + \ al})x
If body force is fB = ax , then exact solution is u(x) =
EA

Constant cross-sectional area A

ayi
Young's modulus E

fB{x) = ax
fS(x) I •R
Figure 3.2 Uniform bar subjected to
body load f (force/unit length) and tip
-
w -
L *
load R

For this problem, there are two types of boundary conditions exist. Prove the resulting approximation
from the Ritz and Galerkin method in the original form are different from each other. Using the all three
base functions of x, x2 , x3 , find approximate solution for the problem by
1. Collocation Method
2. Galerkin’s Method
3. Ritz Method
Plot the solutions together with the exact one in a graph. Discuss the results.

Problem 6

Rework with </> = using

1. Collocation Method
2. Galerkin’s Method
3. Ritz Method
to solve the above problem. Plot <f) —x graph together with exact solution. Discuss about their
advantages and disadvantages.

Problem 7

62
For the system shown, exact solution is given by

u(x) =
-(cat l6) + (P + \a£)x
EA
m >x
b(x) - ax

Numerical values for the variables are L,EA P

a = l,P -10,L = 10,EA = 100.

Using polynomials up to 3rd order as base functions, find an approximated solution by

1. Collocation Method
2. Bubnov-Galerkin’s Method(WRM) where the trial function does not include the essential boundary
conditions.
3. Bubnov-Galerkin’s Method where the trial function includes the force boundary conditions in it.
3. Ritz Method
Plot all the solutions together with the exact one to compare them. Discuss the results.

Problem 8
A tapered bar is subject to a concentrated load P at the tip. The governing equation is given by
d du
dx
AE—
dx
=0 0 <*</.

a) Write the boundary conditions

b)Derive the functional 11 of the variational statement of the problem by integrating the differential
equation by parts.
c)Using just one trigonometric function as base function (hence, only one coefficient to be determined),
find the approximated solution u by (Bubnov) Galerkin method as the original WRM.
d)Using two trigonometric functions, find the approximated solution u by Galerkin method as used for
FEM(Weak form). (* You still need to determine one coefficient of the trial function, since it should
satisfy the forced B.C.)

P
x

63
4. Continuum 2D Elements by Advanced Approach
4.1 Shape Functions in Physical Coordinates
In the derivation of stiffness matrix of a finite element, the displacement distribution within an element is
represented in terms of nodal values by use of the interpolation scheme. For example, in case of 4-node
rectangular element the displacement filed u(x, y) in x-drection is discribed by intepolating 4 nodal
displacements ul,u2,u3,u4 . Extending the idea from the previous article, we can write
4
, v=
i=l i=i

(a-x)(b-y) j-a-x){b-y)
Nx(x,y) = , N2(x,y) =
Aab -Aab
(-a - x)(-b - y)
, N4(x,y) = -
(a x)(-b - y)
N3(x,y) =
Aab Aab
—
(W4) (x3,y3)
4 3
/P
I
y
2b Ux
£x 1 2a 2
(Wi) (x2,y2)

-x, = x2 = x3 = -x4 = a

Characteristics of shape function

i>i;ÿ(xÿ)=i
/=i

2)Ni(xj,yJ) =
1
0
if
if
i
i*j
— j

3) N, (x, >") is bilinear function of X, y

Example 1 3-node rod element (quadratic rod element)

x3 x2
X fl
a • <
3
A
L

64
By using Lagranges’s interpolation polynomials, we get
d,
« = [*, N, AT,] U or W =

d3
with

Nl (x) = —(*2 -x)(x3 -x), N2 x = -(*,- )(x3-x),N3(£) = -—(xl-x)(X2 -x)

1
( )
1
X
1

4.2. Linear Quadrilateral Element

1) Select element type
For a linear quadrilateral element, which is a continuum element, displacement vector is defined by

M = [«i vi •• u4 v4f dtyifty

3

4
d4x’f*x
2

1
2) Select displacement functions
The displacement field in the element can be written using the interpolatinf function;
4
u = 'ZNl(x,y)ul , v = YJN.(x>y)v,
<=i i=i

In matrix notation,
W=M4
where

u(x,y)1 Nx 0 • • JV4 0
M- v(x,y)J ’ 0 Nx •• 0 N4

3) Define the strain/displacement and stress/strain relationships

Two dimensional problems can be categorized as plane stress or strain problem. In plane stress problem,

Oz,T„, Tp are zero by definition and they therefore, does not contribute to strain energy calculation

although there still can exist yz . In the other hand, plane strain problem means yz,yzx,yy2 are zero

by definition and they, therefore, does not contribute to strain energy calculation although there still can
exist <7z . Therefore, rather than using the general three dimensional stress and strain in two dimensional

problems, we will work with three components of stress and strain only that contribute the strain energy,
i.e., in X —y plane

65
The strain components —
x y plane is related to the displacement field by the relation
du dv du dv
£x~fa’ £y~~dy ’ Yxy~~dy+~dx
or in matrix form,

The equation is symbolically written as

£x
£y
r*y ho
dy dx

M-MM.
where

£x £ o
l£}=\£y\’ [d]= 0 A.
3,

Yxy A.
_dy
JL
dx _
Substituting equation into equation, we can get

where

ft o . . ft
ox
0 fL
,
dN
dy
dN{
.• dN,
ox

0
0
BN4
V
dN,
t
dy dx dy dx

For elasticity problem, Hook’s law is given by

where {£„} represents the initial strain and the elastic material constant [D] is

"l-v V 0
E
[A.] = (l V l-v 0 for plane strain
+ v)(l-2v) Q Q l-2v
2
_

66
 1 v 0
v 1 0 for plane stress
0 0

4) Derive the element stiffness matrix and equations.

Virtual work principle is applied to a finite element with unit thickness. The volume integral is changed
to area integral over the element of area A .

£{&}r ={<x}dA { £{*,}r

\A{SU}T{t}dS + {Sd}T{P}
b}dA+
where {6} ,{/} are the body force per unit area, traction per unit length, respectively and is the
externally applied concentrated force vector. They are defined as

«-£}•«={;)ÿ • • P„
Variation of displacement u is only possible through the variation of displacement at the node. With
the variation of the displacement field, strain variation occurs. From the equations, we have

{&/}=[#]{<«} and {<&} = Mo¬

using the expressions, we calculate the left and right terms of equation:

left hand term = l{Ssf {cr} dA

-=Mr(WM-{/.})
( JXMT
MW <« M -ilBf Mft.}
where

ft] = £[ Bf [D][B]dA ,{/,}= £[«r [D\{st}dA

right hand term = £([iV]{<W})r [b}dA+ + {P}
=w(i[Arrft}<«+j[Mrw*+w)
where

Therefore, with substitution of equation, we have from equation

67
which must be hold for all \dd}
We have
[‘M-UW/W/.KM
Normally the second term on the left is known and is put on the right hand side. We therefore get the
element equation

where
dA
and

1/1 - {»}<«+ M «*+ [D)W)dA +{P}

Note the force terms from {&} ,{f},{/?} should be defined as the magnitude per unit thickness of the
element. For example, if the element thickness is t , and a concentrated force with total magnitude of
P0 is applied at node 3 in X direction, then p6= .
P0/t
5) Assemble the element equations to obtain the global equations and introduce the
boundary conditions.
Stiffness matrix Transformation from the local to global coordinates.

6) Solve for the nodal displacements.

Matrix partitioning, Penalty method, Gaussian Method, LU Decomposition, Iteration method.

7) Solve for the element forces(stresses)

M=[z>PM
8)Treatment of distributed loads
For a constraint-strain triangle, a distributed load on an element edge can be can be replaced by the two
statically equivalent concentrated loads acting at the nodes associated with the loaded edge.

q0[N/m] «•'*' M H,)

[A]
2
0 [N]
2
U ?o (Li -[N]
L 2
40ÿ4
f-m
68
9)Types of plane elements
For three node triangular element(linear triangular element), the strain(stress) is constant within an
element(CST). For six node triangular element(quadratic triangular element), the strain(stress)
distribution is linear within an element(LST).

Example 1. Traction on Rectangular element

Calculate the force from traction applied on the edge 2-3.

(w4) (Ws)

S3 y
Ux x2-xi=x3—x2=2a,
S2
y*-yi=y3~y2 = 2b

5,
(Wl) (x2,y2) wo
The nodal force vector equivalent to the applied traction is given by

On the force boundary S = Sl + S2 + S3 + S4 , the tractions are

[° of on S\,S3,S4,
and

where t,=-w„(y1-
Then the equation becomes

UHMTW,*
On the boundary S2 , the shape functions are

Nx(x2,y) = 0, N4(x2,y) = 0 and N2(x2,y) = and N3{x2,y) =

2b —lb
The integrand of the equation becomes

[*r W = [° 0 N2tx 0 N3tx 0 0 of.

Substituting equation into equation and taking integration, we have

69
 » wo(y3 -yf 2w0b
4b2
Similarly,

/ÿ
Is-
wob
3
Finally, the nodal displacement vector is

w0b
{/}- 0 0-ÿ0 3
0 0 0

For an QUAD4 element shown, in order to find the force vector l

due to the traction wx = — ,

use the relation = J| [Aÿ]7 {/}<& and integrate it exactly (It would be easy to work in the

xy coordinates rather than the <%r) coordinates, hence, use Nt(x,y) and do not use a numerical
integration).

4.3 Shape Functions in Parametric Coordinates

The general form of stiffness matrix of a linear quadrilateral element is given by the equation

[*]=£[B(x,y)f [D][B(x,y)]dA
Since the shape of the finite element is not regular, certain difficulties follow from the integration limit.
This can be avoided when the mapping is used. The element or integration domain is mapped onto a

square in the parametric domain. Accordingly the integrand and infinitesimal element

dA = dxdy should be expressed in terms of the parametric coordinates.

4.3.1 The Concept of Mapping

For a two-node ID element, in order to compute the displacement u of point P using interpolation,
we used
u = Nl (x)Wj + N2 (X)U2
Suppose the domain x is mapped into domain £, , which called natural, parametric or intrinsic
coordinates. There would exist a relation or mapping function x = f(ÿ) , which is substituted into
equations to give u in terms of £ ;

70
 u = Nl(4)ul+N2(4)u2,

where iV, (£) = iV, (/(£)) . Although Nt (x), N- (4) are different in their function forms, we
prefer to use N- (4) to Nt (4) to represent the shape function in the parametric coordinates without
confusion. Hence we write

u = Nx(4)ux+N2(4)u2,

Similarly, to find the mapping function X = f(4) (or coordinate transformation) from 4 to X, we
can employ the similar concept. The equation gives the position of point P in X coordinate when
the position is known in 4 coordinate. Since there is one to one correspondence, we have as data set
at each node

£ -1 +1
x X, x2
Linear interpolation is possible and hence we begin with X = aE, + b . Using the data and solving for
a, b and putting them back in the interpolating equation, we get

x = 1~4 x,+ 1+4 x2

— —
= NX(4)X1+N2(4)X2
As we anticipated, the same shape functions result both for the displacement and for coordinate
interpolation. Using the same shape functions for the displacements and coordinates is convenient for the
derivation. Element of this type is called isoparametric element.

Characteristics of shape Function

i)i>,«.i7)=i
i=i

2)NX4j,rjj) =
1 if i~j
0 if i*j
3) Nt (4,t]) is bilinear function of 4, V 2

4.3.2 Shape Functions in Parametric Coordinates for 4-Node Isoparametric,

Quadrilateral Elements
The shape function of 4-node quadrilateral element Nt (4, T]) is bilinear function of 4, *1
4
y
>3 4 ? 3
fV
A X xS
2 2

1 71
Exercise 1
(A) For the 2D 4-node isoparametric quadrilateral element shown, find the x-y coordinates of a point
that is mapped onto the origin of natural coordinate. The coordinates of each node is in the unit of mm
and the displacements are ux = 0,u2= 100/mi, u3 = 200///ft, u4 = 50// /ft
v, = 20///ft, v2 = -50/i/rc, v3 = -50/i/w, v4 = 150/i/ft .
(B) Find the strain and stress components at the point assuming the plane stress conditions.

Material constants are E = 200GPa, V = 0.3 .

y 3(5,5)
4(2,4)

2(6,2)

1(1,1) x

4.4 Change of Variable for the Integral

4.4.1 Calculation of [i?],dL4 for isoparametric element.

To calculate stiffness matrix in parametric coordinates for an isoparametric element, [J9(X, j/)],/£4

should be written in terms of %,TJ . A typical components of [5] is the term like dNx (x, y) / dx .
By change of variables from X, y to %,1) and chain rule, we can write
dNx{x,y) dNtf, 77) dt [ 3N& tj) dV
dx d<%dx dtj dx
where Nx{ÿ,rj) on the right hand side represents Nx (x, y) with x,y replaced by the relation().
Calculation of dd, / dx, dr] / dx , however, is not easy since the relations E, = £(x, 3/) , Tj = rj(x, y)
are difficult to get. On the other hand, assuming Nx is expressed in terms of %,7] and applying
chain rule, we write, omitting Nx ,
d d dx d dy d d dx
dE, dx dE, ’
dy dE, drj dx drj dy drj
in matrix form,

72
 d dx dy d
a? a? dx
d
d dx dy
drj drj dr]
The equation can be written symbolically as
d d
dx
d dx dy d

with
fel-i dr]
w dt dx
dr]
%
dy ’
dr]
d
lax
dx
a
dy

Then, using the inverse of the Jacobian matrix [«/] , we write

a dy dy a
dx 1 a$
a J dX dx a
py. L drj dt J dr]

where Jacobian J is the determinant of the Jacobian matrix [ j] and is given by

j
_dx dy dy dx
dd, dr] dd, dr]

The terms appearing in [ j] and J can be calculated easily, for example

All the components of [f?(x,T)] can be written in terms of pT] if the equation() is applied to

Nt (/ = 1,4). As in the case of N(x)n again we will use the representation [f?(x()j = [5(ÿ)]
without any confusion.

When the finite element in the physical coordinates is mapped onto the square in the parametric

coordinates, the differential area dA is mapped onto dAp . They are related, as proved in appendix, by
dA = JdAp

73
For example, for 2D linear quadrilateral element, Jacobian is —
J dA!dAp =dA/ 4 and for a 3D
Hexahedron : J = dV / dVp .

Finally we have the formulae for stiffness matrix

[*] = [[B&f l l[Bi4l)Y[D]lm)]J*&TI

=[D][B(Q]jdAr

Exercise 1
(A) For the 2D 4-node isoparametric quadrilateral element shown, find the x-y coordinates of a point
that is mapped onto the origin of natural coordinate. The coordinates of each node is in the unit of mm
and the displacements are ux = 0, u2 = lOOÿm, u3 = 200 jum, u4 = 50fim
v, = 20/um, v2 = -50fj.m, v3 = -50fj.m, v4 = 150p.m. .
(B) Find the strain and stress components at the point assuming the plane stress conditions.

Material constants are E - 200GPa, V = 0.3 .

y 3(5,5)
4(2,4)

2(6,2)

1(1,1) x

74
4.5 Numerical Integration
4.5.1 Newton-Cotes Integration Formulas for 1-D

1= £/(*)<&- (/„(#
where fn (x) = a0 + axx + ... + anxn
1) Trapezoidal rule : linear interpolation

I « (b - a) m+m
2
2) Simpson’s 1/3 rule : quadratic interpolation
/(x0) + 4/(x,) + /(x2)
/ « (b-a)
6
where X0 = a, Xj ——
a +b
= - x2=b
Therefore, Newton-Cotes Integration Formulas can be represented as

I= £ f(x)dx £ /„ (x)dx = J wj{xt )

«
#=0

4.5.2 Gauss Quadrature for 1-D

In Gauss Quadrature, the integration points are not spaced equally each other, neither are they
predetermined. Instead, the points are to be determined optimally so that the resulting integration can
get the exact value for a 2nd order polynomial. The formula is given for the integration range of
-1<X<+1.

I
One-point Gauss Quadrature : 7= £ /(x)dx «
M
w,f (xG/ )
where w, = 2, XG1 = 0

Two-point Gauss Quadrature : I =

£ /(x)dx «
;=1
w, /(xG( )

where
— —
,
w, w2 1, xG1 — 1
, xG2
— +
l

Comparison of Simpson’s 1/3 Rule and 2-point Gauss Quadrature:

Both method give exact integration result for a 2nd order polynomial. Simpson’s method need function
evaluation at 3 points whereas only 2 points are need in Gauss Quadrature. We need at least n equally

75
spaced sampling points to integrate exactly a polynomial of order n-1 using the Newton-Cotes method
and at least n unequally spaced sampling points to integrate exactly a polynomial of order 2n-l using the
Gauss Quadrature.

4.5.3 Gauss Quadrature for a 2-D Rectangle

One-point Gauss Quadrature : I= £ £ /(x, y)dxdy «

«=i j=1
wijf (xGi , yGj )
where w, = 4, XG1 =yGl=0

2x2 point Gauss Quadrature : I=

£ £ /(x, y)dxdy «
i=l
I]wijf(xa’
7=1

where wy = l, xGi=±-j=, yGJ = ±-ÿ

3x3 point Gauss Quadrature : I= £ £ /(x, y)dxdy «
1=1 7=1
wyf(xGl , yGJ )

where w22
8
—
= , "V=! hj* 2 and

XG2 = yG2 = 0, XGi i,j* 2

Example 1. Stiffness Matrix of Rectangular Element

n
(w4) (w3) +1

y
Ux -1 +1
4

(Wi) (x2>y2) -1
: Gauss Integration Point

—
-x, = x2 = x3 = x4 = a
-y\=-y2 = y2 = y*=b

76
 M= L
tS«*»T
When using the change of variables between (x, y) coordinate domain and (£, 7]) domain, the four
nodes of the rectangular element should map to the four nodes of the square domain with one to one
correspondence. Applying the one to one correspondence, we have the relations
x = a% , y-btj and dxdy = ab dgdrj

Using the above equations, the integrand is in terms of and integration limit is ± 1.

where [l] = [fi]r[D][fl]

For example, if 2 x 2 point Gauss quadrature is applied with Wy = 1, the stiffness matrix becomes

[k]*abYJYJ[gÿnT1jj\
1=1 7=1

g(<%t , TJ j) can be calculated in the following steps:

1) Write the expressions dNt (x, y)l dx , dNi (x, y)/ dy in terms of £ and rj
For example,
dNl(x,y)/dx-(-b + y)/4ab-(-l + Jj)/4a,
dN{(x, y) / dy = (-a + x) / 4ab = (-1+ g) / 4b , ...

2) At Gauss point (£VTJX) = calculate dNJdx and dNi / dy . For example,

The same results can be obtained by using 2Aÿ(x,_y)/ dx in terms of X and y . For example, the

Gauss integration point (£X,TJX) = (~ corresponds to (Xpj>,) = > which gives

the same results when substituted into dNi (x, y)/ dx in terms of X and y .

77
3) Construct the matrix
£2?(£, ,ÿ,)J
—>/3—75a1 Q

[%„A,)]=
4

0
-75-1
#1
475i>
-V3-1
.. 0
75+1
75+1
475<J
-73-1
# #
475* 475a 475a 475a

4) Construct the matrix [#(£,,77,)] = )] [£>][£(£ = ?7i)]

5) Do the same calculations from step 2 to step 4 for all the integration points.

6) Calculate [£] = abÿ £[#(£, , Vj)]

/=! 7=1

4.5.4 Required Order of Numerical Integration

Numerical integration is needed in the calculation of stiffness matrix K, mass matrix M, initial stress
vector Ri, traction force vector R,. In 3D case, the increase of the numerical integration points requires
the tremendous amount of computation time, while with too few points, there is a possibility of an
element having more modes with zero eigenvalues than the real rigid body modes. Examples are
hourglass modes.

I
I
I.
tl
'\
1
I
ri I
I
i
f\

For instance, 4-node element with 1-point Gauss quadrature or 8 node element with 2x2 point Gauss
quadrature can have the hourglass mode(spurious zero energy mode) shown in the figure. In this case one
or few numbers of these elements can result in unrealistic displacement.

Proof Denoting real mode as{w(}, we have j = |i?| . But since [ÿ]{w/,our} = {0} , we
also have +uhour] = {/?}, which means the mode we get can be a false mode when the
constraint condition cannot restrain the hourglass mode.

When the element assembly process restrains appropriately the deflection mode with zero eigenvalues,
the global stiffness matrix can be avoided from being singular, but in general this phenomena is hard to
control - sometimes, poor control gives large error instead of improving the results. For example, in
frequency analysis of a simple system, a phantom mode can occur which is obviously not a natural

78
frequency of the system. In this case, with the time response of the system alone, it is hard to detect the
inclusion of the phantom mode in the time integration.

The mode corresponding to the zero eigenvalues have zero strain energy.
Proof
Let {wzfT0} be the mode corresponding to the zero eigenvalues, namely

Ml*-}={«}•
The strain energy defined by £/ = —
1 T

Exercise 1. Calculate the definite integral Q= 5X3dx using 2-point Gauss quadrature.

Exercise 2. Determine the nodal force vector for an element with dimension of 4x3, subject to a
distributed force as shown. Use 2-point Gauss Quadrature.

w(x) = 2 + 0. lx2

4
1 4

1
12 3

2 6

79
4.6 Linear Triangular Element
Here, triangular elements will be discussed.

4.6.1 Derivation in Physical Coordinates

1) Shape Functions
Bilinear equation for u(x, y) :
u(x,y) = al+a2x+a3y
Using coordinates (x., y. ) and nodal displacement ui at each node, we have
3
<x,y) = YJNiui
i=1

1
Where Nt(x,y) =— (a, +fiix + yly) and ax=x2y3-y2x3, ft=y2-y3, y, = x3 - x2 , etc.,
2A
and
1 Xi y,
A-l - 1 *2 y2
1 x3 y3
A is the area of the element. Similarly, we can write
3
v(x,y) = 'ÿNivi
i=l
In matrix representation, the displacement field in the element is represented as

{M}=[Aÿ(X,J)]{J}
where

“l
V1
Nt 0 N2 0 N3 0 u2
0 JV, 0 N2 0 N3\’ V2
u3
lV3
2) Strains £x,sy, yxy in terms of displacements u(x,y), v(x,y)

M-MM
where {f} and [5] are defined previously.

Therefore,

M-MM
where

80
 8N3
A o A o A o
SNt 0 SN2 0 0
ax a* a*
av, f.V2 3Aÿ i
M-MM* 0 9) 0 9) 0 9v
2.4
0
* 0 r2 0 y2
W,
3y
3AT,
3-t
aw2
3y
Wj_
&
3ÿ
5y
3Aÿ
&
7i A r2 A x3 A

3) Stress/strain relationships.
For elasticity problem, Hoook’s law is given by

M-MM
4) Stiffness matrix

where dV = Jzr/,4 .
Integrating in z -dir., we have

[*L =ÿ£[5(x,j;)f [D][5(xÿ)]fiL4

Example 1 Stiffness matrix using one-point Gauss Quadrature

Calculate the stiffness matrix using one-point Gauss Quadrature
Using one-point Gauss Quadrature in X — y domain,

[*L ='L[9]>1M<M-'M>]|>P
where |ÿZ?J = [5(xc,_yc)] is evaluated at the Gauss point which is the centroid of the element area.

4.6.2 Derivation in Parametric Coordinates

Area coordinates is used for triangular elements.
“ y Tl=0\
11
5=1
rpl
4 2
k ,4
4
3 £=0 3 i 1
Defining area coordinates by % =XA—A , = —A , shape functions can
T] be written as

81
 Ntf,T])=t, N2tf,Tj)=Tj, N3(4,*r) =1-4-1

Coordinate mapping is done through

*=i>,(£>7K.
/=] 1=1

Displacement interpolation is given by

«(r,
/=!
H,, v(r,z) = £Af,v,
/'=!

Stiffness matrix is given by as before

a d
dx
d
> Or
d dx dy
dr] dr] drj py.
Then,

Mapping, we have

[*]='!,[*(6)f [£][*(£)]•*«,
where, for example, 2D linear triangular element has J = dA!dAp — 2A.

One-point Gauss Quadrature will be used for a triangular element, then

[*L = '0]>][AK =

J = 2A, ld&r,=±

where [/?] = 5(£,?7)| and J are calculated at £ = 77 = 1/3.

Exercise 1 Prove that N, (£, 77) and Nt(x,y) have the same value although they have different
form, for example,

82
 1
Nx(4,ri) = g and N, (x, y) =—(«,+ fixr + yxz) .
2A
Hence we can write
Ni(t,rj) = Ni(x,y)

3 3
Hint: Put x(£,Ti) = YjNixi,
i=i
y(ÿrj) = YJNlyl into N,(x,y) =
i=i
—1 (a, + fitx + yty) to obtain
Nt(x,y) = e, ...

4.6.3 Gauss Quadrature for a Triangle

One-point Gauss Quadrature : I= £/(ÿ,r|)r£4 « w,/(!;,, TJ,)
where w, =1/2, (4,,ÿ) = (1/3,1/ 3)

3-point Gauss Quadrature : I= J f(£„ )dA

T| »
/=1
w,/i(£,, , T)y )

where w, =1/6, (ÿ , r|y ) = (2 / 3,1 / 6), (l/6,l/6), (l/6,2/3) for / = 2,3,4

4
4-point Gauss Quadrature : / =
J /(£,, T|)dL4 «
/=1
w,/(£,, , rjy)
where w, =-27/96, (ÿT],) = (1/3,1/3) and

for / = 23,4 w,= 25/ 96, (ÿ!,rjy ) = (3/5,1/ 5), (1/5,1/5), (1/5,3/ 5)

Tl *1
n

A
4 A A 4 4

4.7 Thermal Stresses and Vibration Problems

4.7.1 Thermal Stresses

83
When there is initail strain, force vector on the righthand side of element equation should include

For temperature rise of T(x,y,t) , the thermal expansion is treated as initial strain.

1
{*„}= 1 \aT={Ilh}aT
0

where a is the thermal expansion coefficient.

If we use the interpolation T = [iV]„, {j1}, it becomes

{f.)ÿ[[Bi[D\{l,t}[N\hdV{T} = a[[Bi[D][B„\c,V{T}
where

where

Exercise 1 A ID element shown use 2nd order interpolation for u and 1st order interpolation for T.
Calculate \Klh\
c/j dÿ d2
r r 2*r
3Q

Tx T2
4.7.2 Vibration
When there is an accerlation of the body, the inertia force can be treated as a body force and force vector
on the righthand side of element equation should include

{M=\A[N)T{b}dA

84
where

{b\ = -p{u} = -p[N]{d(tj\

and, then

where

[m] = pHNf[N]dv

In case of free vibration, there are no other force terms and the element equation becomes

When = {Z)}sinty? is used in the equation, it reduces to an eigenvalue problem

{k}{D} = o)2[m]{D}
with CO
2
being natural frequency of the system and {/)} is the mode shape.

85
4.8 Higher Order Elements
4.8.1 3-Node Truss Element

For Quadratic Element,

In matrix notation,
\d} = \d2

2) Select displacement functions.

u = [Nx
d3
dx

N2 N3]\d2
4

d>
I

LA
4

I
_
We derive the element stiffness matrix for a 3-Node truss element using Principle of Virtual Work and
Minimum Potential Energy Theory

1) Select element type

x

d3,f3
fbid3
P

A,

d2,f2
fbl’h
or u

with
2 2 2
W2(x) = -ÿ-(x,-x)(x3-x), N3ÿ) = -J3;(X1-X)(X2-X)
Similarly,

x = [iV]{x(}

3) Define the strain/displacement and stress/strain relationships.

£ =

where

4ÿ=fe
1rBi1 = dx dx dx dx
Hooke’s Law

(j = Ee or a = E[B\{d)

86
4) Element equation by Potential energy approach
n=u+v

= \lasdv~1 "fA~ l’‘ld*~'E,d,Pi

= \1W [sfiT[S]{rf} Afr- jjrff [JV]r fA-[{4 tdx-{d}T {P}

n=MrgW{‘'|-U)-UHÿl)
where

6) Treatment of force terms

Assume , then

{x)=iMrM4‘}
7) Element equation
Variation od the potential energy should be zero;

(MM-M-M -MH
Finally, we have

MM-M
where

87
4.8.2 Shape functions for 8-node quadrilateral element

4 *i
y\
8
1
1 5 2
Mid-nodes

Nf\4,n)=\d-m-n2)

Corner-nodes

<’«,>?) = ) -|Af’

Characteristics of shape Functions

<=i

2)Ai<ÿl7y) = |
1
0
if
if
i=J
i*j

Example 1. 3-node bar element using one-point Gauss quadrature

1 3 2
x

L/2 L/2

*1 x3 x2

88
 [£]=J,[B(jr)f[D][S(x)]dv
= ljB&f [D][Btf)]Jd
= f >«)]>P«)]ÿ
= 2/"'U
= [ÿ(O)]' [Bp(0)]

= A'2«) = W,(£)=I-<?!

j
_ d* _ v-1 dNj{ÿ,Tj) _~ (l — 2ÿxl (l + 2ÿx2
-2£X3
~d4~ji‘ rf£ 2 2
<fly,
dx
_ 1 aw, _~
~
J d% J
—
l (i— 2£)
2 ’
_ l <flv2 1 1+2ÿ <flv3 l <flv3
~ =
~
dx J d$ J 2 ’ dx J d%
=~ = 2£
J

[£({)] = rÿL
dx dx
*EL±-[-\
dx 2J
+ 24 1 + 74 -*?]

Int = [B{rf[D][B(Z)]AJ = jj[-l + 2£ 1 + 2<T -4ÿf £[-l + 2£ l + 2£ -4£]

H=o = 2(X2A3E-x,) [-1 1 Of [-1 1 0]

1 -1 0
w- A3E -1 1 0
*2 ~*l Q 0 0

Exercise 1.
Find the stiffness matrix of 3-node bar element using two-point Gauss Quadrature

M = !•[*«>] [D][m]M*r
= £'[s(|)]r[n][5«)]ÿ
=([s«)]r[n][B©f/«)) +([B«)]r[n][fi(f)]G(f))

89
Example 2. Calculate the body force terms from gravity in x-dir. for 3-node element which uses
quadratic interpolation for displacement and linear for body force.

["-WjAfM* "here M = [ÿ, AT,,] and [Af] = [AT„ tf„]

2
Nu'(x) = IFÿ2 ~ ~ x) "

Nu (x) = jr(x2-x), N2I (X) = -J(X,-X)

Then,

NldNu NldN2l
N<i2 NMf[Na Nu]dx=£ N2dNv N2dN2l dx

and

wi-MA}
Example 3. Calculate the traction force terms for 2-node element subject to a constant traction
t0 along its axis.

* tflA 1
dx
N2tNu N2,N2t
For example,

«M1 = £ KKdx = j£ (x2 - xfdx| =

«,12 = «,21 = £ NUN2A = £ (*i - x)(x2 -

n,22 = £iv2<ÿ=ÿ£(x1-x)2ÿ=|
w-ftl
90
 *1H3)
Exercise 2 By treating the inertia force as a body force, determine the mass matrix for a 3 node truss
element

4.4.3 Shape Functions for 2D Transient Quadrilateral Elements

*1
+1

PROBLEMS +1
-l
Problem 1. £
L, A, E
1 3 2
-1
r=-1 r=0 r=+1
a) To derive the stiffness matrix for a 3-node 1-dimentional bar element with varying cross sectional area,
write the total potential 11 .
b) Write the virtual work principle in terms of u(x) .
3
c) Write shape function N] ( r ) in the expression of x-directional displacement u(x) = N, (r) di
<= l

d) Use the Ritz method to get the stiffness matrix k= J,EBrB dV where

B= dNx dN2 dN2'

dx dx dx

e) Show that coordinate mapping gives the stiffness matrix in the form k= dr , where

fay, dN2 dN2

B
dr dr dr
with J= —drdx
f) When A(x) = AQ , determine kn , the component of 1st row and 1st column of k by 1 point Gauss
Quadrature.

Problem 2.
Construct a system which has circular cross section and the radius is varying with 2nd order polynomial in
axial direction. A concentrated force is applied at the tip. Assign a suitable numerical value for
geometry and material constants and the force. Use one element with 3-nodes to find a deflection at the

91
tip. 3
2
It k_it
A A
Problem 3.
A
Even though the total node number of a finite element model remains the same, the use of the higher
order element gives better results. Explain why.

o o

a) 8 QUAD4 elements bl 2 OUAD9 elements

Problem 4.

a) Write the shape function for the mid-node #5 of 5-node transient element shown.

b) Write the shape function N\S) (£, Tj) for the comer-node #1 in terms of Aÿ,(4) , JV5(5)
+1
+1
*
Problem 5. A 2D structural system shown is to be analyzed by FEM model compoded of a beam, a
truss, 4 quadrilateral elements as shown.
a) What is the number of global d.o.f.(the dimension of the global stiffness matrix) ?
b) State the displacement boundary conditions.
c) What becomes of the number of global d.o.f. after applying the boundary conditions ?

Pined
Pined

P
<dl
i
Pined Roller
Fixed

92
Problem 6. For a 2D 4-node quadrilateral element shown, nodal displacements are found to be
dlx = 0, d2x = 100/im, d3x = 200pm, dAx = 50pm , dly = 20pm, dly = -50pm,
d3y = -50 jum, dAy = 150pm .
a) Write the shape functions Nt (x), i = 1, 4 in terms of X .
b) Determine v , the displacement y -direction, at point A (2,1)
c) Determine strain component yxy at the same point A.
t y

©ÿ 2
p
-4 4 x x

-2 3
11
d) The nodal force vector contributed by uniform body force fbx =0, fby = pg = const. , is given
by = {fb}dA , where f is the thickness, is the element area. Calculate the

using one point Gauss Quadrature for the element shown.

Problem 7. Plane Strain Problem

A linearly distributed force w(x, y) , with its intensity of w0 = wdz = \x\06 N / m at >' = 0,

are applied on a rectangular block of bxhxt = 40x30x20 mm which is fixed at two edges as
shown. Material constants are E = 200GPa, V = 0.3 .
1) Assume the problem in plane strain condition. Find displacements dix,d3yat node #3 and stresses

crx,cry, Txy at the mid point A. Use one element with the node numbers as shown and one point Gauss

quadrature. Don’t use a commercial FEM program; use calculator only.

2) Find displacements at node #3 and stresses at the mid point A using a commercial finite element
program. Solve it with two models: a plane strain model and 3D model. Use one element in both cases.
3) Solve the problem using multiple elements(for example, 4 x 3, 40 x 30, 400 x 300) with plane strain
model.
4) Discuss about what you learned from this problem.

93
 y
A, fixed
fixed

h = 30mm
•A
tr T"
1-
fixed x 2
fixed
b — 40mm

Problem 8. Plane Stress Problem

A rectangular plate with a hole at the center is being subject to a uniformly distributed force at both ends.
Obtain the stress concentration factor K using a commercial FEM software and show them in a graph
together with the exact solution from the literature and make some discussion about your results. Do the
analysis 4 times with increasing number of the finite elements between 20 -2000. All the dimensions
and the material properties can be chosen at your own choice. Use the symmetry in the analysis.

4o or #0 b

a
Problem 9.
a) Using 1 point Gauss Quadrature, find the area A of an isoparametric element shown.
a) For a 2D 4-node isoparametric quadrilateral element shown, find the x-y coordinates of the point C
that is mapped onto point C'(0.5, 0.5) in natural coordinates. The coordinates of the nodes are in the

unit of [c/w].

b) Nodal displacements are given in the unit of jum by

= 20 100 -50 200 -50 50 150]r and thickness is t-2mm. Find the

strain and stress components at the point A assuming the plane stress conditions. Material constants are

E = 200GPa, v = 0.3 .
c) Calculate [5] at the integration point of one point Gauss quadrature.

d) Calculate the stiffness matrix [&] of the element with one point Gauss quadrature.

e) When the element is fixed at the node 1, 2 and 4 and is subject to a force with

94
Fx = 5KN, Fy = 3KN at node 3, determine the displacement at node 3.

- y
4(2,4)
3(5,5)

4 3
C
•c' £
2(6,2) *
1(M) x 1 2

Problem 10.
a) Code your own computer program using MatLab, Fortran, C++, or whatever to determine the
displacement at node 3 using 4-node isoparametric quadrilateral element with 1 point Gauss quadrature.
Compare your results with those from problem 1 and from a commercial FEM program.

b) Modify your program so that it can use 2 x 2 points Gauss quadrature to determine the displacement at
node 3. Compare your results with those from a commercial FEM program with some comments.

c) Modify your program so that it can use 2x2 elements with 2x2 Gauss quadrature to determine the
displacement at node 3. Compare your results with those from a commercial FEM program with some
comments.

Problem 11.
A tapered bar, fixed to the walls at both ends, experiences a temperature rise. Using 5 linear elements of
the same size, find stresses and the displacements at the nodes by means of a commercial FEM program.
The length of the bar is L- 400mm and the areas at the ends are = 90mm2, = 100mm2 .
As
The material properties are K = 215 J/m-s°C , a = 2xl0~5 / °C , E = 96GPa, V = 0.32 .
The nodal temperature rise are; Tl=20°C, T2 = 50° C, T3 —l 00° C , T4=10°C, T5=0°C.

1 2 3 4
f f
n

95
Problem 12.
Solve the same problem with two of quadratic elements and make some comments

* 2 3 4 5
o 4 O

Problem 13. Find the in-plane natural frequencies and the corresponding mode shapes up to 5th mode
for the steel plate with thickness t = 1mm and material properties

—
E 200GPa, V = 0.3, p = 7.3g / cm3 fixed at one end as shown. Assume plane stress condition.

tH

o r = \0mm
b = 100mm

a = 200mm

96
5. Imposition of Constraints
5.1 Lagrange Multiplier Method
5.1.1 Minimization of a Function with a Constraint
The use of Lagrange multiplier method can be best explained with a problem in which the maximum
value of a function defined by f
— —X2 + y2 with a constraint X2 + y2 = 2x is to be sought.

In this simple example, both x and y cannot be independent variables; rather either one of them can
be an dependent variable since they are related through the constraint equation. In fact, by expressing
y2 from the constraint equation and substituting it into the function f , we can get f in terms of X
only. The condition
therefore, is /max =1/2.
df / dx — 0 , then gives x = 1 / 2 for maximum value value f which,

To apply Lagrange multiplier method to the problem, first, transform the constraint equation in the form
of g(x, >0 =0 and construct a function H(x,y,l) = f(x,y) + l g(x,y) where y is the
Lagrange multiplier. The maximum value of f with the constraint is now given by the maximum of IT
with X, y and 1 as the independent variables.

For the problem at hand, for example, the required function is given by

n = -X2 + y2 + l(x2 + y2 - 2x)

IT and require it to be zero;
To fine the maximum of II , we take variation of
m = -2x5x + 2y8y + 5/l(x2 + y2 - 2x) +1(2x5x + 2ydy - 25x)
= {-2x+l(2x-2)}5x + (2y +12y)5y + (x2+y2-2x)51 = 0 \/Sx, 5y, 5y

Each term must be zero and, hence we have the following simultaneous equations
-2x + l(2x-2) = 0
2y + 12y = 0
x2 +y2 - 2x = 0
The equations can be obtained also from
8U en
—
dx
=0, — =0,
dy dl

The solution of the equations are X = 1/2, y = -yj3/4 , 1= —

1. With the values of X and y , we

have

/max =n max = 1/2.

97
5.1.2 Variational Problem with a Constraint
The Lagrange multiplier method can also be applied to the variational problem with constraint equation.
For example, the isometric problem is to find the equation of the curve with length L which passes
through the fixed points (a, ya ) , (b,yh ) and possesses the maximum area under the curve. The
problem can be put in the form:

Minimize Ay)= [y(x)dx with constraint dx-L = 0

We first construct a functional

Taking variation of II and requiring it to be zero gives

y'Sy
<5n = fcSydx+Syÿfc yjl+(y')2dx-Lj+y £ y/l+y* dx =0

Integrating the third term by parts, we get

b

sn = Ay
yfi+y7Sy a +t 1-—
dx
d
Vi+y2 ,
'ydx + Sy|£ -y/l + yadx - L j = 0
The first term on the right is zero because Sy = 0 at y = a, b and therefore, the second and third term
must be zero and we have a differential equation

d Ay
I —d*{yli+y2 =o

as well as a constraint

£ y]\ + yadx = L .

The solution of the differential equation can be found to be

(x + C,)2 +(y + C2)2 = A2
where C, , C2 are y are the constants to be determined from the constraint equation and other
conditions.

98
5.2 Imposition of Constraints onto a System of Equations
Consider the following problem:
Find the solution of the equation
Kd = F
with
Bd = V
where K is symmetric and B , F , V are the known matrices.

The variational formulation corresponding to it can be given as follows;

For the function defined by
1
n = -drKd-drF
2
, find d so that <5TI = 0 for all d with
Bd = V

5.2.1 Lagrange Multiplier method

Lagrange Multiplier method make it possible to include the constrains in the function FI* :
With the definition

n* =idTKd-dTF + 3J(Bd-V),

find u{ so that <5TI =0 for all ui and Xj

Taking variation, therefore, the problems becomes

K BT d F d
with unknown
B 0 X V X

Where X is the unknown Lagranger multipliers and the reaction force is given by R = BrX, .—
Example 1. ID system with constraints
Consider the system shown. There are three displacement boundary conditions, all of which can be
treated as constraints. The equilibrium equation, after applying the force boundary condition, becomes

3 -3 0
-3 6 -3
4
d2 = 12
fi r P2= 12
d3= 2
0-3 3 d3 A 1 k 2 k 3
k =3

99
with constraints d] - 0 ,d2 = 2 .
The solution for the problem is analytically found to be d2— 3 , f\—~9 , f3=—3.

To apply Lagrange multiplier method, by comparing the equations, we have

3-3 0 4 0
'1 0 o' 0
-3 6 -3 , d = \d2 k F = U2 , B =
0 0 1 •K-2
0-3 3 A 0

In the equation above, Fx and F3 are set to zero, because F represents the prescribed external force
vector.

Since there are only two constraint equations, two Lagrange multipliers are required. It, therefore,
follows

x= A
A
and hence, the augmented equation is

'3 -3 0 1 O' 4 o
-3 6 -3 0 0 d2 12
0 -3 3 0 1 d3 0
1 0 0 0 0 0
0 0 10 0 A 2

Solving, we get
dx= 0 , d2 = 3 , d3 = 2 and =9, =3
Then reaction force is given by

-9]
—
R = BrJi = 0
-3

which means that R is required to impose the constraint conditions on the nodes.

5.2.2 Penalty method

In penalty method, the functional is defined by

n*=|dTKd-dTF + (Bd-V)

100
where a is a diagonal matrix with penalty numbers <2, ?••*? ccm which are the given numbers and must be
large enough compared to the stiffness.

Minimizing II gives the simultaneous equation

(K + BTaB)d = F + BTaV
The reaction force is given by

R = BTaV-(BTaB)d

In the matrix K + BTaB , the penalty numbers can be placed off the diagonal. In this case, if off-
daigonal terms become too large, the matrix can be ill-conditioned.

As can be studied in the discussion about the matrix partitioning method, K is the singular matrix while
the augmented matrices in both methods are not singular. In applying the methods, it should be noted
the reaction forces caused by the imposed constraints should not be considered and hence, the total nodal
force vector is given by F+R.

Example 2. ID system with constraints

Since there are two constraint equations, two penalty numbers are required. We have

a, 0
a=
0 a2
and hence, the augmented equation is

'3 + a, -3 0 dx 0
-3 6 -3 d* = 12
0 -3 3 + a2 4 2a,
When a,=a2=20, dx = 0.40783 , d2 = 3.56334 , <73 = 3.12667 , from the equation, the
reaction force is

R = [-9.1566 0 -22.5334],

When a,=a2=100, <7,=0.08783, d2 = 3.01565, rfj =2.03131, from the equation, the
reaction force is

R = [-8.783 0 -3.131],

101
Example 3. 2D system with constraints
In this example, a rod with pinned condition at one end is to be analyzed. The other end of the rod can
move along the smooth slot. When P is applied as shown, find reaction from the slot and displacement
of the tip.

2 fu
jfc =l 30° 30°
u,f 30° 30°

y
1 1
X p/A p /'V
Solution for the system can be easily obtained by moment equilibrium about the node 2. The reaction
from the slope is N = f3P . Due to its supports at the ends, the rod should experience axial force only.

Hence axial force in the rod is given by \lN2 +P2 = 2P and the deflection is A = 2P / k = 2P .
Simple geometric analysis gives the displacement of node 1 along the slot as A/ sin 30° = 4P, and
finally we have

dXx = 4Pcos60° = 2P

dXy = 4Psin 60° = 2-v/Jp

Applying the finite element technique, the global equation is given by

"

1 -73 -1 73 du fu
J_ -73 3 73 -3 K fly
4 -1 73 1 -73 d2x fu
73 -3 -73 3 rÿ. fy
Boundary conditions for displacements are

rf2,=rf2,=0, V3rf,,-rf„=0

Taking care of the force boundary conditions we have a simultaneous equation for 7 unknowns

102
 1 -y/3 -1 VT 4, JL-AN
2 2 iV

1 -y/3 3 V3 -3
4 -1 >/3 1 -V3 *2, /*,
V3 -3 -V3 3 d2y fly

with the constraints

4,-4,= o, V3d„-rf„=o
The solutions are

N = f3P, dlx=2P, dly = 2-v/Jp

In sloving the equations, we can take care of the first two boundary conditions by the partitioning method.
However, the boundary condition for the type of the last equation can not be applied with the use of the
matrix partitioning method.

Solution 1 Lagrange multiplier method

i -VJ -i VJ" £
2
0 10 0
0
_ 1 -V3 3 >/3 -3 4P
K~4 V3
, F= , B= 0 0 0 1 , v=\0
-1 1 -J3 0
V3 -1 0 0 0
>/3 -3 ->/3 3 0

A
Note that F doesn’t include reaction force N from the slope. Moreover, since the zero displacement
condition at node 1 is taken care of with the constraint equation, the corresponding force components in
F is zero.

103
 1 -A _±
4 0 0 yfe p.
4 4 i
du 2
-A
4
i
4
A
4 -10 0-1 AL
2
_i
4
A4 i
_AI -4 1 0 0 d2x 0
A
4
-i
4 4
i
4
o 1 0 d2y 0
0 0 1 0 0 0 0 o
0 0 0 1 0 0 0 0
V3 -1 0 0 0 0 0 0

Solving the above equation we get

£
dlx = 2P, dly = 2y/3P, d2x=d2y=0,A3=ÿ-P
The reaction force is given by R= Pÿ— \ dL 1
J and the magnitude of the reaction from

the slope is Af = RXx + R2x = \/3P .

Solution 2 Penalty method

a, 0 0
a= 0 a2 0
0 0 a3
where diagonal matrix of penalty numbers ax,...,am must be large enough, but if off-diagonal terms
becomes too large, the matrix can be ill-conditioned.

4 +3“3
_±
4
Ai d,x P.
2

-ÿ-X/3“3 4+“3
A4 dry AL
_1 A i
4+ar
_A t

d2x
2
0
A4
4 4
_2
4
_A 4
3
4+«2
4
4, 0

When ax = a2 = a3 = 20 , the solution is

2.13750P 2

3.65896P V3P
u= 2

-0.05000P P
0 08660 P F*y -'TIP.
When ax = a2 = a3 = 100 , the solution is

104
 _!£
2.02750P
3.50307P
U= =
-0.01000/’ p
0.01732P -yfcp

The reaction force is given by R=

P£— \ -y- 1 ->/3 J . As long as reaction is concerned, exact

solution is obtained.

* When we solve this problem by the two methods, we did not care about the smoothness of the slot. In
fact we focused on the fact that the displacement relationship only. But, surprisingly enough, the result
corresponds to a smooth condition. If there is friction on the slot, the magnitude and direction of the
reaction force would be dependent on the coefficient of friction. In fact, the problem involving the
friction can be solved by a nonlinear analysis. In this linear analysis, therefore, only smooth boundary
can be treated.

Exercise 1 Solve the example 2 by a commercial FEM software.

105
PROBLEMS

Problem 1
Find the solution of the equation using Lagranges’s multiplier method

2-2 0 4 fl
-2 3 -1 dA = 200 with dl = 2d2
0 -1 -1 2 fi
Problem 2
Find the solution d2,fi,f3 of the equation using Lagranges’s multiplier method and penalty method

2-2 0 0 fx
-2 3 -1 d2 200
0 -1 -1 2 fi 3cm 4cm

Problem 3
Find the translation A and rotation (j) of a compound
cube block subject to a force F = 100KN at the center
of rigid plate. Use finite elements only for brass and
r
Rigid Plate

aluminum and use MPC(multipoint constraint).

Aluminu
7cm
I
!B rass
1
Rigid Plate

Problem 4
r Thickness =7cm

For the system shown, find the deflection at A using one element each for aluminum and
brass rod and rigid links for the horizontal bar. The length and area of the bars are
A = 100mm2 , L - 300mm . F = 10KN

Aluminum rod Brass rod

m
Rigid bar

200 mm 200 mm 200 mm

106
F
Problem 5 A bar is hinged to a fixed beam, as shown. Force F is assumed to be large enough to
make node #2 in contact with a smooth stop which is 0.1 units away. Finite element analysis is desired
using a bar and a beam element. The beam element cannot experience an axial deflection and, therefore,
the nodal d.o.f. at node 2 of the bar element is d2Y only. Assume A = 1, E = 1, L = 1 for bar and
El = \,L = 2 for beam.
a) Write the element equation [ÿ]{<i}={F} for the bar element in global coordinate system

when the nodal displacement vector is defined by {d}= \dxx dlY d2Y f"
b) Write the element equation [ÿ]{<i}= {F} for the Euler beam element in global coordinate

system when the nodal displacement vector is defined by {<i}= \d2Y (j)2 d2Y (f)-ÿ .

c) Assemble the stiffness matrices to get a global equation for the system [/£]g{d}g = {iÿ}g .
d) Applying the boundary conditions at nodes 1,3 and partitioning the matrices, find the global

equation in the reduced form [.AT, x ]r{d}r = {ÿ}r , where {d}r = [cl2Y <j)2 J7
e) Finally, apply the Lagrange multiplier method for the constraint d2Y = —0.1 to find the
rotation angle (f>2 at node 2 and the reaction R from the stop.

i
tl
£
,45°

oTT*
3

107
Problem 6 A spring fixed at one end is being subject to a force P at the other end which is restricted
to move along a smooth slot. Find the displacement of the end and reaction force R using
a) Penalty method with a = 100 A:
and k
b) Lagrange multiplier method

y P,
60°

Sol. Lagrange multiplier method

0 0

d2x
T",)# - [ÿ*

2
P

Problem 7 A pulley with outer diameter of 100mm, inner diameter of 49.95mm is shrink-fitted onto a
shaft with diameter of 50.00mm. It is required to find the contact pressure p. Es= 150GPa ,

vs = 0.25 Ep = 200GPa, vp = 0.3

Imagine you are in front of a computer equipped with the commercial program you have used for the fern
homeworks so far. Explain all the procedures you should take in order to get the result, step by step and
in detail. You need to show the command/menu of the program.

108
6. Axisymmetric and 3D Elements
6.1 Axisymmetric QUAD Element
pressure p
A Z

r>
rt
r> Axisymmetric
r* element
!-ÿ
S. 9
4
l 7 r>
\ 3
a
<2H£ 4 iL

tz r

Axisymmeric solution exists only when both the geometry of the body and boundary conditions are
axisymmetirc. When which the axisymmetric solution is expected for a 3D problem, axisymmetric
analysis can be performed. In static problems, the displacement v in 0 -direction is zero and the
displacement fields u,w in r - and z - direction, respectively, are independent of 6 . Hence, we
can write
u = u(r,z), v = 0, w-w{r,z)

When the assumptions on the displacements are applied to the strain-displacement relations, we can write
Yre = 762= 0
du u dw dw du
/rz
dr dz

When the stress components are considered, Tr0, T0Z are zero and, hence, they do not contribute to the
strain energy. The remaining stress components crr,crg,<J2,Trz contribute to the strain energy and
they must be included in the derivation of stiffness matrix. Therefore, rather than using the general three
dimensional stress and strain in the axisymmetric analysis, we use in the derivation of element equation
the four components of stress and strain defined by

109
 °7

°e
. w= £z
£e
Trz Yrz

The strain-displacement relation can be put in matrix form

W-MM
where

r* 01
o
M- I i0 . {“}=
u(r,z)
w(r,z)
A. A.
.Sz dr_

Hooke’s law is given by

M-MM
where

1 1-v 1-v 0
E(1-v) MT 1 1ÿ7 0
(l+v)(l-2v) 1-v 1-v 1 0
l-2v
0 0 0 2(l-v)

In the finite element derivation, linear quadrilateral isoparmetric axisymmetric element will be discussed.
The nodal displacement vector is defined as

M = [«I w, • • M4 Wa]T
where ui,wi represent the nodal displacement in r z - directions, respectively.

The displacement fields U, w in the element are interpolated using the relation

M-MM
where

110
 N, 0 •• N4 0
0 JV, • • 0 NA

Substituting the equation into the equation, we, therefore, have

M-MM
where
dN,
U U 0 SN4 0
dr Sr dr dr
i
SN2 0 8Nt
dz Sz
W-m- o i 0 3_ 0 3. 0
dN,
. dz
3Nj
dr
3N2
dz
dN2
dr
dN3
dz
dN,
dr
dNt
dz
dNt
dr
_
Stiffness matrix
The stiffness matrix is given by the formula

where dV = rdddA is the differential element and the integration is done over the element volume.
Integrating in 9 -dir. can be done analytically since the integrand is independent of 6 . The stiffness
matrix becoms

[*L = f" [D][B]rd9dA = 27z\A[B{r,z)1( [D][B{r,z)]rdA

where dA is the differential area and A represents the area of the axisymmetric finite element in
r—Z plane.

finite element

differential area dA
c •finite element area A
Cn
r

When the formular is described in terms of the parametric coordinates, we can write

[*] = 2»Jÿ [ms,t7)f )] r«,!7) [D\[B(ÿ

dA,

where Ap is the mapped area in parametric coordinates of real element with area A .

Ill
6.2 Axisymmetric Triangular Element
6.2.1 Derivation in Physical Coordinates
1) Displacement interpolation
Consider a three node triangular element. The displacement in r-direction in the element, u(r,z) can
be interpolated with three nodal displacements, ux,u2, U3 using bilinear equation
u(r, z) = ax + a2r + a3z
Using coordinates (rx , Z( ) and nodal displacement u; at each node, we can have
3
u{r,z) = YJNiui,
;=i

where the shape functions are

1
Ni(r,z) = iai + Pf + Yiz)
2A
with

ai ~ r2z3 - z2r3 , 0l=z2-z3, y, =r3-r2,etc.

A represents the area of the element and is given by the formular
1 ri zi
AA 2
1 r2 Z2
1 h

Similarly, the displacement in z-direction in the element, vf(r, z) is interpolated with three nodal
displacements, wx,w2, W3 in the form of
3
Hr,z) =
i=i

We can write the equations in matrix representation as

{H}=[V(r,z)]{rf}
where

[V] = Nx0
0 N2 0 N3 0
Nx 0 N2 0 N3
and

{</} = [«, w, u2 w2 u3 w3f

2) Strains sr, s„ £0,yrz *n terms of displacements u(r, z) , w(r, z)

112
The strain-displacement relation is the same as in the case of QUAD element and they are rewritten here
for convenience;

M-klM
Substituting the equation into the equation, we have

M-[*M
where

A o A o A o
1 0 r, o r2 0 Yi
0 T-+A+2T 0 T+A+2r 0
n A r2 A A
3) Stress/Strain Relationships
The strain-displacement relation is the same as in the case of QUAD element and the same material
constant, equation, is used.

4) Stiffness matrix
The stiffness matrix is given by the formula

[*] =
where dV = rdOdA is the differential element and the integration is done over the element volume.
Integrating in 6 -dir. can be done analytically since the integrand is independent of 0 to give

[*L = 2n JL[B(r’zrf [D][B{r,z)]rdA

where dA is the differential area and A represents the area of the axisymmetric finite element in

—
r z plane.

Example 1. Calculate the stiffness matrix of a traingular element using one-point Gauss Quadrature
—
Using one-point Gauss Quadrature in r z domain,

where is evaluated at the Gauss point (rc, zc) which happens to be the centroid of the element

area and T
— rc.
6.2.2 Derivation in Parametric Coordinates

113
1) Introduction to the area coordinates
A point in the element can be positioned with two of three subareas A{, A2, A3. The corresponding
point in the mapped equilaterial right triangle can be positioned with natural or area coordinates by

A
JJ=AA
If we define the shape functions as N\(£,rj) = % , N2 (£, 77) = 77 , N3(£, T]) = \ — — rj, then the
displacements and coordinates of an isoparametric triangular element are interpolated in the parametric
coordinates as follows;

1=1 i=l

«(& n) = Z N'u>
I=I
» Ji)=H Nw
i=i

5=1
T|=l
4 2
A A
A
3 5=0 3 i i
r

2) Stiffness matrix
The stiffness matrix is given by the formula

[£] = In [B(r, z)f [D][B(r,z)] rdA = In [£(£, 77)f [D][fl(£ 77)] rJdAp
£ £
One-point Gauss Quadrature will be used for a triangular element, then

{ÿ =
where J = 2A, Ap =1/2 and = B(ÿc,rjc) is calculated at %c=Tjc =1/3.

Surface force(Traction)

where t represents a traction force per unit surface area

114
Since ds = rd6dsx , integrating in circumferential direction
(line integral)

Exercise 1. Prove that Nx(ÿ,Tj) and Nt(r,z) have the same value although they have different
form, for example,
1
N1(ÿ,rj) = g and Nx{r,z) =—{ax+pxr + yxz).

and hence we can write

Nfatf) = Ni{r,z)
1
Hint : Put r(Z,ri) = YjNiri’ z(.Z’rl) = TjN'z‘ 'mto Ni(r,z) =-(«, + fj(r + ytz) to obtain
1=1 /=! z/x

NXr,z) = Z, ...

Exercise 2. The force term due to the boundary traction is given by {F}= {t}ds . Calculate
the nodal force vector {F} for the 3-nod triangular element with the traction shown. The coordinates
are in the unit of mm’s.

y
1(0,3)

2(-5,o;
x
3(0,0

wo=10

6.3 3D Solid Elements

Hexaheron element, Tetrahedral element, wedge(triangular prism) element

A
. ...A

115
 f u(x,y,z)
Displacement field in the element : M =|v(x,y,z)
\w{xty,z)
Nodal displacement vector:

M = [w i v, w, • • v„ w„f
Shape functions :

'JV, 0 •• 0 0
0
[N]= 0 Nt • • N„ 0
0
[o 0 JV, • • 0 JV„

Interpolation:

W-MM
Stiffness matrix:

MM*-
where

1 1— V 1— V
0 0 0
l-v 1 1-v 0 0 0
£(l-v) l-v l-v 1 0 0 0
(1 + v)(l - 2v) 0 0 0 2(l-v)
0 0
0 0 0 0 2(l-v) 0
l-2v
0 0 0 0 0 —
2(1 v)

and

i o «'
f- o • • 0
0 i 0

M- i i io
0 0
[8]-WM- 0 aw, • dy
# 0 SNA
IT •
aÿL dN± dNA
# # s_ -L
dy dx dy dx 0
v dz Sy
A 0u ex

116
 i
.

PROBLEMS
Problem 1
Find the translation A of an axisymmetrical block subject to a force F = 100A3V at the center of
rigid plate.

mmmm

Problem 2
A thick-wall cylinder with di = 200mm, da = 240mm, L = 420mm is fixed to the rigid walls at
the both ends.
a) When the internal pressure is pt = 4000 KPa , calculate the stress components <Jr,<Je,CFz,Trz at
the inner radius and the increase of the inner radius Ar. by means of a commercial FEM program. Use
the axisymmetric elements. Compare your results with an analytical solution available in the literature
which is close to this system. E = 200 GPa , V = 0.33
b) When E = 400 GPa , calculate the stress components again. Are their values changed?

Problem 3
The plane strain and axisymmetric QUAD4 element have the same nodal displacement vector , but
their material constant matrix [£)] are different in their dimensions. Explain why.
Problem 4
A pipe of infinite length with wall thickness of dt = 40mm, da = 300mm is subnect to a internal
pressure of pi = 4000 KPa .
a) Calculate the stress components <Tr, <7e, (Tz , Xrz at the inner radius and the increase of the inner
radius Art by means of a commercial FEM program. Use the axisymmetric elements. Compare your
results with an analytical solution in the literatures. E = 200 GPa , V = 0.33
b) When E = 400 GPa , calculate the stress components again. Are their values changed?

117
7. Beam Elements
7.1 Euler Beam Elements
Assumption in Bemoulli-Euler Beam: Plane section originally normal to the neutral axis remains plane
and it does remain normal to the neutral axis. Hence, the deformation of the beam is solely governed by
normal strain from bending moment.

n= - jf <lvdX - jf fbfVdx ~
Jf fbmV >dx ~
X PA Z miPi
(=1
~
i=l

where V " = d2v / dx2 = K is the curvature and v' = J3 is the rotation of the cross section, positive in

counterclockwise. q,fbf , fbm are the distributed force, body force, body moment, respectively, per unit

length. pt , mi are the concentrated force and moment applied to node i . The positive direction of the
force terms are in correspondence with the mating displacement terms, and nn is the number of nodes.

Taking <5TI = 0 gives the weak form

£ v" EI Sv"dx= £ qSvdx + £ fbfSvdx £ fbm8v 'dx-ÿ pi5di -J

+
<=1 /=!

The left-hand term is strain energy increment due to virtual rotation S/3 ;
£ EI v" Sv"dx = £ MSÿÿIx = \MS(d/3) = JMd(Sfi)
M + dM '

df3
M

At x dx
P

Consideration of Convergence
For monotonic convergence, the elements must be complete and the mesh must be compatible.
1. For compatible, in Cm_1 variational problem the displacement function vmust be continuous up to
(m-l)-th order on the element boundary. For example, in case of Euler beam element, the functional
includes the term v" and, therefore v' must be continuous.
2. For complete, the displacement function v should be able to depict the rigid body displacement and
constant strain states.

If V is represented by an interpolating function using nodal displacement only, v' may not be

118
continuous on the element boundary. One method to avoid the discontinuity is to use as the nodal d.o.f
the translation ds and the rotation /? and to interpolate V by means of them.
We, then write

i=i

In matrix form, if we use the same nodal displacement vector of equationQ, equationQ can be

V = MM with [#] = [#, N2 N3 and {</} = [<* A, - A.T

Nodal Values in Physical Space
Consider a two node beam element. The field variable, namely the displacement v(x) can be represented
in the physical coordinates as

V(*) = [AT(*)M win. M=[W, W2 N, JV4], A 2 Aif-

It then follows that the nodal displacements and nodal rotations are defined, respectively, by
dl=v(xl), d2 = v(x2) and /?, = dv(xx )/ dx fi2= dv(x2 )/dx
In case of two node element, Nt ’s are the 3rd order polynomials.

Exercise) Find [V(x)] for two node beam element.

The other method is to constructing V and v' separately. In other words, v is represented by nodal
translations only and v' by nodal rotations only:

v = ZAr,v/> P = YJNA-
In matrix form, if we use the same nodal displacement vector of equationQ, equationQ can be

v = [W,M,

with

[W.] = [V, 0 0 [AT,] = [0 N,

... 0], N1 0 N2 ...
In case of two node element, Nt ’s are the 1st order polynomials. In the expression of II , the 1st and
2nd derivatives of V are replaced by V = /? , v" = /?' .
1

Formulation of Finite Element Equation

In the derivations followed, we consider two node element with fbm = 0 . The vertical displacement is

119
written from equationQ as

v = [N]{d} where [N] = [NX N2 N3 NA] and {d} = [dx (3X d2 02f
Defining 8 = d1 / dx2 , the curvature v" = K can be written as
v" = av.
Substituting equationQ into () gives

V= = [*]M Were [W]= [ÿ

Since £v = [iV]{(W} , Sv" = from equationQ we have

jt£/([s]{w})r([a]{rf}>fc=jt([Ar]{M})r?A+Jt(MW)r4*+{ÿir{r}
where the last two terms of equationO is reduced to single term in the above equation by defining the
nodal force vector

M = [A m\ A mif •

Rearranging the equationQ, we have

{Sd)T £/[sf [B]k{d} ={Sdf ( [Wf

£ £ + £[Af flfdx {p})+

which should be hold for arbitrary . We finally get the element equation

where the stiffiiess matrix and the force vector are defined as

M =!>[*]>}* and

Mapping of deflection

v(x)

A 2 d1( 2
v(x) *(£)
A 1 d,{ 1 A
A
lA A
+i 4
X
L
120
Fig() show the mapping of the deflection curve of a beam element. Suppose the mapping relationship
X = /(£) between E, x = /(£) into v(x) , the deflection
and X are known. By substituting
v(£) in the parametric domain can be obtained, i.e., v(x) = v(f (£)) = v(E,) . Although
v(x),v(£) are different in their function forms, we prefer to use the notation v(£) to v(£) to
represent the vertical deflection in the parametric coordinates without confusion.

Nodal values in parametric space

The deflection v(£) in the parametric domain can be represented as

with
KHX NH N2I K}=[rf, A4 AJ
Since the vertical deflection at the corresponding positions of two domains remain unchanged even after
the mapping, namely, we used as nodal deflection the actual displacement used in equationQ and hence
dx = v(x, ) - v(£ ) , d2 = v(x2 ) = v(£ )

However, we note the slope of the deflection curve is changed after mapping and different symbols

should be used for nodal rotations. Nodal slope = dv(£) / dE, is the slope of the deflection curve

in E, domain.

Shape functions in parametric space

The shape functions J are defined so that they satisfy 4 conditions of equation(),(), and whence the

following 3rd order-polynomials are required

A{=J(1-«2( 2+Z), l-fftf+l).

N2(=i(l+£)2(2-<f), W4{=i(l+4)2d-l)
FigO shows the shapes of the first two shape functions

f*.
jil<fpe=0

sldpe=l
Slofy=0 Slope=0
>
£=-l <r=o 4=1 l = "l £=o
121
For example, the magnitude of is the unit at node #1 and zero at node #2 while the slopes are zero

at both nodes. The slope of N2ÿ is the unit at node #1 and zero at node #2 while the magnitudes are

_
zero at both nodes.

From the relation, we can write

dv(x) dv(£) d£ 1 dv(£)
dx d% dx J d%
where Jacobian is defined by
dx L
J= —
dt 2
Using the definitions of equation()(), we have

Pit, ~ JPi

In representing the deflection V in terms of nodal vectors, it is more beneficial in the further derivations

to use the nodal vector of equation() rather than equationQ since the (3i is the actual rotation while

is not. From equations(X), we can write

4
*€) = [** N2€ N3 Aÿ] A*
d2
= (
>
JN2if N34 JNh] P\
d2
P24 A
The last expression in equation () can be put in the symbolic form as

V«HV(£)]M with [AT«)]=[JV1{ JN„ NH

Comparing equationQ with equationQ, we find [iV(x)] = \N(J"(<ÿ))] = [ iV(£)] •

Stiffness matrix
From equationQ, we can write

dx2 J1 d e
which can be rewritten as

122
 ;
d2 d2
.
with d = —dx .
r- , dr
2
* = d?
Then, [.#(£)], the matrix [i?(x)] in terms of E, , can be obtained by the equation

1
[B(4)] = [5(x)] = 0[tf (*)] =-T0i [AT(fl]

rmulte
1 Ki>' £ 2 df df 2
kEEl
df
Using equationsQ and the relation dx = JdE, = (Z / 2)d£ gives the stiffness matrix

[X] = J[£f [if M&= ™ J>/[«(f)f [m]ds

Full integration of the equationQ gives the stiffness matrix of Euler beam
12 6Z -12 6
4Z2 -6Z 2Z2
sym. 12 -6i
4Z2

Force term from distributed force

Equation() make the nodal force vector from the distributed force be written as a function of E, as in the
equation

jf [N(x)]Tqdx=[l[N(Z)]TqJdt
When the force distribution q is interpolated by the force intensities at the end nodes of two node
element, we can write

with 0?]=K <], {e}={®}

where Qj=q(Xi ) is the nodal values of q and is linear functions of E, . Substitution of

equationQ into equationQ gives

where [F]=ÿ[Nÿ[N]]jdÿ

123
7.2 Timoshenko Beam Element

t
f —dxdv ~Y = P
/ 1

dv P /
/

dx
/
d\ /

dx\y /

v
x
x

For Euler beam, we take the bending effect only into account, excluding the shear effect. A cross
section initially normal to the neutral surface remains straight and normal to the deformed longitudinal
axis, hence, only V is required to describe the behavior of the deformation.

P33/
For Timoshenko beam, the deformation due to shear strain is considered as well as normal strain.
However, the shear strain distribution across the cross section, which is varied in the thickness direction,
is treated to be uniform by adoption of shear correction factor. Hence, plane section originally normal to
the neutral axis remains plane, but it does not remain normal to the neutral axis. Denoting displacement
in y -direction as v(x) , the total rotation of the plane originally normal to the neutral axis is
dv
dx—
P =- Y
where V is the displacement in y -direction and y is the rotation due to shear.

Since the plane section originally normal to the neutral axis remains plane, the deformation, therefore, can
be characterized by two variables, V for the translation and /? for the rotation of the cross section.
Displacement in x axis is then given by u(x, y) = —y/3(x) for a small displacement.

Shear correction facto r(Form factor for shear)

The strain energy per unit length due to shear stress is given by

124
 j

dU
dx
= f—G T(y)2dA
\2
where T(y) is the shear stress distribution across the section and the resultant of the shear stress is the
shear force V .

On the equivalent cross sectional area Ae , a uniform shear stress Te = V / Ae is assumed to exit and
we demand the strain energy be the same as for the actual situations. With the equivalent cross sectional
area Ae , the strain energy per unit length, therefore, can be calculated by
dU Te2Ae V2
= = '
dx 2G 2 AeG
Equating the equations with, we have

v2
4 = {yfdA
Jr A

Shear correction factor is defined as the ratio of the two areas;

k = Ae/A

For a rectangular cross section with height h and width b , the shear stress distribution is found to be

r(y) =
yv_ (ft/2)2-/
~2 A (ft/2)2
Substituting this into eq(), we can get k = 5/ 6.

* 1
A
if 41
I
4
4
I"
i r =VI At
r(y)

Table. Shear correction factor

Cross section k
A Rectangle 5/6
B I-section, Box Section, or Channels A„eb/A(major axis), A(/1.2A(minor axis),
C Circle 9/10
D Thin-walled hollow circular _ 1/2
A= area of the entire section, Aweb=area of the web ht, Af=area of flange, where h is the beam depth and t
is the web thickness

125
Development of 2-D Straight Beam Elements
Strain components are

where £x
strain on the cross section Ae .
du
dx
dB

is the normal strain over the area

dv du dv
sx =- = -y-£L y =-+- =
dx dxdydx

A in X
--
direction and y

Strain energy per unit length from these two quantities is written as
_
p

= yxy is the uniform shear

dx " 2 2
Using the relation () and (), equation becomes
dU
dx S/dA+?f f r2 =
where / is the second moment of area defined by

‘-If*
Since strain energy due to normal stress is accompanied by the angle change dp of two neighboring
cross sections, /?' is used in the first term of equation(), whilst v" is used in the Euler beam since
P' = v" with y = 0 . The curvature is obtained by K = v " for the two types of the beams.

n =U-W
= yjf(£ 'fdx + Jf r2dx - Jf qvdx - fb/v - jf fbmpdx -£ p,d, - J m.Sp
where q,fbf,fbm are the distributed force, body force, body moment per unit length, respectively.

pt , mi are the concentrated force and moment applied to node i . nn is the number of nodes of an
element.

The equation STl =0 gives

El £ p'S(p ')dx + GAk ySydx -

£ fbfSvdx- £ fbmS/3dx - £ qSvdx -J p,Sdi i—1

-±mM = o
i=i

For dynamic problems of a beam with rectangular cross section of b x h , translational and rotational

126
inertia forces can be regarded as the negative of the body force and moment, respectively. We, therefore,
can write

fb = -pbhv/„—
, p6A’#/12.
which, together with shear strain y from equation(), are substituted into equation() to give the weak
form in terms of the two independent filed variables v and /? ;
El£ /?'8{fi yix + GAkÿ(v'-p)(S(v ')-S/3)dx + £ (pbhv)Swdx + £(ÿ-fi)Sj3dx =
f qSvdx + £ pi8di + £
I—1 1=1

Finite element formulation

As V and /? are two field variables, each must be described using shape functions;

v = [JV.]{rf}, /? = K]M
where

[V,] = [W, 0 V, 0 0],

[ÿ[o 0 2 ...
N, iV„J
{rf} = [v, v2 2 fi,... J
Derivatives of V, /3 with respect to X are given by

where

dN„'
M--TJ
1 <#V,
d$
0 0 0 . M = J1 0ÿ0ÿ
dt, dt dt
and J = dx!d% .

Variations of v,j3 and v', /?' are given by

Sv = [N,]{Sd}, Sp = [Nf]{8d}, 5(V)

127
2nd derivatives of v,/3 with respect to t are given by

v= [V,]{rf}, P = [N,,]{d)

Last two terms in equation() can be written as

i=1 i=1

where the nodal force vector is defined by

{/>}=[/>! Pi m2 ...

Substitution of equationQ gives the element equation

where

[M] = pbh['[N„
f [N,]Jd? +

{F}=f>.]Vÿ+M
Element shear locking
When Timoshenko beam element is used for analysis of a thin beam, shear locking can occur.

where a = GAk / El .

For thin beams with h —> 0 , a —> oo . Hence a can be regarded as a penalty number, which
means that the constraint v'— /? = 0 is being applied. In fact, for a thin beam where shear strain is
very small, it follows that v'— J3 = y —>0 analytically. However, in the finite element analysis, if
v {} = y cannot be small due to the characteristics of the chosen interpolating functions of V and
P , then the contributions from the shear strain energy to stiffness can make the analysis results too stiff
than the exact value. The following example shows the case of element shear locking.

128
Ex) Two-node Timoshenko beam element.
Determine the deflection at the tip using a 2-node Thimoshenko beam element.

Considering the imposition of the fixed boundary condition at the wall, we define the nodal displacement

vector as M=[V2 A]

Since v = p=-ÿp2, oj, [AT,]=[o and J= —

dx L
2’
we have L

O], I M
V

W=ÿ]=[o I], | £=+i

V2
A
Then, the stiffness matrix is f—1
0]-[i ¥])r([X 0]-[i
Full Integration gives

'0 o' i. _r G,4£ 4 -2L

L
[K]f = EI
1
+ GAk
4 fl='4ri-2i
Solving the equation

W={“}
gives

ML U2
M-
Although the solution includes the effect of shear deflection, it can not be used to predict the behavior or
thin beam. For example, consider El —>0 for a very thin beam. The solution reduces to

«-${!!
which is stiffer prediction than the Euler beam solution

129
 «-§{!}
In the integration of the stiffness matrix, reduced integration with one-point Gauss quadrature gives

r . ro oi r i -ii GAk 4 -21

L Jr [o i -4-4 4L -2L+ £
2 A

Solving the equation = gives

«-§{{}ÿ
which is the exact solution.

Using k = 0 , Euler beam and Timoshenko beam gives the same result.

In general, two-node Thimoshenko beam Element is not so good since shear deflection is not calculated
exactly. The same applies to a 3-node element whose mid-node is located on the center of the element
and a 4-node element whose mid-nodes are not on thel/Ts of the element.

The conclusion is that the pure displacement-based low-order thick beam, plate and shell elements show
element shear locking. This phenomenum can be solved by using a mixed formulation of interpolating
displacements and transverse shear strain defined by

v=t>,v,.
;=1 ;=1
r=I4 1=1

where yi is the shear strain at the Gauss point i.

PROBLEMS

Problem 1. a) Find the natural frequency of a cantilever with p = 7.6g / cm3 ,

b = 1cm, h = 5cm, L = 50cm, E = 200GPa, v = 0.3 using Euler beam elements with
numbers being 1,2,4,16.
b) Compare the results with those from the analytical method where the natural frequency of the
cantilever is obtained from the assumed static mode shape of a cantilever subject to a concentrated force,

130
 140El
in which fl is the mass per unit length.
11//Z4

1
h h [
L
b

Problem 2. A beam with a retangular crosssection, as shown, is subject to a force P , the stopper is
distanced by A = 20mm . L = 800mm , Plot force-deflection graph until A = 0 .

P.s
18 10 18

I
I
X

©
:.A t ] I10
m
-> 4—
2L L

Problem 3. A plate with 1000 x 500 x 5 mm is reinforced with H-type crosssectional beam along the
middle as shown in the figure. Calculate the deflection at the tip when it is subject to a force 200N.
1) by elementary solid mechanics, Hint : The plate can be substituted by a beam when it’s modulus E is
regarded as E’
2) Two beams
3) plate and a beam

Problem 4. Find the deflection at the tip of the beam using Euler beam elements. Compare the
numerical result with the analytical result. L = 1000mm , E = 200GPa ,
F = -400 j+ 600k [N]
y

10
80 y
F
70
X

z
-H k-
L 10
Units are in millimeters

131
Problem 5. For all the problems, use E = l50GPa, v = 0.30.
1) A beam with (L x b x h) 500 x 5 x 5 mm is fixed at the support and subject to a distributed force
q = 0.5 N / mm at the other end as shown below. Find the displacement at the tip and the bending
stress at the center using a) Euler beam element a) Euler beam elements b) beam elements considering
shear effect(Timoshenko beam) c) plane strain elements d) plane stress elements e) 3D elements f)
analytical solution of Euler beam

2) A beam with (L x b x h) 500 x 5 x 250 mm is fixed at the support and subject to a distributed force
q = 62500 N / mm at the other end as shown below. Find the displacement at the tip and the
bending stress at the center using a) Euler beam elements b) beam elements considering shear
effect(Timoshenko beam) c) plane strain elements d) plane stress elements e) 3D elements f)
analytical solution of Euler beam

3) Compare between your results to make a conclusions about what you have found out from the
problems.

&
y
\

'b
L

Problem 6. A cantilever beam with I-type crosssection has the dimension of

Lxbxhxt = 2000 xl 80x100x15 mm . Calculate the deflection at the tip when it is subject to a
force F = \KN. Use E = l50GPa, v = 0.30.
1) by analytical solution from elementary solid mechanics.
2) by FEM(commercial software)
3) by FEM, but using two beams of T-type cross section.

I
i t
h t

132
8. Plate and Shell Elements
For many problems of continuum mechanics the problem of determining the 3-dimensional stress
distribution is too difficult to solve. In some cases, however, a useful result can be obtained by replacing
the original problem by a 1- or 2-dimensional simplified theory by assuming the strain distribution in the
thickness direction. The corresponding stress distribution is taken care of by analytical or numerical
integration in the thickness direction. The stress distribution in the element plane is treated by finite
element technique. z
y
L
The beam and plate are different from each other. q
\ x
Consider the system shown. If the problem is solved
by beam theory in which the existence of the V
? v/
normal stress components in the lateral y directions is
excluded, the solution 8 = bql? / 3El would be
T
incorrect.
From experiences, we expect the deformed shape would become a cylindrical surface except near the

edges at y = ±Z> / 2 . Normal stress <7y prevents the plate from anticlastic curvature. For the most

region of the plate, £y =0 and the problem is in plane strain situation in X —z plane. Using the

stress-strain relation for the plane strain case and noting <72 =0 , we can obtain the relation

—
<JX =£/(! v2)sx whereas the relation would be <JX = Esx for plane stress situation in x z —
plane. Hence, for the system shown, we can use the elementary analytical solution with E replaced
by E/(1-v2).

We could solve it using the plane strain elements with a large number of elements. But for this situation
it is more economical to use a structural element made from the plate theory.

8.1 Kirchhoff Plate Element

8.1.1 Kirchhoff Plate Theory
y

dr"
q
z
m °x
T*y p
~7_ S IA X
Mid surface

133
The figure illustrates a stress element for plate with the stress components (Jx,(Jy, Txy and distributed

force q . The x, y axes are located on the mid plane of the plate element.
Normal stress in z -direction and transverse shear stresses are neglected and hence, <J, =0 ,
Tyz = Tzx = 0 - Hence, the plate is in the state of plane stress in x —y plane.

Deformation dw
dx
A Z A z
X s' dv

* J >
2 ,W
dx

&
X \S t

Transverse shear deflection is neglected and hence, Y yz = Y zx = 0, but y 0 since Txy & 0 .

z -directional displacement is assumed to be a function of x,y and hence, w = w(x,_y) . Then,

X, y - directional displacements are, respectively,

dw dw
u = -z —
dx
,
dy

Strain -displacement relationship is

du d2w dv d*w du dv d*w
£x~~dx~~ZliT’ £y~~ÿ~~Z~ÿr’ 7xy~~ÿ ~dc~~2Zdxdy +

In matrix form, the equations can be put as —

= z{/e} , where,
ex1
w- «, w- •
d1w
'•V2
v» 2 dwSxdy

The second equation is the definition of curvature of a plate.

Since the stress state for plate corresponds to the plane stress, the constitutive relation is

134
 W=[ACT]{4 where

1 V 0
£
K]= 1-v2 V 1 0
0 0

The weak form for a plate is given by

[{5e}T{o}dV= j[ Swj
where V represents the volume of the plate element, q(x,y) is the normal traction and

{5d}represents the virtual nodal displacement vector and { p\ is the concentrated force vetcor at the

nodes.

Applying the stress-strain relation and the strain-curvature relation, the left hand term becomes

l{Se}T {cr}dV= [{Se}T [D„]{s}dV = \£{SK\T S[D,„]{K}dzdA

With the definition

1 V 0
Eh3
V 1 0
12(1 -v2)
0 0ÿ

the equation becomes

J[SÿqdS +{dd}T {p}

8.1.2 Finite Element Formulation

1) field variable W
In deriving the stiffness matrix, interpolated transverse displacement

w=MM
is used, where

M = [w, PxX PyX Pxl Py2 -J

in which W( is the nodal displacement normal to the plate surface and fixi = w,xi J3yl = w,yX are the

135
nodal rotations.

Then the curvature can be written as = {<9} w , where

{«}- dy1
2-d-2
~ dxdy _
In terms of nodal displacement, we can rewrite it as

The equation becomes

{Sd}T £[sf = {Sd}1 ( £[vf qdS +{p})

where { /?] is nodal force defined by

{p}=[p\ MX1 My\ Pi M*1 My 2 T-

Equating eq() and eq(), we have the element equation

WM-lf)
where stiffness matrix

and the force terms

{F}=£[Af(rfS+{/,}.

1 constant

x y linear

x2 xy y2 quadratic

x3 x2y xy2 y3 cubic

X
4
x3y x2y2
9 9
xy3 y 4 quartic

136
Four-node quadrilateral element has 12 d.o.f.. The order of the complete polynomial for this case is
quartic, which has 16 terms. The use of incomplete polynomial causes the problem of incompatibility
along the edges.

Three-node triangular element has 9 d.o.f.. The complete cubic polynomial has 10 terms. We can add
terms like JC*y,xyi. In this case it is not possible to satisfy continuity of both wand along the
edges.

The numerical results from these elements are poor. Efforts to avoid these problems led to the
developments of many elements.

2) Discrete Kirchhoff Quadratic element(DKQ)

Since the curvature can be written as

0 y
&
fa* s
w= o i
_
: z
3
rs£„
_ JL
dy
JL
dx
8
>
we can write {/f} = [5] 4 1 x

0
* Ox
where [5] = 0 and = [Oy is the rotation of the cross section.

_8y
A A
Bx _
To illustrate the development of four-node quadrilateral element, consider the rotation
8 8

«,=2>A. 0,=2>A<
/=! /'=!

where / = 5, 8 represent the mid points(not nodes).

The equation can be written in matrix form as

...
[JV] = Nx0
0 N2 0 0
where [o}=[o« 0yX ex2 ... 0y6J
Nx 0 N2 ... jVg J

137
In the expression for |/c} in eq(), we use the interpolated rotation 9x,0y which are in terms of 6xi,

6yi which in return can be expressed in terms of wi , /T , ftyj by suitable constraint equations, as

shown here.

In DKQ, transverse displacement w is defined only along the element boundary and it is interpolated as

cubic polynomial using nodal degree of freedom at both ends wi , fixi , J3yi .
16 constraint equations relating 6xi , 6yj to wi , fixi , fiyi . are

- At the comer nodes, rotations 0xi=Pxi, 9yi = ftyi , / = 1,4 (8 equations)

- At the mid points,

a) Tangential slope 9S = w,si where 9S is the rotation from the s-axis (4 equations)
b) Normal slope: 0n5 = +w,n2 )

Here normal slope is assumed to vary linearly along the edge (4 equations)

Writing the equations explicitly for 0xi , 6yi , we have symbolically, the relation

12x9

M = [wi PA Py 1 «i fh P,i -]r

Then the curvature is

12x9 12x9

Stiffness matrix is given by

[K]=l[Bÿ[Dtp][B}dA

138
To this end, stress resultants are used instead of stresses. Examples are beam, plate(flat), shell(curved)
element.

Stress resultant
The stress resultants per unit length are the bending moment

Mx=-fkcrxzdz, My = -ÿ<Jtzdz
and the twisting moment due to horizontal shear forces

=-ij»zdz' z
A
y

In matrix form, the equations!) cane be put as

Mx
My
(I rA
X

, where

Mx
{M} = My •, and {or} = <Ty

139
8.2 Reissner-Mindlin Plate Theory
H
z
A y

z
/
lz gEaEiESr x
*ÿ
X

Normal stress in z -direction is neglected: <JZ = 0 .

The stress resultants per unit length are the bending moment

Mx = - crxzdz, My = -J\fyZfc
and the twisting moment due to horizontal shear forces
z M,
= -fjxyZ<k- ¥y 9>
Vertical shear forces are defined by
& X
rhl 2 fA/2
&=-LT~dz- 8y=~L V*
Defining force vector per unit plate length

{M.P}=[MX M, MV Q, Q,J,
equationsQ can be written as

{Mmp}=-$l[A{°}dz
"z 0 0 0 O' <7*
0 z 0 0 0
[Z]« 0 0 z 0 0
0 0 0 1 0
0 0 0 0 1

Assumptions on displacements

140
 w = w(x,jp)
Similar assumptions on deformation of the cross section is assumed as in the Timoshenko beam. As the
result, the displacements takes the form

U=-zfix
V = -zfiy

1) Strain -displacement(curvature) relationship

In-plane strains
du 0px
dx dx
dv
ey=ir. = -z
dy dy

du dv
Yxy
dy dx dy dx j

Transverse shear strains

dv dw _ (dw
lyz = y
dz dy Kdy

+ — = -— P*
dw du ( dw
Y«= — ox \ ox
OX
n

If we define the strain vector

{£} = [£x £y Yxy Yyz rj,

we can write

M~[z]M
where

z 0 0 0 0 dx
d£y
0 z 0 0 0 By

w- 0 0 z 0
0 0 0 1
0 •
0
w=- Bpx
dx

0 0 0 0
Px-fx
1 Py dy

141
2) Constitutive Relation

where

1 V 0 0 0
V 1 0 0 0
E
and [£>'] = 1-v2 0 0 0 0
0 0 0 Jbv. 0
2
i=y
r- 0 0 0 0 2 J

3) Moment-displacement(curvature) relation

4) Weak form

l{Se}T {v}dV = HSuf {,}dS+{Su}T {p}

(left hand term)=

{.{*}' £;{S*Y [zf [D'][Z\{K}dzdA

={„}dV l
= \A{5K}T[Dÿ{K)dA

KM*[ziÿ'ii2]*
Since

z2 z2v 0 0 0
z2v z2 0 0 0
E
[Z][D’][Z]= 1-v2 0 0 z2if 0 0
1-v
0 0 0 2 0
1-v
0 0 0 0 2

.therefore

142
 1 v 0 0 0 0 0

Eh 3
v 1 0 0 0 [A] 0 0
[A]- 12(1 -v2)
0 0 0 0 0 0
0 0 0 0
12 5 1-y
0 0 0 fG/z 0
0 0 0 0 A2 6 2 0 0 0 0 \Gh

The stress resultants

equationsQ can be written as

5) Finite Element Formulation

%
cbc

ay

W= 3y T
3r -MM
/y-f
0 % O'
o o %, u
M- 0 %y %x Px
-K i o Py
% o l _
and

w
M= A hMM
A
where

Nt 0 0 W2 ... O'
[JV]= o N, o o ... o ,{</}= [A A, A. - Aj
0 0 iVj 0 ... JVj

143
Then

M =[«]W =MM=[*M »i>ere [fl]=[a][tf]

Therefore

[K}=\A[Bÿ[D.P][B\IA

8.3 Decomposition of Stiffness Matrix

If we decompose the stiffness matrix as {icmp j = lÿat

[D*»] [°] |W1

K„}rK]{ÿ!= f*})r
W [ [0] [n,lJlmr

Therefore, the virtual strain energy is from bending and shear. In finite element formulation,

M-(1*M*.DM
, where

0 o'
'

'0 s/3x
"

0 0
o o y* 0 0 0
W-[a.]MP.l- 0 % %x .nd [s,]=[a,]M[a.]= 0 0 0
0 0 0 1 0
0 0 0 % 0 K

8. 4 Shell Element
8.4.1 Membrane Element
Only membrane forces(in-plane forces) exist and there are no bending stiffness. Hence, membrane
element is in plane stress situation in its own plane. For a linear QUAD element, the element equation
becomes

WM-W

144
 a o •% . 0
M-MM- o f •• 0
SWj_
3y
5Wj_
5x
# #
3Aÿ_
5y
*
j3Aÿ_
5jt

M=K 4, 4, 4, d„J
{'W/Wri
where

with traction «-['•'J

and

{p}=\_Pu Ply • • PAX PAyJ

Example
\ I I I II I I III,

F, F,
\A

8.4.2 Shell Element

Shell element has curved surface and therefore, membrane stiffness must be considered
as well as bending stiffness.

y
.x

A
145
 PROBLEMS

Problem 1 The stress and deflection for a built-in circular plate of radius r and thickness t
3 pr2
subject to uniformly distributed loading p are given by crmax = — 4 t2
at the edge and

pr i
Wmax = 64£> at the center. By using plate elements, prove the formula numerically.

rTTT\
Problem 2
A rectangular plate with (Lxbxt = 500 x 250x5 mm ) is fixed at the support and subject to a
concentrated force of F = 500 N as shown in the figure. Es =\50GPa, vs = 0.25 .
1) Find the deflection of the plate at X =L by an analytical solution.
2) Find the deflection of the plate at x=L using commercial FEM software. Use plate elements.
3) The plate is reinforced with a beam with I-type cross section of
—
bxhxt 50x80x8 mm . Find
the deflection at X
4) Do the same as
—in L by an analytical solution.
3) using commercial FEM software. Use plate and beam elements.

y
z
z

L y
\ F
x b
\

\ t
\ I 1 t
h
T t

Problem 3 A plate with one edge fixed at the wall is subject to a distrioutecfcforce of intensity w .
Discuss about the tip deflections that would be obtained with Mindlin element, Kirchhoff element, and

146
Euler beam element and compare them. The mathematical model for beam element is shown also.

w
4£_ t = bw
b

a
b
I I = bt3 !\2

Problem 4 It is required to analyze a cantilever with height h, width b and length L with a) 2D
continuum element, b)3D continuum element, c)Euler beam element, d)Thimoshenko beam element
e)KirchhofF plate element f)Mindlin plate element.
A) Discuss the expected numerical results against the exact solution. Compare also the relative
magnitudes among the numerical results and state the best choice for the analysis. When you discuss,
you should consider the cases with the various relative ratios between h, b, L
B) For each case, if you use 4 elements for the system, what are the total number of nodal degrees of
freedom when boundary conditions
are imposed ? z y

\ X
b
\
±JL
f

147
9. Modeling and Miscellaneous Topics
9.1 Modeling
1) Mesh generation
- large aspect ration, large distortion
- abrupt change in element size
- poor connection; displacement discontinuity along the common element edge
- change in face direction
- missing element
- nodes unattached to any element ; delete it
- nodes unconnected to each other mistakenly; use node merging
- use the refined mesh for the region where the gradient of the field is high; adaptive mesh

2)Use symmetry, periodicity, Substructures

1

[
O [ J o ]

I
3)Use structural element, 2D element

4)Use a special element

5)Rigid body motion

6) Don’t be confused by the stress concentration by a concentrated force.

- St. Venant’s principle

148
10. Conduction Heat Transfer Analysis
10.1 Heat Conduction Equation
Fourier law of heat conduction gives the heat flux
q = -KS7T
where K , T are is the conductivity of the media and temperature, respectively. By applying 1st law of
thermodynamics, we obtain governing equation for heat conduction problem. When K is constant, it
can be written as

KV2T(x,t) + g{x,t) = pc dT(x,t)

dt
where g(x) is the heat source and p, C are the density and heat capacity of the media.

Without any heat generation, the equation can be written as

1 3T{x,t)
v2r(x,o=-
n dt
where T] = K / pc is thermal diffiisivity

For steady state condition, the equation can be written as

KV2T(x) + g(x) = 0

B.C.’s
1) EitherT(x,t) or q(x,f) should be prescribed on the entire boundary T.
2) When a convection exists on the boundary , q is given by
q(x,o = /t(r-rjn,
where n is the outward normal vector on the boundary and h and Tn are the convection heat transfer
coefficient and ambient temperature, respectively.

10.2 Weak Form

Consider the steady state equation().
To derive weak form by Galerkin’s method, multiplying it by ST and integrating over the body, we have

£( KV2T + g)5T dV = 0
Using the relation 8TV2T = V •{STS/ T)-VT- S/ST , we have

[[KS7 .{STS/T)]dV -£[/:vr- VST] dV+ £ gST dV = 0

149
By use of divergence(Gauss) theorem
£ V -\dv = v •n dS and Fourier’s law, the first term of the

equation gives

(1 term)=
J [ATV (6TS7T)]dV = £[K(STVT ) n]dS = -
• • • J[
n dS = q 'STdS

—
where q 1 = q •n is the heat influx to the body.

Since variation and differentiation are interchangeable, we have for the second term of the equation.

(2nd term)= j[[KS/T VST] dV = [[KVT SVT] dV •

Substituting the equationQ into 0, we have the weak form

dV= |gSTdV+

The equationQ can be put in variational form. Namely, The equationQ can be rewritten as

y dV= Jj. q'STdS+ £ gSTdV

By defining the functional for the heat conduction problem as

f
n = [(Wf
the formulation in variation method is given by
dV- [gTdV- [q’TdS
<5T1 = 0 \/5T

Before we derive the finite element, we will consider the heat

q ' can be categorized into a prescribed heat influx qin on boundary and a convective heating on

Th . Therefore we can write

q' = qin-h(T-TJ
where h is the convection heat transfer coefficient.

In case of transient analysis, comparing the equation and the equation, the term pcT can be regarded as
negative heat source or heat sink; hence we can write

g = gs~PcT

where gs is the heat source.

150
Substituting the equation and the equation into the equation, we have

J[(KVT-8VT)dV+ J[pcTST dV+ STdS = [gsST dV + j\sqinSTdS+

10.3 Finite Element Formulation

Defining the nodal temperature vector

{e}=[7; T2 T, r4f,
interpolation of temperature field can be written as

r=[jvM

where [iVj = [./V, N2 N3 Af4] is the shape function.

Stiffness matrix
Gradient of the temperature T(x, y, z) is, then in vector form

d/dx d/dx
S7T = \dldy\T = {d}T with {d} = \dldy
d/dz d/dz

Using the equation(), we have

vr = {d}[AT]{6>} = [5]{0} With[5] = [d][N]

By substituting the equation into the first term on left hand side of the equation, we have

[(KVT-SVT)dV = [K([B\{se})T{[B]{e})dV
={se}{lK{B\T[B}dv){e}={se}[K]{e}
where stiffness matrix is defined as [ K\ = f [5]dV

Mass matrix

151
The second term on left hand side of eq() becomes

pc[tSTdV = pc[(\N]{Se})T[[N\{9})dv
= [sof (HW ={»}r
where mass matrix is defined as [A/]=pc|,[Arf [N\dV

Convection term(Unknown)

[hT8TdS= J[([V]{«})r A([JV]{*})<S = (£ {0})

[AT].*?
-{»}'[*]{«}
where

Heat source
The first term on right hand side of eq() becomes

[g,STdV=[g,([AT]{ÿ})r dV = [g, [A'f dV ={S0}T{Fj

where heat source vector is defined as {F.}-[g.MdV
Prescribed heat influx

[q„STdS= 9([V]{«})rrfS
[ = {«}'' <iS ={»}r{F,}

where

Convection term(Known)

IhZ =(STdS
|.([JV]{«})r hT.ds'j

where {F„} = l[NÿhT„dS

Therefore

{<{[M]{0} + ([*] +[//])M} ={»}' ({F,} +{F,}+{Ft})

152
 [M]{e}+([K]+[//]){f)}={F} where

PROBLEMS

Problem 1
a) When a car is descending a grade at a constant speed —
V 30kph , the brake piston is found to be
exerting a force P = 10KN . Assume the heat generation is uniform all over the friction surface.
Perform the transient analysis to find the temperature rise of the interface at radius rmin = (ro+rj)/2 .
Plot it against time until it becomes steady state. Assume that the side surfaces and the friction surfaces
are subject to the convection with the same convection coefficient h = 23 W I m2 K .
b) From the analysis results, validate that the heat generated is the same as the heat outflux at the steady
state.
c) By constructing a simple ID model, calculate the steady state temperature at the interface and compare
it with the numerical result.

Friction coefficient p = 0.4

Disc Material(Cast Iron)
p = 7.2 g/ cm3, c = 4\9J/kg°C, K = 48.6J/m-s°C, E = \25GPa, v = 0.3, a = 12xl0'6,
Pad Material
p = 2.6g/cm\ c = 1465 J/kg°C, K = \.2J/m-s°C, E = \GPa, v = 0.3, a = 10xI0‘6/°C

r
0.

Problem 2
A plate of 2 x 6 x 0.02 m is fixed with two walls at the ends. The wall #1 is subject to the temperature

rise of A7J = 50° C while an electric resistor exerts heat influx of q = 500W / m2 through the area

of 1 x 0.02 m at the wall #2, the rest of which remains unchanged in temperature, hence A T2 = 0°C .
Material constants are E = 200GPa , V = 0.3 , a=8xlO”6/°C.

153
a) Find temperature distribution assuming steady-state condition. Contour plots will be good.
b) Obtain temperature and von-Mises stress <r ' at the center of the plate
c) Present an analytical prediction to claim your numerical results are correct, however far from the actual
state it may be. (For this purpose, you may need to simplify the problem at your convenience so that you
can get an approximate analytical solution.)

insulated

T g|~ *
2m 1m
i jjE
insulated
m
Tx T2
Problem 3
The stiffness matrix of an axisymmetric element shown in the figure is to be calculted for conduction
heat transfer analysis. Coordinate mapping is used between r —z axis and <z,—t] axis. One point
Gauss quadrature is adopted.
dr dz dr dz
a) Calculate —
df
, -,
dr]
—
df
, -
dr]
at the Gauss point

b) Calculate the numerical value of Jacobian J

c) Calculate dNt dNi / = 1, 4 at the Gauss point

dt' dV’
d) Write the expression for the stiffness matrix [AT] of axisymmetrix element for heat transfer
analysis in terms of [5] = [3][iV] .

e) Write the expressions of all the components of [5] in terms of dNt dNt i = 1,4
dr ’ dz'
f) Calculate dNt
dr
dNt
dz —
i 1, 4 at the Gauss point

g) Put the answers from c) into the expression from e) to determine the numerical values of [5]

y 3(5,5)
4(2,4)

2(6,2)

1(1,1) x

154
Problem 4
a) Using the total potential for heat transfer problem

f
n = {.(Wf dV-[gTd
show that the stiffness matrix of 2-node element shown is given by

dNA2 dNx dN2

dx ) dx dx A2
dV Ai
dNÿdNÿ ( 9N2 V
dx dx l cbc J L

b) Apply one point gauss quadrature to calculate the element K12 of [/C] . Ax = Aÿ 12 = A0

155
11. Nonlinear Finite Element Analysis

Linear system is defined as a system with output proportional to input. Consider a spring fixed to a wall
on one end and subject to a force F on the other end. When the spring coefficient kc is constant,
then the governing equation for the system is given by kcx = F . The solution is easily obtained as
x = F / kc. In general, superposition holds for linear systems. In case of structural analysis problem,
for a system to be linear,
a) displacement-strain relation must be linear ex) w = du/ dx
b) strain-stress relation must be linear ex) <7 = Esx
c) boundary condition must remain unchanged.

When the spring coefficient k is varying with its extension x , the system becomes nonlinear.
Consider, for example, k = ex . Then, governing equation is exx = F . For a given force F , the
solution X can be determined by an iteration method.

11.1 Sources of Nonlinearity

1) Geometric nonlinearity
Small strain is defined by

f
_ 1 du, duj '
,J
2 [dxj dx, j

This is the situation where Green-Lagrangian strain

Eij =I du, , dui !

duk
2|_ dXj dX, dX, dXj
or Cauchy-Eulerian strain

1 34 dUk
|
2 dXj dx, dx, dXj
should be used in the formulations. X( = X, + u,
If dui / dXj and u, are sufficiently small, E,- = Ey = Sy

These strains, therefore, are nonlinear with the displacement. If dui / dXj is sufficiently small,

156
 as in the small strain theory in which the relations is given by

£y = 1[ , duj
2 dXj dxt

ex) Generalized Hooke’s Law

Tu = AEÿSy + 2/jEy

Tu 'ÿ the second Piolar-Kirchoff stress, : Green-Lagrangian strain

Large strain

This is the situation where the displacement gradient dut / dx} is not small enough to ignore the

nonlinear term in Cauchy-Eulerian strain even though the displacement is small.

Large displacement with small strain

Even though the strain is infinitesimal and hence, the material is linear elastic, large displacement
accompanied by rigid body translation and rotations can cause nonlinear relation between strain and

displacement. For this case, Xi & Xt and hence, €y * EtJ

For the system shown, the relation between the applied force F and the deflection A is given by

A Jyi-HA!Vf -l) L
F = 2 AE\ - L
L) VI2+(A/Z)2 ]A

AEE2
1F
When A «L, F= AEÿjÿ or k=
z3
Other example is the large deflection of elastic ruler for which the nonlinear governing equation is given
by

v" M
El’

the solution of which is shown in the figure.

157
Example is snap through of the a button spring

2) Material nonlinearity
This happens when the approximated linear stress-strain relation can be no longer used. This situation is
usually occurs by large strain and therefore in this case geometric nonlinearity should be used as well.
Examples are metal forming, hyperelasticity. The effect, however, sometimes could be considered even
in a small strain situations as in a fatigue analysis from small strain cyclic plasticity of metal. In heat
conduction analysis, temperature dependence of thermal properties can make the system nonlinear.

Example Ramberg-Osgood stress-strain law

\m

. m = 10, cr0 =38000/75/, E = cr0 / €0 =l0x\06 psi, a = 3/7

E E [<T0
Example elastic-perfectly plastic stress-strain law
<7
<7

-ey e
6y

3) Boundary conditions
When the boundary condition is changed as the applied force is increased, such as existence of contact,
friction, the system is nonlinear.

11. 1 Solution of Nonlinear Equation

11.1.1 Newton -Raphson Method
Consider the nonlinear equation /(x) = 0 . Let X. be an approximation of the solution. The better
approximation is x/+1 = X( + Ax for which Ax can be found from the Taylor expansion with center x;

/(*, + Ax) = f{x, ) + / 'O, )Ax + 0(Ax2 ) = 0

We have
y
Ax = -/(x,)//U) y=m
Therefore, the formulae is given by
fix,)
xM =x,- /**,)
X X
The graphical interpretation shown in the figure
x2 X, x0

158
As the iteration goes, the solution converges to the exact solution if the initial value is appropriate. If
not, it can diverge or converge to other solution.

x
Example Find the root of the equation e = x.
f(x) = e~x-x, f\x) = -e~x-l, xM=xt- initial value x0 = 0
-e~x‘-V

i o 1 2 3 4
X 0.000000000 0.500000000 0.566311003 0.567143165 0.567143290
/ 1.000000000 0.106530660 0.001304510 0.000000197 0.000000001

Example Find the root of the equation ex — tan X = 1.5 1

1
ex‘ -tan x( -1.5
/(*) = e* -tan-1 x-1.5, f\x) = ex-{\+x2y\ xM=x-
+ x,2)-' ’
initial value x0 = —7

i 0 1 2 3 4 5
X -7.000 -10.6771 -13.2792 -14.0537 -14.1011 -14.1013
/ -0.702 x 10’1 -0.226 x 10'1 -0.437 x 10'2 -0.239 x 10’3 -0.800 x 10-6 -0.901 x 10‘u

11.1.2 Modified Newton -Raphson Method

Since the recalculation of the slope in each iteration is time consuming, modified N-R method uses the
slope from the first iteration. The figure shows the graphical description of solving a nonlinear equation
g(x) = F. /(x) = g(x) - F y
y = g(x)

F si if

f X

xf\x1 x0
x3 x2

11. 3 Finite Element Solution of Nonlinear System

Consider a nonlinear system for which the equilibrium equation can be written as a single equation
g(u) = F and an initial value that will lead to the solution is hard to guess. For this case, the final

159
force is assumed to be reached by three successively increasing forces, F", n — 1,2,.., N . In each
increment, the equilibrium equation is g(un ) = F" . If the force increments AF” —
= F"+1 F" is
small enough, w"+1 would be close to un . This means that U can be used as an initial value in
solving the nonlinear equation of the next increment g(u"+] ) = F"+l . As initial value w° of the first
increment n = 1 , zero is chosen since the solution u would be close to force-free state.

F, u
F
pN

Fn

F2
F1 t
t\ t2

Example Find the solution of e2“=10.

The final force is assumed to be reached by three successively increasing forces,
2u
—
F" 3, 6, 10. The

equations to be solved are e = F", n = 1, 2, 3 . N-R method gives the iterative equation

e2u‘ -F”
<i = ui ~ 2e2u‘
The table shows the solution is =1.1513 . Note that u]0 = 0.0 for the 1st increment, the
1 2
converged solution u5 of which is used as the initial value u0 for the 2nd increment and so on.

Increment n Fn Iterations i
i 0 1 2 3 4 5
n =1 3.0
u, l
0.0000 3.0056 0.7030 0.5707 0.5498 0.5493
i 0 1 2 3 4 5
n=2 6.0 u] 0.5493 1.0493 0.9172 0.9172
i 0 1 2 3 4 5
n=3 10.0
uj
3
0.9172 1.2158 1.1553 1.1513 1.1513

160
11.4 Implicit Integration in FEM
The nonlinear finite element equation takes the form
[*(«)]{»}={>>(«)}
or
{<!>(«)} = {/>(«)},
where = [AT(«)]{u} is the internal nodal force vector. Consider the situation in which the
external nodal force vector {P} is independent of displacements u, i.e., {P} is prescribed. When
the deflection is large as is usually the case with the nonlinear analysis, it is almost impossible to guess an
initial value |w0| which is close enough to the exact solution to guarantee the convergence.
However, we can overcome this difficulty by noting that the deflection is small with small loading. We
assume the final value {P} is accomplished by successive small load increments {AP} . In this

sense, |P} is assumed to be a specified linear function of time t . The solution at time t is
t+At
used for the initial value in the iterative calculation for {w} at time t + At . The initial value for
the first increment is \u0} = {0}.
In ABAQUS, steps mean cases of analysis procedure options, loading options. In other words, a step is an
event in the course of an analysis. Each step can be broken into increments in which an attempt is made
to find an equilibrium solution through iterations.
increment
For the n -th incremental loading, the equation is
{«><«")} ={««ÿ)} with {«(»”)}=[*(«”)]{“"}
Define which is the force residual or unbalance between internal

and external forces. Solve {'*'("”)} ={0} for {«"} with j un'I as an initial value using
iteration methods of N-R or Tangent stiffness method.

11.4.1 Newton -Raphson Method for Multi-Variables

Assuming |w"| is known at the end of / -th iteration, solve for from the equation

o,
MM ={/*(«")} -{<*>«)}
where
J is the Jacobian, evaluated at the end of i th iteration of n the increment and defined by
a{T} d{®(u)} d{p(u)} d[K] a{P(«)}
[J]
d\u\ d{u} d{u} d{u} d\u\
First two terms compose the tangent stiffness matrix and the 3rd term is called the load stiffness matrix
corresponding to external forces.

Then, the updated solution is given by

H,}=K}+M
Iteration continues until the specified error control criteria are met.

161
11.4.2 Modified N -R Method for Muiti-Variables
Use | J" =
J £./Q ]
for the iterations in n -th increment.

[J;]{K}={W)}-{4>«)}
Then, the updated solution is given by

11.4.3 Tangent Stiffness Method

[j;]=[x«)]
[X;]{A<)={PW)}-{O«)}
Then, the updated solution is given by
K,}=K)+K}
11.4.4 Modified Tangent Stiffness Method
Use

Then, the updated solution is given by

The global Jacobian matrix [ j] is obtained when the material consistent Jacobian matrix J is used in
the calculation of the element stiffness matrix

Convergence Test(Tolerance Set)

1. Force residual

2. Incremental displacement { AM} , |{ Aw|| < SD |{ Au0 1|

3. Incremental strain energy density Ae = A

* Automatic Incrementation control

11.5 Calculation of Jacobian Matrix

Deformation gradient tensor F is defined as

Since X, = Xt + ut , we have the relation

F = I+ D

162
where D is the displacement gradient tensor given by
du du

U>]“ l1
dX dY

G-R strain can be represented in terms of deformation gradient tensor or the displacement gradient tensor
in tensorial form as

E= F - /) = |(D + DT ) +|DT D
It can be verified that

fa dr =tr;s,s:E,d‘v
and Q Sjk is the 2nd Piolar-Kirchhoff stress and 0‘£{j Green-Lagrangian strain. In tensorial form, the

principle of virtual work can be put as

[y8ET:Sd°V = 8W
where : means double contraction of two second order tensor 8ET , S and SW is the virtual work by
external forces.

The left handside of the equation () gives us the required form for internal nodal force vector {<!>(«)] .

A{Q(u)} ,
Since [«/] = in order to find the Jacobian matrix, we consider a finite differentiation in
AM
time on both sides of the equationQ.

From the equationQ, we get

A(£Er) = |[AZ)r-8D + 8DT AD] = 8DT AD

The P-K stress increment is given by
AS = C4 : AE
where C4 is the fourth order tensor.

The finite differentiation of the left-hand side of the equationQ becomes, then

A(ÿÿE7’SÿV)= Jÿ[<5Er: AS + A(£Er):S.]</V = + 8DT-&D:S)ÿd(>V

163
or in matrix form using Voigt notation, we have

j;(,({5£}r[C]{A£} +{5D}r[S]{4B})rf»F
8u_
BX
du du du_
BY
Since D = is not symmetric, in vector form we define jD j = dv.
> and
_lx W_ dX
dv_
far

5n Sn 0 0
5,2 0 0
[s]=
1 J
0 0 su sn
0 oj [sl2 s22_

11.6 Finite Element Formulation

JV, 0 • • N4 0 u(X,Y)
{«} = [iV]{fi?} where [iV] = 0 N, •• 0 N4\’ M= v(X,Y)

'j-
du_
3X ex ol
v
du
# 0
W- dY

dX
0 ar
JhL
Lsr ,° 5r_

W=[s]t"H8][#WG]M »*™ [o]=[a][Jv]

du_
Su_ du_ Sy_ 5X
ex . dX
f\
” ax 0 du_

[*]- t
5K.,
+2~
-0-ÿ0
u ay u
du Bu dv
BY
Bv
BY
_dv_
ax
=[£,]+[£.,]
BY ax J BY BX BY BX. 8v_
dY J

5K.
Bu_
BX 1 0 0 0' ex
Bu_

[*]- BY
By_
0 0 0 1 BY
Bv_
Bu_,Bv. BX
.BY BX. 0 110 dv
dY

164
 r&_
3u_
0 ,-v
0 ax
ax ax 2a.

Kl=f 0
du
. 8Y
djL
ai
du
ax
0
dv
ay
2a.
BY
dv
ax.
ar
JtL
ax
dv_
ay

, where each component of is linear in {#} , for example, the first row and first column

component being [l 0 0 0]{#}.

[«£]=[5£,]+[ÿ].[H]{«»}+i[ÿ({9})][»]+i[ÿ({»})]{»}
but nee

[5£] = [5£,]+[5£„] = ([i/]+[ÿ({0})]){ÿ}

[ÿ]=([ff][G]+[ÿ({S})][G]){W}-[S.]{M}
[A£].[fi„,]{Adj

First term' [ME}\c)md'V ={Sd}T{l[Bj[C][B„,]d°v){Ad}

where w-H+MM)])[c]

Since {£)} = {£?}=[G]{t/} , we have

second term- [r{SD}T

[S]{/U3} dV = ( JJof [S][G] dV){Ad}
Hence,

where [X..]=|„[S„nC][fi.(]dV and [X„] = [S][G] d Y

165
The repetitive equation is

(;K nun + ;KJ«,)AU = '**'J!-JF

Comparing this with ]{AMJ = j/ÿ} — ,

PROBLEMS

Problem 1
Find the deflection-force curve of an elastic cantilever in large deflection by commercial FEM software.
Compare your result with the analytical solution in the lecture note. Assume the force P remains in
y -direction.

Problem 2
A slender circular ring is to be pushed by two opposite force which is large enough to collapse it. The
(/) . The true stress-true strain relation of
ring is made by a wire of circular cross section with diameter

—
the material is shown in the figure. Find F S curve up to S = R/ 3 by commercial FEM software
with material and geometrical nonlinearity considered.

200GPa, v = 0.25, ay = 110MPa, Ep = 30GPa

r = 100/ww, <f) =
5mmÿ=
SI*
/' A\ cr
i* (Ty
1 V ' EP
/> -sy E
£J
E

t F

166
12. Time Integration
12.1 Hyperbolic Problems.
Equilibrium Equations in Dynamic Analysis
Mu+Cu + Ku = R

12.1.1 Newmark Method(Hughes) or Newmark /? -equations(Belytschko)

Three equations are
- Equilibrium Ma„+1+Cv„+1+Kdn+1=R„+1
At2
- Finite difference formula dM+I = dn +At\n + -ÿ— [(1 — 2/?)an + 2/?an+1]
V„+, = v„ +Af[(l - Y)a„ + ya„+1]

Implementations are following;

Define predictors : 1) a„,=d„+irv,+ÿ(l-2«a.

2) vn+1 = v„+A/(l-y)a„
Then, the finite difference formula describing the evolution of the approximate solution become

d„+i=dM+1+ÿAf2a„+i
and
v„+i=v„+1+yAfan+1
When K is constant, these two equations are substituted into the equilibrium equation at time n + 1 to
get the equation

(M+r*/C + /ÿJK)a„, =R„, -Cv„, -Kd„„

which is to be solved for an+1 . Initial condition is —
Ma0 = R0 Cv0 - Kd0 .

When K is a function of d , the acceleration and velocity for these two equations
1
an+I = (d„+i -d„+, ) , v„+1 = vn+1 + (dn+1 - dn+1 )
0At2
are substituted into the equilibrium equation at time n + 1 to get the nonlinear algebraic equation for
d„+1.

167
Unconditional stability is achieved when 2J3>y> —1 .
1 v Q
Conditionally stability is achieved when y> — , /? < , At < —"max
crii

where C0mjol is the maximum frequency of the discretized system, Qcnl is dependent on (3, X and
1
damping of the discretized system £. When % = 0 , Q.cril =
Vr/ 2-1
Properties
Method Type P Y Stability Accuracy
Central difference Explicit 0 1/2 Conditional 2
Undamped trapezoidal rule Implicit 1/4 1/2 Unconditional 2

Example The central difference method

/+A/
This is an explicit method. The solution of u is based on the equilibrium conditions at time t .

Write ,u = -V('-i'u-2'u + ,+i'u), 'u=— ( t~At

u + t+At «)
A/2V 2AtX ’
and put these into the equilibrium equation at time t to obtain
1 ,
-L'M+—
t2
A IAI
C t + At
u = 'R- K— 4'MVU-
t2
A
—t2
'M --
1
— 'C
2 At A
r '-“u
t+At
Solve for u

Example Trapezoidal rule

(3 = 1/4, = 1/2
d„+1=d„+y(v„ + vn+i)
and

v«+i = v„+y(a« +a«+i)

Example Runge-Kutta(2nd order)
From the trapezoidal rule(nonlinear case), in order to avoid iterations to calculate dn+1 ,
M«+l’ r„+i are evaluated using approximated value dn+1 by use of forward Euler method .

a) calculate dn from Mndn = Fn Kndn —

168
b) calculate dn+1 = dn +Afdn
c) calculate Mn+1 K„+l F„+I
’ 5

d) solve for dn+1 from

(Mn+1 +aAfKn+1)d„+I = [M„+, -(l-a)AfK„+I]d„ +AfFn+I

12.2 Parabolic Problems

Equilibrium Equations in Heat Transfer Analysis

Md+Kd=F
where M, K are constant.

12.2.1 Generalized Trapezoidal Family of Methods

Mi„,+Kd„,=F„, (1)

1) Implementation 1
We write

<*n+, = d„ +Afdn+a = d„ +At [(1-a)d„ + ad„+1] (2)

where we defined

<La =(l-a)d„+ad„+1. (3)

Putting equation (2) into equation(l), we have

(M+o.*)d„l=F..,-K[d.+(l-oK] (4)

Solve for dn+1 and then get dn+1 from the equilibrium equation (1).

2) Implementation 2

Eliminating dn+I from equation(1,2), we have the expression for dn+1

(M + aAtK.) d„+1 = [M - (1- a)AfK] d„ +AtFn+a

where

169
 Fn+a=0-«)F„+«Fn+,

12.2.2 Explicit Method

When a. = 0 , the method becomes forward Euler method since we have from equation(2)

d„+i =d„ +Atdn (*)

d(i+1 can be calculated from the reduced form of equation(4)

Md„, = F„, - K(d„ + Art, ) = F„, -Kd„,

which is the equilibrium equation at n + 1. (* ) were used.

This method is also called forward difference method.

12.2.3 Implicit Method

When a * 0 , the method is implicit. When a = 1/2 , it is called trapezoidal rule, midpoint rule,
modified Euler, Crank-Nicolin method and is 2nd order accurate.

d„+, =d„ +Atdn+m and d„+I/2 =(d„ +d„+1)/2 (**)

Then we have

(ÿM+|-Kjd.„=F,.1-K[d,+|4a.
If M, K are the functions of d , nonlinear solution dn+1 from equation(**) and the following
equation is required.

M„,+ d„ +~dn

When a = 1 , it is backward difference method.

d„+, =d„+Afd„+1,

(M +AtK) d„+1 = F„+1 - Kd„ or using (a), Mdn+1 = Fn+1 - Kdn+1

12.2.4 Stability of the Generalized Trapezoidal Method

170
The decomposed equation is represented by
d + Ahd = F
If we investigate the homogeneous equation d + Xhd = 0
Discretization in time gives

Hence, since stability requires |dn+x| < |, we have

i-(i -gyah
-i< <1
1+ aAtAh
when a>j, unconditional stable
2
when a < Y , conditionally stable, hence for stability A t< should be satisfied.
(1-2a)Ah

Convergence : As Ax, At approach zero, the results of the finite difference technique approach the true
solution.
Stability : The errors at any stage of the computation are not amplified but are attenuated as the
computation progresses.

Example

k r-
3T
—dx2 =-
dt
is both convergent and stable if At <

--
The explicit method for the heat conduction equation
1 AY2
2 k
(Carnahan,1969)

171
12. Mixed Formulations
Displacement based formulation
Minimize

n =U{e}T[D]{e}dV- \{u}T{ft}dV- \{u}T{t}dS forall {„}

V V s
with conditions {f} = [£?]{«} and on S (displacement prescribed boundary)

Mixed formulation
Penalty method : Minimize

n=7Z /{*}>]{ \{u}T{,}dS

s}dV-\{u}T{ft}dV-
V V St

V Su
~M)dS
forall {«}, {f}, {/
Or minimize the Hu-Washizu functional

n=\\{e}T[D]Wv-\{«}T{fb}dv-
L V V St

-
V
Jhf (M-WH5
Su

forall {w},{f},{r},{f„}
Or minimize the Hellinger-Reissner functional

n=-\\{AT{D\Wv~ fMrf
L V V St

forall {«},
with the condition |w| = |on S (displacement prescribed boundary)

Example 2-node beam element, linear variation in wand /? and constant shear strain y , uint width,
height h
Hellinger-Reissner functional : element degree of freedom 5 : w, , /?, , w2 , J32 , y c
n= + \{s}T[D\S}{u}dV +
\{‘}r[D]{e}dV
v v

-UK\E °1W+Wr[£
l 2VJL0 GJVJ JLo
°1N
Gjlrj [y
dV +

-je\£xEsx ~\rfir c+r fir)dv +

172
where{£)=H H 41414+ %
<sn = o
-5y cGy c + Sy cGy + y cG5y)dV+ = 0
v

1 -r
+ 1 r
P — \P\ + h7.P1 ~ 2— P\1 "l--
2
P2
, (l-'l (1 +r)
w = nxwx
1
+ h2w2 = —- — w,1 +-W2

-z(l-r)
u = -z(3 = P\ + 2
2 2

In matrix form
_dv
dx
du
dz
_
£x=-ÿ
du dr du
dx dx dr
--—
dr dw du
dx dr dz

£x= 0 L 0
1
L
Yc =Y\
--
\\?1

-*F'
--
L J w2
w,
1 -r
2
—1
/?,1 + w22

-[*.M
L
1+ r
—
2
P22

[Pi]
Wi
I _1zL 1 l+r Pi =
7=
L 2 L 2 w2
[*r]M
[Pi]
Fc = [l]{Fi}
1
n = \{2£-Es‘ ~ -r fir c + Y fir)dv +
v

- jjÿWM)' “(Mir,})r o([i]{r , }) + ([i]{r , })r GJ«, ]M)}dV +

<5n = 0 :
G[\\dV{-,ÿ
E[BMV{d\+lBÿ
+{5r,}r |f[l]r G[B,]dV{d}-pf G[l]rfK{y,}| = 0

173
 K1 KllfM
\KJ Kljlir.}
{*..]= jM E[B.]dV
v
[*..]= J[*JG[IK
V

K]=-Jti]rG[iR
v

Stiffness matrix from mixed formulation

Gh Gh Gh Gh
L 2 L 2
Gh GhL Etf_ Gh GhL Eh3
+
2 4 12Z, 2 4 12Z
M- Gh Gh Gh Gh
L 2 L 2
Gh GhL Etf _Gh GhL Eh3
+-
2 4 121 2 4 121
Stiffness matrix from pure displacement-based formulation

Gh Gh Gh Gh
L 2 L 2
Gh GhL [ Eh3 Gh GhL Eh3
2 3 121 2 6 121
M- Gh Gh Gh Gh
L 2 L 2
Gh GhL_Dl_ _Gh GhL Etf_
2 6 12Z 2 3 +12L.

174
Appendix
Al. Derivation of Jacobian
dr = idx + jdy
dx
= i ——dig +——dij
dx . dy
dx
drj Mdx . dy
dri
dV

= \Eii +
d% d%
+
drj dt]
dr,

= eldÿ + e2dr/
Thereforee, + e2 +
dg dg drj drj

In X —y plane,

on the other hand, dA = dxdy , hence

dxdy = Jdÿdt] °] A.

dy

e2d?i e2d?i
dx

->

175