Finite Element Methods Modupe A.

Onitiri, PhD

FINITE ELEMENT METHODS

The Finite element method (FEM) is by far the most widely used and versatile technique for
simulating deformable solids. FEM is a computer technique for solving partial differential
equation by the element-wise application of the variational method. There are two general types
of FEM in solid mechanics, namely; dynamics and static problems. For dynamic problems, FEM
solves equation of motion for a continuum, usually a more complicated version of F  ma . It

is time-dependent. In the case of static problems, FEM solves equilibrium equations F  0 ;

which in most cases is non-time dependent i.e.
  ij
 bi  0 (40) repeated
 xi
The use of FEM to solve equation (40), which is a homogeneous partial differential equation,
could be difficult to handle numerically. To address this, an equivalent integral formulation is
adopted to replace equation (40), namely;
- the principle of minimum potential energy, and
- the principle of virtual work.

The latter has the advantage of being easily extended to dynamic problems, other stress-strain
laws, and even to problems involving large shape changes. Hence, the principle of virtual work
forms the basis for the FEM while principle of minimum potential energy forms the basis of the
FEM in linear elasticity. The principle of virtual work will be considered in this class.

Finite Element Problem Statement

Given the followings for the isotropic elastic solid in Figure 5;
- the shape of the solid in its unloaded condition, R ,
- the elastic constants for the solid, Cijkl ,

- boundary conditions specifying displacement u i  u i * on a portion S1 and tractions

 ij ni  t j on a portion S 2 of the boundary of R .

To calculate the displacement, u i ; strain  ij and stress  ij satisfying the governing equations of

static linear elasticity;

1
Finite Element Methods Modupe A. Onitiri, PhD

1   ui  u j 
- the strain-displacement equation:  ij     (14) repeated
2   x j  xi 

- the elastic stress-strain law:  ij  Cijkl  kl (42) repeated

  ij
- the equation of static equilibrium: 0 (41) repeated
 xi
- the boundary conditions on displacement and stress:
u i  u i on a portion S1 and tractions  ij ni  t j on a portion S 2 .
*

y
ti  ij n j
S2
ty n
ny n
tx
tz
x
nx
S1 nz

V
z
y (b)
R

e2

e1
x
(a)
Figure 5: (a) Body with specified displacements and
e3
tractions (b) Stress components at a typical point on
z the boundary

Variational Formulation of the Principle of Virtual Work

The principle of virtual work can be used to rewrite the governing equation for the displacement

field in a linear elastic solid in an integral form called the weak form. Advantages of weak

formulation include;

(i) it requires weaker continuity of the dependent variable,

2
Finite Element Methods Modupe A. Onitiri, PhD

(ii) for self adjoint (i.e. even order differential) equations, it always results in a symmetric

coefficient matrix,

(iii) natural boundary conditions of the problem are included in the weak form. Therefore, the

approximation, Ũ, solution is required to satisfy only essential boundary conditions of the

problem.

Considering the equation for static equilibrium (balance equation); equation (40)
  ij
 bi  0 (40) repeated
 xi
But  ij  Cijkl  kl (constituitive equation) (42) repeated

1   ui  u j 
Since  ij     (14) repeated,
2   x j  xi 

equation (40) becomes,

1  u  u 
 ij  Cijkl   k  l   (48)
 2   xl  x k  
Since the elastic compliances must satisfy
Cijkl  Cijlk (49)

  uk 
  ij  Cijkl   (50)
  xl 
Substituting equation (3.3) in equation (40) gives;
   uk 
Cijkl    bi  0 (51)
 xi   xl 
Equation (3.4) is the governing differential equation
Let bi  f i (52)

where  = mass density

bi = body force (force/mass)
fi = body force (force/volume)
and consider a kinematically admissible virtual displacement field

3
Finite Element Methods Modupe A. Onitiri, PhD

wi   ui (53)
Then the weighted integral statement of equation (51) is
    uk  
0   wi Cijkl    f j  dV (54)
v   x i   xl  
Integrating by part the first integral on the RHS of equation (54)
i.e.  u dv  uv   v du
v
s
v

where
 wi 
u  wi ; du  dV 
 xj 
 (55)
   uk   uk 
dv  Cijkl   ; v  Cijkl
 xi   xl   xl 

    uk    uk   u k  wi
  w C i ijkl
 xi

  xl
 dV  wi Cijkl
 
   Cijkl
 xl  s v  xl  x j
dV (56)
v 
Substituting equation (56) in equation (54) gives
 u k  wi   uk 
0    Cijkl dV   wi f j dV  wi Cijkl  (57)
v
 xl  x j v   xl  s

From equation (50), equation (57) becomes

 wi
0     ij
 xj
dV   wi f j dV  wi  ij  s
(58)
v v

Equation (58) shows that integration by part transforms an integral into two new terms, namely;
a boundary term and a different integral.
For the first integral on the RHS of equation (58); since  ij is symmetric

i.e.  ij   ji

 wi 1 w
  ij   ij   ji  i
 xj 2  xj

1   wi  wj 

    ji
2   xj  xi 
ij

4
Finite Element Methods Modupe A. Onitiri, PhD

 wi 
1   w  w j 
  ij   ij   i  

(59)
 xj 
 2   x j  xi
 

where

1   wi  w j 
   = virtual strain field (59a)
2   x j  xi 

From equations (53), equation (59a) becomes

1   ui  u j  
 1   u  u j  
   =  i    =   ij (59b)
2   x j  xi   2   x j  xi  
 
 wi
  ij   ij   ij (60)
 xj

For the expression wi ij s in equation (58); from equation (50) and equation (54),

  ij
 x
v i
dV   ij (61)

 w    w   x
i ij s i s
ij
dV (62)
v i

Using the gradient and divergence theorems, that is,

 Fij
 x
v i
dV   Fij ni dA
s
(63)

the integral on the RHS of equation (62) becomes

  ij
  x
v i
dV    ij ni dA
s
(64)

Substituting equation (64) in equation (62) gives

 w   i ij s   ij ni wi dA (65)
s

Substituting equation (6) and equation (65) in equation (58) we have;

0     ij   ij dV   f i wi dV    ij ni wi dA (66)
v v s

Substituting equation (53) in equation (66) we have;


v
ij   ij dV   f i  ui dV    ij ni  ui dA  0
v s
(67)

5
Finite Element Methods Modupe A. Onitiri, PhD

But
 ij ni  T j  t j (boundary traction)


v
ij   ij dV   f j  ui dV   t i  ui dA  0
v s
(68)

Equation (68) is the integral form (weak form) of the governing differential equation for the
displacement field in a linear elastic solid (equation (51)). It is also the same as the statement of
the principle of virtual work.

t  u
s
i i dA is the boundary condition term of equation (68)

t  u
s
i i dA   t i  ui dA   t i  ui dA
s2 s1
(69)


v
ij   ij dV   f j  ui dV   t i  ui dA   t i  ui dA  0
v s2 s1
(70)

Since u i is specified (that is ui  ui * ) at S1 ,  u i vanishes.

That is  u i  0 at S1

 
v
ij   ij dVo   f j  ui dVo   t i  ui dA  0
v s2
(71)

Equation (71) forms the basis for the Finite Element Analysis (FEA).
From equation (57)

 u k  wi   uk  
 Cijkl  xl  x j
dV   wi f j dV   Cijkl  wi 
 xl   s
(72)
v v 
where the RHS is known as the loading term. It consists of the interior and the boundary loading
terms. The interior loading terms are all those terms in the original differential equation (i.e.
equation (51)) not containing the unknown u k or its derivatives while the boundary terms are all
those terms evaluated on the boundary.
In the absence of body force, equation (72) becomes

 u k  wi   uk  
C ijkl
 xl  x j
dV   Cijkl  wi 
 xl   s
(73)
v 

Construction of Trial Solution

6
Finite Element Methods Modupe A. Onitiri, PhD

The general form of a trial solution is

N
~
U ( x, y, z : a)  o ( x, y, z )   a  ( x, y, z)
i 1
i i (74)

where
o ( x, y, z ) and i ( x, y, z ) are interpolation or shape functions,
~
U ( x, y, z : a) = trial solution which is a function of x, y, z and a,

ai = unknown parameter

Equation (74) is a Lagrange interpolation polynomial; the finite element method requires that
polynomials be written in that form. For a typical element trial solution 3.28) reduces to
N
~
U ( x, y, z : a)   a  ( x, y, z)
i 1
i i (75)

o ( x, y, z ) term is removed because of the essential boundary condition term will be imposed
during the numerical process.
xb  x x  xa
1  ; 2  (76a)
xb  x a xb  x a
 j ( xi )   ji (76b)

Application of an Optimization Criterion

The optimizing criteria determine specific numerical values for each of the parameters a i
. There are two types of optimizing criteria, namely;
(i) methods of weighted residual (MWR)
(ii) Ritz variational method (RVM)

Methods of weighted residual (MWR)

MWR seeks to minimize an expression of error in the differential equation (not the
unknown function itself) i.e.

 w ( x) R( x, c
i j ) dx  0 i 1, 2, ... , N (77)

7
Finite Element Methods Modupe A. Onitiri, PhD

where
wi = set of linearly independent functions.
Special cases are: collocation, subdomain, least square and Galerkin methods. In practical
analysis, the Galerkin and least square methods are usually preferable.

Ritz variational method (RVM)

The Ritz variational method (RVM) or Rayleigh-Ritz method is not a special case of the
MWR. Unlike the MWR which uses the weighted integral formulation, the RVM uses the weak
integral statement. For the Ritz variational method (RVM),
~
(i) when boundary condition is specified as essential U must satisfy the homogeneous
form of the specified boundary conditions of the problem, i.e.
i  wi  0 (78)
~ ~
(ii) U must be complete e.g. if U is an algebraic polynomial, completeness requires
that all terms of the lowest order admissible up to the highest order desired.
(iii)  i must be differentiable as required in the evaluation of the integral statement (weak
form).

In practice, the RVM is more convenient to use compared with WRM because;
(i) the trial solution need only satisfy the essential boundary conditions and not the
natural boundary conditions and the differential equation. The last two are implicitly
contained in the weak integral statement,
(ii) the RVM always yields a symmetric matrix, and
(iii) the required order on the differentiability of the trial or shape functions is reduced.

MWR can be applied to any problem, whether or not it is linear and self-adjoint, whereas, the
RVM is only applicable to those problems for which a variational principle exists (common
situation only when a system of equations is linear and self adjoint). When the Garlekin’s
method is applied to a linear self adjoint differential equation, it agrees exactly with the RVM.

FINITE ELEMENT METHOD EQUATION FORMULATION

Since material is isotropic, equation (73) becomes;

8
Finite Element Methods Modupe A. Onitiri, PhD

L ~  d U~  
L
d U d wi
E dx    E  wi 
 (79)
0
d x d x 
 d x   0
Consider the trial solution, equation (75) for unknown U , i.e.
N
~
U ( x : a)   a  ( x)
i 1
i i (80a)

~ N
dU d
  ai i (80b)
dx i 1 d x
where equation (80a) is displacement and equation (80b) is strain.
Substituting equation (80b) in equation (79), we have;

 d U~  
xb
xb
 N d  j  d wi
  E   a j   
 d x dx    E d x  wi  (81)
xa  j 1 d x  
   xa

but from equation (78), for RVM, wi  i

 xb d  d  j   d U~  
xb
N
  

i
E 
d x a j    E

 i 
 (82)
j 1  x a d x dx  
  d x   xa

where i  1, 2, ..., N ; j  1, 2, ..., N

Expanding equation (82) we have;

 xb d 1 d 1   xb d 1 d  2   xb d 1 d  N   d U~   
xb

     a   1  
 x d x d x  x d x d x  x d x
E d x a  E d x a  ...  E d x   E 
 1
 2
 N   xa 
 a   a   a d x   d x

 d  2 d 1      ~  
xb

 a   d  2 E d  2 d x  a  ...   d  2 E d  N d x  a    E d U    
xb xb xb

 x d x d x  1  x d x  x d x
E d x   2
dx  2 dx  N  d x   xa 
 a   a   a 

    

 xb d  N d 1   xb d  N d  2   xb d  N d  N   d U~   
x b

 d x  a1    d x  a 2  ...    d x  a N    E
 x d x
E E E N  
 x d x  x d x  
 a d x   a d x   a d x   d x   xa 
(83)
In matrix form we have;

9
Finite Element Methods Modupe A. Onitiri, PhD

  ~
  
xb
 xb d 1 d 1 xb
d 1 d 2
xb
d 1 d  N  d U
 a1     E  1  
E
 xa d x d x
dx  dx E dx
dx ...  d x d x 
E d x
   
 d x 
  xa 
xa xa
   
 xb   ~ xb

 d  2 E d 1 d x
xb
d 2 d 2
xb
d 2 d  N    dU  
 d x d x  dx E dx ... x d x E d x d x  

     E
 2   
  2  

 xa xa
dx

a d x   xa 
   
a

         
 xb     
 d  N E d 1 d x d  N d 2 d N d N
xb xb
    
~
dU   
x

 d x  x d x E d x d x
b
E dx ...
dx dx dx a N     E N  

 xa xa a  
  d x   xa 

(84)
xb
d i d i 
Let K ij   d x E d x dx 
xa

~ xb  (85)
 dU   
and Fi    E  i 
 d x   
xa 

Equation (84) becomes
 K11 K12 ... K 1N  a1   F1 
K K1N  a   F 
 11 K12 ...  2  2
    (86)
       
  a N   FN 
 K11 K12 ... K 1N 

In abbreviated matrix and tensor form we have

K  a  F1  (87a)
K ij a j  Fi (87b)

K and Kij are known as stiffness matrix and stiffness term, respectively.

Element equation formulation

Table 3.1 shows that the notation is essentially the same for each element. The only difference
being the numerical values of the subscripts and superscripts, hence, equation needs to be
derived for only one of the elements; the resulting equation can then be used for all the other
elements simply by substituting the correct numerical values of the scripts and superscripts for
each element in turn.
Therefore, from equations (3.40) and (3.41) we have;

10
Finite Element Methods Modupe A. Onitiri, PhD

 K11( e ) K12 
(e)
a1  F1 
(e)

 (e)      (e)  (88)

 K11 K12  a 2  F2 
(e)

where superscript (e) identifies the stiffness and load terms as belonging to an element.

Table 3.1: Connectivity table for mesh

Local node 1 Local node 2

Element
becomes node becomes node
(1) 1 2
(2) 2 3
(3) 3 4
(4) 4 5
(5) 5 6
(6) 6 7
(7) 7 8
(8) 8 9
(9) 9 10
(10) 10 11

For a typical element trial solution in one dimension, equation (75) can be written as
N
~
U ( e ) ( x : a)  a 
(e)
i i ( x) (89)
i 1

where
x2  x x  x1
1( e )  ; 2 (e)  (90)
x 2  x1 x 2  x1
From equation (90),
d 1 d 2
(e) (e)
1 1
 ;  (91)
dx x2  x1 dx x2  x1
Substituting equation (91) in the stiffness term in equation (85) we have

11
Finite Element Methods Modupe A. Onitiri, PhD

 1   1  1 x 2  x1 
x2

K 11      E    dx  
x1 
x 2  x1   x 2  x1  2 x 2  x1 

 1   1  1 x 2  x1 
x2

K 12    
  E   dx  
x1 
x 2  x1   x 2  x1  2 x 2  x1 
 (92)
 1   1  1 x 2  x1 
x2

K 21     E   dx  
x1 
x 2  x1   x 2  x1  2 x 2  x1 

x2
 1   1  1 x 2  x1 
K 22     E   dx  
x1 
x 2  x1   x 2  x1  2 x 2  x1 

From equations (85) and considering the relationship

 j ( xi )   ji (76b) repeat

the boundary load terms become;

~
 d u~ ( e )   d u~ ( e )   d U (e)  
F1     E  1 ( x 2 )    E  1 ( x1 )   E  
 d x x  d x  x 
 d x 
 x1 
 (93)
2 1

~
 ~
du  (e)
 ~
du  (e)
 d U (e)  
F2     E   2 ( x 2 )    E   2 ( x1 )     E 
 d x x  d x  x  d x  
 x2 
2 1

Substituting equations (92) and (93) in equation (88) we have

  ~
 1 x 2  x1 1 x 2  x1  d U (e)  
  
2 x  x  a1     E d x 
2 x 2  x1       x1 
 2 1
     (94)
 1 x 2  x1 1 x 2  x1  ~
a    d U (e)  
 2 x  x 2 x 2  x1   2     E 
 

   x2 
2 1 dx

Imposition of boundary conditions in equation (94) gives the solution to the unknown a i .
But

 ~ x2
d U (e) 

  E    ( e ) ( x : a) (95)

 dx  x1

where equation (95) is a constitutive equation.

Substituting equation (80b) in equation (95), we have;
N  d i ( e )   d  (e) d 2
(e)

 ( e ) ( x : a)   E    ai   E  1 a1 
  dx a 2


(96)
i 1  d x   dx 
From equation (91),

12
Finite Element Methods Modupe A. Onitiri, PhD

 
 ( e ) ( x : a)   E  
1
a1   1
a2  (97)
 x 2  x1 x2  x1 

a a 
  ( e ) ( x : a)   E  1 2  (98)
 x2  x1 

REFERENCES

1. Birley, A. W.; Heath, R. J. and Scott, M. J., (1988), Plastics Materials Properties and

Applications, 2nd ed., Blackie Academic and Professional, London.

2. Bower, Allan F., (2010), Applied Mechanics of Solid, CRC Press, New York.

3. Chandrupatla, Tirupathi R. and Belegundu, Ashok D., (2002), Introduction to Finite

Elements in Engineering, 3rd edition, Prentice Hall, New Jersey.

4. Bathe, Klaus-Jürgen, (1982), Finite Element Procedures in Engineering Analysis,

Prentice-Hall, Inc. New Jersey.

5. Burnett, David S., (1987), Finite element analysis from concept to application, Addison-

Wesley Publishing Company, Boston.

6. Reddy, J. N., (2006), An Introduction to the Finite Element Method, 3rd edition,

McGraw-Hill International, New York.

7. Heinbockel, J. H., (1996), Introduction to tensor calculus and continuum mechanics,

Trafford Publishing, Indiana.

13