0% found this document useful (0 votes)
10 views112 pages

5 FEM Formulation

Uploaded by

kakanda.alphonse
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
10 views112 pages

5 FEM Formulation

Uploaded by

kakanda.alphonse
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 112

Chapter 3

Finite Element Method of


Elastic Mechanics Problems
Contents

1 Summary

2 Two-dimensional Poisson equations

3 General format of finite element method

4 Plane problems of elastic mechanics

5 Further discussions
I. Summary
Summary
The variational principle is generally applicable in elastic mechanics, so we’ll
discuss the finite element analysis in terms of this method.

From the direct solution of variation problems we know that if the trial
function is close enough to the real one, the result will be rather accurate.
It can be called as the classical variational method, which actually helps
solving some engineering problems.
However as discussed before, selecting trial functions will be very difficult
once the boundary conditions are relatively complex.
Furthermore, the trial functions vary with different problems, so it’s not
convenient for computer coding.

The finite element method develops to overcome these disadvantages.

The finite element method first divides the continuum into a finite number of
elements. In the mathematic term, it divides the solving domain into
subdomains and adopt an unified displacement function for every subdomain.
Summary

This type of displacement function is equivalent to the trial functions in the


classical variation method but it’s set for the elements, which may not satisfy
the deformation compatibility condition between one another. So this make it
much easier to establish a displacement function.
Take the triangle plane element for example. It can satisfy the deformation
compatibility condition within an element, and displacement boundary
condition on the boundary but it doesn’t satisfy the deformation compatibility
at the junctions between each other, because the deformation is continuous
but the partial derivatives are not continue.
Thus the problem will be associated with each element rather than the global
continuum.

It can be proved that the results we get through the finite element method will
converge to the real displacement of the continuous body once the
displacement function satisfies a certain condition.

We’ll first discuss the question from a mathematical term.


II. Two-dimensional
Poisson equations
2.1 The weak form of equivalent integral
We’ve solved the two-dimensional Poisson equation by the classical variation
method before. (the Galerkin method)
Now let’s look at the general expression of Poisson equation in a plane
bounded domain Ω.
 u  f

u   u0 ( x, y )
2 2
Δ is Laplace operator  2  2
x y
This is obviously the most simple case of the first type of the “boundary
value problem” , the elliptic equation, which is quite common in physics or
mechanics. For example the membrane equilibrium problem under the out-of-
the-plane external force and the steady temperature field problem.

The Poisson equation will become the Laplace equation without the
action of external force.
2.1 The weak form of equivalent integral

We can directly obtain its weak form of equivalent integral by the Galerkin
method.
As mentioned before, there’s usually a quite high demand of continuity of the
trial function in this situation and we can transform it into the corresponding
weak form of equivalent integral after integration by parts.
Closure includes the
    domain and its
Suppose boundary which we
u( x, y)  C 2 () C1 () call total here
That is u(x,y) is 2nd order continuously differentiable in domain Ω and is first
order continuously differentiable in the entire domain.
Then for an arbitrary  ( x, y)  C1 ()

The blue 


(u  f ) dxdy   (u ) dxdy   ( f ) dxdy
 
part is
  
Green     (u x )  (u y )  dxdy   u x x  u y y  dxdy   f  dxdy
 x y  
Formula  
u
   ds   u x x  u y y  dxdy   f  dxdy  0
 n  
2.2 Element Analysis
We’ll discuss the two-dimensional finite element method in detail and
compare it with the one-dimensional ordinary differential equation boundary
problem.

 Element division
Divide the domain Ω into a series of triangle, rectangular, quadrilateral and curved
polygon elements. The triangle elements are the most simple and flexible. It can
approximate the boundary well, so triangle element division is widely adopted.
We usually call the points of triangle the nodes. And then number them. Denote the
nodes as Pi ( xi , yi )(i  1,..., NP), the elements as ek (k  1,..., NE) .

Pay attention to the following items when numbering the nodes.


1.The node must be the vertex of adjacent elements but not the inner point on
the boundary.

对 错
2.2 Element Analysis
 Element division
2.Avoid large obtuse angle.

好 差

3. The meshes should be fine where the gradient of u(x,y) varies intensely;
and sparse where the gradient varies slightly.
4. The numbering of the elements can be arbitrary but the arrangement can
largely influences the band width of the global stiffness matrix. The
principle is that keeping the difference of the neibourghing nodes minimal.
Example:For the triangular division shown as follows

How would you number them to have small band width?


2.2 Element Analysis
Pm

 Interpolation polynomial
We still adopt the linear interpolation whose general expression
is as follows in the two-dimensional situation. Pj
u  ax  by  c Pi
It has 3 undetermined coefficients.
The value of u at 3 points should be used to determine the form of the
interpolation polynomial in every element. And usually we’ll take the 3 vertexes of
the element. Let the value of u at the node Pi be u i , which means u(x i,y i)= u i
( i=1,……,NP )。
The three vertexes of an arbitrary element e are Pi, Pj, Pm. Let e = ∆ PiPjPm ,
which are in an anticlockwise order.
If the value of interpolation function respectively equals u i , u j , u m at three
vertexes,then a, b, c should satisfy
 axi  byi  c  ui

 ax j  by j  c  u j
ax  by  c  u
 m m m
2.2 Element Analysis

Interpolation polynomial
(the Cramer law)

1  yj 1 ym 1 yi 1 
a  ui  uj  um 
2 e  ym 1 yi 1 y j 1 

xi yi 1
1  xj 1 xm 1 xi 1  where 2 e  x j yj 1
b  ui  uj  um 
2 e  xm 1 xi 1 x j 1 
xm ym 1

1  xj yj xm ym xi yi 
c  ui  uj  um 
2 e  xm ym xi yi x j y j 
2.2 Element Analysis
 Interpolation polynomial
Pi, Pj, Pm are anticlockwise, so ∆e is positive and is the area of
the triangle element e = ∆PiPjPm.
Substitute it into the general form of linear function and we can
get the interpolation function of element e.
u  Ni ( x, y)ui  N j ( x, y)u j  Nm ( x, y)um
Where
1  yj 1 xj 1 xj yj 
Ni   x y 
2 e  ym 1 xm 1 xm ym 
1
  ai x  bi y  ci 
2 e
The expressions of Nj,Nm and the constants aj、bj、cj 、 am、
bm、cm can be obtained by rotating the subscripts of the
expression above.
2.2 Element Analysis
 Interpolation polynomial
Suppose
 e  ui u j um  ,  N    Ni
T
Nj Nm 
Similar to the one-dimensional situation we have u   N  e in the
element e.

The gradient vector of u can be expressed as

 u   N i N j N m 
 x   x x x

u      
 u   N i N j N m 
e

 y   y   
 y y 
1  ai a j am 
 b b   e   B  e
2 e  i j bm 
B is a 2×3 constant matrix
2.2 Element Analysis
 Interpolation polynomial

Ns(xt,yt) (s,t=i,j,m) is the linear interpolation basis function of element e.


u  N i ( x, y )
The properties are as follows

1
1. Ns(xt,yt) (s,t=i,j,m) is first order polynomial in element e
Pm
Pi
2. Ns(xt,yt)=δst (s,t=i,j,m)
Pj
3. In geometry u = Ni(x,y) represents a plane passing through (xi, yi, 1),(xj, yj,
0),(xm, ym, 0) in the space x,y,u.
Conclusion is the same for the basis function u = Nj(x,y) and u = Nm(x,y).

4. Due to the uniqueness of linear interpolation, the interpolation of a linear


function is the function itself. So we have the following identical equations
1  Ni  N j  N m
x  xi Ni  x j N j  xm N m
y  yi N i  y j N j  ym N m
2.2 Element Analysis

 Interpolation polynomial Pm
5. Consider the linear transformation
1  Ni ( x, y), 2  N j ( x, y) Pi
Because
Nm  1  1  2 Pj
The reverse transformation is
x  ( xi  xm )1  ( xi  xm )2  xm 2

y  ( yi  ym )1  ( yi  ym )2  ym
B

The Jacobi determinant is 1

 ( x, y ) xi  xm x j  xm
  2 1
 (1 , 2 ) yi  ym y j  ym O 1 A

The element e becomes a standard triangle OAB on the plane λ1-λ2 after the
transformation. Image of the point Pi is A(1,0), image of the point Pj is B(0,1) and
image of the point Pm is O(0,0).
We’ll further discuss this property in the following chapter of isoparametric element.
2.2 Element Analysis

 Interpolation polynomial
6.On every side of the element e Ns(xt,yt) (s,t=i,j,m) is the first order function of
the parameter t of the arc length.
Take the side PP i j for example
Let Pi (t  0), Pj (t  l  PP
i j )

t (t  l)
Then Ni ( x, y ) |P P  1 
i j
l Pj
N j ( x, y ) | P P 
t (t 0)
i j
l Pi
Nm ( x, y) |PP  0
i j

The third equation holds because because Pm


Ni  N j  Nm  1
2.2 Element Analysis

 Element stiffness matrix and element load vector


Rewrite the equivalent integral weak form of the boundary problems of Poisson equations
into the following equation according to the elements divided.
NE NE

  (u 
n 1 en
x x  u y y )dxdy   f  dxdy
n 1 en
Now let’s use the function value at the nodes to replace every term of the equation above.
Let the element e=∆PiPjPm. The function value of u(x,y), φ(x,y) at the node Ps(s=i,j,m) is
respectively us, us*.

According to u   B e we have:


 (u   u y y )dxdy    u dxdy
T
x x
e e

  ( B  * )T ( B  e )dxdy


e
e

    ( B  B)dxdy       k   
* T T * T
e e e e e
e
2.2 Element Analysis
 Element stiffness matrix and element load vector

 k e is a 3×3 element stiffness matrix

 k e   [ B]T [ B]dxdy   e [ B]T [ B]


e

 kiie kije kime 


 
  k eji k ejj k ejm 
 kmie kmje e 
kmm
 
The stiffness coefficients of the element is
 N s N t N s N t 
k  e 
e
 
   
st
 x x y y 
1
 (as at  bs bt )
4 e
2.2 Element Analysis
 Element stiffness matrix and element load vector

For the right part of the weak form of the equivalent integral
NE NE

  (u 
n 1 en
x x  u y y )dxdy   f  dxdy
n 1 en

    fdxdy

We have T
f dxdy ( )
e e

  ([ N ]{ *}e )T fdxdy  { *}Te {F}e


e
 Fi e 
Where  
{F }e   [ N ]T fdxdy   Fje 
e  Fme 
 
The equation above is element load vector.
2.3 Global integration
 Global stiffness matrix and the global load vector

Substitute the expressions of element stiffness matrix and element load vector into the
integral form of the element.
NE NE

  (u 
n 1 en
x x  u y y )dxdy   f  dxdy
n 1 en

We should expand { }e ,{F}e ,[k ]e to a NP-dimensional vector and a NP×NP square


matrix for easier superposition.
The method is similar to the one-dimensional situation and the difference is that the
global number may not be ordinal because the order of three vertexes Pi,Pj,Pm must
be anticlockwise in the element e. For example when j<i<m
   
F e   k e
k e
k e 
 j  jj ji jm 
   
The imaginary points  e  
represent element 0 { F }e  F
 i  {k }e   k e
ij k e
ii k e
im 
   
 e  e e e 
 Fm   kmj kmi kmm 
   
   
2.3 Global integration
 Global stiffness matrix and the global load vector

We can then superpose them after expanding :


NE NE

 en en en  }en {F}en
{
n 1
* T
} [ k ] { }  { * T

n 1

NE NE
{ } ( [k ]en ){ }  { } ({F}en )
That is * T * T

n 1 n 1
Thus

{ *}T ([K ]{ } {F})  0


Where NE NE
[ K ]  [k ]en [ F ]   [ F ]en
n 1 n 1
2.3 Global integration
 Global stiffness matrix and the global load vector

We can then superpose them after expanding :


NE NE

 en en en  }en {F}en
{
n 1
* T
} [ k ] { }  { * T

n 1

NE NE
{ } ( [k ]en ){ }  { } ({F}en )
That is * T * T

n 1 n 1
Thus

{ *}T ([K ]{ } {F})  0


Where NE NE
[ K ]  [k ]en [ F ]   [ F ]en
n 1 n 1
2.3 Global integration
 Global stiffness matrix and the global load vector
It can be proved that [K] is both symmetric and nonnegative definite.
Actually for the quadric form
{ }T [ K ]{ }  0
Because NE
{ } [ K ]{ }   { }Ten [ K ]en { }en
T

n 1

  [ B]{ } dxdy
NE
   [ B ]{ }en
T
en
n 1 en

NE
   u dxdy
2

n 1 en
The function u under the sign of integration is a linear interpolation function of
which the values at three vertexes of the element en are ui ,uj and um respectively.
The following discussion is similar to the one-dimensional problem.
There’s no constraint condition for the function φ(x,y) on the boundary, so {δ*} is an
arbitary NP dimensional vector. Then we can obtain the linear equations.
{ *}T ([ K ]{ } {F})  0  [K ]{ }  {F}
2.3 Global integration
 Global stiffness matrix and the global load vector
One only needs to arrange the elements in element stiffness matrix and the
element load vector in the order of their subscripts, then the global stiffness
matrix and the global load vector can be obtained. We should then deal with
the constraints:

From the final linear equations


ui (i  1,..., NE)
From the interpolation polynomial we can obtain the expression of the
approximate solution
u( x, y)  [ N ]{ }en  ui Ni ( x, y)  u j N j ( x, y)  um Nm ( x, y)
Where en  PP
i j Pm (n  1,..., NE)
It should be noticed that the stiffness matrix is usually in band and sparse
because the basis functions of the finite element usually consist of low order
piecewise polynomial functions, which brings much convenience for
discretization and numerical solution.
III. Finite element form
of elastic mechanics
3.1 Matrix expression of the basic equations
Now let’s discuss the application of the finite element method in elastic
mechanics. The finite element method originates from the method of matrix
structural mechanics, but its real attractions is that it successfully resolves the
continuum body (field) problems.
The conventional finite element method of elastic mechanics or finite
element method for short is based on the principle of virtual displacement.
The process is the same for all kinds of problems and one only needs to
change the governing equations, coordinates, displacement and strain to the
corresponding expressions.
Here we’ll adopt the most simple linear tetrahedron element to learn the
derivation of finite element method for elastic problems in 3D space. In the
next chapter we’ll discuss the derivation of elastic plane problems.
Actually the 3D problem is more generic than the plane problem and the
derivation process is almost the same. As we have already learnt something
about the finite element form of the plane problems before when discussing
the two-dimensional Poisson equations, we’ll start with the 3D problems in
this chapter.
We mainly adopted the component form and tensor form when
reviewing the elastic mechanics. And now we’ll use the matrix form for the
convenience of expression and derivation.
Let’s first look at the according notations such like the matrix operator.
3.1 Matrix expressions of the governing equations
Strain, displacement and stress are usually expressed in the form of
matrix in the elastic 3D problems.
 x   x 
   
u   y  y
    z    z 
u  v       
 w  xy   xy 
   yz   yz 
   
 zx   zx 

u,v,w respectively represent the displacement in the direction of x,y,z.


εx, εy, εz and σx, σy, σz respectively represents the normal strains and stresses
in the direction of x,y,z.
γxy, γyz, γzx and τxy, τyz, τzx respectively represents the shear strains and
stresses of the plane.
3.1 Matrix expression of the governing equations

Now introduce the matrix operator


T
  
 x 0 0
y
0
z 

    
  
  0 0 0
 y x z 
   
 0 0 0 
 z y x 
  2G   0 0 0
Then the basic equations are as follows     2G  0 0 0 
Geometric equation  
ε  [] u      2G 0 0 0 
(61) (63) (31)
Equilibrium equation D 
 0 0 0 G 0 0 
[ ] σ  f  0
T

(36) (61) (31)  0 0 0 0 G 0


Constitutive equation
σ  D ε  
(61) (66) (61)  0 0 0 0 0 G
The constitutive matrix D of the elastic body can be expressed as above using the
Lame constants.
3.1 Matrix expression of the governing equations
Then the equilibrium equation can be expressed as
[] σ  f  [] D ε  f  [] D [] u  f  0
T T T

(36) (61) (31) (36) (66) (61) (31) (36) (66) (63) (31) (31)

The boundary conditions are


Boundary of displacemnet
uS u

Boundary of external force
cos(n, x) 0 0 cos(n, y ) 0 cos(n, z ) 
 0 cos( n, y ) 0 cos( n, x ) cos( n, z ) 0 σ  T
 
 0 0 cos(n, z ) 0 cos(n, y ) cos(n, x) 
The total functional of the minimum potential energy can be accordingly
expressed in the form of matrix
1 T
   ε Dε dV   uT f dV   uT T dV
2  S
Explanation: The equation of virtual work, the principle of virtual displacement
and the principle of minimum potential energy can be all used to derive the
discrete form of the finite element.
3.2 Discretization and the establishment of the element

The member system can be naturally discretized according to its structural


form. We can suppose the solving domain to be a polyhedron which can be
totally represented by the polyhedron elements.

The tetrahedron element is the most simple one among the three-dimensional
elements. So we use it to discretize the solving domain. After meshing, we
have the elements en(n=1,……,NE) and the nodes(vertexes of the
tetrahedron)are Pi(xi,yi,zi) (i=1,……,NP).  ui 
 

Suppose that the displacement u at every node Pi is i  vi 
u
w 
 i
Denote an arbitrary element e whose four vertexes are Pi , Pj , Pm, Pl as e =
(Pi , Pj , Pm, Pl).
Pl
z
Pm
y Pi
x Pj

The order follows the right hand rule, which means when Pi ,Pj and Pm rotate
in the direction of the right hand Pl is in the direction of thumb.
3.2 Discretization and the construction of element
A linear interpolation can be determined in the element according to the value of
displacement at four vertexes.
u  α1x  α2 y  α3 z  α4
Where αi(i=1,…,4) is a 3×1 column matrix. It should satisfy
us  α1xs  α2 ys  α3 zs  α4 (s  i, j, m, l )
Similarly to the plane triangle element one can use the Cramer rule to solve αi.
Substitute it into the equation above and we can get
u  Ni ui  N j u j  N mu m  Nl ul   N   e
Where (31)
(312) (121)
1 x y z 1 xi yi zi 1 xi yi zi
1 1 xj yj zj 1 1 xj yj zj 1 1 xj yj zj
Ni  , , Nl  , Ve 
6Ve 1 xm ym zm 6Ve 1 xm ym zm 6 1 xm ym zm
1 xl yl zl 1 x y z 1 xl yl zl
Ve is the volume of tetrahedron e. Pl
xi  xm yi  ym zi  zm
1
6
1
6
 1
6

Ve  V六面体  Pmi  Pmj Pml  x j  xm y j  xm z j  xmP m
Pj
. xl  xm yl  xm zl  xm Pi
3.2 Discretization and the construction of the element

{a}e is the displacement vector of the nodes.

ae  ui wl 
T
vi wi uj vj wj um vm wm ul vl
[N] is the matrix of the interpolation function.

 Ni 0 0 Nj 0 0 Nm 0 0 Nl 0 0
 
  
N  0 N i 0 0 N j 0 0 N m 0 0 N l 0 
 0 0 Ni 0 0 Nj 0 0 Nm 0 0 Nl 

  Ni I 3 N j I3 N mI3 Nl I 3 
where
I3 is a 3rd element matrix.
1
Ns   as x  bs y  cs z  d s  (s  i, j, m, l )
6Ve
as,bs,cs,ds 为are the coefficients related to the geometric position of the nodes.
3.2 Discretization and the construction of the element
For example for the node i 1 yj zj 1 xj zj
ai   1 ym zm , bi  1 xm zm ,
1 yl zl 1 xl zl
1 xj yj xj yj zj
ci   1 xm ym , d i   x j ym zm
1 xl yl xj yl zl
The rest as,bs,cs,ds (s=j,m,l) can be written in the similar way.
It should be noticed that the permutation of the three-dimensional problems are not
totally the same as the plane problems because it cannot assure the right hand rule.
For example j,m,l,I doesn’t obey the rule.

So it should be corrected as: If the permutation starts with a even number( e.g. j
equals 2) then it obeys the left hand rule.
Pl
1 ym zm 1 yl zl 1 yi zi
a. j  1 yl zl , am   1 zi zi , al  1 y j zj Pm
Pi
1 yi zi 1 zj zj 1 yj zm Pj
3.3 Solving process
Substitute the displacement of the element in the form of interpolation into
the geometric equation and the physical equation
u   N ae
then the strain and the stress can be expressed as
ε     u     N  ae  B ae
(61) (31) (612)
(63) (63) (312) (121) (121)

σ  D ε  D B ae
(61) (66) (61) (66) (612)
(121)
where
B    Ni I3 N j I3 NmI3 Nl I3   Bi Bj Bm Bl 
After the calculation of partitioned matrix we have
 as 0 0 
T
 N s N s N s 
 x 0 0 0   0 b 
 y z   s 0 
 N s N s N s  1  0 0 cs 
B s    N s I3   0 0 0    ( s  i, j , m, l )
.  y x z  6Ve  sb a s 0 
 N s N s N s   0 cs bs 
 0 0 0   
 z y x   cs 0 as 
3.3 Solving process
The potential energy of a discretized model equals to the sum of potential
energy of all the elements. We can then obtain the total potential energy of a
discrete model that
 1 T 
     e
p ε Dε dV   uT f dV   e uT T dS 
e  
e 2 e S
e

 1
           
T T T
B ae D B ae dV   N ae f dV   e N ae T dS 
e  
e 2 e S

 T
   ae
e 
1 T
e 2 B DB dV a 
e 

 e
a
T
e  e N T
f dV   a T
e   e 
Se
NT T dS 

Let
Ke   BT D B dV P fe   NT f dV
(1212) e (126) (66) (612) e (123) (31)
(121)

PSe   e NT T dS P e  P fe  PSe
S (123) (31)
(121)

Ke and Pe can be respectively called the element stiffness matrix and the element
equivalent load column matrix.
3.3 Solving process
Substitute the two equations above into the expression of the potential energy
and we have
1
   
   ae K e ae   ae P fe   ae PSe
2 e
T

e
T

e
T
 
  a K a    a P 
1 T e T e
e e e
2 e e

Assemble the equation and we can get the global stiffness matrix and the load vector.
1
  aT Ka  aT P
2
Because the unknown variables of the total potential energy in the discrete form are
the displacement of the structure at the nodes. Then according to the variation
principle the functional gets its stationary value when its first variation is zero.

0
a
Then we can get the solving equation of the finite element method.
Ka  P
So this is the general principle of the finite element solving equations based on the
principle of minimum potential energy of elastic mechanics.
The specific calculation involves the forming of element stiffness matrix, column
matrix of element equivalent nodal loads, integration of global stiffness matrix
and global load column matrix as well as the introducing of the given boundary
3.3 Solving process
In conclusion the element stiffness matrix is
 K iie K ije K im e
K ile 
 (33) (33) (33) (33) 
Ke Ke Ke K e 

1  (33)ji (33)jj (3jm jl 

Ke  

BT DB dV   e 
3) (3 3)
(1212) e 2
K mi K mj K mm K ml 
e e e

 (33) (33) (33) (33) 


 Ke Ke Ke K e 
 (33)li (33)lj (3lm ll

 3) (33) 
For the linear triangle element B is the matrix of constants. So
 BTs DBtVe
e
K
(33) st

 as at (  2G )  (bs bt  cs ct )G asbt   bs at G as ct   cs at G 
1  
 b a   a b G b b (   2G )  ( a a  c c )G b c   c b G
36Ve  
s t s t s t s t s t s t s t

 cs at   as ct G cs bt   bs ct G cs ct (  2G )  (bsbt  as at )G 
( s, t  i, j , m, l )
For the element of high degree B is not a constant so numerical integration is
usually needed.
3.3 Solving process
Calculate the integral to obtain the equivalent load vector

Pfe   NTandf dV PSe   e NT T dS


e (123) (31) S (123) (31)
(121) (121)
For the simple distribution, volume coordinate will be more convenient.
For the more complex load distribution numerical integral will be needed.
The integration of global stiffness matrix and the global load vector is similar to the
situation before.
Now let’s discuss the method “multiply the diagonal elements with large numbers”
introduced by the boundary conditions.
When the displacement of the node is given a j  a j , then the equation j needs to be
modified as K a  K a    K a   K a   K a
j1 1 j2 2 jj j jn n jj j

The magnitude of α can be around 1010, so we have


 K jj a j   K jj a j  a j  a j
If there are several given displacements, modify them in order.
This method is applicable to any given displacement (zero or nonzero).
When introduce the boundary condition by this method, the order of the equation and
the order of nodal numbering will not be changed, which is quite convenient for
programming. That’s why it’s often used in the finite element method.
IV. Plane problems of elastic mechanics
The elastic plane problems refer to the plane stress problem and the plane
strain problem. Let’s look at a complete process of derivation for further
understanding.
Express with the form of matrix

x   x 
u     
Displacement u  Strain ε    y  Stress σ  
 y
v   xy 
 
 xy 
 

u,v is the displacement in the direction of x and y, εx 、εy and σx 、σy


respectively represent the formal strain and the formal stress in the direction
of x and y. γxy and τxy respectively represent the shear strain and shear stress.
 
 0
The corresponding differential operator of matrix is  x 

[ ]   0 
 y 
 
 
 y x 
um
Still take the triangle element for example.
m( xm , ym ) vm
This element has 6 degrees of freedom (DOF)
of the nodal displacement.
Assemble all the node displacement to a column matrix uj
q  u1 v1 u2 v3 
e T
v2 u3 ui j( x j , y j )
y
(61) vj
v
i ( xi , yi )i
Also assemble the forces of the nodes to a column matrix. O x
T
P 
e
 P1x P1 y P2 x P2 y P3 x P3 y 

(61)

One can use the column matrix of node displacement and the other relevant
interpolation functions to express all the mechanical parameters
u(xi,yi)、σ(x,y) 、 ε(x,y) 、 Πe
using function interpolation, geometric equations, physical equations and the
calculation formula of potential energy.
4.1 Element analysis
 Expression of the element displacement field
If express the displacement in the directions of x and y, then the displacement of each
direction can be determined by 3 nodal displacement. So there are 3 undetermined
coefficients in each direction.
Considering the simplicity, completeness, continuity and the uniqueness of the
coefficients, the displacement mode is
u  x, y   1   2 x   3 y

v  x, y    4   5 x   6 y
Due to the nodal condition at the point (xi,yi) we have
u  xi , yi   ui 

  i  1, 2, 3
v  xi , yi   vi 

Then the undetermined coefficients are
u1 x1 y1 1 u1 y1
1 1
1 
1
u2 x2 y2 
1
 a1u1  a2u2  a3u3  2  1 u2 y2   b1u1  b2u2  b3u3 
2A 2A 2A 2A
u3 x3 y3 1 u3 y3
1 x1 y1
1 x1 u1 1
3 
1
1 x2 u2 
1
 c1u1  c2u2  c3u3  A  1 x2 y2
2A 2A 2
1 x3 u3 1 x3 y3
4.1 Element analysis
 Expression of the element displacement field
Similarly we have
1
4 
1
 a1v1  a2v2  a3v3   5   b1v1  b2v2  b3v3 
2A 2A
1
6   c1v1  c2v2  c3v3 
2A
where x2 y2 
a1   x2 y3  x3 y2 
x3 y3 
1 y2 

b1   y2  y3   指标轮换 
1 y3 
1 x2 
c1    x2  x3 
1 x3 

Rewrite the displacement function in the form of node displacement using the
equations above and we have:
u  x, y   N1  x, y  u1  N 2  x, y  u2  N 3  x, y  u3

 v  x, y   N1  x, y  v1  N 2  x, y  v2  N 3  x, y  v3
4.1 Element analysis
 Expression of the element strain field
According to the geometric equations of the elastic plane problems.
 u    
   0
  x   x   x 
   v     u  x, y  
ε  x, y     y     0    [ ] (2u1)
(31) y  x  v  x, y   (32)

 xy  
   u v    

  
 
 y x   x x 
Substitute the displacement with the interpolation function and we have
ε  x, y   [] u  [] N ( x, y) qe  B ( x, y) qe
(31) (32) (21) (32) (26) (61) (36) (61)

Substitute the expression of N(x,y)


b1 0 b2 0 b3 0
1    
B ( x, y )  [] N ( x, y)   0 c1 0 c2 0 c3    B1 B2 B3 
(36) (32) (26) 2A
c1 b1 c2  (32) (32) (32) 
b2 c3 b 
4.1 Element analysis
 Expression of the element stress field
According to the physical equations of the elastic plane stress problems.
 
 x  1
 0  x 
  E   
σ  x, y    y   2 
1 0  y   D ε
(31) 1   
(33) ( 31)
 xy  1    xy 
  0
0 
 2 
 E  
The coefficient (E,μ) should be changed to   as to the plane
 1  1  
2
strain problem.

Substitute the expression of strain and we have


σ  x, y   D B ( x, y) qe  S qe
(31) (33) (36) (61) (36) (61)

Where S is called the matrix of stress function.


4.1 Element analysis
 Expression of the element potential energy
We have expressed the three fundamental variables(u,ε,σ)
of the element by the column matrix qe of the node
displacement. Substitute it into the expression of the element
potential energy and we have
1
e   σ ε    
T T T
dV f u dV T u dS
2  e
 e
S

   
T
1
 q eT  e BT DB dV q e  N f dV   N T dS
T T
qe
2   e
S

1 eT e e
 q K q  P eT q e
2
Where Ke(x,y) is the element stiffness matrix
Ke   D B dV  t  e BT DB dA
T
B
(66) e (63) (33) (36) A

t is the thickness of the plane stress problem


4.1 Element analysis
 Expression of the element potential energy
For the linear triangle element with three nodes the matrix B is a constant
coefficient matrix. So the element stiffness matrix can be written as
 k11 k12 k13 
K e  B D B tA  k 21 k 22 k 23 
T
(66) (63) (33) (36)
k 31 k 32 k 33 
Where the sub- matrix is
Et  k1 k3 
k rs  B r D B s tA 
T

(22) (23) (33) (32) 4(1   2 ) A  k2 k4 

The equivalent node load is


P e   e NT f dV   NT T dS
(61)  S

 t f dA  t 
T T
N N T dl
e (62) (21) S (62) (21)
4.1 Element analysis
The stiffness equation of the element
Take the first order extremum of the element potential energy to the node displacement.
Then the element stiffness equation is
q  P
e e e
K
(66) (61) (61)
Notice:
The displacement field of the triangle element with three nodes is linear. So
the B(x,y) and S(x,y) obtained above are both constant coefficient matrixes.

So the stress and strain at any point in the element is constant and the triangle
elements with three nodes will become the element of constant stress or strain.

The meshes should be more intensive in the area where the strain gradient
(stress gradient) changes sharply. Otherwise the real situation of the strain and
stress cannot be described and it will cause a big error.

After the element analysis there will be the conventional integration of the
global stiffness, integration of the load vector and then solve the problem.

The key point of the finite element method is to express the element field
variables in a function form. Now let’s look at a simple example.
4.2 Examples
 The finite element analysis of cantilever plane problems
A cantilever is applied a concentrated force F at the right end. The elasticity modulus of the material
is E, the Poisson’s ratio=1/3, the thickness of the cantilever is t.
Please solve the displacement of every node and the reaction at support according to the plane stress
problems. 2m y F /2
Pi
3 ji m 1

2
Pj
1m
1

4 j 2
mO i
x
(1)Structural discretization and numbering F /2

Discretize the structure, the numbering of elements and nodes is shown in the picture. There are
2 triangle elements with 3 nodes. The force F moves to the nodes 1 and 2 in terms of the static
equivalent principle.
The column matrix of the node displacement:
q  u1 v1 u2 v4 
T
v2 u3 v3 u4
The column matrix of the external force at the nodes:
T
 F F 
F  0  0  0 0 0 0
 2 2 
The column matrix of the bearing reaction of the constraint: T
R  0 0 0 0 R3 x R3 y R4 x R4 y 
The global column matrix of the node load:
T
 F F 
P  0  0  R3 x R3 y R4 x R4 y 
 2 2 
4.2 Examples
 The finite element analysis of cantilever plane problems
(2)Description of every element
When the local node numbers of two elements are as follows then they have
the same stiffness matrix.
2m y F /2
Pi
3 ji m 1

2
Pj  2 2
1m
 1 0 0 1  
1 3 3
4
 
 0 1
j 2
mO 1 2 2
 
i
0
x
 3 3 3 3
F /2
 
 k ii k ij k im   0 2 4 4 2
0  
  9 Et  3 3 3 3
K (1)  K (2)   k ji k jj k jm  
32  2 2 
k mi k mj k mm   0 0 4  4 
  3 3 
 2 4 2 7 4 
 1    
 3 3 3 3 3 
 2 1 2 4 13 
  3 
3

3
4
3 3 
4.2 Examples
 The finite element analysis of cantilever plane problems
(3)Establish the global stiffness matrix
Assemble the element matrixes according to the displacement freedom degree to
get the global stiffness matrix.
The assembling process can be written as
2m y F /2
Pi
3 ji m 1

2
Pj
1m
1

4 j 2
mO i
x
F /2

K K (1)
K (2)

The corresponding positions of every element stiffness submatrix in the global


stiffness matrix are as follows.
4.2 Examples
 The finite element analysis of cantilever plane problems
(3)Establish the global stiffness matrix
Then we can obtain the global stiffness equation
4.2 Examples
 The finite element analysis of cantilever plane problems
(4)The treatment of the boundary conditions and the solution of the equation
In this problem the displacement boundary conditions are u3=0、 v3=0、 u4=0、
v4=0. Substitute them into the global equation. Delete the 5~8 lines of the node
displacement that are already known.
2m y F /2
Pi
3 ji m 1

2
Pj
1m
1

4 j 2
mO i
x
F /2

Why delete the 5~8


lines?
4.2 Examples
 The finite element analysis of cantilever plane problems
2m y F /2
(5)Calculation of the bearing reaction. Pi
3 ji m 1
The node displacements are 2
F
u1 v2   1.88 8.99 1.50 8.42
T T Pj
v1 u2 1m
Et 1

Substitute it into the global stiffness equation 4


mO i
j 2

x
and we can get the bearing reactions that F /2
9 Et  2 4 
R3 x  
 1u  v1  v2   2 F
32  3 3 
9 Et  2 1 4 
R3 y    u1  v1  u2   0.07 F
32  3 3 3 
9 Et  2 
R3 x  
 2u  v2   2 F
32  3 
9 Et  2 1 
R3 x    u 2  v2   1.07 F
32  3 3 
So the bearing reactions above constitute a balanced system of force with the
external force.
4.3 Conclusion of the solving process
Not let’s review each step of solving the problems with the finite element
method.
Take the plane problems with constant body force (e.g. the gravity) and linear
surface force for example. And further embody the method from the aspect of
programming.
 Divide the area into triangle elements

The size of the element is arbitrary to some extent. However one should take
the computer capacity, speed and the computational accuracy requirement
into consideration.
So the following points should be obeyed
1. 1.The node must be the vertex of adjacent elements but not the inner point
on the boundary.
2. Avoid large obtuse angle.
3. The meshes should be intensive where the gradient of u(x,y) varies
intensely;and sparse where the gradient varies slightly.
4. The number of the elements can be arbitrary but the quality can largely
influences the band width of the global stiffness matrix. The principle is that
the smaller the maximum of the absolute value of the differences between
adjacent element numbers, the better.
4.3 Conclusion of the solving process
 Give the raw information according to the meshes
1. Coordinates of the nodes:
(xi,yi) i=1,……,NP
2. Information of the element:
Provide every element with a three-node number(n1,n2,n3) in an
anticlockwise order and also with the number of the material types.
3. Characteristics of the material :
Give the elastic constants and the density of the constant body force of
every material type.
4. Information of the concentrated force:
The node number and the magnitude of the concentrated force.
5. Information of the surface force:
Give the numbers of two ends and the load density of every line element
under the surface force.
6. Information of the constraints:
The node number of the constraints and the constraint situations in the
directions of x and y.
4.3 Conclusion of the solving process
 Establish the solving equation
1. Zero setting:
Kij=0,Fi =0 (i=1,……,NP)
2.Element analysis:
Provide the element information including the element numbering, the
node numbering, the coordinates, material and the body force. Calculate
the element stiffness matrix and the equivalent node load vector.
Scanning the elements one by one and then assemble the element stiffness
to get the global stiffness matrix.
One can only calculate a half of the global matrix due to its symmetry.
And even the elements outside the half-band width can be ignored.
3. Line element analysis:
Calculate surface force to get its equivalent node load vector.
4. Concentrated force:
Extract the node number and the value of the concentrated force one by one
and then accumulate directly.
We can then obtain the global stiffness K and the global load vector P.
4.3 Conclusion of the solving process
 Treatment of the constraints
The geometric boundary conditions are needed to constrain the displacement
of the rigid body system in the displacement field problem. to eliminate
the singularity of the global stiffness matrix.
We usually adopt the following two methods: “change the diagonal elements
to 1” and “multiply the diagonal elements by large number” .
 Solve the linear equations
It should be noticed that solving the linear algebraic equations requires a
lot of calculation.
We should make full use of the symmetry, the sparsity and the banded
distribution of the stiffness matrix and there has already developed
several efficient methods.
 Solve the stress
We can then obtain the stress ad strain after getting the displacement  un1 
solution. The stress in every element is constant for the line elements v 
  2G  0   an1 0 an 2 0 an 3 0   n1 
here, which is 1 
σ  Dε  D    N  e      2G  u n 2 
0   0 bn1 0 bn 2 0 bn3   
2 e  v
 0 0 G   bn1 an1 bn 2 an 2 bn 3 an 3   n 2 
 
So the discontinuity of the stress between elements can be solved by later  un3 
 vn 3 
process.
4.4 Examples

Example1
A homogeneous square sheet with the thickness t=1cm has an
uniform tension q=106N/m at both top and bottom sides. The
elasticity modulus E=1000, the Poisson’s ratio   1/ 3 . Ignore the
self weight and solve its stress components by the finite element
method.

q
y q
y 2 2
4 3
2m 2
x
1
x
1 2
q=106N/m
1 2
4.4 Examples

1. Establish the mechanical model


Solve:
In this example the length and width of the structure are considerably
larger than the thickness. The load is in the plane of the sheet and is
uniformly distributed along the thickness. So this is a plane stress
problem and we can extract ¼ of the structure due to the symmetry.
2. Structure discretization node
The ¼ of the structure can be discretized into coordinate1 2 3 4
two triangle elements. The node number, the x 0 1 1 0
element division and the coordinate are given in the
y 0 0 1 1
picture2. The coordinate figure of every node is
given in table1. Table 1
3. Solve the element stiffness matrix
1) Calculate the difference between the node coordinates and the area of
the elements.
Element1(i、j、m 1,2,3)
b1  y2  y3  1 b2  y3  y1  1 b3  y1  y2  0
4.4 Examples
c1  x2  x3   0 c2  x3  x1   1 c3  x1  x2   1

1 
1
2
 2

b2c3  b3c2  11  0   1 
1 1
2
2) Solve the element stiffness matrix
Calculate the constants that will be used first.
1  1 Et 9E E 9E
  
2 3 4(1   ) 16
2
2(1   )
2
8
Then
 
9E 
 1   1 
1
 0  0
1
  1  0 
1
 0   1
K11    1
1 3 3 3

16   0   1  1   1 0 0  0    1  1 
1
3 3 3 
9 E 1 0 
  1
16 0
 3 
4.4 Examples
 1   1  4 2
9 E  1   0  
3 K 1  9 E  3
 
K12    1
1 3
1
K13    1
1 9 E
 22  2
3
4
16   ; 16  0 ; 16   
3 3  3   3 3
 1 1
9 E  3  1  1 2 3
K 23    1
1 3 K 1 

9 E
3 0
16  0 1
33
16   1 ;   1 1
 3 
1 0  1 3 0
3
So the stiffness matrix of element 1 is  1 1  1
1 1
   0
K 11 1 K 12 1 K 13 1   3 3 3 3 
9E 
 4 2 1 1
1  
K 66  K 21  K 22  K 23    
1 1 1
3 3 3 3 2
K 1 K 1 K 1  16  4 1 
 31 32 33   对  1
 3 3 
 称
1
0
 3  3
 1 
4.4 Examples
The stiffness matrix will be just the same when the order of
node number 341 of element2 corresponds to 123 of element 1.
3 4 1

 1 1
 1 0  1 0
3 3
 1 1 1 1  3
   0
 3 3 3 3 
9E 
4 2 1 1
 
K 626   3 3 3 3
16  4 1  4
 对  1
 3 3 
 称
1
0
 3 
 1 
1
4.4 Examples
4. Assemble the global stiffness matrix
The global stiffness matrix is
K11 1 2 
 
K 88 
  211 2
K 1
K 22 1 
K  K32 1 K33 12 
 31 2 2
 K 41  K 43 2 K 44  
Because [Krs]=[Ksr]T and the order of node number 123 of
element 1 corresponds to 341 of element2. Then we have
3E 3 0
K11  1
 K 33 
2

16 0 1

K 21  1 2

 K 43   K 12  
1 T

3E   3 1 
16  1  1

K 31  1 2

 K 13   K 13  
1 T

3E  0  1
16  1 0 
4.4 Examples
3E  4  2
K 22 
1
 K 44 
2

16  2 4 

K 32 
1 2

 K 14   K 23  
1 T

3E   1 1 
16  1  3

3 E 1 0 
K 33 
1
 K 11 
2

16 0 3

K 31 
2 1

 K 13   K 13  
2 T

3E  0  1
16  1 0 

K 41 
2 1

 K 23   K 14  
2 T

3E   1 1 
16  1  3
4.4 Examples

So the global stiffness matrix is

4 
0 4 对 
 
 3 1 4 称 
 
1  1  2 4
K 88 
3E  
16  0  2  1 0 4 
 
 2 0 1 3 0 4 
1 1 0 0 3 1 4 
 
 1  3 0 0 1  1  2 4
4.4 Examples
6. Introduce the constraint conditions and modify the stiffness matrix
Substitute the constraint conditions: u1 =v1=0;v2=0;u4=0 and the column
matrix of equivalent node forces F  0 0 0 0 0 q / 2 0 q / 2T into the
stiffness equation K    F. Delete the line and column 1,2,4,7 in [K]
according to the displacement 0 then the stiffness equation will become
4  u2   0 
  u   0 
3E   1 4  3  
    
16  1 0 4  v
 3  q / 2 
   
 0 1  1 4 v4  q / 2 
So the displacement at nodes can be obtained
u2 u3 v3 v4    q / 3E  q / 3E q / E q / E
T T

Thus    q / E0 0 1/ 3 0 1/ 3 1 0 1


T
4.4 Examples

7. Calculate the element stress matrix and solve the element stress
Solve the element stress matrix [S]1、[S]2 and then solve the
stress components of every element.
 0 
 0 
 x   3 0 3  1 0 1    0  0
  
 1   y   S 1 1    1 0 1  3 0 3      
3 E q / 3 E 3q
 8 / 3  q 1
  8
      0  8  0  0
 xy   0 1 1 1 1 0   q / 3 E     
 
 q / E 

 q / 3 E 
 q/E 
 x  3 0  3 1 0  1    0  0
  3E   
  0   3E 8q /(3E )  q 1
 2   y   S      
2 2

8 
1 0 1 3 0 3 q / E  8    
  0 1 1  1  1 0     0  0
 xy   0     
 
 0 

The element stress can be regarded as the stress value at the centroid.
4.4 Examples
Ex.2
The discretized structure of a plane stress problem is shown in the Fig. 3.
(a) shows the discretized global structure. (b) shows the structure of
element 1, 2, 4. (c) shows the structure of element 3. Please try to solve
the displacement at the nodes, the element stress and the element strain by
the finite element method. (The Poisson ratio   0, the element
thickness t=1).
y Yi vi Yi vi Ym vm
p i a m Xm
Xi Xi
1 ui ui i um
3
1 a a
2 3
uj 1,2,4
3
2 4 x Xj Xm Xj
5 6 m um j
4 j a uj
Yj v j Ym vm Yj v j

Fig.3
4.4 Examples
Solution:
First determine the coefficients bi , b j , bm , ci , c j , cm that will
be needed when solving the element stiffness and the area A.
For the elements 1,2 and 4 we have
bi  0, b j   a, bm  a
ci  a, c j   a, cm  0
A  a2 / 2
For element 3 we have:
bi  a, b j  0, bm  a
ci  0, c j  a, cm  a
A  a2 / 2
4.4 Examples

Then solve the element stiffness matrix.


The stiffness matrix of elements 1, 2 and 4 is
i j m

1 0 1 1 0 1
0  i
 2 0  2 0 0 
1, 2 , 4  E  1 0 3 1  2  1
k    j
4  1  2 1 3 0  1
0 0 2 0 2 0
  m
 1 0 1 1 0 1 
4.4 Examples

The stiffness matrix for element 3 is


i j m

2 0 0 0 2 0 
0  i
 1 1 0  1  1 
3  E 0 1 1 0  1  1
k    j
40 0 0 2 0  2
 2  1  1 0 3 1
  m
 0  1  1  2 1 3 

Relationship between the node numbers of every element and


the global numbers is shown in the table 2.
4.4 Examples

Table 2

Corresponding relation table between element node numbers and the global node number

单元号 1 2 3 4

节点号 节 点 总 编 号
I 1 2 2 3
j 2 4 5 5
m 3 5 3 6
4.4 Examples

Extend the element stiffness matrix to a 12*12 matrix according to


the total number of the nodes and the corresponding relationship
of the node number.
1 2 3 4 5 6
1 0 1 1 0 1 0 0 0 0 0 0
0 2 0 2 0 0 0 0 0 0 0 0
1

 1 0 3 1 2  1 0 0 0 0 0 0
  2
 1  2 1 3 0  1 0 0 0 0 0 0
0 0 2 0 2 0 0 0 0 0 0 0
  3
1 1
 
1 E 1 0 0 1 0 0 0 0 0 0
k12
0 0 0 0 0 0 0
12
4 0 0 0 0 0
  4
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
  5
0 0 0 0 0 0 0 0 0 0 0 0
 
0 0 0 0 0 0 0 0 0 0 0 0
6
 0 0 0 0 0 0 0 0 0 0 0 0
4.4 Examples
1 2 3 4 5 6
0 0 0 0 0 0 0 0 00 0 0
0 0 0 0 0 0 0 0 0 0 0 0
1

0 0 1 0 0 0 1 1 0 1 0 0
  2
0 0 0 2 0 0 0 2 0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
  3
E 0 0 0 0 0 0 0 0 0 0 0 0
k122 12  
4 0 0 1 0 0 0 3 1  2  1 0 0
  4
0 0 1  2 0 0 1 3 0  1 0 0
0 0 0 0 0 0 2 0 2 0 0 0
  5
0 0 1 0 0 0 1 1 0 1 0 0
 
0 0 0 0 0 0 0 0 0 0 0 0
6
0 0 0 0 0 0 0 0 0 0 0 0
4.4 Examples
1 2 3 4 5 6
0 0 0 0 0 0 0 0
0 0 0 0
0 0 0 0 0 0 0 0 0 0 0 0
1

0 0 2 0 2 0 0 0 0 0 0 0
  2
0 0 0 1 1 1 0 0 1 0 0 0
0 0  2 1 3 1 0 0  1 0 0 0
  3
E 0 0 0 1 1 3 0 0  1  2 0 0
k123 12  
4 0 0 0 0 0 0 0 0 0 0 0 0
  4
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 1 1 1 0 0 1 0 0 0
  5
0 0 1 0 0 2 0 0 0 2 0 0
 
0 0 0 0 0 0 0 0 0 0 0 0
6
0 0 0 0 0 0 0 0 0 0 0 0
4.4 Examples
1 2 3 4 5 6

0 0 0 0 0 0 0 0 0 00 0
0 0 0 0 0 0 0 0 0 0 0 0 
1

0 0 0 0 0 0 0 0 0 0 0 0
  2
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 1 0 0 0 1 1 0 1
  3
E 0 0 0 0 0 2 0 0 0 2 0 0
k1212  
4

4 0 0 0 0 0 0 0 0 0 0 0 0
  4
0 0 0 0 0 0 0 0 0 0 0 0
0 0 0 0 1 0 0 0 3 1  2  1
  5
0 0 0 0 1  2 0 0 1 3 0  1
 
0 0 0 0 0 0 0 0 2 0 2 0
6
0 0 0 0 1 0 0 0 1 1 0 1 
4.4 Examples

Add the extended element stiffness matrixes together and the


global stiffness matrix K can be obtained.
1 2 3 4 5 6

1 0 1 1 0 1 0 0 0 0 0 0
0 2 0  2 0 0 0 0 0 0 0 0  1
 
 1 0 6 1  4 1 1 1 0 1 0 0
  2
  1  2 1 6  1  2 0  2 1 0 0 0 
0 0  4 1 6 1 0 0  2 1 0 1
  3
4
E1 0 1  2 1 6 0 0 1  4 0 0
K   k12e12  
e 1 4 0 0 1 0 0 0 3 1  2 1 0 0
  4
0 0 1  2 0 0 1 3 0 1 0 0
0 0 0 1  2 1  2 0 6 1  2  1
  5
0 0 1 0 1  4 1 1 1 6 0  1
 
 0 0 0 0 0 0 0 0  2 0 2 0  6
 0 0 0 0 1 0 0 0 1 1 0 1 
4.4 Examples

So the general equation of the structure is:

R  K
Where
R  0  P 0 0 0 0 0 0 0 0 0 0
T

  u1 v1 u2 v2 u3 v3 u4 v4 u5 v5 u6 v6  T

Considering the boundary conditions:


u1  u2  u3  v4  v5  v6  0
4.4 Examples

Eliminate the singularity through multiplying the diagonal elements


by large number and the structural general equation is
1 1015 0 1 1 0 1 0 0 0 0 0 0  u1   0 
    
 0 2 0 2 0 0 0 0 0 0 0 0   v1   P 
 1 0 6  1015 1 4 1 1 1 0 1 0 0  u 2   0 
    
 1 2 1 6 1  2 0 2 1 0 0 0   v2   0 
 0 0 4 1 6 1 0 0 2 1 0 1  u3   0 
    
E 1 0 1 2 1 6 0 0 1 4 0 0   v3   0 
  
4 0 1 3  1015 2 1  u 0
 4   
0 0 0 0 1 0 0

 0 0 1 2 0 0 1 3  1015 0 1 0 0   v4   0 
    
 0 0 0 1 2 1 2 0 6 1 2  1  u5   0 
 0 0 1 0 1  4 1 1 1 6  1015 0  1   v5   0 
    
 0 0 0 0 0 0 0 0 2 0 2 0  u6   0 
 0
 0 0 0 1 0 0 0 1 1 0 1 1015  v6   0 
4.4 Examples

So the displacement at the nodes is:


 u1   0 
v   3.252 
 1  
u 2   0 
   
v
 2   1 . 252 
u3   0.088 
   
 v3  P  0.374 
    
u
 4 E  0 
 v4   0 
   
u5   0.176 
v   0 
  5
 
u6   0.176 
v   0 
 6  
4.4 Examples

Substitute the node displacement into the equation and we can et


the element stress and strain.

Element 1: u1 
v 
x  0 0  a 0 a 0     0.088 
1

1   1   u2  P  
    y   2 0 a 0  a 0 0       2.000 
  a v
 2 Ea  0.880 
  
 a 0  a  a 0 a 
  
xy
u3 
 
 v3 

 x  1 0 0    x   0.088 
1       P 
   y   E 0 1 0    y    2.000 
      a  0.440 
 xy   0 0 0.5  xy   
4.4 Examples

Element 2:
u2 
v 
x  0 0  a 0 a 0     0.176 
2

2    1   u4  P  
    y   2 0 a 0  a 0 0       1.252 
  a v
 4 Ea  0 
 xy  
 a 0  a  a 0 a 
 u   
 
5

 v5 

 x  1 0 0    x   0.176 
    P 
 2    y   E 0 1 0    y    1.252 
    a 
 
xy

 0 0 0.5
 
 xy  0 
4.4 Examples

Element 3:
u2 
v 
x   a 0 0 0 a 0  2   0.088 
3    1   u5  P  
   y   2  0 0 0  a 0 a     0.374 
  a v
 5 Ea  0.614 
 xy  
 0  a  a 0 a a 
 u   
 
3

 v3 

 x  1 0 0    x   0.088 
    P 
 3   y   E 0 1 0    y    0.374 
    a 
 
xy

 0 0 0.5
 
 xy  0.307 
4.4 Examples

Element 4:
u3 
v 
x  0 0  a 0 a 0     0 
3

4    1   u5  P  
    y   2 0 a 0  a 0 0       0.374 
  a v
 5 Ea  0.264 
 xy  
 a 0  a  a 0 a 
 u   
 
6

v6 

 x  1 0 0    x   0 
    P 
 4    y   E 0 1 0    y    0.374 
    a 
 
xy

 0 0 0.5
 
 xy   0.132 
V. Further discussions
We’ve already obtained basic understanding of the fundamental
principles and equations of the finite element method, which
provide necessary preparations for the applications to the linear
elastic analysis of the typical continuum system.

The finite element method is mainly characterized by the explicit


physical meanings of its variables and the matrix, which is quite
necessary for us to further understand and master the essence of
finite element method.

On the other hand solving the complicated problems aims to get


highly accurate result. But high accuracy will undoubtedly
increase the computational complexity sharply.

That’s why both the precision and efficiency should be taken


into consideration and the preliminary theory of error should be
introduced.
5.1 Characteristics of element stiffness matrix
Now let’s first discuss the properties of the element stiffness matrix. We can
establish a governing equation of an element by the principle of minimum
potential energy. e e
K a P
e

For the linear triangle element ae and Pe can be represented respectively


a e  [ui vi u j v j um vm ]T  [a1 a2 a6 ]T
P e  [ Pix Piy Pjx Pjy Pmx Pmy ]T  [ P1 P2 P6 ]T
So the equation can be expanded as follows
 K11 K12 K16   a1   P1 
K K 22 K 26   a2   P2 
 21
   a3   P3 
     
   a4   P4 
   a5   P5 
    
 K 61 K 62 K 66   a6   P6 
5.1 Characteristics of element stiffness matrix
The equation above is the equilibrium equation of the element nodes: There is
an equilibrium equation on each direction of x and y for every node. So there
are totally 6 equilibrium equations for 3 nodes.
The left part of the equation expresses the internal force of the element nodes
represented by the node displacement. The right part expresses the equivalent
element nodal force of the external load.
Let a1=1(ui=1),a2=a3=…=a6=0. According to the equilibrium equation we
have  K11   P1 
K  P 
 21    2 
   
   
 61  a1 1  P6 
K
The equation indicates the physical meaning of the first column:
The first column elements mean the magnitude of the node force applied in
the direction of displacement when a1=1 and the displacement of other nodes
equals zero.
We can compare it to the stiffness concept in the displacement method of
structural mechanics.
5.1 Characteristics of element stiffness matrix
The elements are obviously in an equilibrium state under the forces of nodes.
So the sum of the forces in the direction of x and y should equals zero.
x : K11  K 31  K51  0
y : K 21  K 41  K 61  0

The physical meanings of other column elements of the element stiffness


matrix can be obtained in the same way.
So the physical meaning of an arbitrary element Kij in the element stiffness
matrix is:
the magnitude of the node force that should be applied to the node i when the
node j of the element has a unit displacement and the displacements of other
nodes equal zero.
If the stiffness of the element is high, a big node force will be needed for a unit
displacement of the node.
So every element in the element stiffness matrix represent the magnitude of the
element stiffness that is the stiffness coefficient.
5.1 Characteristics of element stiffness matrix

The characteristics of the element stiffness matrix can be concluded as follows


 Symmetry
The symmetry of the matrix has already been discussed in the weighted
residual method and the Galerkin method of variation principle.
The essence is the same as for the finite element method based on the
principle of minimum potential energy.

 Singularity
Similar to the physical meaning of the element stiffness when aj=1(j=1,2,…,6)
and the displacements of other nodes equal zero we have
K1 j  K 3 j  K 5 j  K j1  K j 3  K j 5  0
K2 j  K4 j  K6 j  K j 2  K j 4  K j 6  0

That is to say only three lines (or columns) of the 6×6 stiffness matrix of the
linear triangle element are independent, so the matrix is singular.
5.1 Characteristics of element stiffness matrix
 Sparsity and band distribution
When a continuous body is discretised into finite elements, it’s obvious that
there are only a few related elements around a node, which is called the
support region of the shape function of this node.
So a node is related to only a few nodes around it through the adjacent
elements. That’s why despite the large number of elements and nodes or the
high degree of the global stiffness matrix, the non-zero elements are very few.
And this is the sparsity of the stiffness matrix.

If the nodes are numbered reasonably, these non-zero elements will centralize
in a banded region. And this is the band distribution of the stiffness matrix.

 In conclusion, the coefficient matrix of the finite element equation posses the
following characteristics: symmetry, singularity, sparsity and band
distribution of the non-zero elements.
When solving the large-scale equations, one should pay attention to these
characteristics for a higher efficiency. The singularity which will vanish after
the introducing of boundary conditions.
5.2 Requirements of convergence
 The convergence criteria of finite element method
We’ve known that the finite element method can be regarded as a special form
of the Ritz method when solving the differential equation. What’s different is
that the trial function of the finite element method is defined in the element
(subdomain) but not the global domain.

So we can compare the convergence requirement of the finite element method


to that of the Ritz method.
The convergence requirement of the Ritz method is: the trial function must be
complete and continuous. And the approximate solution of the Ritz method must
converge to the exact solution when n increases.
Then what about the convergence requirement of the finite element method?

As for the finite element method the global functional of the field function is
summed by the element functional. If the interpolation function of the element
( the trial function) is a complete polynomial, then the solution can be exact in a
finite element.
However the trial function of the finite element method can only have finite
terms, so the solution is an approximate one.
5.2 Requirements of convergence
 The convergence criteria of finite element method
So when the solution of finite element method will converge to an exact one
as the size of the element tends to zero.

Take an undetermined scalar field function for example. The differential


equation is
L( )  b  0
The corresponding functional is
1 
    C ( )C ( )   b  d 

 2

Suppose the functional П contains φ and its derivative with a degree up to m,
then the approximate solution must be at least an m-order polynomial.
If the trial function is a complete p-order polynomial, it must satisfy p≥m,
and the trial function is
  0  1 x   2 x 2    p x p

d m p!
 m !  m  (m  1)!  m1    p x p m
. dx m
( p  m)!
5.2 Requirements of convergence
 The convergence criteria of finite element method
It can be seen that φ is a complete p-order polynomial. So all its derivatives
with a degree up to m contain the constant term.
When the size of an element tends to zero, the derivative with a degree up to
m will tend to a constant. So the functional of every element will probably
converge to its accurate solution.
If the trial function also satisfies the continuity requirement, the functional of
the global system will converge to its exact value. So the solution of the finite
element method converges to the accurate value. That is to say it’s
convergent.
So the requirements of convergence are as follows.
1. Requirement of completeness
If the highest derivative degree of the functional field function is m, then one
of the conditions that the solution converges is that the trial function of the
element field function is at least a complete m-degree polynomial.
In other words the trial function must contain itself and the terms of which the
derivatives with a degree up to m are constants.
The element is complete if its interpolation function satisfies the requirements
above.
5.2 Requirements of convergence
 The convergence criteria of finite element method
When the trial function is a polynomial, the function inside the element
satisfies the requirement of continuous obviously.

The element is Cm-1 order continuous if the trial function is a mth-order


polynomial. So we should pay attention to the continuity on the interface
between the elements. So here comes another convergence criteria.

2. Requirements of compatibility
If the highest order of the functional derivative is m, the trial function on the
interface between the elements must have Cm-1 order continuity. That is to say
the function on the interface must have the continuous derivative with a
degree up to m-1.
If the interpolation function of the element satisfies the requirement above,
the element is compatible.
When the element is both complete and compatible, the solution of the finite
element method converges.
So the solution converges to the exact value when the size of the element
tends to zero.
5.2 Requirements of convergence
 The convergence criteria of finite element method
It should be pointed out that
There are usually many kinds of errors of the finite element solution. So the
meaning of “converge to the exact value” is that the discretization error tends
to zero.

The discretization error means:


the error that emerges when we use the element trial function to approximate
the field function of the global domain as a continuous solving domain is
discretize into infinite subdomains (elements).

Another main error of the finite element method is caused by the limited
significant digits of computer including:
1.Rounding error:Randomly caused by rounding off.
2.Truncation error:Caused by the truncation of the actual digits due to the
limited significant digits of computer.
5.2 Requirements of convergence
The physical meaning of the convergence criteria
Take the plane problem for example.
The functional
1 T
 ep   ε Dε dV   uT f dV   e uT T dV
e 2 e S

contains the first order derivative of displacement u and v, so m=1.


The convergence criteria 1 requires that the interpolation function or the
displacement function is at least the first order complete polynomial of x,y.
Displacement and the terms where the first order derivative is constant
represent the corresponding displacement mode to the rigid body
displacement and the state of constant strain.
The element deformation actually contains the rigid body displacement and
the element strain also tends to be constant strain when the size is becoming
infinitely small.
So the completeness requirement means that the finite element solution
consisting of the interpolation function must reflect the rigid body
displacement and the constant strain state of the element.
5.2 Requirements of convergence
 The physical meaning of the convergence criteria
The convergence criteria 2 requires that the displacement function must
has C0 order continuity.
That is to say that the zero order derivative of the displacement function is
continuous or it is continuous on the interface between the elements.
It is obvious that the interpolation function of a linear triangle element
satisfies both the completeness requirement and the coordination requirement.
So the solution of this kind of element can converge.
We should notice that:
The functional usually contains first order derivative for the two-dimensional
or three-dimensional elastic mechanics problems. The approximate
displacement function is required to be C0 order continuous, which means it
only requires the function itself be continuous on the interface between the
elements.
However if the derivative in the functional is higher than first order ( the
derivative is second order for the problems of beam, plate and shell), then it
requires the trial function has a first order or higher than first order
continuous derivate on the interface between the elements, which means it
has a continuity of C1 or higher.
5.2 Requirements of convergence
 Patch test and the compatibility of the element
There is a very simple method to decide whether a new displacement mode of the
element satisfies the convergence requirement, which is called the patch test.
The steps are as follows:
Assemble a small piece of elastic body by several elements. And make sure that at
least one internal node is completely surrounded by the elements. The internal node
can be changed to get an arbitrary shape of the element.
Apply the displacement corresponding to the state of constant strain or an external
force to the nodes on the boundary. The internal node has neither the application of
the external force nor the displacement constraints.

If the element displacement, stress and strain agree well with the given constant strain
state of the elastic body, then the element has passed the patch test.
This method is first proposed by B.M.Irons and has already been proved to be the
sufficient condition of the convergence.
5.2 Requirements of convergence
 Patch test and the compatibility of the element
Sometimes the compatibility requirement can be relaxed. If the element
passes the patch test, then the finite element solution can still converge to the
right answer.
This kind of element can be called the non-compatible element.
Some non-compatible elements even possess better accuracy than the
compatible elements. But they are regarded as “the illegal element” in the
past because they fail to satisfy the compatibile condition.
Why some of the non-compatible elements have better accuracy than the
compatible ones?
It is because:
The structure of the finite element obtained by the assumed displacement
field is usually more rigid than the practical one. However the non-
compatible elements permit the separation or overlap between the elements,
which makes the structure of the finite element become more flexible.
These two opposite effects can offset each other and sometimes we can get
better results.
5.3 Properties of the finite element solution
 Estimation of the solving accuracy
Still take the plane problem for example to study the solving accuracy and
convergence rate of the elements.
The displacement field of the element can be expanded into the following
series.
 u   u 
u  ui    x    y 
 x i  x i

If the size of the element equals h, then Δx and Δy in the equation above are
the magnitude order of h.
If the displacement function of the element adopts the p-order complete
polynomial or it can approximate the first p order polynomial of the Taylor
series mentioned above, then the error of the displacement solution u will be
O(hp+1) order.
In terms of the triangle element with 3 nodes, because the interpolation
function is linear where p=1, then the error of u is approximately O(h2) order.
And we can also predict that the rate of convergence is also O(h2) order.
5.3 Properties of the finite element solution
 Estimation of the solving accuracy
So if we halve the size of all the elements, the error u will become (1/2)2=1/4
of the previous error.

The deduction is also applicable for the error estimation of strain, stress and
the strain energy as well as the rate estimation of convergence.

For example when the strain is based on the m order derivative of


displacement, its error will be O(hp-m+1) order. When we adopt the plane
triangle linear element, we have p=m=1. So the strain error is estimated to be
O(h) order.

If the strain energy is expressed by the quadratic component of strain, the


error is estimated to be O(h2(p-m+1 )) order. For example the error of strain
energy is O(h2) order as for the plane triangle linear element.
5.3 Properties of the finite element solution
 Estimation of the solving accuracy
The finite element solution of the coordinate elements that satisfies the
completeness and compatibility requirements is monotonously convergent
when the size of the element h→0. So we can extrapolate the result obtained
from twice meshing to improve the accuracy.
(1)
For the plane triangle linear element, if the result after first meshing is ui ,
then further halve the element size we can get the result ui(2).
Because u (1)  u O(h 2 )
i i

 ui O ( h / 2  )
(2) 2
u i

Then we can get better approximate solution by extrapolating

ui   4ui(2)  ui(1) 
1
3
It should be noticed that the error above is only limited to the discretization
error of meshing.
As discussed before the practical errors should contain the error of
numeric calculation.
5.3 Properties of the finite element solution
 The lower bound of the displacement element solution
The displacement element is the finite element based on the principle of
minimum potential energy whose basic unknown quantity is displacement.
We can get the properties of the lower bound of the deviation between the
approximate and accurate solution .
The discretization form of the total potential energy of the system.
1
  aT Ka  aT P
2
We can get the solving equation of the finite element method according to
δП=0
Ka  P
Substitute it into the expression of the total potential energy
1 T 1 T
  a Ka  a Ka   a Ka  U
T

2 2
The equation above indicates that the total potential energy of the system
equals to the opposite number of strain energy in the equilibrium state.
5.3 Properties of the finite element solution
 The lower bound of displacement element solution
So
  min  U  Umax
The assumed approximate displacement mode is usually
different from the accurate solution in the finite element method,
so the total potential energy of the system we get will be bigger
than the real one.

Denote the total potential energy, total strain energy, the


stiffness matrix and the node displacement of the finite element
method as follows
, U , K , a
The corresponding accurate solutions are
, U , K , a
5.3 Properties of the finite element solution
 The lower bound of the displacement element solution
Because
  U    U  U  U
which means
aT Ka  aT Ka
For the accurate solution we have
Ka  P
For the approximate solution we have
Ka  P
Thus
aT P  aT P
5.3 Properties of the finite element solution
 The lower bound of the displacement element solution
Because aT P  aT P
It can be seen that the reason why the approximate strain energy is smaller
than the accurate is that the displacement of the approximate solution is
smaller than that of the accurate one overall.
So the displacement solution obtained by the displacement element is not
bigger than the accurate solution overall. So the solution has a lower bound.
The explanation of the lower bound is
Every element is originally one part of the continuous body with infinite
degrees of freedom.
After the displacement function of the element is assumed, the degree of
freedom will become finite expressed by the node displacement. The
displacement function constrains and limits the element deformation so the
element stiffness is stronger than that of the practical continuous body. Thus
the global stiffness matrix increases and the discrete stiffness matrix is larger
than the practical one. So the approximate displacement solution is smaller
than the accurate one oveal.
5.4 Conclusions
We discussed the basic principles and the equations of the finite element method
through the examples of elastic static problems. The main points are as follows
1. Establish the theoretical basis of finite element method and select the type of
element
The principle of minimum potential energy and the displacement element whose basic
unknown quantities are the node displacements are most commonly used. Almost all
the current general programs of finite element analysis take the displacement element
as the main form of element.
2. Establish the FEM model
Mainly include the choose of element form, meshing and the setting of boundary
conditions, which requires a basic understanding here.
3. Establish the characteristic matrix and the solving equation of finite element
method
The element characteristic matrix includes the interpolation function matrix, the strain
matrix and finally we should establish the element stiffness matrix and the load vector
to obtain the governing equation of finite element method.
4. Choose the numerical method and solve the equation of finite element method
We’ll discuss this in more details later on.
5. About the convergence of finite element solution
Provide the requirements for completeness that must contain the rigid body
displacement and the displacement mode of constant strain. Meanwhile the element
boundary must satisfy the requirement for continuity compatability.
5.4 Conclusions

After establishing the basis of finite element method we’ll discuss how to set up
elements and the corresponding element analysis in the following chapter without any
complete description of solving process. The establishment of element shape function is
one of the most crucial and skillful steps.

One of the most important characteristics of FEM is to adopt the interpolation


function as the displacement mode.
The element nodes we discussed before are all the corner nodes of the element. And
as we use the linear interpolation the computational accuracy is not very good. One of the
effective methods to improve the FEM accuracy is to increase the accuracy of
interpolation ( high order interpolation)

Generally speaking when use the finite element method to solve problems we should
first consider and determine the element form and the corresponding interpolation and
then carry on the rest steps according to the routine process.

Here we recommend two mathematical software for more efficient calculation


 Matlab:numerical calculation expertise
 Mathematica:symbolic computation expertise
谢谢!

You might also like