5 FEM Formulation
5 FEM Formulation
1 Summary
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 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
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.
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 ()
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) .
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
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.
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
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).
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
( 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
(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*.
( B B)dxdy k
* T T * T
e e e e e
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
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
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
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
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:
(36) (61) (31) (36) (66) (61) (31) (36) (66) (63) (31) (31)
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 (31)
(312) (121)
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
ae 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 ae
then the strain and the stress can be expressed as
ε u N ae B ae
(61) (31) (612)
(63) (63) (312) (121) (121)
σ D ε D B ae
(61) (66) (61) (66) (612)
(121)
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 ae D B ae dV N ae f dV e N ae T dS
e
e 2 e S
T
ae
e
1 T
e 2 B DB dV a
e
e
a
T
e e N T
f dV a T
e e
Se
NT T dS
Let
Ke BT D B dV P fe NT f dV
(1212) e (126) (66) (612) e (123) (31)
(121)
PSe e NT T dS P e P fe PSe
S (123) (31)
(121)
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
ae K e ae ae P fe ae 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
(33) (33) (33) (33)
Ke Ke Ke K e
Ke
BT DB dV e
3) (3 3)
(1212) e 2
K mi K mj K mm K ml
e e e
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
x x
u
Displacement u Strain ε y Stress σ
y
v xy
xy
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 [ ] (2u1)
(31) y x v x, y (32)
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
(31) (32) (21) (32) (26) (61) (36) (61)
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
(66) e (63) (33) (36) A
t f dA t
T T
N N T dl
e (62) (21) S (62) (21)
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
(66) (61) (61)
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)
2
Pj
1m
1
4 j 2
mO i
x
F /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
1
2
2
b2c3 b3c2 11 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 66 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 626 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 88
211 2
K 1
K 22 1
K K32 1 K33 12
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
4
0 4 对
3 1 4 称
1 1 2 4
K 88
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 / 2T 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
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
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
2 0 0 0 2 0
0 i
1 1 0 1 1
3 E 0 1 1 0 1 1
k j
40 0 0 2 0 2
2 1 1 0 3 1
m
0 1 1 2 1 3
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
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
k1212
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
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
E1 0 1 2 1 6 0 0 1 4 0 0
K k12e12
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
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
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.
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.
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.
d m p!
m ! m (m 1)! m1 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.
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.
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
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.
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.
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.
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.