EM 424: Finite Elements

Principle of Virtual Work and the Finite Element method

Consider a uniaxial load problem where a bar of variable cross sectional area carries a distributed body force, f x , along its length as well as several discrete loads, Pa , Pb as shown The bar is fixed at its left end and we will let Ua , Ub be the displacement at the applied loads Pa , Pb , respectively. Then the principle of virtual work (or minimum potential energy) says that

= xxe xx dV f xux dV PaUa PbUb = 0

(1)

Ua

Ub

Pa
Fig.1

P fx

Now, suppose we break this bar into M elements and let lm be the length of the mth element, Um , Um +1 ( m = 1,..., M + 1) be the end displacements of the mth element (called nodal displacements) , and Fm , Fm +1 (m = 1,..., M + 1) be the nodal forces acting at the ends (nodes) of each element. For the problem shown above the force Pb at the right end acts on the (M+1)th node and we will assume that the elements are arranged so that the other force Pa acts at the pth node which has the nodal displacement Up . For this problem there are only two non-zero applied nodal forces Fp and FM+1 , but for now we will continue to include all the other possible nodal forces as well.

EM 424: Finite Elements

Up = U a

lm

M+1

= Ub

F =P
p

F
a

M+1

=P

Um+1 , F m+1 U , Fm
m

Fig.2

For the displacement in the mth element, we will assume that the displacement is a simple function which varies linearly between the two end (nodal) values Um ,Um +1 . This function is shown in Fig. 3.

ux Um

U m+1

x
Fig. 3
m) Even though the displacement , u( x , in the mth element only depends on the nodal displacements at the end points of that element, we can include all the other nodal displacements as well by defining

EM 424: Finite Elements

M +1 j =1

m) u( = x

H(
j

m)

Uj

(m = 1,..., M)

(2)

where
1 j = m lm (m ) Hj = j= m+1 lm all other j 0

(3)

and where 0 < < lm . The expression of Eq. (2) is equivalent to simply writing for the element
m) u( = U m 1 + U m +1 x lm lm = Um +

(Um +1 Um )
lm

(4)

which is the displacement profile shown in Fig. 3. Using Eq. (2) the virtual displacement in the mth element can be written in terms of the virtual changes of all the nodal values as
m) u( = H(jm )U j x j =1 M +1

(5)

By differentiation Eq. (2), we can also obtain the strain in the mth element in a similar form
e xx
( m)

u( ) M + 1 ( m ) = x = J j Uj j =1
m

(6)

where
1 l m 1 (m ) Jj = lm 0 j=m j = m +1

(7)

all other j

EM 424: Finite Elements

which is just equivalent to expressing the strain as a constant in each element given by

e xx =

( m)

Um +1 Um lm

(8)

The virtual change of this strain, therefore, is also

m) e xx = J ( j U j

( m)

M +1 j =1

(9)

If we place all these results into the principle of virtual work for this problem (Eq. (1)) and use the fact that the stress is given by xx = Eexx we obtain
M +1 ( m ) M +1 ( m ) E J U J U dV j m k k j m =1 j =1 Vm k =1
M

M +1 m M +1 = f x H (j ) U j dVm + Fj d j U j m =1 Vm j =1 j =1
M

(10)

where the sum over m is the result of adding up the distributed work-energy terms over all the elements and the constants d j are defined to eliminate all the nodal forces Fj except for those where applied forces act, e.g. for our case

1 for j = p, Fp = Pa d j = 1 for j = M + 1, FM +1 = Pb 0 for all other j

(11)

Note that strictly speaking since we allowed the nodal displacement U1 to vary here (we have kept U1 ) , the reaction force at that node should be included in the non-zero Fj terms since that reaction would do work, just like the applied loads. But since later we will enforce the boundary condition U1 = 0 , we can omit this reaction force in anticipation of the fact that it will not enter later since it does no virtual work. Interchanging the orders of summation in Eq. (10) and combining all the terms then gives
M +1 j =1

Uj E J (j m )J k(m)Uk dVm fx H(j m )dVm + Fj dj = 0

m =1 Vm
k =1 m=1 Vm

M +1

(12)

EM 424: Finite Elements

However, since we assume the Uj are all arbitrary, each of the terms multiplying these virtual displacement variations must vanish, to give
M +1 (m) (m ) M ( m) + Fj d j E J J U dV = f H dV j k k m x j m m =1 Vm k =1 m =1 Vm
M

(13)

which is just a set of linear equations for the unknown nodal displacements of the form
M +1 k =1

K jkU j = R j + Fj d j

(14)

where the stiffness matrix, K, is given by

M ( m ) ( m) K jk = EJ j J k dVm Vm m =1

(15)

and the body force term is

M m) Rj = fx H( j dVm m=1 Vm

(16)

We can write Eq. (14) out in matrix form as

K11 ... ... K M +11 ... K1 M +1 U1 R1 F1d1 ... ... ... ... = + ... ... ... ... ... K M +1 M+1 UM +1 RM +1 FM +1d M+1 ...

(17)

...

Before this system of equations can be solved, we must impose the displacement boundary condition U1 = 0 . This can be done by simply eliminating the first equation (since U1 = 0 this first equation should not have appeared in the first place anyway) and forcing all the terms involving U1 to be zero. This is equivalent to zeroing out the first row and column of terms in Eq. (17) to obtain

EM 424: Finite Elements

6 0 0 0 U R Fd K2 M +1 2 2 2 2 ... ... ... ... = + ... ... ... ... ... KM +1 M +1 UM+1 RM +1 FM +1d M +1 0

0 0 0 K22 ... ... 0 0 K M+12

... ... 0 ...

(18)

...

which is system of M equations in the remaining M unknown nodal displacements that can then be solved. Once all the nodal displacements are found, then Eqs. (2) and (6), in conjunction with Hookes law, can be used to find the displacements, strains, and stresses throughout the bar. Note that if some other displacement was specified in the bar as a non-zero value, then that constraint could also be handled in much the same way as the boundary condition U1 = 0 . For example, if the displacement at the third node was specified to be 0, then this condition could be incorporated by zeroing out all the terms in the U3 = U third row of Eq. (18) and moving the third column terms involving U3 to the right side of the equations as known terms, i.e. 0 0 0 0 0 K22 0 K24 0 0 0 0 0 K42 0 K 44 ... ... ... ... 0 K 0 K M +14 M+1 2 0 0 0 0 K2 M +1 U2 R2 F2 d2 K23U 0 0 0 0 0 = + K4 M +1 U4 R4 F4 d4 K43U ... ... ... ... ... ... KM +1 M +1 UM +1 RM +1 FM +1 dM +1 K M +13U ... ... 0 ... 0 (19) to give a resulting system of M-1 equations to be solved for the remaining M-1 unknowns.

EM 424: Finite Elements

Example Consider the problem of a bar of constant cross-sectional area A and Youngs modulus E q loaded by uniform distributed load q0 lb/unit length ( f x = 0 ) . By breaking this bar into A three equal length finite elements, solve for the nodal displacements, assuming that the strains are constant in each element. Compare the predictions for displacement and stress with the exact solution for this problem.

q0

The three elements contain four nodes labeled as shown below. In forming up the stiffness matrix elements, it is convenient to consider the individual contributions first from each element and then sum over all elements. Thus, we will write each element

U1

U2 q0

U3

U4

L/3
contribution as

L/3

L/3

K jk = EJj J k dVm
Vm

( m)

( m ) ( m)

(20)

so that the total stiffness matrix is

m) K jk = K ( jk m =1 M

(21)

EM 424: Finite Elements

Similarly, for the body force

m) m) R( = fx H( j j dVm Vm

Rj = Rj
m=1

(m )

(22)

Then, for the first element, taking its length as l ( l = L / 3 ) we have

l l 1 EA 1 (1 ) (1 ) (1) K11 = EA J1 J1 d = EA d = l 0 0 l l l l 1 EA 1 (1 ) (1 ) (1) K12 = EA J1 J 2 d = EA d = l 0 0 l l

K21 = EA J2 J1 d = K12
0

(1 )

(1 ) (1)

(1 )

(23)

l l 1 EA 1 (1 ) (1 ) (1) K22 = EA J2 J 2 d = EA d = l 0 0 l l all other K jk = 0

and for the distributed load on this element

l l ql ( 1) R1 = q0 H1(1)d = q0 1 d = 0 2 l 0 0 l l ql ( 1) (1) R2 = q0 H2 d = q0 d = 0 2 0 0 l 1) all other R ( j = 0

(240

Since the calculations for the other two elements involve similar evaluations, we only quote the end results below For the second element

EA l EA (2 ) (2 ) K23 = K32 = l
(2 ) (2 ) K22 = K33 =
2) all other K ( jk = 0

(25)

q0 l 2 (2 ) all other R j = 0
(2 ) (2 ) R2 = R3 =

EM 424: Finite Elements

and for the third element

(3 ) (3 ) K33 = K44 = (3 ) (3 ) K34 = K43

EA l EA = l
(26)

3) all other K ( jk = 0

q0 l 2 ( 3) all other R j = 0
(3 ) ( 3) R3 = R4 =

If we add up all these element contributions then the set of simultaneous equations has the form
EA l EA l 0 0 EA l EA EA + l l EA l 0 0 EA l EA EA + l l EA l q0 l U 1 2 q0l q0 l 0 U2 + 2 = 2 EA U q0l q0 l + 3 2 l 2 EA q0 l U l 4 2 0

(27)

where we have shown how all of the element contributions add together to give the final result. Applying the boundary conditions U1 = U 4 = 0 means that we must cross out the first and fourth rows and columns and solve the remaining two equations in two unknowns given by

2 AE l AE l

AE U q0l 2 l = 2 AE U q l 0 3 l q0 l 2 q0 L2 = . AE 9 AE

(28)

which has the solution U2 = U3 =

We can compare these results with the exact solution for this problem which is easy to obtain. The reader should verify that the exact solution for the displacement function and its specific values at the nodal locations are given by

EM 424: Finite Elements

10

q0 Lx q0 x 2 2 AE 2 AE L 2 L q L2 ux = ux = 0 3 3 9 AE ux (x ) =

(29)

so that the finite element solution values at the nodes are, in fact, exact.

AE u x /q 0 L

Exact solution Finite Element solution

x/L

The exact behavior of the axial stress for this problem also is easily obtained as

xx =

q0 L q0 x 2A A

(30)

EM 424: Finite Elements

11

xx A/q 0 L

Finite element solution

Exact solution

x/L

It is obvious that the finite element solution does a better job of representing the displacements than the stresses. This is to be expected since the stresses involve derivatives of the approximating displacement functions.

EM 424: Finite Elements

12

Finite Element Solution for 10 elements: If we simply increase the number of elements to ten, the solution for the displacement becomes much closer to the exact solution

AE u x /q 0 L
0.14

0.12

0.1

0.08

0.06

0.04

0.02

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

x/L

and the same is true for the stresses:

EM 424: Finite Elements

13

xx A/q 0 L
0.5 0.4 0.3 0.2 0.1 0 -0.1 -0.2 -0.3 -0.4 -0.5

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

x/L