2.092/2.

093 Finite Element Analysis of Solids & Fluids I Fall 09

Lecture 15 - Solution of Dynamic Equilibrium Equations

Prof. K. J. Bathe MIT OpenCourseWare

In the last lecture, we described a physical setup that demonstrates the technique of Gauss elimination. We
used clamps on each DOF and removed one clamp for one step of Gauss elimination.

u1

u2


u3
u4

should be positive, and should remain positive.

Our rule: Remove clamps one at a time, in the order we would perform Gauss elimination. If there is a
clamp seeing no more stiness after having removed some clamp(s), the structure is unstable.

Example

K=





All diagonal terms are positive. However, there will be a zero diagonal entry after Gauss elimination has
been performed for the 3rd DOF.

1
Lecture 15 Solution of Dynamic Equilibrium Equations 2.092/2.093, Fall 09

Solution of dynamic equilibrium equations

Consider a system with n DOFs:
+ CU + KU = R(t)
MU (1)

FI

with initial conditions

U t=0 = 0 U ; U t=0 = 0 U

The term CU will be discussed later. Our methods for solving (1) are:
Mode superposition: We rst transform the equation and then integrate.
Direct integration: We integrate the equation directly!
First, lets transform Eq. (1). Assume we use

U (t) = P X (t) (2)

nn n1

The function P is independent of time. Substitute this into Eq.(1) to obtain

+ P T CP X + P T KP X = P T R
P T MP X (A)

The best P matrix would diagonalize the matrix, thereby decoupling the equations. To obtain a wonderful
P , consider
MU + KU = 0 (free vibration)
U = sin (t t0 )
Then,
2 M sin (t t0 ) + K sin (t t0 ) = 0 (a)
For (a) to hold,
K = 2 M
K 2 M = 0

Let 2 = . We have a generalized eigenvalue problem. We must have det (K M ) = 0, and we nd the
solution for from the roots of the characteristic polynomial

p() = a0 + a1 + a2 2 + . . . + an n

Find the eigenvalues 1 , 2 , . . ., n from p() = 0 and then the eigenvectors 1 , . . . , n from

(K i M ) i = 0

Then, normalize i so that it satises Ti M i = 1. We now have (see Chapters 2, 10)

0 12 22 . . . n2

for 1 for 2 for n

2
Lecture 15 Solution of Dynamic Equilibrium Equations 2.092/2.093, Fall 09
Each i represents a mode shape, and we have

Ti M j = ij

where ij is the Kronecker delta, so we call i M -orthogonal (or M -orthonormal, because Ti M i = 1).
In turn, this yields
Ti Kj = i2 ij
Physically,

Consider 1 :
T1 M 1 = 1
T1 K1 = 12
The strain energy in the beam is 12 T1 K1 = 12 12 . By orthonormality, also,

T2 M 1 = 0

T2 M 2 = 1
and
T2 K2 = 22
Consider this simple case, for which we must solve K = 2 M :

M =
0



Then
1
M = 2 K = K


0
0
2
1 = .
A non-trivial solution is = 0, =
0
0
Note: 12 = 0 for rigid body motion. (No strain energy!)

3
Lecture 15 Solution of Dynamic Equilibrium Equations 2.092/2.093, Fall 09
Now lets use P = [ 1 ... n ]. Then, (A) becomes

12

zeros
+ P CP X +
X T .. T
X = P R

.
zeros n2

For now, lets assume no damping. (If C = 0, there is no damping and the equations are decoupled.) Then,
we have
X + 2 X = T R
nn

12

zeros x1
; 2 = .. ; X = ...

= 1 2 ... n

.
zeros n2 xn
So, we have
i + i2 xi = Ti R
x (i = 1, . . . , n)
0 0
As always, we need the initial conditions xi , x i to solve.

4
MIT OpenCourseWare
http://ocw.mit.edu

2.092 / 2.093 Finite Element Analysis of Solids and Fluids I

Fall 2009

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.