FREE VIBRATION OF A BAR – EIGENVALUE

PROBLEM

•THE DYNAMIC BAR PROBLEM

•SEPARATION OF VARIABLES ANALYSIS
•FEM FORMULATION FOR THE EIGENMODES
•ACCURACY ESTIMATES
•STUDY OF WAVE EQUATION FOR GIVEN WAVE
NUMBER – A DETAILED CASE STUDY
 EA = CONSTANT

x=L

DISTRIBUTED FORCE f(x,t)

ρA u( x, t )
F ( x, t ) F ( x + ∆x, t ) FREE-BODY
DIAGRAM
f ( x, t )∆x
x x + ∆x

EQUATION OF MOTION
dF 0< x< L; t >0
+ f ( x, t ) = ρA u
dx
INITIAL CONDITIONS:

u ( x,0) = u0 ( x) ; u ( x,0) = v0 ( x)

BOUNDARY CONDITIONS:

du
u (0, t ) = 0 ; EA | x =L = 0
dx

UNDAMPED MOTION OF THE BAR

FREE VIBRATION PROBLEM DISTRIBUTED FORCE
f(x,t)=0
•THE COMPLETE DYNAMICAL PROBLEM IS CALLED
AN IBVP PROBLEM
•THE FREE VIBRATION OF THE BAR IS CAUSED DUE
TO THE INITIAL PERTURBATION GIVEN TO THE BAR
(INITIAL DISPLACEMENT OR VELOCITY)

dF d  du 
= ρA u ⇒  EA  = ρA u
dx dx  dx 

USE SEPARATION OF VARIABLES

u ( x, t ) = U ( x) F (t )
 PUTTING THIS IN THE DIFFERENTIAL EQUATION

d 2U
EA 2 F (t ) = ρAU ( x) F(t )
dx
GIVES
d 2U Negative to ensure that
E dx 2 F
= = −ω 2 the solution does not
ρ U F blow up with time

Oscillatory motion with

THE CONTINUOUS EIGENVALUE natural frequency ω
PROBLEM
E
U ′′ + ω 2U = 0 0< x<L
ρ
U (0) = 0 ; U ′( L) = 0
ASSUME U ( x ) = e iα x

E ρ
− α +ω = 0 ⇒ α = ±
2 2
ω = ± kω
ρ E

U ( x) = A cos(α x) + B sin(α x)

FROM THE B.C. AT x=0 U (0) = 0 ⇒ A = 0

FROM B.C. AT x=L U ′( L) = 0 ⇒ cos(α L) = 0

π
⇒ α n L = (2n − 1) ; n = 1,2,  , ∞
2
THE nth MODE SHAPE (EIGENFUNCTION) IS GIVEN BY

 πx
U n ( x) = sin(α n x) = sin  (2n − 1) 
 2L 
ALSO…
E
⇒ ωn = αn nth NATURAL
ρ FREQUENCY

THE COMPLETE SOLUTION IS GIVEN BY

∞
u ( x, t ) = ∑ (Cn cos ωnt + Dn sin ωnt )sin(α n x)
n =1

Cn , Dn OBTAINED FROM INITIAL CONDITIONS

•GENERALLY, ONE IS INTERESTED IN ONLY THE FIRST
FEW NATURAL FREQUENCIES AND CORRESPONDING
MODE SHAPES.
•CANNOT BE EXPLICITLY OBTAINED IN MOST CASES,
WHERE THE SPATIAL DIFFERENTIAL EQUATION IS NOT
EASY TO SOLVE.
•NOTE THAT HERE THE MODES ARE DISTINCT. IN
PRACTICAL SYSTEMS, THE MODES CAN BE REPEATING
OR CLUSTERED AROUND A GIVEN FREQUENCY.
•IN ACOUSTIC PROBLEMS, MODE SHAPES FOR SPECIFIC
FREQUENCY BAND IS REQUIRED.

HOW TO USE FEM TO SOLVE FOR THIS?

START WITH THIS…..

EA U ′′ + ω 2 ρA U = 0 0 < x < L
U (0) = 0 ; ( EAU ′) | x =L = 0

GET WEIGHTED RESIDUAL FORMULATION

r ( x) = EAU ′′ + ω 2 ρ A U = 0

∫ (EAU ′′ + ω ρA U ) w( x)dx = 0
L L

∫ r ( x) w( x)dx = 0 ⇒ 2

x =0 x =0
 INTEGRATE BY PARTS ONCE…

L L

∫ ∫ ρ AU w dx + [EAU w]x=0
x=L
EA U ′ w′ dx = ω 2
′
x =0 x =0

AGAIN C0 ELEMENTS WILL WORK !

1 2 3 4 5

N N
U ( x) = ∑ δ i φi ( x) ; w( x) = ∑ β i φi ( x)
i =1 i =1
 THIS WILL GIVE THE STANDARD FORM

[ K ]{δ } = ω [ M ]{δ }
2

STIFFNESS MATRIX MASS MATRIX

GENERALIZED EIGENVALUE PROBLEM

•THE FEM SOLUTION WILL APPROXIMATE THE FIRST N

EIGENVALUES AND MODE SHAPES OF THE SOLUTION
•ONLY THE FIRST (APPROXIMATELY) N/2 EIGENPAIRS ARE
ACCURATE. ACCURACY DECREASES WITH INCREASE IN
FREQUENCY
 ω FE ≥ ωl
l
; eigenpair ω FEl , {δ (l ) }

The l th mode shape U FEl ( x) = ∑ δ i(l ) φi ( x)

i =1

HOW ACCURATE IS THE SOLUTION?

| ω FEl − ωl | ≤ Cl || U l − U FEl ||2

DEPENDS ON HOW WELL THE MODE SHAPE CAN BE

REPRESENTED BY THE BASIS FUNCTIONS AND MESH
EXAMPLE TO SHOW USE OF FEM

X1 = 0 X2 =1 X3 = 2

U(0)=0
3
U FE ( x) = ∑ δ i φi ( x) δ1 = 0
i =1

THE MATRIX PROBLEM NOW TO BE SOLVED IS

2 1
EA  2 − 1 3 6
 {δ } = (ρAh ω )
2
{δ }
h − 1 1 1 1 
 6 3 
Stiffness matrix Mass matrix
 SOLVING THIS EIGENVALUE PROBLEM GIVES

Material data taken h = 1, EA = 1, ρA = 1

ω FE = 0.806 (0.785)
1
ω FE = 2.815 (2.355)
2

EXACT ONES IN PARENTHESIS

ERROR = 2.7% AND 16.34%

•NOTE THAT THE FEM VALUES ARE HIGHER!!!

•THE ERROR IN THE EIGENVALUE INCREASES AS
THE EIGENVALUE NUMBER INCREASES
•LET US LOOK AT THE MODE SHAPES
 0.707   0.707 
ω FE ⇒ {δ }
(1)
=  ; ω FE2 ⇒ {δ }
( 2)
= 
1.000  − 1.000 
1

Note that these give exact nodal values of the actual mode shape
exact
exact

FASTER CONVERGENCE AS WE HAVE

SEEN

A FINER MESH OR A HIGHER ORDER APPROXIMATION WILL

GIVE A BETTER RESOLUTION OF THE MODES (AGAIN, HALF
THE NUMBER WILL BE QUITE ACCURATE)
 THE REVERSE PROBLEM ….

WHAT IF WE ARE INTERESTED IN OBTAINING THE

MODE SHAPE OF A PARTICULAR FREQUENCY?
EA U ′′ + ω 2 ρA U = 0 0< x<2
π
U (0) = 0 ; U | x = 2 = 1; ω = ;
4
EA = 1 ρA = 1
3
U FE ( x) = ∑ δ i φi ( x) FEM SOLUTION
i =1 WITH SAME MESH
THE WEAK FORM IS THE SAME, i.e.

[ K ]{δ } = ω [ M ]{δ }
2
WITH

δ1 = 0 δ3 = 1

THIS GIVES δ 2 = 0.694

NOT EXACT
ANYMORE!!!

EXACT=0.707
THIS IS A FEATURE OF THESE COMPUTATIONS. THE
NODAL FE VALUES ARE NOT EXACT. IN FACT, THE
FE SOLUTION THINKS IT IS RIDING ON A SLOWER
MOVING WAVE (NUMERICAL DISPERSION!!!) WITH
FREQUENCY ω .
•ALL THE FEATURES OF THE FE SOLUTION SEEN IN THIS
EXAMPLE HOLD FOR ALL THE MODELS/DOMAINS THAT YOU
CAN COME ACROSS.
•IN TWO AND THREE DIMENSIONS, THE MODE SHAPES GET
AFFECTED BY THE GEOMETRY OF THE DOMAIN AND HAVE
THE SINGULAR (STRESS CONCENTRATION) BEHAVIOR THAT
WE SAW IN THE ELASTICITY PROBLEM. HENCE, FOR THESE
CASES ALSO, WE HAVE TO USE THE PROPERLY REFINED
MESHES.
•THE NATURAL FREQUENCIES CONVERGE TO THE EXACT
ONES FROM ABOVE.
•THE FREQUENCY OF THE WAVE THAT THE FE SOLUTION
RIDES, CONVERGES TO THE EXACT ONE FROM BELOW.
THE ONE-DIMENSIONAL AXIAL WAVE PROBLEM

2 UNIFORM MESH
d u
2
+ k 2
u = f = 1 (let' s say) WITH MESH-SIZE
dx h. (linear elements)
u ( 0) = u ( L ) = 0

k 2h2
1− FE SOLUTION
cos(k h) = 3 CORRESPONDS TO λ2π
k 2h2 ANOTHER WAVE h= =
1+ WITH k
n kn
6
MESH SIZE GIVEN
∆k IN TERMS OF n
k = k − ∆k δ= ELEMENTS PER
k WAVELENGTH
•FE SOLUTION TRAILS THE EXACT SOLUTION
•THE NODAL VALUES ARE NOT EXACT

•FOR n=4, THE ERROR δ = 9 %

•FOR n=5, THE ERROR δ = 5.6 % ERROR IN THE
WAVE FREQUENCY
•FOR n=9, THE ERROR δ = 1.9 %
(Dispersion)

•IN 2D OR 3D, THE EFFECT OF UNSMOOTHNESS OF

GEOMETRY, AND MATERIAL BOUNDARIES, WILL
MAKE THIS DEMAND MORE SEVERE !!!