2 - Boundary Element Tearing and Interconnecting Methods

Advanced Vibrations
Advanced Vibrations
Hello:
Sign In
Sylabus
Homework, Projects, Exams, Grading
Office Hours
Facilities
http://me.unm.edu/~djsegal/
Calibration
Calendar
11/30/98 Copyright Dan Segalman, 1998

/home/djsegal/UNM/VibCourse/slides/Lecture1.frm
Advanced Vibrations
Overview
Advanced Vibrations
Core Elements
Frequency Domain Analysis

Energy Methods
Continuous Systems
Systems of Many D.O.F.s & Eigenanalysis

Lumped Mass Single DOF System
Elements:
Advanced Vibrations
Spring: F = kx
k
c Damper: F = cẋ
m Mass: F = mẋ˙

Assembled System:
Advanced Vibrations
k F
m
Applied force F ( t ) is balanced by acceleration of mass,

stroke of damper, and displacment of spring:
mẋ˙ + cẋ + kx = F ( t )

Simplification by Mass Normalization
Advanced Vibrations
Dividing through by the mass (we shall do this again later with multi-dof
systems):
2
ẋ˙ + 2ζω n ẋ + ω n x = f ( t )
where
c k 2 F (t)
---- = 2ζ , ---- = ω n , and ---------- = f ( t )
m m m
Recall: ω n is the undamped natural frequency;
and ζ is fraction of critical damping. (Note: ζ is pronounced “zeta”.)

About the Differential Operator
Advanced Vibrations
2
d d 2
Note that the operator L = 2
( ) + 2ζω ndt ( ) + ω n ( ) has the
dt
properties:
L is Linear: L ( f + g ) = L ( f ) + L ( g )
and L is time-translation invarient: L ( x ( t + T ) ) = L ( x ) t+T
This means that transform methods work for this sort of problem. We
shall discuss this more next time.

An Additional Special Normalization
Advanced Vibrations
Nondimensionalization of equation for single DOF system:
Let τ = ωn t and let y ( τ ) = x ( τ ⁄ ω n ) , then

2 2
dy 1 dy 1
= -----2- dx , = -----2- d x
dτ ωn d t t = τ ⁄ ωn dτ
2
ωn d t 2 t = τ ⁄ ωn
f ( τ ⁄ ωn ) F ( τ ⁄ ωn )
and y'' ( τ ) + 2ζy' ( τ ) + y ( τ ) = --------------------- = ----------------------
2 k
ωn
Say F ( t ) = Ak sin ( ωt ) , then
 ω
y'' ( τ ) + 2ζy' ( τ ) + y ( τ ) = A sin τ -
-----
 ω n

Nondimensionalization
Advanced Vibrations
Consider the steady state solution of
y'' ( τ ) + 2ζy' ( τ ) + y ( τ ) = sin ( τα )
where α = ω ⁄ ω n , we anticipate a solution of the sort
y ( τ ) = H̃ ( α ) sin ( τα – ψ̃ ( α ) )
If we solve forH̃ ( α ) and ψ̃ ( α ) for all α for a fixed ζ we have the

steady state solution for all single DOF systems with that value of ζ .

Advanced Vibrations
Please read the first chapter of

Meirovitch
Life will be easier for you to read
one chapter ahead.

Homework Number 1
Advanced Vibrations
1. Solve the transient Single DOF problem

y'' ( τ ) + 2ζy' ( τ ) + y ( τ ) = sin ( τα ) numerically using MATLAB
until steady state for values of α equal to 0 to 2.0 in increments of 0.1.
2. Plot H̃ ( α ) versus α
3. plot ψ̃ ( α ) versus α
4. Do this for the three values of ζ : 0.01, 0.1, 1.0
Hint: Let v = y' and solve v' + 2ζ 1 v = sin ( ατ )

y –1 0 0

Slides of Lecture 2
Advanced Vibrations
Today’s Class
• Homework
• Assembly of Systems from Components
• Convolutions with Indicial and Impulse

Response Functions
• Beginning of Fourier Analysis and the Complex

Operators
Copyright Dan Segalman, 1998

11/30/98
1 /home/djsegal/UNM/VibCourse/slides/Lecture2.frm
Recalling Homework Number 1
Advanced Vibrations
1. Solve the transient Single DOF problem

y'' ( τ ) + 2ζy' ( τ ) + y ( τ ) = sin ( τα ) numerically using MATLAB
until steady state for values of α equal to 0 to 2.0 in increments of 0.1.
2. Plot H̃ ( α ) versus α
3. plot ψ̃ ( α ) versus α
4. Do this for the three values of ζ : 0.01, 0.1, 1.0
Hint: Let v = y' and solve v' + 2ζ 1 v = sin ( ατ )

y –1 0 0

11/30/98
Solution
Having finished reading Chapter 1 of text
Advanced Vibrations
1
The transfer function H̃ is H̃ = --------------------------------
2
-
1 – α – i2ζα
The magnitude of H̃ is determined from
= H̃ H̃ ∗ = ----------------------------------------------
2 1
H̃
2 2 2
( 1 – α ) + ( 2ζα )
and the phase is determined from
Im ( H̃ ) ( H̃ – H̃ ∗ ) 2αζ
ψ̃ ( α ) = atan ---------------- = atan ------------------------- = atan --------------2-
Re ( H̃ ) i ( H̃ + H̃ ∗ ) 1–α
These are plotted as follows:

11/30/98
Plots from Homework 1
Magnitude of Transfer Function
Advanced Vibrations
H̃
ζ=0.01
ζ=0.1
ζ=1.0
α=ω/ωο

11/30/98
Plots from Homework 1
Phase of Transfer Function
Advanced Vibrations
ζ=0.01
ζ=0.1
ζ=1.0
arg ( H̃ )
α=ω/ωο

11/30/98
Conclusions from Homework 1
Advanced Vibrations
• There are advantages to frequency analysis. We

shall see more of this later.
• It takes a slightly damped system a very long

time to reach steady state.

11/30/98
Synthesis of Systems
Advanced Vibrations
Recall that it was asserted that:

Intrinsic # DOF of System = #Masses X DOF/Mass
Consider the following system
u1 u2
M
K1 K2
Does this system have two degrees of freedom?

Should it have just one?
How do we reconcile this?

11/30/98
Advanced Vibrations
Resolution is achieved through the elimination of massless degrees of

freedom - statically.
u1 u2
M
K1 K2
F1 , F2 F2 , FM M
As shown here, F1 and F2 are applied forces: F 1 = K 1 u1 ,

F 2 = K 2 ( u 2 – u 1 ) . FM is the force applied to the mass F M = Mu̇˙2
Balancing forces at a virtual mass at location x1 to deduce F1 = F2.
Balancing forces at the actual mass at location x2, F M = –F 2

11/30/98
Advanced Vibrations
u1 u2
M
K1 K2
F1 , F2 F2 , FM M
K2 K 1K 2
From F 1 = F 2 , u 1 = -------------------- u 2 and F 1 = F 2 = -------------------- u 2
K1 + K2 K1 + K2
K 1K 2
From F M = – F 2, Mu̇˙2 = – -------------------- u 2 which is our desired single
K1 + K2
degree-of-freedom system.
11/30/98
Stating the Obvious
Advanced Vibrations
• The reaction forces on massless subsystems is

equal and opposite to the forces applied to them.
• Note that extra degrees of freedom are removed

by balancing forces and equating displacements.
• Constitutive expressions relate applied forces to

resulting kinematics.

11/30/98
Some More Complex Cases
Advanced Vibrations
Returning to the liquid-like second order system

u1 u2
FM
M M
C K
C F1
F2 F2
Recall the synthesis rules:

• The displacements at location 1 are implicitly continuous.
• The forces balance: F 1 = F 2 , and F 2 = – F M . Remember

that the constitutive equations relate applied force to resulting
kinematics.

11/30/98
Advanced Vibrations
Returning to the liquid-like second order system

u1 u2
FM
M M
C K
C F1
F2 F2
Recall the constitutive rules:

• F 1 = cu̇ 1
• F 2 = K ( u2 – u1 )
• F M = Mu̇˙2

11/30/98
Advanced Vibrations
Solving the system: u1 u2

FM
cu̇ 1 = K ( u 2 – u 1 ) and C K
M M
K ( u 2 – u 1 ) = – Mu̇˙2 . C F1
F2 F2
Setting λ = c ⁄ K , we solve the

t
–( t – τ ) ⁄ λ dτ
first to find: u1 ( t ) = ∫e u 2 ( τ ) -----
λ
0
which is substituted in to the dynamics equation to obtain the
t
 –( t – τ ) ⁄ λ dτ
governing equation Mu̇˙2 + K  u 2 – e

∫
u 2 ( τ ) ----- =
λ
0.
0
This is tedious.

11/30/98
Using Laplace Transforms
Advanced Vibrations
cu̇ 1 = K ( u 2 – u 1 ) and K ( u 2 – u 1 ) = – Mu̇˙2
Becomes: csU 1 ( s ) – cu 1 ( 0 ) = K [ U 2 ( s ) – U 1 ( s ) ] and

2
K [ U 2 ( s ) – U 1 ( s ) ] + s MU 2 ( s ) = su̇ 2 ( 0 ) + u 2 ( 0 )
After solving out U 1 ( s ) , we have
2 cs cu 1 ( 0 )
s M + K ---------------- U 2 ( s ) = K ---------------- + M [ su̇ 2 ( 0 ) + u 2 ( 0 ) ] .
K + cs K + cs
It is good practice to normalize:
2 2 λs 2 λs
s + ω n --------------- U 2 ( s ) = ω n --------------- + [ su̇ 2 ( 0 ) + u 2 ( 0 ) ]
1 + λs 1 + λs

11/30/98
Another Problem
Advanced Vibrations
Remember that we were told that

Intrinsic # DOF of System = #Masses X DOF/Mass
That was a lie.

11/30/98
Another Problem
Advanced Vibrations
So far today, we have considered homogeneous systems - those

without imposed forces.
Lets consider the following two systems:
System 1 u System 2 u1 u2
M M
Where boundary forces or displacements are applied

at massless degrees of freedom, we often retain
those degrees of freedom.
Often we resolve out degrees of freedom associated

with the masses.
11/30/98
Another Problem
Advanced Vibrations
System 1 System 2 u1 u2
K K
M c M
c K2
Governing equations for System 1: Mu̇˙ + cu̇ + Ku = F
For System 2: Mu̇˙1 + cu̇ 1 + K u 1 = K 2 ( u 2 – u 1 )
F = K 2 ( u2 – u1 )
Lets solve these subject to homogeneous initial conditions

11/30/98
Impulse Response
Advanced Vibrations
Solve System 1: Mu̇˙ + cu̇ + Ku = F :

2
Using Laplace transforms, U (s) = ( F ) ( s ) ⁄ ( Ms + cs + K )
Transforming back to the time domain,
t
u(t ) = ∫ h ( t – τ )F ( τ ) dτ .
0
1 –1  1 
where h ( t ) = -----  ---------------------------------------
-
M  s + 2ζω n s + ω n
2 2
h is the impulse response function for this system.

11/30/98
Impulse Response Function
Advanced Vibrations
If we have a system whose governing equation in response to a

imposed force is L ( u ) ( t ) = F ( t ) , having a Laplace transform
G(s) (u)(s) = ( F ) ( s ) , then
–1
h = (G)
is the impulse response function for that system. The resulting

t
displacement is u(t ) = ∫ h ( t – τ )F ( τ ) dτ
0
Note: h solves L ( h ) ( t ) = δ ( t )

11/30/98
The Integral of Duhamel
Advanced Vibrations
Similarly, if we define the indicial response function g:

t
u(t ) = ∫ g ( t – τ )Ḟ ( τ ) dτ + F ( 0 )g ( t )
0
g ( t ) solves the equation L ( g ) ( t ) = H ( t ) , where H is the Heavyside

step function.

11/30/98
Homework Number 2
Advanced Vibrations
Part A. Determine the impulse response function h ( t ) for System 1.

Assume that the system is under-damped ( ζ < 1 ).
Part B. Determine the indicial response function g ( t ) for System 1.

11/30/98
Homework Number 3
Advanced Vibrations
Part A. Resolve out the degree of freedom u1 associated with the mass
in System B to in favor of u2. Obtain a governing equation in a single
degree of freedom u2. You may leave this in Laplace transform space.
Part B. Determining the impulse response function for this problem.

This function lives in the time domain. Assume that the system is
under-damped.

11/30/98
Concepts from Complex Variable Analysis
A Reminder
Advanced Vibrations
General Notions
Complex number z = x + iy
Real and Imaginary operators: Re { x + iy } = x , Im { x + iy } = y
Complex Conjugate: ( x + iy )∗ = x – iy
1 1
Observe: Re { z } = ( z + z ) and Im { z } = - ( z – z∗ )
--
- ∗ ----
2 2i
Complex Function of a Complex Variable: f ( z ) = u ( z ) + iv ( z )
An Analytic Function has a Unique Derivative:
∂f ∂f ∂u ∂v ∂v ∂u
= ⇒ = and = –
∂x ∂ ( iy ) ∂ x ∂y ∂x ∂y

11/30/98
A Reminder
Advanced Vibrations
About Analytic Functions
z = x + iy is analytic
n
z is analytic if n is an integer
αz
e is analytic
if f 1 and f 2 are each analytic, then so are
α f 1, f 1 + f 2, f 1 f 2 , f 1 ⁄ f 2
if f is analytic, then so is f '

if Re ( f ) is constant then Im ( f ) is also constant.
if Im ( f ) is constant then Re ( f ) is also constant.
11/30/98
Slides of Lecture 3
Advanced Vibrations
Today’s Class
• Homework
• Fourier Analysis and the Complex Operators

12/7/98
Homework Number 2
Advanced Vibrations
Part A. Determine the impulse response function h ( t ) for System A.

Assume that the system is under-damped ( ζ < 1 ).
Part B. Determine the indicial response function g ( t ) for System A.
System A
K
M
c

12/7/98
Solution to Homework 2
Part A
Advanced Vibrations
The impulse response function satisfies
t
u(t ) = ∫ h ( t – τ )F ( τ ) dτ .
0
1 –1  1 
By Laplace Transforms, h ( t ) = -
----  ---------------------------------------
-
M  s + 2ζω n s + ω n
2 2
Our task is to invert this laplace transform.

12/7/98
Part A
Advanced Vibrations
2 2
We begin by noting that s + 2ζω n s + ω n = ( s – r 1 ) ( s – r 2 )
2 2
where r 1 = – ζω n + iω n 1 – ζ and r 2 = – ζω n – i ω n 1 – ζ
Note further that these are complex conjugates. This is no surprise
since the roots of a polynomial of real coefficients are combinations of
real numbers and complex conjugate pairs.
We now expand as partial fractions
1 1 A B
---------------------------------------
- = --------------------------------------
- = -----------------
- + --------------------
-
2
s + 2ζω n s + ω n
2 ( s – r 1 ) ( s – r 1∗ ) ( s – r 1 ) ( s – r 1∗ )
We solve for A and B : 1 = A ( s – r 1∗ ) + B ( s – r 1 ) to find
A = 1 ⁄ 2iIm ( r 1 ) and B = – 1 ⁄ 2iIm ( r 1 )

12/7/98
Part A
Advanced Vibrations
1 –1  1 
h ( t ) = -----  ---------------------------------------
-
M  s + 2ζω n s + ω n
2 2
1 –1  1 1 
= ---------------------------
M2iIm ( r 1 )  ------------
- – ---------------- 
î s – r 1 s – r 1∗ 
1  r 1 t r 1∗ t 
= ---------------------------  e – e 
M2iIm ( r 1 ) î 
Re { r 1 }t
e  iIm { r 1 }t iIm { r 1∗ }t 
= ---------------------------  e –e 
M2iIm ( r 1 ) î 
– ω n ζt
e 2
= ------------------------------- sin ( ω n 1 – ζ t )
2
Mω n 1 – ζ
12/7/98
Partial Solution to Homework 2
Part B
Advanced Vibrations
Here we begin by recalling that

t
u(t ) = ∫ g ( t – τ )Ḟ ( τ ) dτ + F ( 0 )g ( t )
0
By taking the Laplace transform:
{ u } = { g } ( s { F } – F ( 0 ) ) + { g }F ( 0 )
= s {g} {F }
but we recall that {u} = {h} {F },
and we conclude that {g} = {h} ⁄ s

12/7/98
Partial Solution to Homework 2
Part B
Advanced Vibrations
For this problem,

1 1 1 1 1 
{g} = --
- {h} = --------------------------- --
-  ------------- – ---------------- 
s M2iIm ( r 1 ) s î s – r 1 s – r 1∗ 
We expand this into partial fractions (yuck):
1 1 1  1 ⁄ r 1 1 ⁄ r 1 1 ⁄ r 1∗ 1 ⁄ r 1∗
---  ------------- – ----------------  = – ----------- + ------------- + -------------- – ----------------
s î s – r 1 s – r 1∗  s s – r1 s s – r 1∗
It is up to the students to complete the assignment.

12/7/98
Part A
Advanced Vibrations
Part A. Resolve out the degree of freedom u1 associated with the mass
in System B to in favor of u2. Obtain a governing equation in a single
degree of freedom u2. You may leave this in Laplace transform space.
Part B. Determining the impulse response function for this problem.

This function lives in the time domain. Assume that the system is
under-damped.
System B u1 u2
K
c M
K2

12/7/98
Part A
Advanced Vibrations
Mu̇˙1 + cu̇ 1 + K u 1 = K 2 ( u 2 – u 1 )
System B u1 u2
K
M
c K2 F = K 2 ( u2 – u1 )
Lets rewrite the first equation as Mu̇˙1 + cu̇ 1 + K u 1 = F which is

solved by Laplace transform:
{F }
{ u 1 } = -----------------------------------------------
2 2
- so u 1 = h 1 *F
M ( s + 2ζω n + ω n )
– ω n ζt
e 2
where h 1 ( t ) = ------------------------------- sin ( ω n 1 – ζ t )
2
Mω n 1 – ζ

12/7/98
Part A
Advanced Vibrations
System B u1 u2
K
Mu̇˙1 + cu̇ 1 + K u 1 = F
M
c K2 F = K 2 ( u2 – u1 )
We solve the second equation for u2 :
F
u 2 = ------ + u 1
K2
F 1
= ------ + h 1 *F ⇒ h 2 = h 1 + ------ δ
K2 K2
=  h 1 + ------ δ *F
1
 K2 

12/7/98
A Reminder
Advanced Vibrations
General Notions
Complex number z = x + iy
Real and Imaginary operators: Re { x + iy } = x , Im { x + iy } = y
Complex Conjugate: ( x + iy )∗ = x – iy
1 1
Observe: Re { z } = ( z + z ) and Im { z } = - ( z – z∗ )
--
- ∗ ----
2 2i
Complex Function of a Complex Variable: f ( z ) = u ( z ) + iv ( z )
An Analytic Function has a Unique Derivative:
∂f ∂f ∂u ∂v ∂v ∂u
= ⇒ = and = –
∂x ∂ ( iy ) ∂ x ∂y ∂x ∂y

12/7/98
A Reminder
Advanced Vibrations
Examples of Analytic Functions
z = x + iy is analytic
n
z is analytic if n is an integer
αz
e , sin ( αz ) and cos ( αz ) are analytic

12/7/98
Identifying Analytic Functions,
Some Useful Rules
Advanced Vibrations
if f 1 and f 2 are each analytic, then so are
α f 1, f 1 + f 2, f 1 f 2 , f 1 ⁄ f 2
if f is analytic, then so is f '

if Re ( f ) is constant then Im ( f ) is also constant.
if Im ( f ) is constant then Re ( f ) is also constant.

12/7/98
Analytic Functions,
Some More Useful Rules
Advanced Vibrations
The Re{ } operator distributes over addition: (trivially.)

Re { A + B } = Re { A } + Re { B }
If f ( α ) is a complex function of a real parameter, α , then the Re and

derivative operator commute: (a little harder to see.)
d d f 
Re { f ( α ) } = Re  
dα î dα 

12/7/98
Analytic Functions,
Advanced Vibrations
The integral over a cycle of two harmonic functions: (more difficult)
ω
ω ------
1
------ ∫ Re { Ae }Re { Be } dt = --- Re { AB∗ }
2π iωt iωt
2π 0 2
1
= --- [ Re { A }Re { B } + Im { A }Im { B } ]
2

12/7/98
Analytic Functions,
Advanced Vibrations
When you have the trick to prove the above, you can show the
following:
ω ∞ ∞
ω 0 2π-
-----  iω 0 pt   iω 0 qt 
------ ∫ Re  ∑ A p e Re  ∑ B q e dt
2π 0 î p=0  î q=0 
∞
1
∗
= Re { A 0 B 0 } + ---
2 ∑ Re { Aq Bq∗ }
q=1

12/7/98
About Real (Actual) Forces
and Displacements
Advanced Vibrations
Forces and displacements are real. They may be scalar, vector, or

algebraic, but they are not complex. On the other hand, we sometimes
define complex quantities to represent, indirectly, periodic forces:
iωt
f ( t ) = Fe in NOT a force
The following could be forces:
iωt iωt – i ωt – i ωt
Re { Fe } , Im { Fe } , Re { Fe } , Im { Fe }
– i ωt
+ F∗ e
iωt
or Fe
This last form will be especially useful when we deal with complex
modes.
In the above, F could be a complex scalar, a complex vector, an array of

complex scalars, or an array of complex vectors. We will consider the
more complicated cases later in this course.

12/7/98
Application to Damped Harmonic Oscillator
Advanced Vibrations
Consider the simple, single-degree-of-freedom oscillator shown below:

c
m Pcos(ωt+θ)
k
x(t)
The equation of motion for the mass is:
mẋ˙ + cẋ + kx = P cos ( ωt + θ ) which we can express in

exponential form by
iωt
P cos ( ωt + θ ) = Re { Fe }
where Re { F 1 } = P cos ( θ ) and Im { F 1 } = P sin ( θ ) .

12/7/98
Solution in Frequency Space
Advanced Vibrations
Lets postulate that the steady state displacement has the same form:
iωt
u ( t ) = Re { X e }
In order to evaluate the time derivatives of x , we must observe that the
Re{ } and derivative operators commute:
d iωt  d iωt  iωt

ẋ ( t ) = Re { X e } = Re  X e  = Re { iωX e }
dt î dt 
2 iωt
and similarly, ẋ˙( t ) = Re { – ω X e }

12/7/98
Substituting back into the
equation of motion:
Advanced Vibrations
2 iωt iωt
mRe { – ω X e } + cRe { iωX e }
iωt iωt
+ kRe { X e } – R e { Fe } = 0
We observe that the Re operator distributes over addition:

2 iωt
Re { [ ( – ω m + iωc + k )X – F ]e } = 0
There are several easy ways to prove that the complex argument of the
iωt
Re operator in the above equation is zero. Then the coefficient of e
is also zero:
2
[ ( – ω m + iωc + k )X – F ] = 0

12/7/98
Solution
Advanced Vibrations
Solving for X ( ω ) from which we find that
F
X = --------------------------------------------
2
-
( – ω m + iωc + k )
F⁄k F
- = --- H ( ω )
= ----------------------------------------------------------------
2
1 – ( ω ⁄ ω n ) + 2iζ ( ω ⁄ ω n ) k
2
where ω n = k ⁄ m and ζ = c ⁄ ( 2m ) . H ( ω ) is the magnification
factor.
The text discusses the magnification factor, its magnitude, where the
magnitude is largest, and the role of damping - ζ - in calculating Q

12/7/98
Limiting Cases
Advanced Vibrations
Let’s examine limiting cases:
ω ⁄ ω n → 0 ⇒ F = kx : this is statics, force is in phase with

displacement
2
ω ⁄ ω n → ∞ ⇒ F = – x mω : mass loading is opposite to
displacement
2
1 1–ζ
ω = ω n ⇒ x = --- 1 – i ------------------ F :
2 ζ
For small damping, F = 2ζiX

12/7/98
Products and Integrals
Advanced Vibrations
The importance of the rigor introduced above is illustrated in the

calculation of the time-averaged rate of energy dissipation.
1 T
D = --- ∫ ẋ ( τ ) f ( τ ) dτ
T 0
1 T  F iωτ 
= --- ∫ Re  iω --- H e Re { Fe } dτ
iωτ
T 0 î k 
ω 2
= – ------ Im ( H ) F
2k
where T = 2π ⁄ ω .

12/7/98
Fourier Analysis
Advanced Vibrations
System A We again consider the damped linear

oscillator. This time we consider an
K F(t) applied force
M ∞
c
 ipω 0 t 
F ( t ) = Re  C pe ∑ 
î p=0 
(The “Re” operator was neglected in equations 1.86, 1.87of the text.)
We deduce the coefficients of the Fourier series from integration
against the basis functions:
2π ⁄ ω 0 2π ⁄ ω 0
ω0 ω0 – iω 0 q
C 0 = ------
2π ∫ F ( t ) dt ,and C q = ------
π ∫ F ( t )e dt for 0<q
0 0
Please verify this yourselves!

12/7/98
Response of the System
Advanced Vibrations
Exploiting linearity and our previous analysis of damped linear

systems,
∞
 ipω 0 t 
x ( t ) = Re  ∑ ( C p ⁄ k )H ( ω p )e 
î p=0 
1
where H ( ω ) = ---------------------------------------------------------------------
2
-
[ 1 – ( ω ⁄ ω n ) ] + i2ζ ( ω ⁄ ω n )
This is the key result of Fourier Analysis.

12/7/98
Let’s Calculate some Averages
The Easy Way: Re-Examine Page 16
Advanced Vibrations
Lets calculate the square of RMS force:

2
F =
ω ∞ ∞
ω 0 2π-
-----  iω 0 pt   iω 0 qt 
------ ∫ Re  ∑ C p e Re  ∑ C q e dt
2π 0 î p=0  î q=0 
∞
2 1
= C 0 + ---
2 ∑ C q C q∗
q=0
∞
1
∑
2 2
= C0 + --- Cq
2
q=0

12/7/98
Advanced Vibrations
Lets calculate the square of RMS displacement:

2
x =
ω ∞ ∞
ω 0 2π-
-----  Cp iω 0 pt   Cq iω 0 qt 
------ ∫ Re  ∑ ------ H ( ω p )e  Re  ∑ ------ H ( ω q )e  dt
2π 0 î p=0 k  î k 
q=0
2 ∞
C
=  ------ + -------2-
1
∑ C q C q∗ H ( ωq )H ( ωq )∗
0
 k
2k q=0
2 ∞
 C  1
∑
0 2 2
= ------ + -------2- C q H ( ωq )
 k
2k q=0

12/7/98
Advanced Vibrations
Lets calculate the square of RMS dissipation rate:
F ẋ =
ω ∞ ∞
ω 0 2π- 
----- iω 0 pt   Cq iω 0 qt 
------ ∫ Re 
2π 0 ∑ C pe Re  ∑ iω 0 q -----
k
- H ( ω q )e  dt
î p=0  î q=1 
∑ C q C q∗ Im { H ( ωq ) }
1
= – ------
2k
q=0

12/7/98
Slides of Lecture 4
Advanced Vibrations
Review of Topics of Students’ Choice

11/30/98
Slides of Lecture 5
(Presented over two periods)
Advanced Vibrations
Today’s Class: Energy Methods
• Lagrange Equations
• Hamilton’s Principle
There are two energy methods we shall cover in this class. Together
they can form the basis of deriving governing equations and discretized
governing equations for almost any system we encounter
In this class and next, we shall derive the Lagrange equations and
practice applying them. After that we shall examine Hamilton’s
principle.

11/30/98
Lagrange Equations: Theme
Advanced Vibrations
Up to now We have assembled the governing equations multi-body

systems using:
• force balance
• displacement continuity
• perspicacity
We had to be smart.
Lagrange Equations Provide a Recipe

We do not have to be so smart.
We have broader choice of degrees of freedom - called generalized
coordinates.
We can accommodate constraints, implicitly or explicitly.
Certain invariance properties hold (explained later).

11/30/98
Lagrange Equations: Derivation Outline for
Systems of Particles
Advanced Vibrations
Newton’s Law of Motion for a Particle:
∂
F = p
∂t
where p is linear momentum.

Assuming constant mass and in the special case of a orthogonal
Cartesian coordinate system,
3
∂ ∂
F =
∂t
p = mv = m
∂t ∑ ẋ˙n bn =
n=1
where b n are unit basis vector that are constant in time and space. This
permits us to write a simple expression involving the components:
F n = mẋ˙n
11/30/98
Advanced Vibrations
The big lesson here is that

∂ ˙
F = mx => F n = mẋ˙n
∂t
for orthogonal, Cartesian coordinate systems
What about other coordinate systems?

11/30/98
Advanced Vibrations
Consider a particle whose position, g,

is expressed in terms of a polar
coordinate system: g = re r ( θ ) g
where the basis vectors e r ( θ ) and eθ ( θ )
er ( θ ) r
e θ ( θ ) are related to the standard θ
orthogonal cartesian coordinate
system by: e r ( θ ) = cos ( θ )i + sin ( θ ) j and

e θ ( θ ) = – sin ( θ ) i + cos ( θ ) j
The particle moves, so the coordinates r and θ of the particle are

functions of time: r = r ( t ) and θ = θ ( t ) .

11/30/98
Advanced Vibrations
A little arithmetic shows that
∂ ∂
e r ( θ ) = θ̇e θ ( θ ) and that e θ ( θ ) = – θ̇e r ( θ )
∂t ∂t
Recalling that g = re r ( θ ) we derive

˙ ˙˙ 2
g = ṙe r + rθ̇e θ and g = ( ṙ˙ – rθ̇ )e r + 2ṙθ̇e θ
Expressing any imposed force in this systems as F = F r er + F θ eθ

we find that Newton’s Law gives us
2
F r = ṙ˙ – rθ̇ and F θ = 2ṙθ̇

11/30/98
Advanced Vibrations
2
What do we observe from F r = ṙ˙ – rθ̇ and F θ = 2ṙθ̇ ?
• Because this is not an orthogonal Cartesian coordinate system:
F r ≠ ṙ˙ and F θ ≠ θ̇˙
• The form of the scalar equations of motion depends on the
coordinate system used
• We had to be smart to deduce the equations of motion

11/30/98
Advanced Vibrations
Lets find a method that does not require us to be so smart. One such
method is the use of Lagrange Equations.
• The derivation shown here is not that of the text, but it is the more
standard approach.
• Students are not responsible for reproducing this derivation, but

you will see it again in your studies of applied mechanics. It
would be good to be able to follow the derivation.
• The derivation shown here is for systems of particles. The

extension to systems of rigid bodies and of continua will be
accepted on faith.
• The way to get comfortable with Lagrange equations is through

practice.

11/30/98
Advanced Vibrations
We consider a system of N particles and set of n generalized

coordinates { q k } . It is assumed that these coordinates:
• are enough to describe the configuration of the system of
particles completely at any time, i.e.
x k ( t ) = f k ( q 1, q 2, …q n, t )
 ∂ 
• are independent, i.e. Rank  f k = n
î ∂ ql 
• are appropriately differentiable
∂x k ∂
(We sometimes simplify the notation: = f k. )
∂ qr ∂ qr

11/30/98
Advanced Vibrations
We differentiate our kinematic relations to derive an expression for

particle velocity:
˙
 ∂ f  q̇ + ∂ f
xk ( t ) = ∑  ∂ qr k r ∂ t k
r
The rates q̇ r are sometimes called speeds. Note that
∂ ˙ ∂ ∂ ˙ ∂
xk = f k , so we conclude that xk ( t ) = x k . We shall
∂ q̇ r ∂ qr ∂ q̇ r ∂ qr
use this later.
With our expression for particle velocity we can construct kinetic
energy in terms of the generalized degrees of freedom.
1
T = --- ∑ m k  ∂ x  q̇ + ∂ x ⋅  ∂ x  q̇ + ∂ x
2 k ∑  ∂ qr k r ∂ t k ∑  ∂ qs k s ∂ t k
r s
11/30/98
Lagrange Equations: Derivation Outline
for Systems of Particles
Advanced Vibrations
The partial derivative of kinetic energy on speed is called the

generalized momentum:
∂T  ˙ ∂ ˙
pr =
∂ q̇ r
= ∑ mk  xk ⋅ ∂ q̇r xk .
k
Because Newton’s equations involve the rate of change of momentum,
we are motivated to take the following time derivative:
dp r d ∂T  ˙˙ ∂x˙k  
˙ d ∂x k
˙
=   = ∑ k  k ∂ q̇r ∑ k k ⋅ d t  ∂ q̇r
m x ⋅ + m x .
dt d t  ∂ q̇ r k k
We shall simplify each of the summations on the right hand side of the
above equation.

11/30/98
Advanced Vibrations
 ˙˙ ∂x˙k 
Simplifying ∑ m k  x k ⋅ 
k  ∂ q̇ r
˙
∂x k ∂x k
Recalling that = ,
∂ q̇ r ∂ qr
 ˙˙ ∂x˙k   ˙˙ ∂x k  ∂x k
∑ mk  xk ⋅ ∂ q̇r =  ∑ mk xk ⋅ ∂ qr = ∑ F k ⋅ ∂ qr = F r
k k k
Fr is called the Generalized Force associated with the generalized

coordinate qr .

11/30/98
Advanced Vibrations
˙
˙ d  ∂x k  ˙ d  ∂x k 
Simplifying ∑ m k x k ⋅   = ∑ mk xk ⋅ ,
d t  ∂ q̇ r 
d t ∂ qr
k k
We note that the derivative in the summation is:

2 2
d  ∂x k  ∂ xk ∂ xk

d t ∂ qr 
=
∂ q r ∂q s∑q̇ s + ∑
∂ q r ∂t
s s
∂ ∂x k ∂x k ∂
=
∂ qr ∑ ∂ qs s ∑ ∂ t
q̇ + =
∂ qr
xk and obtain
s s
˙ d  ∂x k  ˙ ∂ ˙
∑ mk xk ⋅ d t  ∂ qr = ∑ mk xk ⋅
∂ qr
xk
k k
∂ mk ˙ ˙ ∂T
= ∑
∂ qr k
-----
2
- [ xk ⋅ xk ] =
∂ qr
11/30/98
Advanced Vibrations
dp r ∂T
We combine all the above to find that = Fr + .
dt ∂ qr
d  ∂T  ∂T
The more common form of this is – = Fr
 
d t ∂ q̇ r ∂ q r
The above are Lagrange’s equations.
• They hold in any system of coordinates.
• They reduce down to the component form of Newton’s Law for

orthogonal, Cartesian system.
• They are an invariant recipe.
• They accommodate constraints. This is shown later.

11/30/98
Lagrange Equations:
First Example Problem
Advanced Vibrations
Consider the pendulum shown, consisting of a mass

suspended by a massless rod. The only degree of y
freedom is θ . Lets proceed with the steps of the O x
recipe.
1. Express configuration in terms of the generalized
θ Tension T
coordinates: x = R sin ( θ )i – R cos θ j
R
2. Evaluate the Jacobian. These terms are really
∂
important. x = R cos θi + R sin θ j
∂θ m
mg
3. Write the applied forces: F = – mg j – T ( x ⁄ x )
∂
4. Calculate generalized forces: Fr = F ⋅ x = – mgR sin θ .
∂θ

11/30/98
Lagrange Equations:
First Example Problem
Advanced Vibrations
5. Calculate kinetic energy:

m ˙ ˙ m ∂ ∂  2 m 2 2
T = ---- ( x ⋅ x ) = ---- x ⋅ x θ̇ = ---- R θ̇
2 2 ∂θ ∂θ 
 2
d  ∂T  2
6. Plug and Chug = mR θ̇˙ = – mgR sin θ
d t  ∂ θ̇
Observe
• We did not have to make any smart observations to evaluate the
governing equation
• The tension on the rod did not play any role in the equation for θ̇˙ .
This is because forces of constraint do not act in the directions of
motion, so they do no work. We shall see this some more later.

11/30/98
Degrees of Freedom and Constraints:
Several Approaches
Advanced Vibrations
1. Because we have freedom in choosing our generalized degrees of

freedom, we can exploit the constraints to minimize the number of
degrees of freedom.
2. Where our constraints are holonomic (of the sort
h ( q 1, q 2, …q n ) = 0 ) we can use the constraints to resolve out
extra degrees of freedom before calculating kinetic energy or
generalized forces.
We shall practice these later.

11/30/98
Degrees of Freedom
Potential Energies
Advanced Vibrations
Sometimes, we can simplify the calculations of

generalized forces when they derive from a potential
energy. At the moment, we can identify several sorts
of force:
conservative external force - such as gravity
conservative internal force - such as strain energy
non-conservative external force - generally the case
non-conservative internal force - damping elements

11/30/98
Degrees of Freedom
Potential Energies
Advanced Vibrations
Potential energies give rise to conservative forces. The forces depend

only on current configuration and can be written in terms of gradients
of the corresponding potential energy.
∂
F = –∇ V = – V
∂x
Note the nice feature of the chain rule for conservative forces:
∂ ∂V ∂ ∂V
Fr = F ⋅ x = – ⋅ x = –
∂ qr ∂ x ∂ qr ∂ qr
This facilitates the formulation of the equations.

11/30/98
Lagrange Equations:
First Example Problem Revisited
Advanced Vibrations
Consider the pendulum shown, consisting of a mass y

suspended by a massless rod. The only degree of O x
freedom is θ . Lets proceed with the steps of the
receipt.
1. Again, express the configuration in terms of θ Tension T
generalized coordinates:
R
x = R sin ( θ )i – R cos θ j
∂
2. Evaluate the Jacobian. x = R cos θi + R sin θ j m mg
∂θ
3. There are only conservative forces F = – ∇(mgy)

from which we deduce
∂ ∂
Fθ = – mgy = – ( – mgR cos θ ) = – mgR sin θ
∂θ ∂θ

11/30/98
Advanced Vibrations
4. Calculate kinetic energy:

m ˙ ˙ m ∂ ∂  2 m 2 2
T = ---- ( x ⋅ x ) = ---- x ⋅ x θ̇ = ---- R θ̇
2 2 ∂θ ∂θ 
 2
d  ∂T  2
5. Plug and Chug = mR θ̇˙ = – mgR sin θ
d t  ∂ θ̇
Note that by using the potential energy, we were able to ignore

irrelevant forces (the tension in the rod).

11/30/98
Advanced Vibrations
Another Example
Consider the following system involving both
spring tension and gravity. y
1. Again, the location of the mass is expressed in O x
terms of the generalized coordinate θ ,
R
x = R sin ( θ )i – R cos θ j k
2. Evaluate the Jacobian
m
3. The generalized forces are calculated. This
time we only need to consider potential
2
energies: V = mgy + k y ⁄ 2 .

11/30/98
Advanced Vibrations
We use our recipe for generalized force:

∂V ∂ 2 2
Fθ = – = – [ – m gR cos θ + kR cos θ ⁄ 2 ]
∂θ ∂θ
2
= – mgR sin ( θ ) + kR sin ( 2θ ) ⁄ 2
2 2
4. Again, the kinetic energy is T = m ( R θ̇ ) ⁄ 2
2 2
5. Plug and chug: mR θ̇˙ = – mgR sin θ + kR sin ( 2θ ) ⁄ 2

11/30/98
Lets do and EASY one
Advanced Vibrations
Consider the following system involving both a

torsional spring tension and gravity.
K
1. Again, the location of the mass is expressed
R
in terms of the generalized coordinate θ ,
θ
x = R sin ( θ )i – R cos θ j . m
2. Evaluate the Jacobian
3. The generalized forces are calculated. This
time we only need to consider potential
2
energies to do this: V = mgy + K θ ⁄ 2 .
We use our recipe for generalized force:
∂V ∂ 2
Fθ = – = – [ – m gR cos θ + K θ ⁄ 2 ]
∂θ ∂θ
= – mgR sin ( θ ) – K θ

11/30/98
Advanced Vibrations
2 2
2
5. Plug and chug: mR θ̇˙ = – mgR sin θ – K θ

11/30/98
Advanced Vibrations
Lets Do One with Damping

Here we consider a another simple problem. It
contains both conservative and dissipative forces:
K
1. Again, the location of the mass is expressed in
terms of the generalized coordinate θ ,
R
x = R sin ( θ )i – R cos θ j θ
2. Evaluate the Jacobian. m
3. The generalized forces are calculated. This time µ
we to consider potential energy as well as a
force due to viscous damping and add the
C D
forces: F θ = F θ + F θ . As before, the force
due to the potential energy V = mgy is

11/30/98
Advanced Vibrations
C ∂V ∂
Fθ = – = – [ – m gR cos θ ]
∂θ ∂θ
= – mgR sin ( θ ) .
The force due to damping must be calculated in the brute force manner:
D D ∂ ∂
F θ = F ⋅ x = – µẋi ⋅ x
∂θ ∂θ
= – ( µθ̇R [ cos θi ] ⋅ R [ cos θi + sin θ j ] )
2 2
= – µθ̇R cos θ
2 2
2 2 2
5. Plug and chug: mR θ̇˙ = – mgR sin θ – µθ̇R cos θ

11/30/98
That Was Hard
Advanced Vibrations
There must be an easier way to handle viscous

dissipation. That will be covered in the next lesson.
We shall address more of these dissipative problems
then.

11/30/98
Some Easy Conservative Problems
The simplest problem
Advanced Vibrations
The potential energy is trivially found:

1 2 k
V = --- ku . The generalized force is
2
∂V
Fu = – = – ku . u
∂u
m 2
The kinetic energy is just as easily found: T = ---- u̇ , so
2
d  ∂T  ∂T
– = mu̇˙ = – ku

d t ∂ u̇  ∂u
That was easy. Lets do another.

11/30/98
Some Easy Non-Conservative Problems
The simplest problem with a force.
Advanced Vibrations
The potential energy is trivially found:

1 2 k Fe(t)
V = --- ku . The generalized force is
2
∂V
Fu = – + F e = – ku + F e ( t ) . u
∂u
m 2
The kinetic energy is just as easily found: T = ---- u̇ , so
2
d  ∂T  ∂T
– = mu̇˙ = – ku + F e ( t )

d t ∂ u̇  ∂u
That was not so difficult. Lets do another just slightly more difficult.

11/30/98
A still-simple problem.
Advanced Vibrations
k1 m1 k m2
2
u1 u2
1 2 1 2
The potential energy is easily found: V = --- k 1 u 1 + --- k 2 ( u 2 – u 1 ) .
2 2
The generalized forces are
∂V
F u1 = – = –k 1 u1 –k 2 ( u1 – u2 ) and
∂ u1
∂V
F u2 = – = –k 2 ( u2 – u1 ) .
∂ u2
11/30/98
Advanced Vibrations
m1 2 m2 2
The kinetic energy is just as easily found: T = ------ u̇ 1 + ------ u̇ 2 , so
2 2
d  ∂T  ∂T
– = m 1 u̇˙1 = – k 1 u 1 – k 2 ( u 1 – u 2 ) and
 
d t ∂ u̇ 1 ∂ u 1
d  ∂T  ∂T
– = m 2 u̇˙2 = – k 2 ( u 2 – u 1 )
 
d t ∂ u̇ 2 ∂ u 2
m1 u̇˙1 k 1 + k 2 –k 2 u1
In matrix form, + = 0
m 2 u̇˙2 –k 2 k2 u2 0
Thus are developed mass and stiffness matrices
Lets do it again.

11/30/98
A still-simple problem.
Advanced Vibrations
k1 m1 k m2
2
u1 u2
k3
The potential energy is

easily found:
1 2 1 2 2
V = --- k 1 u 1 + --- k 2 ( u 2 – u 1 ) + k 3 u 2 . The generalized forces
2 2
∂V
are F u = – = – k 1 u 1 – k 2 ( u 1 – u 2 ) and
1 ∂ u1
∂V
F u2 = – = –k 2 ( u2 – u1 ) – k 3 u2 .
∂ u2
11/30/98
Advanced Vibrations
m1 2 m2 2
The kinetic energy is just as still: T = ------ u̇ 1 + ------ u̇ 2 , so
2 2
d  ∂T  ∂T
– = m 1 u̇˙1 = – k 1 u 1 – k 2 ( u 1 – u 2 ) and
 
d t ∂ u̇ 1 ∂ u 1
d  ∂T  ∂T
– = m 2 u̇˙2 = – k 2 ( u 2 – u 1 ) – k 3 u 2
 
d t ∂ u̇ 2 ∂ u 2
m1 u̇˙1 k1 + k2 –k 2 u1
In matrix form, + = 0
m 2 u̇˙2 –k 2 k 2 + k 3 u2 0
This is still easy.

11/30/98
Lets do a funny one
Advanced Vibrations
m1 k m2 m3
1 k2
The potential energy is u1 u2 u3

again easily found:
1 2 1 2
V = --- k 1 ( u 2 – u 1 ) + --- k 2 ( u 3 – u 2 ) . The generalized forces are:
2 2
∂V
F u1 = – = –k 1 ( u1 – u2 )
∂ u1
∂V
F u2 = – = –k 1 ( u2 – u1 ) – k 2 ( u2 – u3 )
∂ u2
∂V
F u3 = – = –k 2 ( u3 – u2 )
∂ u3
11/30/98
Advanced Vibrations
m1 2 m2 2 m3 2
The kinetic energy is: T = ------ u̇ 1 + ------ u̇ 2 + ------ u̇ 3 .
2 2 2
The Lagrange equations now give the matrix form
m1 u̇˙1 k1 –k 1 u1 0
m2 u̇˙2 + – k 1 k 1 + k 2 – k 2 u 2 = 0
m3 –k 2 k2 u3 0
This is still easy.

11/30/98
About Mass and Stiffness Matrices
Advanced Vibrations
What can we conclude from our limited experience?

• the mass matrix is symmetric
• the mass matrix is nonsingular (so long as each d.o.f. has mass)
• the mass matrix is positive definite
• the stiffness matrix is symmetric
• the stiffness is positive, semi-definite (rigid body motions).

11/30/98
A Summary of the Recipe
Advanced Vibrations
1. Express the cartesian coordinates in terms of the generalized d.o.f.s

2. Calculate the partial derivatives of position w.r.t. generalized d.o.f.s
3. Express the potential energies in terms of the generalized d.o.f.s
4. Evaluate the generalized forces in terms of the generalized d.o.f.s
5. Evaluate the kinetic energy in terms of the generalized d.o.f.s
d  ∂T  ∂T
6. Substitute into the famous – = Fr
 
d t ∂ q̇ r ∂ q r

11/30/98
Lecture 5, Homework 1
Advanced Vibrations
k
R
The disk shown has radius R, mass m, and mass moment of inertia J.
The disk rolls without slipping and is restrained by a spring of stiffness
k. Use the method of Lagarange’s equations to derive an equation of
motion for the displacment, u, of the center of gravity of the disk.
Hint: Rθ̇ = u̇ .
11/30/98
Advanced Vibrations
Derive the equations of motion for the following 2-d.o.f. problem:
m1
u
θ
L
m2

11/30/98
Next Time
Advanced Vibrations
We shall compare solution to homework 2B.

We shall go over today’s problems again.

11/30/98
The Time after Next Time
Advanced Vibrations
Small Displacements and Linearization

Dissipation Potential
More on Mass and Stiffness Matrices

11/30/98
Slides of Lecture 6
Advanced Vibrations
Today’s Class:
• Solution to Homework 1 of Lecture 5.
• Short Quiz
• Comparison of answers to Homework 2B
• Practice with Lagrange Equations

11/30/98
Advanced Vibrations
k
R
The disk shown has radius R, mass m, and mass moment of inertia J.
The disk rolls without slipping and is restrained by a spring of stiffness
k. Use the method of Lagrange’s equations to derive an equation of
motion for the displacement, u, of the center of gravity of the disk.
Hint: Rθ̇ = u̇ .
11/30/98
Lecture 5, Homework 1: Solution
Advanced Vibrations
1 2 1 2
Kinetic Energy: T = --- mu̇ + --- J θ̇
2 2
Substituting Rθ̇ = u̇ into the expression for kinetic energy and

performing the appropriate derivatives:
d  ∂T  2
= ( m + J ⁄ R )u̇˙
d t  ∂ u
1 2 ∂V
Potential Energy: V = --- ku yields force: F u = – = – ku
2 ∂u
∂V
The potential energy yields a force Fu = – = – ku
∂u
2
Putting it all together: ( m + J ⁄ R )u̇˙ + ku = 0
11/30/98
Short Quiz:
Advanced Vibrations
Consider the problem of an inverted

pendulum. Derive the equation of x = R ( sin θi + cos θ j )
motion of this structure in terms of the
angle θ. m
Note that the torsional spring serves to R

restore (center) the pendulum but that
gravity serves to exacerbate any θ
displacement from center.
Remember to write T = … and

V = … κ

11/30/98
Resolution of Homework 2B
Advanced Vibrations
Determine the indicial response System 1

function g ( t ) for System 1. Recall that K
the indicial response function satisfies M
c
t
u(t ) = ∫ g ( t – τ )Ḟ ( τ ) dτ + F ( 0 )g ( t )
0
We take the Laplace transform to
find Lu = ( Lg ) ( sLF – F ( 0 ) ) + ( Lg )F ( 0 ) = ( Lg ) ( sLF )
Recall that the step response satisfies Lu = ( Lh ) ( LF ) from which
we conclude that Lg = ( Lh ) ⁄ s and that
t
g(t ) = ∫ h ( τ ) dτ
0
11/30/98
Advanced Vibrations
From an earlier homework, we found
1 –1  1 
h ( t ) = -----  ---------------------------------------
-
M  s + 2ζω n s + ω n
2 2
– ω n ζt
e 2
= ------------------------------- sin ( ω n 1 – ζ t )
2
Mω n 1 – ζ
We integrate this to find
t – ω n ζτ
e
∫
2
g(t ) = ------------------------------
- sin ( ω n 1 – ζ τ ) dτ
2
0 Mω n 1 – ζ

11/30/98
Advanced Vibrations
After a little integration, this becomes.
g(t ) =
– ω n ζt  2 ζ 2 
1–e  cos ( ω n 1 – ζ )t + ------------------ sin ( ω n 1 – ζ )t
 1–ζ
2 
----------------------------------------------------------------------------------------------------------------------------------------
K
Simplifications have included the observations that
2 2 2 2 2
( ω n ζ ) + ( ω n 1 – ζ ) = ω n and that Mω n = K

11/30/98
Remember to do the second homework
given in Lecture 5.
Advanced Vibrations
m1
u
θ
L
m2
Express kinematics in terms of generalized degrees of freedom:
x 1 = ui and x 2 = ui + L sin θ i – L cos θ j .
The potential energy is just that due to gravity: V = – m 2 gL cos θ

11/30/98
Next Time
Advanced Vibrations
• Small Displacements and Linearization
• Dissipation Potential
• More on Mass and Stiffness Matrices

11/30/98
Slides of Lecture 7
Advanced Vibrations
Today’s Class:
Conservation of Energy
Small Displacements and Linearization
Dissipation Potential

11/30/98
Mechanical Energy
Advanced Vibrations
Consider a system composed of N particles. For each of which,
d
mk vk = F k .
dt
We contract each side with the particle velocity and sum over particles:
 dv k
∑ mk  vk ⋅ d t  = ∑ F k ⋅ vk
k k
 dv k d mk
Note that ∑ m k v k ⋅ ∑ d t 2 ( vk ⋅ vk ) = d t
dT
= -----
- and that
 dt 
k k
∑ F k ⋅ vk is the rate at which work is applied to the system.

k

11/30/98
Mechanical Energy
Advanced Vibrations
Always,
dT d
= ( Work Done to System )
dt dt
But for Conservative Systems,
∂V dx k
∑ k k ∑
dT dV
= F ⋅ v = – ⋅ = –
dt k k ∂x
dt dt
k
from which we conclude that
d
(T + V ) = 0
dt

11/30/98
Lets Apply Conservation of Energy
To a Simple System.
Advanced Vibrations
A pendulum is released from rest at an angle of θ0 .

Calculate the maximum velocity achieved by the pendulum.
R
T 0 = 0 and V 0 = – mgR cos θ 0 , so
Initially
θ
T 0 + V 0 = – mgR cos θ 0
m
The potential energy is minimized when the pendulum is at
the bottom of its swing. At that point, the kinetic energy
must be at its maximum.
T θ=0 +V θ=0 = T0 + V0 ⇒ T θ=0 = mgR ( 1 – cos θ 0 )
and v max = 2gR ( 1 – cos θ 0 )

11/30/98
Solution to Homework Assigned in
Previous Lecture
Advanced Vibrations
m1
u
θ
L
m2
Express kinematics in terms of generalized degrees of freedom:
x 1 = ui and x 2 = ui + L sin θ i – L cos θ j .
The potential energy is just that due to gravity: V = – m 2 gL cos θ

11/30/98
Advanced Vibrations
The generalized forces are:
∂V ∂V
Fu = – = 0 and F θ = – = – m 2 gL sin θ .
∂u ∂θ
The kinetic energy is

1 ˙ ˙ 1 ˙ ˙
T = --- m 1 x 1 ⋅ x 1 + --- m 2 x 2 ⋅ x 2
2 2
1 2 1 2 1 2
= --- m 1 u̇ + --- m 2 ( u̇ + Lθ̇ cos θ ) + --- m 2 ( Lθ̇ sin θ )
2 2 2
1 2 1 2 2
= ( m 1 + m 2 )u̇ + m 2 Lθ̇u̇ cos θ + m 2 L θ̇
--
- --
-
2 2

11/30/98
Advanced Vibrations
Lets calculate the appropriate partial derivative from this
∂T
= 0 , and
∂u
d  ∂T  d
= [ ( m 1 + m 2 )u̇ + m 2 Lθ̇ cos θ ]
dt  ∂ u̇  dt
2
= ( m 1 + m 2 )u̇˙ – m 2 Lθ̇ sin θ + m 2 Lθ̇˙ cos θ
∂T
= – m 2 Lθ̇u̇ sin θ and
∂θ
d  ∂T  d 2
= [ m 2 ( Lu̇ cos θ + L θ̇ ) ]
d t  ∂ θ̇ dt
= m 2 L ( u̇˙cos θ + Lθ̇˙ – u̇θ̇ sin θ )
11/30/98
Adding all the Ingredients
Advanced Vibrations
2
( m 1 + m 2 )u̇˙ – m 2 Lθ̇ sin θ + m 2 Lθ̇˙ cos θ = 0
2
and m 2 Lu̇˙cos θ + m 2 L θ̇˙ = – m 2 gL sin θ
These are two messy equations to solve.

11/30/98
Lets Simplify
Advanced Vibrations
0
2
( m 1 + m 2 )u̇˙ – m 2 Lθ̇ sin θ + m 2 Lθ̇˙ cos θ = 0
2
and m 2 Lu̇˙cos θ + m 2 L θ̇˙ = – m 2 gL sin θ
Lets introduce the assumption that θ and θ̇ are each so small that we
can ignore powers and products of them. From this we get the small-
angle formulae:
3 5 2 4
θ θ θ θ
sin θ = θ – ----- + ----- … ≈ θ and cos θ = 1 – ----- + ----- … ≈ 1
3! 5! 2! 4!

11/30/98
We now have
Advanced Vibrations
( m 1 + m 2 )u̇˙ – ( m 2 Lθ̇˙ ) = 0 and

2
m 2 Lu̇˙ + m 2 L θ̇˙ = – m 2 gLθ .
In matrix form, this is
m1 + m2 m2 L
u̇˙ + 0 0 u = 0
m2 L m2 L
2
θ̇˙ 0 m 2 gL θ 0
Which is our canonical form for linear the undamped vibration problem

11/30/98
Steady State Solution
Advanced Vibrations
Lets see what kinds of solution we might get if we postulate steady-
u = Re  û e iωt  . Substituting this

state solutions of the form  
θ î θ̂ 
into the matrix equation, we find
 2 2 
 –ω ( m1 + m2 ) –ω m2 L û iωt 
Re  e  = 0
 2
–ω m2 L – ω m 2 L + m 2 gL θ̂
2 2  0
î 
from which we deduce that
2 2
–ω ( m1 + m2 ) –ω m2 L û
= 0
– ω m 2 L + m 2 gL θ̂
2 2 2 0
–ω m2 L
11/30/98
Steady State Solution
Advanced Vibrations
2 2
–ω ( m1 + m2 ) –ω m2 L û
The equation = 0 has a
– ω m 2 L + m 2 gL θ̂
2 2 2 0
–ω m2 L
non-trivial solution only if
 2 2 
 – ω ( m 1 + m 2 ) – ω m2 L 
det   = 0
 2
–ω m2 L
2 2
– ω m 2 L + m 2 gL 
î 
2 2
m 2 ω L [ m 1 Lω + ( m 1 + m 2 )g ] = 0
which has two solutions.

11/30/98
Solutions
Advanced Vibrations
2
• Solution 1: ω = 0 . Substituting this into the matrix equation
û
we see = 1 . This is translation at constant speed.
θ̂ 0
( m 1 + m 2 )g
2
• Solution 2. ω = – ---------------------------- . Substituting this into the
m1 L
1
û
matrix equation, we find = ( m 1 + m 2 )g . Lets discuss
θ̂ – ----------------------------
m2 L
the significance of this.

11/30/98
Linearization of Vibration Problems.
Perturbation about Static Equilibrium
Advanced Vibrations
Say we have derived our non-linear vibration equations
d  ∂T  ∂T ∂V e ∂x k
∑
e
– + = Fk(t) ⋅ = F r ( { q }, t ) .
 
d t ∂ q̇ r ∂ q r ∂ q r ∂ qr
k
We linearize in the following steps:
s
1. Solve for a stable equilibrium configuration { q } . The system will
be in equilibium if ∂V = 0 for each r and will be in
∂ qr {q} = {q }
s
stable equilibrium if V is at a local minumum.

11/30/98
Perturbation about Static Equilibrium
Advanced Vibrations
s
2. express { q } = { q } + { ∆q } , and substitute into the governing
equation. In this substitution, we use the Taylor series expansion:
∂f
f ( { q } + { ∆q } ) = f ( { q } ) + ∑  
s s
∆q r
 ∂ q r
{q }
s
r
3. delete all products and powers of ∆q r and ∆q̇ r .

4. group terms to form the mass and stiffness matrix.

11/30/98
Lets try this out on some simple problems
Advanced Vibrations
Consider the problem of an inverted

m 2 2 x = R ( sin θi + cos θ j )
pendulum. T = ---- R θ̇ ,
2 m
κ 2
V = --- θ + mgR cos θ , yielding R
2
2 θ
mR θ̇˙ + κθ – mgR sin θ = 0 .
(We have done this before),
Say κ < mgR , then θ = 0 is no

κ
longer a configuration of static
equilibrium. (How do we know? The
negative stiffness is a clue.)

11/30/98
Examine Potential Energy
for different κ ⁄ mgR
Advanced Vibrations
3.8
1.4
8 3.6
3.4
3.2
V/mgR
3 1.2
V/mgR
V/mgR
6 2.8
2.6
2.4 1
2.2
4
2
1.8
0.8
1.6
2 1.4
1.2
1 0.6
–3 –2 –1 0 1 t 2 3 –3 –2 –1 0 1 t 2 3 –3 –2 –1 0 1 t 2 3
θ θ θ
κ =2mgR κ =mgR κ =mgR/2
We see that the location θ = 0 is not a stable equilibrium for

κ < mgR

11/30/98
Linearization
Advanced Vibrations
s s
1. We now solve for static equilibrium: κθ – mgR sin θ = 0
This may be done numerically:
sin(θ)/θ
κ
-----------
mgR
s
sin θ κ
Solving ------------
s
- = ----------
-
θ mgR
θ
s
From the above plot we may pick off the value of θ that satisfies
the equilibrium equation.
11/30/98
Formulation of Linearized Equations
Advanced Vibrations
s
2. substitute { q } = { q } + { ∆q } into governing equation
2 ˙ s s
mR ∆θ̇ + κ ( θ + ∆θ ) – mgR sin ( θ + ∆θ ) = 0 , and expand
with Taylor series
2 s s s
mR ∆θ̇ + κ ( θ + ∆θ ) – mgR ( sin ( θ ) + cos ( θ )∆θ )
˙
2 ˙ s 0 s s
= mR ∆θ̇ + [ κθ – mgR sin ( θ ) ] + κ∆θ – mgR ( cos ( θ )∆θ )
2 ˙ s
= mR ∆θ̇ + ( κ – mgR cos ( θ ) )∆θ = 0
Note that the equation for static equilibrium appears and is identically
zero.
3. delete all products and powers of ∆q r and ∆q̇ r . Done.

4. group terms to form the mass and stiffness matrix. Done.

11/30/98
The Compound Pendulum Problem
Lets derive the governing equations
Advanced Vibrations
Kinematics: x 1 = R 1 ( sin θ 1 i – cos θ 1 j ) .

R1
x 2 = R 1 ( sin θ 1 i – cos θ 1 j )
and . θ1
+ R 2 ( sin θ 2 i – cos θ 2 j )
˙
Velocities: x 1 = R 1 θ̇ 1 ( cos θ 1 i + sin θ 1 j ) m1
R2
˙
x 2 = R 1 θ̇ 1 ( cos θ 1 i + sin θ 1 j ) θ2
and
+ R 2 θ̇ 2 ( cos θ 2 i + sin θ 2 j )
m2

11/30/98
Compound Pendulum
Advanced Vibrations
m1 2 m2 2 2
T = ------ ( R 1 θ̇ 1 ) + ------ [ ( R 1 θ̇ 1 ) + ( R 2 θ̇ 2 ) ]
Kinetic Energy: 2 2
+ m 2 ( R 1 θ̇ 1 ) ( R 2 θ̇ 2 ) cos ( θ 2 – θ 1 )
Lets do the exercise:
∂T 2
= ( m 1 + m 2 )R 1 θ̇ 1 + m 2 R 1 R 2 θ̇ 2 cos ( θ 2 – θ 1 )
∂ θ̇ 1
d  ∂T  2
= ( m 1 + m 2 )R 1 θ̇˙1 + m 2 R 1 R 2 θ̇˙2 cos ( θ 2 – θ 1 )
d t  ∂ θ̇ 1 2
+ m 2 R 1 R 2 θ̇ 2 θ̇ 1 sin ( θ 2 – θ 1 ) – m 2 R 1 R 2 θ̇ 2 sin ( θ 2 – θ 1 )
∂T
= m 2 R 1 R 2 θ̇ 2 θ̇ 1 sin ( θ 2 – θ 1 )
∂ θ1
11/30/98
Compound Pendulum
Advanced Vibrations
∂T 2
= m 2 R 2 θ̇ 2 + m 2 R 1 R 2 θ̇ 1 cos ( θ 2 – θ 1 )
∂ θ̇ 2
d  ∂T  2
= m 2 R 2 θ̇˙2 + m 2 R 1 R 2 θ̇˙1 cos ( θ 2 – θ 1 )
d t  ∂ θ̇ 2 2
– m 2 R 1 R 2 θ̇ 2 θ̇ 1 sin ( θ 2 – θ 1 ) + m 2 R 1 R 2 θ̇ 1 sin ( θ 2 – θ 1 )
∂T
= – m 2 R 1 R 2 θ̇ 2 θ̇ 1 sin ( θ 2 – θ 1 )
∂ θ2

11/30/98
Compound Pendulum
Advanced Vibrations
Potential Energy:
V = m 1 gy 1 + m 2 gy 2 =
= – m 1 gR 1 cos θ 1 – m 2 g [ R 1 cos θ 1 + R 2 cos θ 2 ]
∂V ∂V
= g ( m 1 + m 2 )R 1 sin θ 1 and = gm 2 R 2 sin θ 2
∂ θ1 ∂ θ2
Put them all together:
2
( m 1 + m 2 )R 1 θ̇˙1 + m 2 R 1 R 2 θ̇˙2 cos ( θ 2 – θ 1 )
2
– m 2 R 1 R 2 θ̇ 2 sin ( θ 2 – θ 1 ) + g ( m 1 + m 2 )R 1 sin θ 1 = 0
2
m 2 R 2 θ̇˙2 + m 2 R 1 R 2 θ̇˙1 cos ( θ 2 – θ 1 )
2
+ m 2 R 1 R 2 θ̇ 1 sin ( θ 2 – θ 1 ) + gm 2 R 2 sin θ 2 = 0
and
11/30/98
Homework
Advanced Vibrations
1. The above derivation was done on the

fly. Please check that it is correct. R1
2. Using the methods discussed above, θ1

identify the equilibrium configuration
and linearize the equations of motion
about that configuration.
m1
R2
θ2
m2

11/30/98
Rayleigh Damping
Advanced Vibrations
For the special case of viscous damping,

there exists a dissipation potential that
facilitates the formulation of the equations of
motion. Lets look at the example shown.
Recall that x = R ( sin θi – cos θ j ) The

θ restraining force on the pendulum is found
D
R to be F = – µẋ 1 i = – µRθ̇ cos θi . The
corresponding generalized force associated
with the θ direction is
D D ∂ 2 2
µ
m F θ = F ⋅ x = – µR θ̇ ( cos θ )
∂θ
The governing equation is found to be
2˙ 2
mR θ̇ + Rmg sin θ = – µ ( R cos θ ) θ̇

11/30/98
Rayleigh Dissipation Potential
Advanced Vibrations
There is another way to calculate the generalized force associated with

the dashpot.
˙D 2
The rate of dissipation in the dashpot is D = F ⋅ x = µẋ 1 . We now
˜D ∂ ∂ µ 2 2
F θ = – (D ⁄ 2) = – (--- ( R cos θ ) θ̇ )
∂ θ̇ ∂ θ̇ 2
2
evaluate = – µ ( R cos θ ) θ̇
˜D D
Notice that F θ = F θ .
Lets discuss why this should not surprise us. Consider virtual work.

11/30/98
Rayleigh Dissipation Function
Advanced Vibrations
The Rayleigh dissipation function provides a convenient method for

calculating the generalized force associated with dashpots.
The recipe is as follows
1. calculate the rate of dissipation, D , in terms of the speed in

cartesian coordinates.
2. convert that expression to one involving generalized coordinates

and generalized speeds.
3. make the partial with respect to each generalized speed and

multiply by –( 1 ⁄ 2 )

11/30/98
Rayleigh Dissipation Function
Warning
Advanced Vibrations
Though the Rayleigh dissipation function is easy to use, its uses are
limited.
• Only in a minority of circumstances is the damping associated
with discrete dashpots.
• Usually, damping is associated with a material dissipation. We

will discuss this later if time permits.

11/30/98
Slides of Lecture 8
Advanced Vibrations
Today’s Class:
Review Of Homework From Lecture 7

Hamilton’s Principle
More Examples Of Generalized Coordinates
Calculating Generalized Forces Via Virtual Work

11/30/98
Homework from Lecture 7
Advanced Vibrations
In the most recent class, we derived the governing

equations for the compound pendulum. The homework R1
assignment was to verify the derivation and to linearize the θ1

resulting equations.
m1
R2
θ2
We found the governing equations to be
m2
2
( m 1 + m 2 )R 1 θ̇˙1 + m 2 R 1 R 2 θ̇˙2 cos ( θ 2 – θ 1 )
2
2
m 2 R 2 θ̇˙2 + m 2 R 1 R 2 θ̇˙1 cos ( θ 2 – θ 1 )
2
and

11/30/98
Advanced Vibrations
1. Establish the location of stable equilibrium about which to

linearize. By observation, that location is θ1 = θ2 = 0 .
0 0
s s
2. Substitute θ 1 = θ 1 + ∆θ 1 and θ 2 = θ 2 + ∆θ 2 into the
governing equations and expand with Taylor series.
1
2˙
( m1 + m 2 )R 1 θ̇ 1 + m 2 R 1 R 2 θ̇˙2 cos ( θ 2 – θ 1 )
θ2-θ1 θ1
2
1
2˙
m 2 R 2 θ̇ 2 + m 2 R 1 R 2 θ̇˙1 cos ( θ 2 – θ 1 )
θ2-θ1 θ2
and 2
11/30/98
Advanced Vibrations
3. Delete all terms involving powers and products of θ 1 , θ 2 , θ̇ 1 , and

θ̇ 2
2˙
( m1 + m 2 )R 1 θ̇ 1 + m 2 R 1 R 2 θ̇˙2
2 0
– m 2 R 1 R 2 θ̇ 2 ( θ 2 – θ 1 ) + g ( m 1 + m 2 )R 1 θ 1 = 0
and
2˙
m 2 R 2 θ̇ 2 + m 2 R 1 R 2 θ̇˙1
2 0
+ m 2 R 1 R 2 θ̇ 1 ( θ 2 – θ 1 ) + gm 2 R 2 θ 2 = 0

11/30/98
Advanced Vibrations
4. Group terms to form the mass and stiffness matrix.
2
( m1 + m 2 )R 1 m 2 R 1 R 2 θ̇˙1 g ( m 1 + m 2 )R 1 0 θ1
+ = 0
m2 R1 R2
2
m2 R2 θ̇˙2 0 gm 2 R 2 θ 2 0
Note that both the mass and stiffness matrices are symmetric and
positive-definite.
The mass equation will be positive so long as we avoid massless
degrees of freedom, that is we always want to choose our degrees of
∂T
freedom so that ≠ 0.
∂ q̇ r
The stiffness matrix will always be non-negative definite. If there are no
rigid-body modes, it will be positive definite. We shall discuss this more
later.

11/30/98
Advanced Vibrations
We consider a mechanical system whose configuration at any time is

characterized by the N generalized coordinates { q } . The system is
A
subject to potential energy V and additional forces { F r } and evolves
over the interval ( t 1, t 2 ) according to the Lagrange equations
d  ∂T  ∂T ∂V A
– + = F r for each r .
 
d t ∂ q̇ r ∂ q r ∂ q r
We can imagine the evolution of
the system configuration over that q3
interval by picturing the motion of q(t2)
a point whose coordinates are
q(t)
{ q ( t ) } in an N-dimensional
Cartesian system. q2
q(t1)
q1

11/30/98
Advanced Vibrations
We contract these scalar equations with test functions η r which are as

yet undetermined except for the conditions η r ( t 1 ) = η r ( t 2 ) = 0 for
each r, and then sum them.
N t2
d  ∂T  ∂T ∂V
∑∫
A
– + – F r η r ( t ) dt = 0
d t  ∂ q̇ r ∂ q r ∂ q r
r = 1 t1

11/30/98
Advanced Vibrations
Integration by parts yields

N t2
∂T ∂V
∑∫
A
– + – Fr η r ( t ) dt
∂ qr ∂ qr
r = 1 t1
N t2
d   ∂T   ∂T
+ ∑∫  
d t ∂ q̇ r
ηr –
 ∂ q̇ r
η̇ r dt = 0
r = 1 t1
Recalling that η r ( t 1 ) = η r ( t 2 ) = 0 , the above integral simplifies
N t2
 ∂T ∂V ∂T 
∑ ∫  – ∂ qr + ∂ qr – F r ηr ( t ) –∂ q̇r η̇r dt = 0
A
r = 1t î 1


11/30/98
Advanced Vibrations
The above is true for all test functions ηr .
η r ( t ) = q̃ r ( t ) – q r ( t ) q3
Let
= δq r ( t ) q(t2)
where { q̃ ( t ) } is another path {q(t)}
from { q ( t 1 ) } to { q ( t 2 ) } “near” ~
{q(t)} q2
the path taken by { q ( t ) } . q(t1)
q1
Our integrals can now be written:
N t2
  ∂T ∂T  ∂V 
∑ ∫  –  ∂ qr δqr + ∂ q̇r δq̇r + ∂ qr – F r δqr dt = 0
A
r = 1t î 1

Observe that the terms involving potential energy are a complete
differential.
11/30/98
Advanced Vibrations
Rearranging the above:

t2 N
   ∂T δq + ∂T δq̇  + – ∂V + F A δq  dt
∫  ∑ î  ∂ q r r ∂ q̇ r r ∂ qr r r

t r=1
1 t2
= ∫ ( δT – δV + δW ) dt = 0
t1
A
Where δW ( t ) = F r ( t ) ( q̃ r ( t ) – q r ( t ) ).
This form of Hamilton’s principle asserts that the actual path is one
t2
about which
∫ ( δT – δV + δW ) dt = 0
t1

11/30/98
Hamilton’s Principle: Special Case
Advanced Vibrations
A
For the special case where the generalized forces Fr are prescribed
loads, we can define the “Potential Energy of Loading”
∑
A
A = – F r qr
r=1
T
and the “Total Potential Energy” is V = V + A.
In this case, Hamilton’s Principle becomes: The true path in

configuration space of the system makes the quantity
t2
∫
T
J = ( T – V ) dt stationary.
t1

11/30/98
Hamilton’s Principle: Example
Beam Bending Equation
Advanced Vibrations
The strain energy in an Euler-

p(x)
Bernoulli beam is
L 2 2 M0 ML
1 ∂ y 
V = --- ∫ EI  2  dx . The
2
0
∂x  Q0 QL
L
potential energy of loading is
L
A = M 0 ∂y – M L ∂y – Q 0 y ( 0 ) + Q L y ( L ) – ∫ p ( x )y ( x ) dx
∂x 0 ∂x L
0
L
1  ∂y 2
2 ∫
And the Kinetic Energy is T = --- m
 ∂t 
dx where m is the mass
0
per unit length of the beam.

11/30/98
Advanced Vibrations
Lets evaluate the virtual quantities, beginning with Kinetic Energy:

L L
∂ 2 ∂y 2
δT = --- ∫ m ( y + δy ) dx – --- ∫ m   dx
1 1
2 ∂t 2 ∂t 
o 0
L
 ∂y  ∂ 
≅ ∫m δy dx
∂t  ∂t 
0 and
t2 t2 L
  ∂y  ∂  
∫ δT dt = ∫  ∫ m  ∂ t   ∂ t δy dx dt
t1 t 0 2
L t2
 ∂  ∂y  
= ∫ m ∫ δy – ẏ˙δy dt dx
 ∂t ∂t 
 
0 t1

11/30/98
Advanced Vibrations
t2 L 0 t2 L 2
t2 ∂y
∫ δT dt = ∫ m [ ẏδy ] t1 dx – ∫ ∫ m
∂t
2
δy dx dt
t1 0 t1 0
Strain Energy:
L 2 L 2 2
1  ∂  2
1 ∂ y 
δV = --- ∫ EI  2 ( y + δy ) dx – --- ∫ EI  2  dx
2 ∂x 0
 2 ∂x  0
L
 ∂2 
≅ ∫ EI y''  2 δy dx
0
∂x 
L
∂ L
= EI  y'' δy + ∫ EI y δy dx
L IV
– EI ( y'''δy )
 ∂x  0
0
0
11/30/98
Advanced Vibrations
The strain energy term becomes:
t2
∫ δV dt =
t1 t2 L
∂ L
∫ EI  y'' δy + ∫ EI y δy dx dt
L IV
– EI ( y'''δy )
 ∂x  0
0
t1 0

11/30/98
Advanced Vibrations
In a similar manner, we find the contribution from Potential Energy of

Loading
t2
∫ δA dt
t1 t2
∂ ∂
= ∫ M0
∂x
δy
0
– ML
∂x
δy
L
– Q 0 δy ( 0 ) + Q L δy ( L ) dt
t1
t2 L
–∫ ∫ p ( x )δy dx dt
t1 0

11/30/98
Advanced Vibrations
We can group terms.
We start with the terms involving δy ( 0 ) , δy ( L ) ,

∂ ∂
δy 0 , and δy
∂x ∂x L
t2
∂ ∂
∫ M0
∂x
δy 0 – ML ∂x
δy L – Q 0 δy ( 0 ) + Q L δy ( L ) dt
t1
t2
∂ L
+∫ EI  y'' δy
L
– EI ( y'''δy ) dt = 0
 ∂x  0
0
t1

11/30/98
Advanced Vibrations
From which we deduce [ Q 0 – E I y''' ( 0 ) ]δy ( 0 ) = 0 .
A geometric boundary condition specifying y ( 0 ) implies that

δy ( 0 ) = 0 . If displacement is not specified there, then
Q 0 = EI y''' ( 0 ) . This is a “natural” boundary condition.
Similar interpretations are made of [ Q L – E I y''' ( L ) ]δy ( L ) = 0 ,
∂ ∂
[ M 0 – E I y'' ( 0 ) ] δy = 0 and [ M L – E I y'' ( L ) ] δy = 0,
∂x 0 ∂x L

11/30/98
Advanced Vibrations
Matching terms in the spacial integral we have
t2 L
∫ ∫ [ mẏ˙ + EI y
IV
– p ]δy ( x, t ) dx dt = 0
t1 0
from which we conclude that
IV
mẏ˙ + EI y = p ( x, t )

11/30/98
Advanced Vibrations
Hamilton’s principle is general and always works, though sometimes it

is hard to evaluate.
In particular, note how Hamilton’s Principle is used to derive the partial

differential governing equations.
Also, we saw how to define the potential energy of loading and to use
that with Hamilton’s principle. We will see that we can also use it in with
Lagrange’s equations.

11/30/98
Advanced Vibrations
Change of Subject.
The following is an introduction to the method of

ASSUMED MODES.

11/30/98
More On Generalized Degrees of Freedom
Distributed Displacement
Advanced Vibrations
Lets consider an Euler Bernoulli

beam simply supported at each
end.
Initially, we assume that all
forces are conservative. We
postulate a displacement L
distribution of the sort
2
x(L – x) x (L – x)
y ( x, t ) = A 1 ( t ) -------------------
2
- + A 2 ( t ) ---------------------
3
-
L L
= A1 ( t ) f 1 ( x ) + A2 ( t ) f 2 ( x )
We shall derive Lagrange equations for the evolution of A 1 ( t ) and

A 2 ( t ) . These are our generalized degrees of freedom.
11/30/98
Advanced Vibrations
Kinetic Energy:
L L
1 1 2
T = ∫ mẏ dx = ∫ m [ Ȧ 1 ( t ) f 1 ( x ) + Ȧ 2 ( t ) f 2 ( x ) ] dx
2
--
- --
-
2 2
0 0
1 2 2
= --- [ ( Ȧ 1 ) I 1 + 2 Ȧ 1 Ȧ 2 I 2 + ( Ȧ 2 ) I 3 ]
2
L L
mL mL
∫ 1 ∫ 1 2
2
where I 1 = m f ( x ) d x = -------- , I 2 = m f ( x ) f x d x = -------
-
30 60
0 0
L
mL
∫ 2
2
and I 3 = m f ( x ) d x = --------
-
105
0

11/30/98
Advanced Vibrations
Strain Energy:
L L
1 1
V = --- ∫ EI ( y'' ) dx = --- ∫ EI [ A 1 ( t ) f 1'' ( x ) + A 2 ( t ) f 2'' ( x ) ] dx
2 2
2 2
0 0
1 2 2
= --- [ ( A 1 ) I 4 + 2 A 1 A 2 I 5 + ( A 2 ) I 6 ]
2
where
L L
EI 4 EI 4
I 4 = ∫ EI f 1'' ( x ) dx = -------- ∫ EI f 2'' ( x ) dx = --------
2 2
3
-, I6 = 3
- and
0 L 0 L
L
EI 2
I5 = ∫ 1
EI f '' ( x ) f 2 '' ( x ) d x = --------
L
3
-
0

11/30/98
Advanced Vibrations
Lagrange Equations:
d  ∂T  ∂T ∂V
– + = Ȧ˙1 I 1 + Ȧ˙2 I 2 + A 1 I 4 + A 2 I 5 = 0
d t  ∂ Ȧ 1 ∂ A 1 ∂ A 1
and
d  ∂T  ∂T ∂V
– + = Ȧ˙1 I 2 + Ȧ˙2 I 3 + A 1 I 5 + A 2 I 6 = 0
d t  ∂ Ȧ 2 ∂ A 2 ∂ A 2
1 1
--- --- ˙
mL 2 EI 4 2 A 1
4 Ȧ 1 + -----
In matrix form: -------- -3 = 0 .
15 1 1 Ȧ˙2 L 2 4 A 2 0
--- ---
4 7
Note that both matrices are symmetric, positive definite.
11/30/98
Generalized Forces, Calculated by
Method of Virtual Work
Advanced Vibrations
Recall that the generalized force associated with the generalized

∂x n
coordinate q r is F r = ∑ F n ⋅
n
∂ qr
We examine the incremental work associated with increments of qr :
∂x n
δW = F r δq r = ∑ F n ⋅ δq r = ∑ F n ⋅ δxn
n
∂ q r n
The generalized force associated with the generalized coordinate qr is
δW
F r = --------
δq r

11/30/98
Advanced Vibrations
Lets consider an Euler Bernoulli

p(x)
beam simply supported at each
end. M0 ML
We consider moments M0 and
ML applied at the ends and a
distributed traction applied L
along the length of the beam.
We postulate a displacement distribution of the sort
2
x(L – x) x (L – x)
y ( x, t ) = A 1 ( t ) -------------------
2
- + A 2 ( t ) ---------------------
3
-
L L
= A1 ( t ) f 1 ( x ) + A2 ( t ) f 2 ( x )
Lets calculate the generalized forces associated with the generalized

coordinates A 1 and A 2 .

11/30/98
Advanced Vibrations
The work done to the structure by the forces acting

through a virtual displacement δy = ( δ A 1 ) f 1 ( x ) , is
L
 
δW = δ A 1  – M 0 f 1' ( 0 ) + M L f 1' ( L ) + ∫ p ( x ) f 1 ( x ) dx
 0

L
 
so F A1 =  – M 0 f 1' ( 0 ) + M L f 1' ( L ) + ∫ p ( x ) f 1 ( x ) dx
 0

F A2 can be calculated similarly.

11/30/98
Homework for Lecture 8
A numerical experiment with linearization
Advanced Vibrations
Many times we have derived the equations for a spring

κ
θ̇˙ +  ----------2 θ + --- sin θ = 0 .
g
reinforced pendulum:
 R  κ
mR
R
2
The linearized form is θ̇˙ + ωL θ = 0 where
θ
2 κ g
ωL = ----------2 + --- . m
mR R
We use the linearized frequency to non-dimensionalize the time

parameter. Define τ = ω L t , define φ ( τ ) = θ ( τ ⁄ ω L ) , and define
g⁄R
α = ---------
2
-
ωL

11/30/98
continued
Advanced Vibrations
2 2 2
dφ dθ 1 dφ
Then
2
= ------
2 2
and
2
+ [ ( 1 – α )φ + α sin φ ] = 0
dτ d t ωL dτ
1. Solve numerically the dimensionless governing equation for the
initial conditions: φ ( 0 ) = π and dφ = 0 over the period

dτ 0
( 0, 6π ) for the three cases: α = 0 , α = 1 ⁄ 2 , and α = 1 .
π
2. Do the same as above but for the initial conditions φ ( 0 ) = --- and
6
dφ = 0
dτ 0
3. Compare and discuss the your results for parts 1 and 2.

11/30/98
Advanced Vibrations
Next Time
Example problems students choice and discussion of

past material.
Discussion of mid-term exam.

11/30/98
Slides of Lecture 9
Advanced Vibrations
Another Example of Linearization
Another Example of Lagrange Equations with

Generalized Degrees of Freedom
In-Class Practice of Problems of Students’s Choice

11/30/98
More Linearization
Advanced Vibrations
Derive the governing

equations for x, y1, & y2
for the car shown. m, J x
Then linearize this set L L
of three equations. y1 y2

11/30/98
More Linearization
Solution
Advanced Vibrations
y2 – y1
Define the angle θ so that sin θ = ---------------- . This is the rotation of the
2L
car about its c.g. Also, define the vertical displacement
y = ( y 1 + y 2 ) ⁄ 2 of the c.g.
m 2 m 2 J 2
Express the kinetic energy: T = ---- ẋ + ---- ẏ + --- θ̇
2 2 2
k 2 k 2
The potential energy is V = mgy + --- y 1 + --- y 2
2 2
Derive the governing equation for each of x , y 1 , and y 2 :
d  ∂T  ∂V
+ = mẋ˙ = 0

d t ∂ ẋ  ∂x
11/30/98
More Linearization
Solution
Advanced Vibrations
∂T ∂T ∂y ∂T ∂θ ∂T ∂x
=    +    +   
∂ ẏ 1  ∂ ẏ   ∂ y 1  ∂ θ̇  ∂ y 1  ∂ ẋ   ∂ y 1
mẏ J θ̇ 1
= ------
- – ------ -------------------------------------
2 2L y – y 2
1 –  ----------------
2 1
 2L 
m ( ẏ 1 + ẏ 2 ) J ( ẏ 2 – ẏ 1 )
= -------------------------- – ------ --------------------------------2-
2 2L y 2 – y 1

1 – ----------------
 2L 

11/30/98
More Linearization
Solution
Advanced Vibrations
d  ∂T  m ( ẏ˙1 + ẏ˙2 ) J ( ẏ˙2 – ẏ˙1 )

= -------------------------- – ------ --------------------------------2-

d t ∂ ẏ 1  2 2L y 2 – y 1

1 – ----------------
2  2L 
2J ( ẏ 2 – ẏ 1 )
– -------------2- ----------------------------------------
2
-
2
( 2L ) 1 –  --------------- y 2 – y 1
-
 2L  and similarly
d  ∂T  m ( ẏ˙1 + ẏ˙2 ) J ( ẏ˙2 – ẏ˙1 )

= -------------------------- + ------ --------------------------------2-

d t ∂ ẏ 2 2 2L y 2 – y 1

1 – ----------------
2
 2L 
2J ( ẏ 2 – ẏ 1 )
+ -------------2- ----------------------------------------
2
-
2
( 2L ) 1 –  --------------- y 2 – y 1
-
 2L 
11/30/98
More Linearization
Solution
Advanced Vibrations
∂V
= 0
∂x
∂V ∂V ∂y ∂V
=    +  
∂ y1  ∂ y   ∂ y 1  ∂ y 1
mg
= ------- + k y 1
2
∂V ∂V ∂y ∂V
=    +  
∂ y2  ∂ y   ∂ y 2  ∂ y 2
mg
= ------- + k y 2
2

11/30/98
More Linearization. Solution
Put it all together
Advanced Vibrations
d  ∂T  m ( ẏ˙1 + ẏ˙2 ) J ( ẏ˙2 – ẏ˙1 )

= -------------------------- – ------ --------------------------------2-

d t ∂ ẏ 1 2 2L y 2 – y 1

1 – ----------------
2
 2L 
2J ( ẏ 2 – ẏ 1 ) mg
– -------------2- ----------------------------------------
2
- = – ------- – k y 1
2
( 2L ) 1 –  --------------- y 2 – y 1 2
-
 2L 
d  ∂T  m ( ẏ˙1 + ẏ˙2 ) J ( ẏ˙2 – ẏ˙1 )

= -------------------------- + ------ --------------------------------2-

d t ∂ ẏ 2 2 2L y 2 – y 1

1 – ----------------
 2L 
2
2J ( ẏ 2 – ẏ 1 ) mg
+ -------------2- ----------------------------------------
2
-
2
= – ------- – k y 2
( 2L ) 1 –  --------------- y 2 – y 1 2
-
 2L 
11/30/98
More Linearization. Solution
Linearize
Advanced Vibrations
m ( ẏ˙1 + ẏ˙2 ) J mg
-------------------------- – ------ ( ẏ˙2 – ẏ˙1 ) = – ------- – k y 1
2 2L 2
and
m ( ẏ˙1 + ẏ˙2 ) J mg
-------------------------- + ------ ( ẏ˙2 – ẏ˙1 ) = – ------- – k y 2
2 2L 2
Please verify the above.
Though it is common to start linearizing earlier in the process,

premature linearization can sometimes lead to error. We shall see this
later.

11/30/98
Another Example of Lagrange Equations
With Assumed Modes
Advanced Vibrations
Consider the cantilevered p(x)

beam shown at right. The
beam has bending
modulus EI, and a spring
of stiffness k at it end.
There is a distributed load EI, m
p(x,t) imposed as shown.
k
We postulate a defomation
espressed as L
2 2
x x (L – x)
y ( x, t ) = A 1 ( t ) -----2 + A 2 ( t ) ---------------------
3
-
L L
= A1 ( t ) f 1 ( x ) + A2 ( t ) f 2 ( x )
We shall derive equations for the evolution of A 1 ( t ) and A 2 ( t ) . Note

that, by construction, A 1 ( t ) and A 2 ( t ) have dimensions of length.
11/30/98
More Lagrange Equations
With Distributed Displacement
Advanced Vibrations
Lets calculate the Kinetic Energy:
L
m
∫ ---2- ( ẏ )
2
T ( A 1, A2 ) = dx
0
1 2
= --- [ A ( t ) I 1 + 2 A 1 ( t ) A 2 ( t )I 2 + A 2 ( t )I 3 ]
2 1
where
L
mL
∫ m( f 1( x))
2
I1 = dx = --------
5
0
L
mL
I2 = ∫ m f 1 ( x ) f 2 ( x ) dx = --------
30
and
0
L
mL
∫ m( f 2( x))
2
I3 = dx = ---------
105
0
Advanced Vibrations
Lets calculate the Potential Energy:
L
EI k
∫ -----2- ( y'' )
2 2
V ( A 1, A2 ) = dx + --- ( y ( L ) )
2
0
1 2
= --- [ A ( t ) I 4 + 2 A 1 ( t ) A 2 ( t )I 5 + A 2 ( t )I 6 ]
2 1
L
∫ EI ( f 1 ′′ ( x ) )
2 2
I4 = dx + k ( f 1 ( L ) )
0 3
where = 4EI ⁄ L + k ,
L
I5 = ∫ EI f 1 ′′ ( x ) f 2 ′′ ( x ) dx + k f 1 ( L ) f 2 ( L ) , and
0 3
= – 2EI ⁄ L
L
∫ EI ( f 2 ′′ ( x ) )
2 2 3
I6 = dx + k ( f 2 ( L ) ) = 4EI ⁄ L
0
Advanced Vibrations
Lets finally calculate the generalized forces due to the applied pressure.
The virtual work associated with a virtual displacement δ A 1 is

L L
 ∂y 
δW 1 = ∫ p ( x, t ) δ A dx = ∫ p ( x, t ) f 1 ( x ) dxδ A1
 ∂ A 1 1
0 0
L
=> F 1 ( t ) = δW 1 ⁄ δ A 1 = ∫ p ( x, t ) f 1 ( x ) d x
0
Similarly
L
F 2 ( t ) = δW 2 ⁄ δ A 2 = ∫ p ( x, t ) f 2 ( x ) d x
0
11/30/98
Advanced Vibrations
Put it all together to find
I 1 I 2 Ȧ˙1 I 4 I 5 A1 F1(t )
+ =
I 2 I 3 Ȧ˙2 I 5 I 6 A2 F2(t )
Plugging in integrated values:
1 1 3
--- ------ F1(t )
EI  4 + --------- – 2 A 1
30 Ȧ˙1 + ----- kL
mL 5 3 
- EI  =
1 1 ˙ A2 F2(t )
------ --------- Ȧ 2 L –2 4
30 105

11/30/98
Next Time
Advanced Vibrations
Review Of Homework From Lecture 8
A Short Quiz On Linearization.
Linearization Of Lagrange Equations
Properties Of Resulting Matrix Equations

11/30/98
Slides of Lecture 10
Advanced Vibrations
Today’s Class:
Review of Homework from Lecture 8

A short quiz on linearization.
Linearization of Lagrange Equations
Properties of Resulting Matrix Equations

11/30/98 1
A numerical experiment with linearization
Advanced Vibrations
Many times we have derived the equations for a spring

κ
θ̇˙ +  ----------2 θ + --- sin θ = 0 .
g
reinforced pendulum:
 R  κ
mR
R
2
The linearized form is θ̇˙ + ωL θ = 0 where
θ
2 κ g
ωL = ----------2 + --- . m
mR R
We use the linearized frequency to dimensionless the time parameter.

g⁄R
Define τ = ω L t , define φ ( τ ) = θ ( τ ⁄ ω L ) , and define α = ---------
2
-
ωL

11/30/98 2
continued
Advanced Vibrations
2 2 2
dφ dθ 1 dφ
Then
2
= ------
2 2
and
2
+ [ ( 1 – α )φ + α sin φ ] = 0
dτ d t ωL dτ
1. Solve numerically the dimensionless governing equation for the
initial conditions: φ ( 0 ) = π and dφ = 0 over the period

dτ 0
( 0, 6π ) for the three cases: α = 0 , α = 1 ⁄ 2 , and α = 1 .
π
2. Do the same as above but for the initial conditions φ ( 0 ) = --- and
6
dφ = 0
dτ 0
3. Compare and discuss the your results for parts 1 and 2.

11/30/98 3
Solution
Using Matlab
Advanced Vibrations
Using state space formulation:
– α sin ( x 2 ) – ( 1 – α )x 2
x = φ̇ , then ẋ =
φ x1
This appears as a function ‘pendu1.m’ provided to matlab:

function yprime = pendu1(t,x)
global alpha;
yprime = [ -alpha.*sin(x(2)) - (1-alpha).*x(2); x(1)];
return;

11/30/98 4
Results
Calls to Matlab
Advanced Vibrations
T = [0:299]*(6*pi)/200; %define time array
%
phi_max = pi; % Initial Condition φ=π
%
alpha = 1;
[t,x] = ode23(’pendu1’, T , [0 phi_max]); %Runge-Kuta Integration
x1 = x;
%
alpha = 0.5;
[t,x] = ode23(’pendu1’, T , [0 phi_max]);
x2 = x;
%
alpha = 0.0;
x3 = x;
%
plot(t,x1(:,2), t,x2(:,2), t,x3(:,2));
print pi -depsc
%
%
phi_max = pi/6;
%
alpha = 1;
x1 = x;
%
alpha = 0.5;
x2 = x;
%
alpha = 0.0;
x3 = x;
%
plot(t,x1(:,2), t,x2(:,2), t,x3(:,2));
print pi6 -depsc
11/30/98 5 /home/djsegal/UNM/VibCourse/slides/
Solution for φ0=π
Advanced Vibrations
3
m
2
1
R
0
−1
κ
−2
−3
θ
−4
0 5 10 15 20 25 30
α=1 α=0.5 α=0.0 When the pendulum starts

displaced nearly vertically,
the restoring force due to
gravity is also near zero.
The more of the restoring force that is due to the torsional string, the
more that the response will appear to be harmonic.
11/30/98 6
Solution for φ0=π/6
Advanced Vibrations
0.6
0.4
κ
0.2
R
0
When the
pendulum starts θ
−0.2 only a little
distance from the m
−0.4 static equilibrium,
the ‘small angle’
−0.6
approximation is very good.
−0.8
0 5 10 15 20In that case, the linearity of
25 30
α=1 α=0.5 α=0.0 the result is nearly

independent of whether the
restoring force is due to the spring or to gravity.
Legitimate linearization requires that the linearized terms be nearly
equal to the corresponding nonlinear terms throughout the
deformation.
11/30/98 7
Short Quiz
Advanced Vibrations
This quiz should take about 15 minutes.

11/30/98 8
Formal Linearization of Lagrange Equations
Derivation of System Matrices
Advanced Vibrations
In the limiting cases of small deflection ( q r « 1 ), small rates q̇ r « 1 ,

the Lagrange equations simplify.
We begin by looking at kinetic energy.
mn ˙ ˙ mn  ∂x˙n ∂x˙n
T = ∑ ------ x n ⋅ x n = ∑ ------ ∑ ∑  ⋅  q̇ r q̇ s
2 2 r s  ∂ q̇ r ∂ q̇ s
n n
Again, we make use of one of the core observations in the derivation of
˙
∂x n ∂x n
the Lagrange equations: =
∂ q̇ r ∂ qr
mn  ∂x n ∂x n 1
T = ∑ ------ ∑ ∑ ⋅ q̇ r q̇ s = ∑ ∑ --2- q̇r q̇s M̂ rs ( { q } )

2 r s ∂ qr ∂ qs 
n r s

11/30/98 9
Advanced Vibrations
In the limit of small displacement about the configuration of stable

s
equilibrium { q } , the mass matrix becomes
m n   ∂x n  ∂x n 
= M̂ rs ( { q } ) = ∑ ------ 
s
M rs ⋅ .
2   ∂ q r {q }
s  ∂ q s
{ q }
s
n
s
Note that M rs = M̂ rs ( { q } ) is still symmetric.
Important: In the limit of small displacement,
2
1 ∂T
T = ∑ ∑ 2 r s rs rs ∂ q̇r ∂q̇s
--
- q̇ q̇ M & M =
r s

11/30/98 10
Advanced Vibrations
Observe that the mass matrix is positive semi-definite. Consider an

t
array of generalized speeds { q̇ } . The corresponding kinetic energy is
1
∑ 2 rs r q̇s
t t t
T ( { q̇ } ) = --
- M q̇
r, s
1 t t m n   ∂x n  ∂x n 
= ∑ --- q̇ r q̇ s ∑ ------  ⋅ 
  ∂ q r  ∂ q s
{ q }
r, s
2 2 {q }
s s
n
mn   ∂x n  ∂x n  t t
= ∑ ------ ∑ ∑  ⋅  q̇ r q̇ s
2 r s   ∂ q r {q }
s  ∂ q s
{ q }
s
n
m n ˙t ˙t ˙t
= ∑ ------ y n ⋅ y n ≥ 0 where y n =  ∂x n
∑  ∂ qr
t
q̇ r
2 r {q }
s
n

11/30/98 11
Advanced Vibrations
The meaning of the mass matrix.
Consider the “acceleration” force

d  ∂T  d ∂ 1
= --- ∑ ∑ M st q̇ s q̇ t = ∑ M rs q̇˙s .
d t  ∂ q r d t  ∂ q̇ r 2 s t 
s
Next, consider column vector of generalized accelerations that are all

T
zeros except for a “1” on the r’th row: { q̇˙} r = 0 . 1 . . 0 . The
resulting “acceleration forces” seen by each of the other generalized
degrees of freedom are the r’th column of M .

11/30/98 12
Example of Mass Matrix
Advanced Vibrations
Lets consider a
L1 L2 L3 L4 structure similar to
the disk-shaft
system in
E, Js Meirovitch.
Ignoring the
kinetic energy of
the shaft, the
J1, θ1 J2, θ2 J3, θ3 kinetic energy of
the systems is
1 2 1 2 1 2
T = --- J 1 θ̇ 1 + --- J 2 θ̇ 2 + --- J 3 θ̇ 3 .
2 2 2
2 J1 0 0
∂T
Our mass matrix is [ M ij ] = = 0 J2 0
∂ θ̇ i ∂θ̇ j
0 0 J3
11/30/98 13
Another
Example of Mass Matrix
Advanced Vibrations
Consider a
cantilevered beam for
which we postulate a
displacement field EI, m
with two free
generalized L
coordinates:
y ( x, t ) = A 1 ( t ) f 1 ( x ) + A 2 ( t ) f 2 ( x )
The kinetic energy in the beam will be
L
m 2
T ( A 1, A 2 ) = ∫ ---- ( ẏ ) dx
2
0
1 2
= --- [ Ȧ ( t ) I 1 + 2 Ȧ 1 ( t ) Ȧ 2 ( t )I 2 + ( 2 Ȧ 2 ( t ) )I 3 ]
2 1
L L
∫ m( f 1( x)) ∫ m( f 2( x))
2 2
where I 1 = dx , I 3 = dx
0 0
L
and I 2 = ∫ m f 1 ( x ) f 2 ( x ) dx
0
Another Example of Mass Matrix
Continued
Advanced Vibrations
2
∂T I1 I2
Our mass matrix is [ M ij ] = = m
∂ Ȧ i ∂ Ȧ j I2 I3

11/30/98 15
Advanced Vibrations
Consider the potential energy in the vicinity of a configuration of stable

equilibrium. In an equilibrium configuration, the sum of the potential
∂V
forces in each direction is zero 0 = – .
∂ qr
Taylor series expansion for V is

2
V ( { q } ) = V 0 + --- ∑ ∑ ∂ V
1
∆q r ∆q s + H.O.T. .
2 r s ∂ q ∂q
r s { qs }
2
We define the stiffness matrix to be K rs = ∂ V .
∂ q r ∂q s {q }
s
Note that, by construction, K is symmetric.

11/30/98 16
Advanced Vibrations
Potential energy in terms of Stiffness. In the limit of small strains,

1
V = V 0 + --- ∑ ∑ K rs q r q s
2 r s
where we have redefined the generalized coordinates to be zero at the
equilibrium configuration. Usually, we set the datum so that V 0 = 0 .
Because stability requires that potential energy be at a local minimum,
∑ ∑ K rs qr qs ≥ 0 for all { q } . This is an assertion that the stiffness

r s
matrix is positive semi-definite.

11/30/98 17
Advanced Vibrations
The meaning of the Stiffness matrix.
Consider the generalized force

∂ 1
Fr = – --- ∑ ∑ K st q s q t = – ∑ K rs q s .
∂ qr  2 s t 
s
Next, consider column vector of generalized displacement that are all

T
zeros except for a “1” on the r’th row: { q } r = 0 . 1 . . 0 . The
resulting force seen by each of the other generalized degrees of
freedom are the r’th column of K .

11/30/98 18
An Example Problem of Involving the
Stiffness Matrix
Advanced Vibrations
Lets consider a
L1 L2 L3 L4 structure similar to
the disk-shaft
system in
G, Js Meirovitch.
J1, θ1 J2, θ2 J3, θ3
The strain energy in the shaft is computed in terms of the differences in

the rotations at the disks
2 2 2 2
1 θ ( θ – θ ) ( θ – θ ) θ
2 1
V = --- GJ s -----1 + ----------------------- 3 2
- + -----3 .
- + -----------------------
2 L1 L2 L3 L4

11/30/98 19
An Example Problem of Involving the
Stiffness Matrix
Advanced Vibrations
Our stiffness matrix is
1 1 1
----- + ----- – ----- 0
L1 L2 L2
2
∂V 1 1 1 1
[ K ij ] = = GJ s – ----- ----- + ----- – -----
∂ θ i ∂θ j L2 L2 L3 L3
1 1 1
0 – ----- ----- + -----
L3 L3 L4
Note that the i, j th element of K is the torque felt on disk i due to a

unit rotation imposed on disk j .

11/30/98 20
Another Example Involving the
Stiffness Matrix
Advanced Vibrations
Consider a
cantilevered beam for
which we postulate a
displacement field EI, m
with two free
generalized L
coordinates:
y ( x, t ) = A 1 ( t ) f 1 ( x ) + A 2 ( t ) f 2 ( x )
The strain energy in the beam will be
L
EI
V ( A 1, A 2 ) = ∫ ------ ( y′′ ) dx
2
2
0
1 2 2
= --- [ A ( t ) I 4 + 2 A 1 ( t ) A 2 ( t )I 5 + A 2 ( t ) I 6 ]
2 1
L L
∫ EI ( f 1 ′′ ( x ) ) ∫ EI ( f 2 ′′ ( x ) )
2 2
I4 = dx , I 6 = dx
0 0
L
and I 5 = ∫ m f 1 ′′ ( x ) f 2 ′′ ( x ) dx
0
Another Example of Mass Matrix
Continued
Advanced Vibrations
2
∂V I4 I5
Our stiffness matrix is [ K ij ] = = EI
∂ Ai ∂ A j I5 I6

11/30/98 22
Matrix Multiplication
A short review
Advanced Vibrations
T T
Say a and b are column vectors of length N , then b a = a b are
N
defined to be ∑ ar br . This is the inner product of algebraic vectors.
r=1
Say A is an N by N matrix, then the matrix vector product Aa is

N
 
defined so that the r th element of Aa is  ∑ Ars as 
î s=1 
N N
 
Combining the above two definitions: a Ab =  ∑ ∑ A rs a r b s
T
r = 1 s = 1 

11/30/98 23
Matrix Multiplication
A short review
Advanced Vibrations
The product of two matrices is defined in a similar way. Say B is also

an N by N matrix also, then the product AB is defined as
N
( AB ) rs = ∑ Art Bts
t=1
The transpose of a matrix is that which is obtained by reversing the

T
order of the indices: ( A ) rs = ( A ) sr . The transpose of products is
N
∑
T T T
found to be ( ( AB ) ) rs = ( AB ) sr = A st B tr = ( B A ) rs . In
t=1
T T T
matrix notion: ( AB ) = B A

11/30/98 24
Flexibility Matrix
(Lets start using matrix notation)
Advanced Vibrations
In those cases where there are no rigid body motions, the stiffness
matrix is nonsingular, and we can define a flexibility matrix
–1
a = K .
In this case we observe that since F = – Kq , q = – aF .

The strain energy may now be expressed in terms of force
1 T 1 T 1 T
V = --- q Kq = --- ( – aF ) K ( – aF ) = --- F aF
2 2 2
Sometimes vibration problems are formulated in terms of the flexibility
matrix, though the use of displacement-based finite elements makes
flexibility formulations decreasingly popular.
Discussion of the flexibility matrix is presented here for completeness.
We shall do little more with it in this class.

11/30/98 25
Matrix Transformations
Advanced Vibrations
Say that we have derived the following matrix representation of a linear

system: Mq̇˙ + Kq = F where the column vector F contains all
externally applied load.
Now, say we select another set of generalized coordinates with which
we can specify the first set: q = Tβ . Lets re-derive our governing
equations in terms of β.
Kinetic Energy:
1 T 1 T 1 T T
T = --- q̇ Mq̇ = --- ( T β̇ ) M ( T β̇ ) = --- β̇ ( T MT )β̇
2 2 2
1 T 1 T 1 T T
Strain Energy: V = --- q Kq = --- ( Tβ ) K ( Tβ ) = --- β ( T KT )β
2 2 2
Potential Energy of Loading:
T T T T
A = – q F = – ( Tβ ) F = – β ( T F )
11/30/98 26
Matrix Transformation
Lagrange Equation in New System
Advanced Vibrations
d  ∂T  ∂T ∂V
0
∂A
 – + = –
d t  ∂ β̇ r ∂ β r ∂ β r ∂ βr
becomes M̃β̇˙ + K̃ β = F̃
T T T
where M̃ = T MT , K̃ = T KT , and F̃ = T F
Note that the mass and stiffness matrices remain symmetric and at
least positive semi-definite.
T
If T is a nonsingular, square matrix, the transformation T AT is called
a congruence transformation of A.

11/30/98 27
Advanced Vibrations
We often use such transformations

when introducing constraints to a
system. Here we consider two
independent pendulums. The kinetic
and potential energies are
m1 2 2 m2 2 2
T = ------ R 1 θ̇ 1 + ------ R 2 θ̇ 2 and R1
2 2 R2
V = – mgR 1 cos θ 1 – mgR 2 cos ( θ 2 ) F1 F2
m1
The potential energy of loading is m2
A = –F 1 R1 θ1 – F 2 R2 θ2
The linearized equations of motion are
2
m1 R1 0 θ̇˙1 mgR 1 0 θ1 F 1 R1
+ =
0
2
m2 R2 θ̇˙2 0 mgR 2 θ 2 F 2 R2

11/30/98 28
Lets add a constraint
Advanced Vibrations
Consider the effect of connecting the

two masses with a massless rod. This
has the effect of requiring
R 1 θ 1 = R 2 θ 2 , permitting us to write
θ1 1
= θ1 . R1
θ2 R1 ⁄ R2 R2
F1 F2
m1
1 m2
Define T = . Then
R1 ⁄ R2
2
m1 R1 0 1
M̃ = 1 R 1 ⁄ R 2 = ( m 1 + m2 )R 12
0
2
m2 R2 R1 ⁄ R2

11/30/98 29
With a constraint
Advanced Vibrations
m 1 gR 1 0 1
K̃ = 1 R 1 ⁄ R 2
0 m 2 gR 2 R1 ⁄ R2
2
R1
= m gR + m g -----
1 1 2 R-
2
The stiffness matrix is
and the applied force is
F 1 R1
F̃ = 1 R 1 ⁄ R 2 = R1 ( F 1 + F 2 )
F 2 R2

11/30/98 30
With a constraint
Advanced Vibrations
The resulting equation for θ 1 is
2
 R 1
[ ( m1 + m2 )R 12 ]θ̇˙1 +  m 1 gR 1 + m 2 g ------ θ 1 = R 1 ( F 1 + F 2 )
 R 2
The addition of a constraint reduced the number of active degrees of

freedom by one.

11/30/98 31
Next Time
Advanced Vibrations
Short Orientation
LAB TOUR

11/30/98 32
Advanced Vibrations
Today’s Class:
Eigen Analysis
This topic brings together our experience with matrix transformations

and with integrating systems of equations to motivate the search and
use of generalized eigen solutions.

11/30/98 1
Integration of Linearized Equations
Advanced Vibrations
Say that we have derived governing equations for a vibrating system

M ẋ˙ + Kx = F
To integrate this numerically, we might pose it in a state space
˙
formalism:
M 0 ẋ = – K ẋ + F .
0 M x –K x 0
This can be rearranged to ẏ = – A y + F̃

–1
where A = M 0 K and F̃ = F .
0 M –K 0
We have integrated these equations numerically in the past without a lot
of difficulty, but this approach can be impractical for large problems.

11/30/98 2
Integration of Linearized Equations
Advanced Vibrations
For large problems, integrating ẏ = – A y + F̃ , which involves

inverting the mass matrix once, but performing large matrix-vector
multiplies at every time-step can be daunting. We would like to reduce
the size of the algebraic problem that must be solved at each time step.
Modal analysis lets us do that.
Further, modal analysis provides additional advantages:

• providing a compact representation of the system dynamics
• facilitating modal truncation for model reduction
• providing a simple route to calculating transfer functions

11/30/98 3
A Digression Why Do We Not Worry
About The Singularity Of M?
Advanced Vibrations
Usually the mass matrix is constructed to

x1 x2 x3
be nonsingular, but it can happen that
one introduces massless degrees of m1 m2
freedom. Consider the following k1 k2 k3 k4
structure. It has two masses, but three
degrees of freedom. The node connecting the two springs is massless.
m1 2 m2 2
The kinetic energy is T = ------ ẋ 1 + ------ ẋ 2 . The mass matrix is
2 2
2 m1 0 0
∂T
M ij = so M = 0 0 0 , which is singular. We
∂ x i ∂x j
0 0 m2
T
demonstrate singularity by observing that a Ma = 0 when
T
a = 0 10

11/30/98 4
Why Do We Not REALLY Worry
Advanced Vibrations
We would really like to condense the degree of freedom x 2 out of the

problem. With the aid of the potential energy, we obtain an equation for
x 2 in terms of x 1 and x 3 .
k1 2 k2 2 k3 2 k4 2
V = ----- x 1 + ----- ( x 2 – x 1 ) + ----- ( x 3 – x 2 ) + ----- x 3 .
2 2 2 2
Equilibrium of the node at x 2 requires that
∂V
0 = = – k 1 x 1 + ( k 1 + k 2 )x 2 – k 3 x 3
∂ x2
k1 k3
from which we can solve for x 2 : x 2 = ---------------- x 1 + ---------------- x 3 .
k1 + k2 k2 + k3
(Compare the above to the second row of the stiffness matrix.)
11/30/98 5
Why Do We Not REALLY Worry
Advanced Vibrations
We use the above constraint to reduce the size of our system:
x1 1 0
x1 k1 k3
x2 = T where T = ---------------- ---------------- .
x3 k1 + k2 k2 + k3
x3
0 1
˙˙
T x1 T x1 T
Our new governing equations are T MT + T KT = T F
x3 x3
where M , K , and F are the mass matrix, stiffness matrix, and applied
force of the original problem.
In the rest of this development, we shall assume that the mass matrix is
positive definite.
11/30/98 6
Eigen Analysis
Finding a de-coupled model for the system
Advanced Vibrations
Now that we know a lot about the effects of performing a linear

transformation of our variables, we can pose the following question:
Given a system defined by mass matrix M , mass matrix K , and dis-
placement variables x , does there exist a transformation matrix T a
T T
new set of variables x = Tβ and such that T MT and T KT are
each diagonal?
The answer will be YES, but we shall have to prove it by demonstrating

the transformation matrix that does the trick. We shall find that this
matrix is one whose columns are the eigenvectors of the M , K system.
An eigen-pair (eigen solution) of the generalized eigen-problem of M

and K are an eigenvalue λ and an eigenvector x such that
– λMx + Kx = 0
An eigen-pair is often represented as ( λ, x ) .
11/30/98 7
Eigen Analysis
Does a solution exist?
Advanced Vibrations
How do we know that even one eigen-solution exists? Consider the

following extreme value problem.
From among all column vectors X that are not identically zero, find the
T
X1 K X1
column vector X 1 such that R 1 ( X 1 ) = ---------------------- is at a minimum.
T
X1 M X1
T
Let λ 1 = R 1 ( X 1 ) then for any y such that y M y > 0 ,
T
d ( X 1 + αy ) K ( X 1 + αy )
---------------------------------------------------------
- = 0.
d α ( X + αy ) M ( X + αy ) T
1 1 α=0

11/30/98 8
Eigen Analysis
Does a solution exist?
Advanced Vibrations
After a bunch of linear analysis, and exploiting the symmetry of K and

M , we find that the above condition requires that
T
y ( K X 1 – λ 1 M X 1 ) = 0 for all nonzero y . From which we conclude
that K X 1 – λ 1 M X 1 = 0 . This proves the existence of the lowest
eigenvalue.
The existence of the next higher eigenvalue is proven by considering a
slightly modified minimization: Let
T
X KX . An evaluation of the appropriate
λ 2 = R 2 ( X 2 ) = min ----------------
T
0 < X MX X MX
T
T
X 1 MX = 0
derivative will show that K X 2 – λ 2 M X 2 = 0 . In this manner, we

demonstrate the existence of as many eigen-solutions as the dimension
of K and M .
11/30/98 9
Eigen Analysis
Orthogonality where λ a ≠ λ b
Advanced Vibrations
Following is the standard proof that the eigenvectors are orthogonal, or

can be selected to be orthogonal to each other with respect to the mass
and the stiffness matrix. Say:
• that λ a and x a are an eigen pair to the M and K system
• that λ b and x b are another eigen pair to the M and K system
• and that λa ≠ λb
Then λ a M x a – K x a = 0 and λ b M x b – K x b = 0 .

11/30/98 10
Eigen Analysis
Orthogonality where λ a ≠ λ b
Advanced Vibrations
λ a M x a – K x a = 0 and λ b M x b – K x b = 0 .
Observe that:
T T
1. λa xb M xa – xb K xa = 0
T T
2. λb xa M xb – xa K xb = 0
Subtracting one from the other, and exploiting symmetry of M and K ,

T T
we see that ( λ a – λ b )x a M x b = 0 . Since λ a ≠ λ b , x a M x b = 0 .
T
From either 1.) or 2.), we see also that xa K xb = 0 .

11/30/98 11
Eigen Analysis
Orthogonality where λ a = λ b
Advanced Vibrations
Say:
• that λ a and x a are an eigen pair to the M and K system
• that λ b and x b are another eigen pair to the M and K system
• and that λ a = λ b = λ but that x a ≠ x b
Then λ a M x a – K x a = 0 and λ b M x b – K x b = 0 .
T T
Define x a = x a – cx b where c = ( x a M x b ) ⁄ ( x b M x b ) .
T
Note that ( λ, x a ) is still an eigen-solution, that x a M x b = 0 , and that
T
xa K xb = 0.
We have a new, equivalent, but orthogonal pair of eigen vectors.
11/30/98 12
Eigen Analysis
Orthogonality and Normalization
Advanced Vibrations
The equation λMx – Kx = 0 leaves the vector x undetermined in

magnitude. If ( λ, x ) works, then so does ( λ, αx ) for any scalar α .
T
We normalize the eigenvectors so that x k M x k = 1 for each eigen
vector x k . This is Mass Normalization. These vectors are orthonormal
with respect to the mass matrix.
T
Note that because λk M xk – K xk = 0 , xk K xk = λk

11/30/98 13
Eigen Analysis
Significance of Orthogonality
Advanced Vibrations
Lets consider the solution x ( t ) to M ẋ˙( t ) + Kx ( t ) = F ( t ) and

N
expand x(t ) = ∑ βk ( t )xk where xk is the k’th normalized
k=1
eigenvector, corresponding to the k’th eigen value λk .
We note that this expression is actually a change of variables
{ x } = P { β } where the columns of P are the normalized
eigenvectors. See what this does to the mass and stiffness matrices.
T T
P MP = I and P KP = Λ
where I is the N by N identity matrix and Λ is the diagonal matrix of

eigenvalues. Because M and K are positive definite, 0 < λ k for each
k.
11/30/98 14
Eigen Analysis
Advanced Vibrations
The matrix P is called the MODAL MATRIX.

If the eigenvalues are all unique, and the columns are ordered
according to the order of the eigen values, then the modal matrix is
unique.
If there are repeated eigenvalues, then the columns corresponding to
repeated eigenvalues in one version of the modal matrix must span the
space of the corresponding columns in any other version.
The β k are called MODAL COORDINATES. We shall use them quite a lot
as generalized coordinates.

11/30/98 15
Eigen Analysis
System Solution
Advanced Vibrations
In modal coordinates, the governing equation becomes
T
β̇˙ + Λβ = P F = F̃
We can solve for these modal coordinates, β k , individually. We can
solve for them numerically, and sometimes we can achieve analytic
solutions.
The solution in physical coordinates can be recovered from
x ( t ) = P β( t )
Lets examine free vibration: β̇˙ + Λβ = 0 . This has solution

 i λk t 
β k ( t ) = Re  A k e  . We now identify the eigenvalues in terms of
î 
2
natural frequencies: λk = ωk .
11/30/98 16
Eigen Analysis
Another Motivation
Advanced Vibrations
We have motivated the search for eigen-solutions by looking for the

congruence transform that would diagonalize both the mass and the
stiffness matrix.
We can also motivate the search for eigen-pairs by asking ourselves:

Given the system M ẋ˙ + Kx = 0 , with some set of initial condi-
N
tions, can we express the solution as x(t ) = ∑ βk ( t )xk for some
k=1
set of basis vectors xk ?
The answer is YES, and we have demonstrated the character of the
eigenvectors x k in the proceeding slides.

11/30/98 17
Eigen Analysis
Another Motivation
Advanced Vibrations
N
Given M ẋ˙ + Kx = 0 , postulate a solution x ( t ) = ∑ βk ( t )xk
k=1
N
From which ∑ β̇˙k ( t )M xk + βk ( t )K xk . Consider the case where
k=1
β m ( 0 ) = 0 for all m ≠ k , then β̇˙k ( t )M x k + β k ( t )K x k = 0 .
 iω k t 
We expect a solution β k ( t ) = Re  A k e  , which yields
î 
2
– ω k M x k + K x k = 0 : the algebraic eigenvalue problem.
11/30/98 18
Example Problem
Advanced Vibrations
Consider the system of two identical, freely

-hanging pendula shown. There are two
degrees of freedom, so the mass an θ1 θ2
stiffness matrices will each be 2x2, and
there will be exactly two eigen-pairs. R R
Lets first derive the linearized governing k
m 2 2 m 2 2
equations: T = ---- R θ̇ 1 + ---- R θ̇ 2 , m m
2 2
k 2 2
V = – m gR cos θ 1 – m gR cos θ 2 + --- R ( sin θ 2 – sin θ 1 )
2
mR
2
0 θ̇˙1 mgR +
2
kR – k R θ1
2
to find + = 0
θ̇˙2 mgR + kR θ 2
2 2 2
0 mR –k R

11/30/98 19
Example Problem
Advanced Vibrations
 2 θ1 
Lets look for solutions  ω ,  to the eigen-problem:
 θ2 
mR
2
2
0 θ1 mgR + kR
2
– k R
2 θ1
–ω + = 0
2 θ 2 2 θ
0 mR 2 –k R mgR + kR 2
In dimensionless form, this is
1
2 0 θ1 1 + γ – γ θ1
–α + = 0
0 1 θ2 –γ 1 + γ θ2
2 2
where α = ω R⁄g and γ = Rk ⁄ mg

11/30/98 20
Example Problem
Advanced Vibrations
Combine the two matrices
1+γ–α
2
–γ θ1
= 0
–γ 1+γ–α
2 θ2
A non-trivial solution requires the determinant to be zero:

2 4
( 1 + 2γ ) – 2α ( 1 + γ ) + α = 0 .
2
This yields solutions α = 1 + γ ± γ = 1, 1 + 2γ .

11/30/98 21
Example Problem
Advanced Vibrations
We substitute these back into the matrix equation to solve for θ1 & θ2 :
γ – γ θ 1 = 0 for α 2 = 1 , from which we see

–γ γ θ2
θ 1 = θ 2 . This is the pendula swinging in unison.
– γ – γ θ 1 = 0 for α 2 = 1 + 2γ , from which we

–γ –γ θ2
see θ 1 = – θ 2 . This is the pendula swinging counter
to each other.

11/30/98 22
Example Problem
Advanced Vibrations
Lets normalize these modes with respect to the mass matrix to

calculate our transformation matrix:
1
-------
1 0 1 = 2 => we must scale this mode: x = 2
11 1
01 1 1
-------
2
1
-------
1 0 1 = 2 => we must scale this mode: x = 2
1 –1 2
0 1 –1 1
– -------
2

11/30/98 23
Example Problem
Advanced Vibrations
If we guess the eigen-modes first, then we can deduce the eigenvalues

from the eigen equation.
1 1+γ–α
2
–γ θ1
Given x 1 ∼ , and = 0 , we find
1 –γ 1+γ–α
2 θ2
2
1–α = 0 =>
2
α1 = 1 .
2 0
1–α
2
Similarly, given x 2 ∼
1 , we find
1 + 2γ – α = 0 =>
–1 2 0
– 1 – 2γ + α
2
α 2 = 1 + 2γ .
11/30/98 24
Homework
Due Monday
Advanced Vibrations
L1 L2 L3 L4 Lets consider a structure similar to the

disk-shaft system in Meirovitch.
E, Js
J1, θ1 J2, θ2 J3, θ3

Ignoring the kinetic energy of the shaft:
• Derive the linear equations for this system
• Write the matrix eigen-equation
• Assuming thatJ 1 = J 2 = J 3 and that

L 1 = L 2 = L 3 = L 4 , write a dimensionless version of the
matrix eigen equation.
• Using Matlab, find the eigen solutions.
• Plot modes as in Fig 4.1.
• Write the modal matrix. Remember to normalize w.r.t. mass

matrix.
11/30/98 25
What We Might Have Learned Up to Now
Advanced Vibrations
Governing Equations For Assembled Systems

Complex Representation For Force And Displacement In Steady State
RMS Quantities And Energy Dissipation In Steady State
Impulse Response Functions
Laplace Solution To Simple Systems
Lagrange Equations
Calculation Of Generalized Forces By Virtual Work And Potential
Energy Of Loading
Assumed Modes Method
Conservation Of Mechanical Energy
Linearization
A Little About Eigen-analysis

11/30/98 26
Advanced Vibrations
Today’s Class:
Review of Eigen-Analysis
Solution to Homework of Lecture 11
More Eigen-Analysis

11/30/98 1
Recall
Advanced Vibrations
In Lecture 11, we found that the governing equations M ẋ˙ + Kx = F

are diagonalized by the transformation x = Pβ , where the columns of
2
P are the eigenvectors v n solving ( – ω n M + K )v n = 0 .
Those eigenvectors are mutually orthogonal with respect to the Mass
and Stiffness matrices. Further, we scale those columns so that they
are orthonormal with respect to the mass matrix:
T
P MP = I
2
In that case,
T
P KP = ωn
When a structure is correctly modeled by a finite number of discrete

2 T T
degrees of freedom, then ω 0 = min ( ( x Kx ) ⁄ ( x Mx ) )

11/30/98 2
Homework
Due Today
Advanced Vibrations
L1 L2 L3 L4 Lets consider a structure similar to the

disk-shaft system in Meirovitch.
G, Js
J1, θ1 J2, θ2 J3, θ3

Ignoring the kinetic energy of the shaft:
• Derive the linear equations for this system
• Write the matrix eigen-equation
• Assuming thatJ 1 = J 2 = J 3 and that

L 1 = L 2 = L 3 = L 4 , write a dimensionless version of the
matrix eigen equation.
• Using Matlab, find the eigen solutions.
• Plot modes as in Fig 4.1.
• Write the modal matrix. Remember to normalize w.r.t. mass

matrix.
11/30/98 3
Homework
Solution
Advanced Vibrations
In Lecture 10, we derived stiffness matrix from the potential energy

2 2 2 2
1 θ ( θ – θ ) ( θ – θ ) θ
2 1
V = --- GJ s -----1 + ----------------------- 3 2
- + -----3 .
- + -----------------------
2 L1 L2 L3 L4
1 1 1
----- + ----- – ----- 0
L1 L2 L2
2
∂V 1 1 1 1
[ K ij ] = = GJ s – ----- ----- + ----- – -----
∂ θ i ∂θ j L2 L2 L3 L3
1 1 1
0 – ----- ----- + -----
L3 L3 L4
4
Note that J s is a polar moment of inertia of the area ( πR ⁄ 2 ). The
units of polar moment of inertia are Length4.
11/30/98 4
Homework
Solution
Advanced Vibrations
We derived the mass matrix from the kinetic energy
1 2 1 2 1 2
T = --- J 1 θ̇ 1 + --- J 2 θ̇ 2 + --- J 3 θ̇ 3 .
2 2 2
2 J1 0 0
∂T
[ M ij ] = = 0 J2 0
∂ θ̇ i ∂θ̇ j
0 0 J3
Here J 1 , J 2 , and J 3 are mass moments of inertia. For a disk,

J M = ρJ A ∆H where ρ is mass density, J A is polar moment of
inertia, and ∆H is the disk thickness. The units of J M is are
Mass*Length2.

11/30/98 5
Homework
Solution
Advanced Vibrations
The governing equation is
1 1 1
----- + ----- – ----- 0
L1 L2 L2
J1 0 0 θ̇˙1 θ1
1 1 1 1
0 J2 0 θ̇˙2 + GJ s – ----- ----- + ----- – ----- θ2 = 0
L2 L2 L3 L3
0 0 J 3 θ̇˙3 θ3
1 1 1
0 – ----- ----- + -----
L3 L3 L4

11/30/98 6
Homework
Solution
Advanced Vibrations
L 1 = L 2 = L 3 = L 4 = L and that
Assuming that
J1 = J2 = J3 = JM ,
1 0 0 θ̇ 1 GJ 2 – 1 0 θ 1
˙
s
JM 0 1 0 θ̇ 2
˙ + --------
- – 1 2 – 1 θ2 = 0
L
0 0 1 θ̇˙ 0 –1 2 θ3
3
The corresponding algebraic eigen-problem is
1 0 0 θ 1 GJ 2 – 1 0 θ 1
2 s
–ω J M 0 1 0 θ2 + --------
- – 1 2 – 1 θ2 = 0
L
0 0 1 θ3 0 –1 2 θ3

11/30/98 7
Homework
Solution
Advanced Vibrations
2 2JML
There is only one dimensionless group to form: Let α = ω ---------- .
GJ s
Our algebraic equation is now
 1 0 0 2 – 1 0  θ1
 2 
 – α 0 1 0 + –1 2 –1  θ2 = 0
 
 001 0 –1 2  θ3
The eigen-solutions are
 1⁄2   1⁄ 2   1⁄2 
     
 2 – 2, 1 ⁄ 2  ,  2, 0  ,  2 + 2, – 1 ⁄ 2 
     
 1⁄2   –1 ⁄ 2   1⁄2 
11/30/98 8
Homework
Solution
Advanced Vibrations
1⁄2 1⁄ 2 1⁄2
The MODAL MATRIX is P = 1⁄ 2 0 ( –1 ⁄ 2 )
1 ⁄ 2 –1 ⁄ 2 1⁄2
T T
Test diagonalization: P MP = P IP = I
2– 2 0 0
T
and P KP = 0 2 0
0 0 2+ 2

11/30/98 9
Homework
Solution
Advanced Vibrations
Mode Shapes
Mode 1
Lowest Energy
Mode 2
Mode 3
Highest Energy

11/30/98 10
Systems with Rigid Body Modes
Advanced Vibrations
Because we know that the modes are orthogonal with respect to the
mass matrix, we may use known eigenvectors to reduce the size of the
remaining eigen-problem. This is particularly useful for systems with
easily recognized rigid body modes.
Consider the system shown. The
k k
m m m m
mass matrix is M = m
m
k –k
and the stiffness matrix is K = – k 2k – k . We see right away that a
–k k
T
rigid body mode is v 1 = 11 1 . (Observe K v 1 = 0 .)

11/30/98 11
Advanced Vibrations
We know that there are two remaining eigen modes and that they are
orthogonal to the rigid body mode:
v 1T Mx = m 1 1 1 x = m ( x 1 + x 2 + x 3 ) = 0 . This provides a
constraint that can be used to reduce the order of the system.
x 2 = – ( x 1 + x 3 ) defines a transformation matrix
x1 1 0 x
x2 = – 1 – 1 1 = Tx .
x3
x3 0 1

11/30/98 12
Advanced Vibrations
M = T MT = m 2 1 = T KT = k 5 4
T T
Resulting in and K
12 4 5
The eigenvectors of this reduced system are
1⁄ 2
v2 = 1 ⁄ 2 => v2 = T v2 = – 2 and
1⁄ 2
1⁄ 2
–1 ⁄ 2
v 3 = – 1 ⁄ 2 => v 3 = T v 3 = 0
1⁄ 2
1⁄ 2

11/30/98 13
Eigen Example
Another Assumed Modes Problem
Advanced Vibrations
Lets consider a simply

supported beam. We shall
approximate the deformed
shape by a low-order
polynomial satisfying the
geometric boundary conditions y ( 0, t ) = y ( L, t ) = 0 :
N k N
 x  x
y ( x, t ) = ∑ A k ( t ) --- 1 – -- =
 L  k ∑ Ak ( t ) f k ( x )
k=1 k=1
We shall use the A k ( t ) as generalized coordinates and solve for the

eigenmodes and frequencies in terms of them.
It is important to note that these polynomials can at best be

approximations for the vibration modes of this structure.

11/30/98 14
Eigen Example
Mass Matrix
Advanced Vibrations
The kinetic energy is

L N N 1
m 2 mL
T = ∫ ---2- ẏ dt = -------- ∑ ∑ Ȧi Ȧ j ∫ f i ( sL ) f j ( sL ) ds
2
0 i=1j=1 0
N N
mL 2
= -------- ∑ ∑ Ȧ i Ȧ j --------------------------------------------------------------------------
2 (i + j + 1)(i + j + 2)(i + j + 3)
i=1j=1
T
= ( mL ) Ȧ M Ȧ
2
where M ij = --------------------------------------------------------------------------
(i + j + 1)(i + j + 2)(i + j + 3)

11/30/98 15
Eigen Example
Mass Matrix
Advanced Vibrations
The potential energy is

L
EI 2
V = ∫ -----2- y'' dt
0 2 2
1
∂ f i  ∂ f j
N N
EI 
= --------4- ∑ ∑ A i A j ∫  2 ( sL )  2 ( sL ) ds
2L i = 1 j = 1 0
∂s  ∂s 
EI T
= --------
-A KA
4
2L
2
2ij ( i 2
+ j – 6 ( i + j ) + 3ij + 7 )
where K ij = --------------------------------------------------------------------------- for i+ j>1
(i + j – 1)(i + j – 2)(i + j – 3)
and K 11 = 4
11/30/98 16
Eigen Example
Lagrange Equation
Advanced Vibrations
The governing equation is
( mL )M Ȧ˙ +  -----3- K A = 0
EI
 
L
The corresponding algebraic eigenvalue is achieved by postulating that
iωt
A ( t ) = Re { A 0 e } problem is
 – ( ω 2 mL ) M +  -----
EI  
- K A0 = 0
    3
L
2 3
2 ω ( mL )L
We nondimensionalize this by setting α = -------------------------- and the
EI
2
eigen problem is now ( –α M + K ) A0 = 0

11/30/98 17
Eigen Example
Case N=1
Advanced Vibrations
1 1
M = 0.0333 , K = 4.000 .
2
Eigen solution α 1 = 120 and P 1 = [ 5.4772 ]
2
This asserts that the frequency α 1 = 120 is associated with the
y ( x ) =  ---  1 – --- .
x x
postulated shape
 L  L
From Lecture 11, we know that this is an upper bound for the square of
the first natural frequency.

11/30/98 18
Case N=3
Advanced Vibrations
0.0333 0.0167 0.0095 4.0000 2.0000 2.0000

3 3
M = 0.0167 0.0095 0.0060 K = 2.0000 4.0000 4.0000
0.0095 0.0060 0.0040 , 2.0000 4.0000 4.8000
2
α 1 = 97.47 4.4458 -14.4914 28.6397
3
Solution: α 2 = 2520 , P = 4.7965 28.9828 -132.7290
2
2
α3 = 17375 -4.7965 -0.0000 132.7290
2
This asserts that ω 1 = 97.47 is associated with the shape
2 3
 x  x   x
y ( x ) = 4.45 --- + 4.80 --- – 4.80 ---   x
1 – --- and is still an
 L  L   L    L
upper bound for the square of the first natural frequency.

11/30/98 19
Convergence with Of Eigenvalues
in Assumed Modes Problem
Advanced Vibrations
It is interesting to examine how the calculated natural frequency

Table 1:
Terms α12 α22 α32 α42 α52
1 120
2 120 2520
3 x97.47 2520 17400
4 x97.47 1572 17370 76188
5 x97.41 1571.8 x8161 76188 258502
Exact x97.409 1558.5 x7890 24937 x60880
depends on the number of assumed modes. There is a reason for this

that we will explore soon.
Note that even with the lowest order guess for the mode shape, we were
able to get a reasonable estimate for the first natural frequency.

11/30/98 20
Mode Shapes: Look at Approximation for
the First Mode for Each Order
Advanced Vibrations
As more polynomials are

admitted, the computed
1.4
mode shape appears to
converge on the exact
1.2
solution. In this case the
exact solution is 1
 x
y ( x ) = sin π -
--
0.8
 L 0.6
0.4
0.2
0 0.2 0.4 0.6 0.8 1

x
N=1
N=3
N=5
11/30/98 21
An Important Distinction
About Eigenmodes and Eigenvectors
Advanced Vibrations
4.4458
In the above, X = 4.7965 is an exact eigenvector of the above
-4.7965
algebraic eigen problem. (MATLAB says so!).
BUT
2 3
 x  x   x
y ( x ) = 4.45 --- + 4.80 --- – 4.80 ---   x
1 – --- is very much
 L  L   L    L
NOT the exact first mode of the simply supported beam, even though it
is the best approximation that can be achieved with a quartic
polynomial.

11/30/98 22
A More General Principle
Rayleigh Quotient
Advanced Vibrations
For any elastic system, the lowest natural frequency is achieved by that
displacement field that minimizes the ratio of potential to kinetic energy,
 V ( y )
λ = min ------------
 T ( y )
T (( y) > 0)
In the above, the field y is used as both the displacement and the
velocity field.
The above relation was already proven for cases of point masses, but
holds true for continuous systems. The minimization is over all
displacement fields which are square integrable and sufficiently
differentiable.
Restriction to a smaller set of displacement fields will yield an upper-
bound for the first eigen value.
We shall explore this more as we focus on continuous systems later on.

11/30/98 23
Advanced Vibrations
Next Time
Post-Mortem on MidTerm
Modal Damping
Modal Strain Energy
Transfer Functions

11/30/98 24
Lecture 13
Advanced Vibrations
TODAY
Solutions To Midterm
Eigen Solutions to the Differential Equations

Damping Matrices
Modal Damping

11/30/98 1 /home/djsegal/UNM/VibCourse/slides/Lecture13.frm
Mid-Term Take Home Exam
Problem 1. Assumed Modes
Advanced Vibrations
Consider the cantilevered Beam

shown here. There is a time- P(t)
dependent load applied at the end
and a dashpot placed at the center of
the span.
EI, m µ dy
Assume a single deformation mode: F = µ
dt
 x 2
y ( x, t ) = A 1 ( t ) --- . L/2
 L
• Derive the governing equation L
for A1 ( t ) .
• Assuming that P is harmonic, calculate the complex
magnification factor for the tip displacement as a function of the
frequency.
• Calculate the time-averaged rate of energy dissipation as a

function of frequency and force amplitude.
Solution to Problem
Advanced Vibrations
2

y ( x, t ) = A 1 ( t ) -
x
-- .
P(t)
 L
L
m 2  x 4 Lm 2 EI, m µ dy
T = ∫ ---- Ȧ --- dx = ---------- Ȧ F = µ
2  L 2⋅5 dt
0 L/2
L
EI 2  2  2 4EI 2
V = ∫ ------ A -----2 dx = --------3- A L
2 L  2L
0

Solution to Problem
Advanced Vibrations
Applied force: δW = Q P δA = Pδy ( L ) ⇒ Q P = P

Damping force:
Ȧ δA µ Ȧ
δW = Q D δA = F D δy  --- = – µ  --- ------ ⇒ Q D = – -------
L
 2  4 4 16
Lm ˙ µ Ȧ 4EI
The Lagrange equation is -------- Ȧ + ------- + --------- A = P ( t )
5 16 3
L
In more familiar terms, this is M Ȧ˙ + C Ȧ + KA = P

Lm µ 4EI
where M = -------- , C = ------ , and K = ---------
5 16 3
L

Solution to Problem
Advanced Vibrations
iωt
Assume that P ( t ) = Re { P 0 e } , then we expect the resulting
iωt
displacement will be A ( t ) = Re { A 0 e } . Substitution into the
governing equation yields
2 iωt
Re { ( [ – ω M + iωC + K ] A 0 – P 0 )e } = 0
2
from which we conclude that [ – ω M + iωC + K ] A 0 = P 0
 ω 2 ω P0
which is re-arranged to 1 – ------ + 2iζ ------ A 0 = ------
 ω 0 ω0 K
2 K 20EI C µL 5
where ω 0 = ----- = -----------4- and ζ = --------------- = ------- -----------
M mL 2Mω 0 64 mEI

Solution to Problem
Advanced Vibrations
Solving for A 0 in terms of P 0 ,
P0
A 0 = ------ H ( ω )
K
1
where H ( ω ) = ------------------------------------------------ is the Magnification Factor.
 ω 2 ω
1– ------ + 2iζ - -----
 ω 0 ω0

Solution to Problem
Advanced Vibrations
The average rate of energy dissipation is
2π ⁄ ω
ω
D = ------
2π ∫ P ( t )ẏ ( L, t ) dt
0
2π ⁄ ω
ω
∫
iωt iωt
= -
----- Re { P 0 e }Re { iωA 0 e } dt
2π
0
1 1
--
- ∗
= Re { ( P 0 ) ( iω A 0 ) } = --- Re { ( P 0∗ ) ( iωP 0 H ( ω ) ⁄ K ) }
2 2
3 2 3 2
L P0 L P0
= ω ----------------- Re { iH ( ω } = – ω ----------------- Im { H ( ω }
8EI 8EI

Problem 2.
Advanced Vibrations
Consider the three pendula shown.

• Calculate the governing equations θ1 θ3
for the three angles θ 1 , θ 2 , & θ 3
in terms of the parameters shown. R R
(Ignore vertical components of
R/2
spring extension.)
m m
• Linearize the equations, evaluat-
ing the mass and stiffness matri- m
k k
ces.
• Impose the constraint that θ 1 = θ 3 , writing the constraint matrix.

Write the mass and stiffness matrices in the reduced system.
• Calculate the eigenmodes and frequencies in the reduced system.
• Express those eigenmodes in terms of the full set of displacement

degrees of freedom, θ 1 , θ 2 , & θ 3 .

Solution
Advanced Vibrations
Derivation of governing equations:
2 k R 2
V = --- --- sin θ 2 – R sin θ 1 + ---  --- sin θ 2 – R sin θ 3
k  R
2  2 2  2
R
– mgR cos θ 1 – mg - cos θ 2 – mgR cos θ 3
-- .
2
∂V 2 θ 2
= mgRθ 1 + kR θ 1 – -----
∂ θ1  2
2
∂V R kR
= mg --- θ 2 + --------- ( θ 2 – θ 1 – θ 3 ) and
∂ θ2 2 2
∂V 2 θ 2
= mgRθ 3 + kR θ 3 – ----- after linearization.
∂ θ3  2
Solution
Advanced Vibrations
m 2 2 m  R 2 2 m 2 2
T = ---- R θ̇ 1 + ---- --- θ̇ 2 + ---- R θ̇ 3
2 2  2 2
Yielding the governing equations
1 0 0 θ̇ 1
˙
( 1 + γ ) –γ ⁄ 2 0 θ1
R
--- 0 1 ⁄ 4 0 θ̇˙2 + – γ ⁄ 2 ( 1 + γ ) ⁄ 2 – γ ⁄ 2 θ 2 = 0
g
0 1 θ̇˙ 0 –γ ⁄ 2 ( 1 + γ ) θ3
3
where γ = ( kR ) ⁄ ( mg )

Solution
Advanced Vibrations
T  iωt T
Setting θ1 θ2 θ3 = Re  e A1 A2 A3 
î 
The resulting algebraic eigen problem becomes
1 0 0 A1 ( 1 + γ ) –γ ⁄ 2 0 A1
2
–α 0 1 ⁄ 4 0 A2 + –γ ⁄ 2 ( 1 + γ ) ⁄ 2 –γ ⁄ 2 A2 = 0
0 1 A3 0 –γ ⁄ 2 ( 1 + γ ) A3
2 2
where α = Rω ⁄ g

Solution
Advanced Vibrations
The constraint that θ 1 = θ 3 yields the constraint equation
A1 10
A1
A2 = T where T = 0 1 . The new matrices are
A2
A3 10
2(1 + γ ) –γ
T
K = T KT = 1 and
–γ --- ( 1 + γ )
2
M = T MT = 2 0
T
0 1⁄4

Solution
Advanced Vibrations
The eigen problem in this reduced system is
A1 2(1 + γ ) –γ A1
–α 2 0
2
+ 1 = 0.
0 1 ⁄ 4 A2 –γ --- ( 1 + γ ) A 2
2
2
2 ( 1 + γ ) – 2α –γ
det 2 = 0
1 α
–γ --- ( 1 + γ ) – ------
2 4
4
α 3 2
This has a characteristic equation ------ – --- ( 1 + γ )α + ( 1 + 2γ ) = 0
2 2

Solution
Advanced Vibrations
3 2 1 2
The solutions are α = --- ( 1 + γ ) ± --- 1 + 2γ + 9γ . Substituting
2 2
this back into the algebraic eigen value problem
– (1 + γ ) −
2 1
+ 1 + 2γ + 9γ –γ
A2
1
--- [ ( 1 + γ ) −
2
–γ + 1 + 2γ + 9γ ] -----
A1
-
8
= 0
0
2
A2 1 + γ − 1 + 2γ + 9γ
yields ------ = – ------------ + ----------------------------------
A1 γ γ
Solution
Map these modes back to original space
Advanced Vibrations
A1 10
A1
A2 = T where T = 0 1 , so our solutions are
A2
A3 10
1
 α 2 = 3--- ( 1 + γ ) ± 1--- 1 + 2γ + 9γ 2 , 1 + γ 1 + 2γ + 9γ
2
 2 2  – -----------
- −
+ ----------------------------------
γ γ
1

Examine Solution
Advanced Vibrations
Consider case where gravity dominate over spring forces: γ = 0 . The

3 1 2
two solutions are α = --- ± --- :
2 2
A1
= 1
2
α = 1 corresponds to and
A2 0
A1
= 0
2
α = 2 corresponds to
A2 1
as expected.

Examine Solution
Advanced Vibrations
Consider case where spring forces dominate over gravity: γ → ∞ . The

 3 3
2
two solutions are α = --- ± --- γ :
 2 2
A1
= 1
2
α = 0 corresponds to (rigid body motion), and
A2 2
2 A1 1
α = 3γ corresponds to =
A2 –4
as expected.

Solutions to Governing Equations
Advanced Vibrations
Lets consider the solution x ( t ) to M ẋ˙( t ) + Kx ( t ) = F ( t ) and

N
expand x(t ) = ∑ βk ( t )xk where xk is the k’th normalized
k=1
2
eigenvector, corresponding to the k’th eigen value ω k . This can be
expressed in matrix form as x ( t ) = Pβ ( t ) where P is modal matrix.
T T
This yields the system β̇˙ + Dβ = P F = where D = P KP is
the diagonal matrix of eigen values.

Advanced Vibrations
Lets solve for the β via Laplace Transforms:

2 2
( s + ω k ) β k = sβ k ( 0 ) + β̇ k ( 0 ) + ( k ( t ) ) ( s )
This has the solution
t
sin ( ω k τ )
β k = β k ( 0 ) cos ω k t + β̇ k ( 0 ) sin ω k t + ∫ k ( t – τ ) ---------------------
- dτ
ωk
0
–1 –1
where the vector β̇ ( 0 ) = P ẋ ( 0 ) and β ( 0 ) = P x ( 0 )

With Damping
Advanced Vibrations
The above development provided the solutions for problem with no

damping. The more general problem is where the damping is
represented by a general matrix C . In general, this will yield complex
modes, and we shall study that case later.
There is a class of damping matrices for which the elastic eigen modes
are preserved. These problems are much more tractable. Because
T T
P MP = I and P KP = D we may conclude that
–T –1 –T –1
M = P I P and K = P DP .
–T –1
This suggests looking for damping matrices C = P χP where χ
is also a diagonal matrix consisting of terms 2ζ k ω k on the diagonal.

Damping Matrices
Advanced Vibrations
We examine our dynamics equation
M ẋ˙ + C ẋ + Kx = F .
We diagonalize with the modal matrix to find
T
β̇ + χβ̇ + Dβ = P F =
˙
Again, we solve for the β via Laplace Transforms:

2 2
( s + 2ζ k ω k s + ω k ) β k = sβ k ( 0 ) + β̇ k ( 0 ) + k ( t ) for each k.

Solution
Advanced Vibrations
–1  1  – ζωt sin ( ω 1 – ζ 2 t )
Exploit that  -----------------------------------------
- = e --------------------------------------
 s + 2ζ k ω k s + ω k 
2 2
ω 1 – ζ2
–1  s  d  –ζωt sin ( ω 1 – ζ 2 t )
and that - =  e
 ----------------------------------------- --------------------------------------
 s + 2ζ k ω k s + ω k 
2 2 dt ω 1 – ζ2 
to solve for βk ( t ) .

Transfer Functions
Synthesized from Modal Response
Advanced Vibrations
The whole system is represented in Laplace space by
d T
( x ) = P ( β ) = PG P (F )
d
where G (s) is a diagonal matrix, whose diagonal terms are
d 1
G kk = -----------------------------------------
2
-
2
s + 2ζ k ω k s + ω k
we have assumed homogeneous initial conditions.
Say we want to find the transfer function from force DOF n to
displacement DOF m. Then
modes
∑
d
G mn ( s ) = P mk G kk P nk
k=1
Modal Damping
Advanced Vibrations
The assumption made above, that the damping matrix is diagonalizable,

is the assumption of modal damping.
Note that modal damping includes the special cases of proportional
damping:
M ẋ˙ + ( aM + bK )ẋ + Kx = F ( t )
is diagonalized by the modal matrix. x = Pβ as
β + ( aI + bD )β̇ + Dβ = .

Modal Damping Terms
Advanced Vibrations
We see the advantages of assuming modal damping, the question

becomes: “How do we deduce that damping ζ k associated with each
mode?”
There are three standard answers.
1. Experimentally. Plot frequency response, identify peaks with eigen
frequencies, and determine fraction for critical damping from the
half power points.
T
2. Where a real damping matrix is known, compute P CP and
discard the off-diagonal terms
3. Where the damping mechanism can be modeled, set the model

damping to reproduce the dissipation ratio associated with modal
vibration. This is the modal strain energy method.
Examples and discussion follow.

Next Time
Advanced Vibrations
More on Modal Damping and the Modal Strain Energy Method.

An Introduction to Linear Viscoelasticity.

Lecture 14
Advanced Vibrations
TODAY
More on Damping Matrices and Modal Damping

Linear Viscoelasticity

Modal Damping Terms
RECALL
Advanced Vibrations
We saw the advantages of assuming modal damping, the question

became: “How do we deduce that damping ζ k associated with each
mode?”
There are three standard answers.
1. Experimentally. Plot frequency response, identify peaks with eigen
frequencies, and determine fraction for critical damping from the
half power points.
T
2. Where a real damping matrix is known, compute P CP and
discard the off-diagonal terms
3. Where the damping mechanism can be modeled, set the model

damping to reproduce the dissipation ratio associated with modal
vibration. This is the modal strain energy method.
Examples and discussion follow.

Damping from Experimental Data
Advanced Vibrations
Experimental measurements yield impulse hammer

plots of acceleration versus
frequency. Ideally, each peak is
accelerometer
associated with a distinct natural
mode of the system.
We say in the first portion of the test
that we could deduce the damping
of a single degree-of-freedom by
examination of the “half power
points”. ζk= ∆ω/ωk
We can perform the same analysis |a|
for each of the modal peaks that are
manifest from the experimental data,
yielding a fractions for critical
damping for those modes.

Modal Damping Terms
from Full Damping Matrix
Advanced Vibrations
Say we are give a system
2 1 ẋ˙1 + 2 0 ẋ 1 + 2 – 1 x 1 = 0
1 2 ẋ˙2 0 3 ẋ 2 –1 2 x2 0
In Matlab,
[P,D] = eig(k,m) :Find the generalized eigen solution:
PP =P/sqrt(P’*m*P) :Normalize the modal matrix
Cm = PP’*c*PP ;map the damping matrix to modal coordinates
2.5000 -0.2887
-0.2887 0.8333
C = 2.500 0
d
The diagonal of this matrix is
0 0.833
Modal Damping Terms
from Full Damping Matrix
Advanced Vibrations
The full system in modal coordinates is
1 0 β̇˙1 + 2.500 0 β̇ 1 3 0 β1
+ = 0
0 1 β̇˙2 0 0.833 β̇ 2 0 0.333 β 2 0

Modal Damping Terms
By Modal Strain Energy Method
Advanced Vibrations
Given a system with mass matrix M and stiffness matrix K ,

1. compute the modal matrix P . Each eigen mode corresponds to a
column of P.
2. compute the energy dissipation per cycle when the system is
forced to undergo the deformation of the k’th eigen mode at the k’th
eigen frequency
3. Calculate the maximum strain energy in that cycle.
4. Calculate the ratio of the dissipation to energy storage
5. Set the modal dissipation to reproduce the same ratio in modal

coordinates.
This requires and example

Modal Damping Terms
Advanced Vibrations
Consider the system to the right, the k k k

mass and stiffness matrices are m m
µ
M = m 1 0 and K = k 2 –1 . The algebraic eigen problem is

01 –1 2
 m 10 2 – 1  x1
 – ω ----
2 +  = 0 , having eigen solutions
 k 01 –1 2  x2
   
---- = 1 , 1 ⁄ 2 
  ω2  m and   ω  ---- = 3 ,
1⁄ 2 
2 m
  k      k   
 1⁄ 2   –1 ⁄ 2 

Modal Damping Terms
Advanced Vibrations
The energy dissipation associated with the dashpot is
2π ⁄ ω
∫
2
D = µ ( ẋ 2 – ẋ 1 ) dt
0
 
x1  iω1 t 1 ⁄ 2 
For the first mode, = Re  Ae ,
x2  1⁄ 2 
î 
 iω 1 t 
ẋ 2 – ẋ 1 = Re  iω 1 Ae ( 1 ⁄ 2 – 1 ⁄ 2 )  = 0 => ζ 1 = 0
î 
The second mode will be more interesting.

Modal Damping Terms
Advanced Vibrations
The energy dissipation associated with the dashpot is
2π ⁄ ω
∫
2
D = µ ( ẋ 2 – ẋ 1 ) dt
0
 
x1  iω2 t 1 ⁄ 2 
For the second mode, = Re  Ae ,
x2  –1 ⁄ 2 
î 
 iω 2 t  iω 2 t
ẋ 2 – ẋ 1 = Re  iω 2 Ae ( 1 ⁄ 2 + 1 ⁄ 2 )  = 2ω 2 Re { iAe }
î 
The dissipation is D = 2µω 2 π ( A∗ A ) .

Modal Damping Terms
Advanced Vibrations
The energy storage is

T
k iω 2 t 2 1 ⁄ 2 2 –1 1 ⁄ 2
V = --- ( Re { Ae } )
2
–1 ⁄ 2 –1 2 –1 ⁄ 2
3k iω 2 t 2
= ------ ( Re { Ae } )
2
3k
and the max storage is V max = ------ A∗ A .
2
D 2µω 2 π ( A∗ A ) 4 µω 2
The dissipation ratio is ------------ = --------------------------------- = --- ----------
V max 3k 3 k
------ A∗ A
2

Modal Damping Terms
Advanced Vibrations
The corresponding analysis in modal coordinates is
2
β̇˙2 + 2ζω 2 β̇ 2 + ω 2 β 2 = 2
iω 2 t
Here, we assume that β 2 = Re { Be }.
The energy dissipation per cycle is
2π ⁄ ω 2
2 π
∫ 2ζω 2 ( β̇ 2 ) dt = ------ ( 2ζω 2 )ω 2 ( B∗ B ) =2πζω 2 ( B∗ B )
2 2
D =
ω2
0
2
ω2
The max energy stored per cycle is V max = ------ ( B∗ B ) so the
2
dissipation ratio is 4πζ .

Modal Damping Terms
Advanced Vibrations
We now set those damping ratios equal,
4 µω 2 µω 2
--- ---------- = 4πζ 2 => ζ 2 = ----------
3 k 3πk
Our two modal equations are:
2 2
β̇ 1 + ω 1 β 1 =
˙
1 and β̇ 2 + 2ζ 2 ω 2 β̇ 2 + ω 2 β 2 =
˙
2

Introduction to Linear Viscoelasticity
Advanced Vibrations
Here we restrict our attention to rubbery polymers.

Such polymers have Poisson’s ratio τ
almost 0.5; we can ignore volumetric d
strains. In what follows, we restrict h
attention to problems for which all
relevant deformations are in shear.
Engineering shear strain is defined
The theory of linear viscoelasticity as γ = d/h
relates the current stress to the
history of strain. The dependence is linear and time-translation
invariant. This requires a constitutive equation of the sort:
t
σ(t ) = ∫ G ( t – τ )γ̇ ( τ ) dτ
–∞
where G ( τ ) , the relaxation modulus, is a function of time

characteristic of the material. There are sophisticated laboratory
methods of measuring the relaxation modulus.
Oscillatory Strain
Advanced Vibrations
For oscillatory deformation, the strain can be expressed

iωt iωt
γ ( t ) = A sin ( ωt + ψ ) = Im ( γ 0 e ) so γ̇ ( t ) = Im ( iωγ 0 e )
We expect that the resulting stress will harmonic at the same
iωt
frequency: σ ( t ) = Im { σ 0 e }.
These are substituted into the constitutive equation to find
σ 0 = γ 0 ( G′ ( ω ) + iG′′ ( ω ) ) where
∞ ∞
G′ ( ω ) = ∫ ωG ( τ ) sin ( ωτ ) dτ &G′′ ( ω ) = ∫ ωG ( τ ) cos ( ωτ ) dτ
0 0

Oscillatory Strain
Advanced Vibrations
The stress associate with γ ( t ) = A sin ( ωt + ψ ) is

AG′ ( ω ) sin ( ωt + ψ ) + AG′′ ( ω ) cos ( ωt + ψ )
Common laboratory techniques measure G′ and G′′ directly.
The storage modulus G' ( ω ) is an indicator of the ability of the polymer

to store mechanical energy. The loss modulus G'' ( ω ) is an indicator of
the ability of the polymer to dissipate mechanical energy.

Superposition
Advanced Vibrations
Linear viscoelastic materials have the property of being linear.

The stress resulting from the superposition of strain histories is the
superposition of the stresses associated with those histories.
If γ (t) = ∑ Ak sin ( ωk t + ψ k ) , then

k
τ(t ) = ∑ Ak G' ( ωk ) sin ( ωk t + ψ k )

k
+ ∑ A k G'' ( ω k ) cos ( ω k t + ψ k )
k

Energy Dissipation and Storage
Advanced Vibrations
At a given frequency ω , the energy dissipation per cycle per unit

volume is
2π ⁄ ω
ED = ∫ τ ( t )γ̇ ( t ) dt
0
2π ⁄ ω
= A2 ∫ [ G' ( ω ) sin ωt + G'' ( ω ) cos ωt ]ω cos ( ωt ) dt
0
2
= πA G'' ( ω )
The max energy stored per unit volume in a cycle is estimated as

1 1 2
E S = --- max [ ( A sin ωt ) ( AG' ( ω ) sin ωt ) ] = --- A G' ( ω )
2 2
Energy Dissipation and Storage
Advanced Vibrations
The loss factor η of either a structure or a material is defined by

ED
η ( ω ) = ------------- . For a linearly viscoelastic material,
2πE S
η ( ω ) = G′′ ( ω ) ⁄ G′ ( ω ) = tan δ ( ω )
The general forms
of the storage
modulus and loss
log G” log G'
modulus are G’
illustrated in the
figure
Loss modulus is
maximum where
storage modulus is G”
steepest
log ω

Solution of Transient Problems
Advanced Vibrations
Rigorous solution of transient problems is much more difficult. Using

assumed modes, virtual work, and Lagrange Equations, we derive the
following form of equation for structures of elastic and viscoelastic
material
t
M ẋ˙ + Kx + ∫ Γ ( t – τ )ẋ ( τ ) dτ = F (t)
–∞
where the matrix Γ is constructed of the relaxation moduli of the

viscoelastic materials in the structure. This equation is very hard to
solve in the time domain.
The form is simplified a little by resort to Laplace
2
transforms: s MX ( s ) + ( K + s ( Γ ) )X ( s ) = L(F ) ,
but inversion of this system of equations is usually impractical.

Viscoelasticity is Easier
in the frequency domain
Advanced Vibrations
– ω M X ( ω ) + K X ( ω ) + Γ∗ ( ω )X ( ω ) = F ( ω )
2
Some finite element computer codes, including NASTRAN, can address

this problem in frequency space. Again, inversion to the time domain
can be a problem, but we often have an intrinsic interest in the
frequency domain representation itself.
Correspondence Principle (Part One)

The governing equations in frequency space for a linearly viscoelastic
structure are those of the corresponding elastic structure, but with
elastic material properties replaced by complex material properties.
Correspondence Principle (Part Two)

This part is not as useful to us here. It observes that the solution to a
quasi-static viscoelastic problem can be obtained by inversion of the
Laplace transform mapped from the solution of a corresponding elastic
problem.

Steady State Oscillation
Advanced Vibrations
Steady State, harmonic response of a x

single degree of freedom system: F0sinωt
k
Consider the system shown. The mass m
sits on a pad of thickness h, and area
A. At the imposed frequency, the G’, G”, h, A
storage and loss moduli are G’(ω) and
G”(ω), respectively.
k x(t) F0sinωt
We anticipate that the steady state
motion of the mass will be m
x ( t ) = X sin ( ωt + φ ) . Fv(t)
Fv(t)
In complex notation,
iωt
F ( t ) = F 0 Im { e } and
i ( ωt + φ ) iωt iφ
x ( t ) = X Im { e } = Im { X 0 e } where X 0 = X e

Vibration of Linearly Viscoelastic Structures
Advanced Vibrations
The force applied to the polymer is

XA
F v ( t ) = -------- ( G' ( ω ) sin ( ωt + φ ) + G'' cos ( ωt + φ ) )
h
A iωt
= --- Im { X 0 e ( G' ( ω ) + iG'' ( ω ) ) }
h
The equation of motion of that mass is
F0sinωt
mẋ˙ = F 0 sin ωt – kx ( t ) – F v ( t ) . k x(t)
m
Fv(t)
Making all the appropriate Fv(t)
substitutions, we find
2 A
– ω mX 0 = F 0 – k X 0 – X 0 --- ( G' + iG'' )
h
Note the similarity to the elasto-dynamics problem.
Advanced Vibrations
2 A
– ω mX 0 = F 0 – k X 0 – X 0 --- ( G' + iG'' )
h
From which we can solve for the complex frequency response function
for normalized displacement as a function of frequency:
X0 1
------ ( ω ) = ------------------------------------------------------------------------------------------------------
F0 [ k + ( A ⁄ h )G' ( ω ) – mω ] + i ( A ⁄ h )G'' ( ω )
2
X0 1
Note that ------ ( ω → ∞ ) = -------------- , as it should. This is mass loading.
F0 –ω m
2
X0 1
Also note that ------ ( 0 ) = -------------------------------------- , as it should. This is the
F0 k + ( A ⁄ h )G' ( 0 )
static response.

A Material
Advanced Vibrations
Lets examine the response of a real system. The following plot for
polyisobutylene is taken from “Vibration Damping” by Nashif et. al.
E’=3G’
tan(δ)

Advanced Vibrations
ω 5
We can fit the above plots reasonably well in 0.1 < ------ < 10 by
2π
ω 3.3
Log ( 3G' ( ω ) ) = Log ( E' ( ω ) ) = 6.2 + 0.0112  Log  ------  or
  2π 
  ω   3.3
5.72 + 0.0112 Log ------
  2π 
G' ( ω ) = 10 and
 ω   ω  2
Log ( tan δ ( ω ) ) = – 1 + 0.56Log ------ – 0.069 Log ------ so
 2π   2π 
G'' ( ω ) = G' ( ω ) tan δ ( ω )

 ω   ω  2   ω   3.3
4.72 + 0.56Log ------ – 0.069 Log ------ + 0.0112 Log ------
 2π   2π    2π 
= 10
Advanced Vibrations
The above
7
approximations 10
show the correct
behavior at low
frequencies
6
10 G’
5
10
G”
4
10
3
10
-2 -1 0 1 2 3 4
10 10 10 10 10 10 10
f=ω/2π

Advanced Vibrations
Lets assign properties to the problem x

we discussed above. F0sinωt
k
Say our mass is 0.10m x 0.10m in area, M
0.1m high, and made of steel.
G’, G’’, h, A
2
A = ( 0.1m ) ( 0.1m ) = 0.01m
3
M = ( 0.1m ) ( 0.1m ) ( 0.1m ) ( 7700kg ⁄ m ) = 7.7kg
Set h = 0.01m and k = 3E6N ⁄ m
Then A ⁄ h = 1.0
We shall calculate the fraction of critical damping by two methods:
• Examination of half-power band width. This reflects the true
material properties.
• Modal strain energy. This is approximate.

Fraction of Critical Damping in
Viscoelastic Structure
Advanced Vibrations
Lets calculate the damping by examination of the half power points.

−6
x 10
2.5
Case h=0.01 so A/h=1.0

2
ζ=0.069
X0/F0
1.5
0.5
0
0 20 40 60 80 100 120 140 160 180
f (Hz)
Fraction of Critical Damping in
Viscoelastic Structure
Advanced Vibrations
Lets consider
x 10 another case:
−5
2.5
2
Case h=0.10 so A/h=0.1
ζ=0.0084
X0/F0
1.5
0.5
0
0 20 40 60 80 100 120 140 160 180
f (Hz)
Advanced Vibrations
The above treatment is that for a steady state oscillatory response.
To solve this system in the time domain, we must

• solve an integro-differential equation (usually not a good idea)
• approximate this problem by a second order system with

constant coefficients. This is what modal strain energy was made
for.

Solution of Transient Problems
Advanced Vibrations
In order to solve transient problems involving linearly

viscoelastic materials, we need a simpler approach.
The approach that has developed over the last four
decades is:
• Map the elastic problem to mode space
• Estimate modal damping via the modal strain

energy method
• Integrate the resulting modal equations
• Map displacements back to spacial degrees of

freedom
We shall see some examples using modal strain energy to estimate
modal damping for viscoelastic structures.
Damping from Modal Strain Energy
Advanced Vibrations
Lets see what damping model we get from the modal strain energy
method applied to our earlier problem.
1⁄2
For the elastic system, ωn = ( k ⁄ m ) = 624 /sec =>
f n = 100 Hz. Lets look at the energy dissipation per cycle of this
single degree of freedom system when oscillating at its resonant
frequency at amplitude B:
2 2 5
E D = πB ( A ⁄ h )G'' ( ω n ) = πB ( 4.74 ×10 N ⁄ m )
The energy stored per cycle is
1 2 1 2 6
E S = --- B ( k + G' ( ω ) ( A ⁄ h ) ) = --- B 3.07 ×10 N ⁄ m
2 2
ED
2ζ = η = ------------- = 0.134 => ζ = 0.067
2πE S
Approximation with 2nd Order System
Advanced Vibrations
Our approximate model for the viscoelastic system is
2
β̇˙ + 2ζω n β̇ + ω n β = F ⁄ M
We convert back to spacial coordinates:
M ẋ˙ + 2ζ KM ẋ + Kx = F ( t )
In order to recover the static response, the stiffness we use is that

computed at frequency ω = 0 :
6
K = k + ( A ⁄ h )G' ( 0 ) = 3.54 ×10 N ⁄ m .
We just found the found the fraction of fraction of critical damping
ζ = 0.067 , and the mass remains M = 7.7kg

introduction to Linear Viscoelasticity
Approximation with 2nd Order System
Advanced Vibrations
Lets compare this approximation to the actual system for the case of
iωt
steady state harmonic excitation: F = Im { F 0 e } and
iωt
x = Im { X 0 e } yielding
1
X 0 ⁄ F 0 = ----------------------------------------------------------------
2
-
[ ( K – Mω ) + i2ζω KM ]
Lets compare this approximation to the full viscoelastic system.

Comparison of Linear Viscoelastic Model
with Second Order System
Advanced Vibrations
−6
x 10
2.5
viscoelastic system
second order
2
approximation
X0/F0
1.5
0.5
0
0 20 40 60 80 100 120 140 160 180
f (Hz)
Advanced Vibrations
Matlab code for the above:

% Calculate response of systems in slides page 28 and page 35 % find the half power points
% of Lecture 14 cross=0;
% spring stiffness and the mass for i=2:Points
k=3e6 if( (Ha(i)-Hmax/2^0.5)*(Ha(i-1)-Hmax/2^0.5) <0 )
m=7.7 cross = cross+1;
% ic(cross) = i;
% Repeat current value of Ah end
Ah end
% %
% specify the frequency range to examine %mark these points
fmin = 10.0; fmax = 180.0; f0 = f(i0); f1 = f(ic(1)); f2 = f(ic(2));
Points = 1001; h0 = Ha(i0);h1 = Ha(ic(1)); h2 = Ha(ic(2));
f(i) =fmin + (fmax-fmin)*(i/Points); %
w(i) = f(i)*2*pi; %%evaluate fraction of critical damping
% zeta = (f2 - f1)/(2*f0)
% Interpolating formulae for storage and loss moduli %
gp(i) = real(10.0^(5.72 + 0.0112*(log10(f(i)))^3.3)); %%try modal strain energy method
gpp(i) = real(10.0^(4.72+0.56*log10(f(i)) -0.069*(log10(f(i)))^2 + Ed = pi*Ah*gpp(i0);
0.0112*(log10(f(i)))^3.3)); Es = 0.5*(k+gp(i0)*Ah);
% Driving point frequency response. A complex function of eta = Ed/(2*pi*Es);
frequency zetaE = eta/2
H(i) = 1.0/(k+Ah*gp(i)-m*w(i)^2 + j*Ah*gpp(i)); %
end % Plot the magnitude of driving point frequency response and
% % mark the points
Ha = abs(H); % magnitude of the driving point frequency p1 = plot(f,Ha,’r-’, f0,h0,’b+’, f1,h1,’b+’, f2,h2,’b+’);
response %
% %
% Power is proportional to the square of displacement %plot the modal approximation to the frequency response
amplitude. for i=1:Points
% We shall find the half power points HM(i) = 1.0/(k+Ah*gp(1)-m*w(i)^2 + j*(2*zetaE*m*w(i)*(k/
ic = [1 Points]; % array of indices of half power points m)^0.5));
i1 = 1; i2=Points; HMa(i) = abs(HM(i));
Hmax = max(Ha); end
% find the frequency where Ha is maximum % Plot it and compare with the ViscoElastic model
i0 = find( abs(Ha - Hmax)<1.0e-10); p2 = plot(f,Ha,’r-’, f, HMa, ’g.’);

Summary of Modal Strain Energy Method
Advanced Vibrations
1.Perform eigen solution of linear elastic system, cal-

culating modes and frequencies.
2.For each mode, calculate the elastic energy ES
stored in the system when it is deformed to the
extreme of each mode. For viscoelastic parts of the
structure, use the storage modulus at those eigen
frequencies.
3.For each mode, calculate the energy ED dissipated
per cycle.
ED
4.For each mode, set ζ k = -------------
2πE S

Next Time
Advanced Vibrations
Introduction to Frequency Response Methods

Lecture 14.5
Advanced Vibrations
COMPLEX MODES

2/10/99 1 /home/djsegal/UNM/VibCourse/slides/Lecture14_5.frm
State Space Formulation
Advanced Vibrations
What happens when we cannot assume modal damping?

As usual, we assume a second-order system with constant coefficients:
M ẋ˙ + C ẋ + Kx = F ( t ) ,
Put this into a state space formulation:
˙
M 0 ẋ + C K ẋ = F ( t )
0 M x –M 0 x 0
Each of the if the original problem has N degrees of freedom, the new
system has 2N degrees of freedom.

Change of Variables
Advanced Vibrations
Define = M 0 . Since M is nonsingular, so is .

0 M
Define y = ẋ , substitute the change of varibles into the state-space

x
–1
governing equation, and premultiply by toobtain
ẏ – Ay = F̃ where A = –
–1 C K and F̃ =
–1 F (t)
–M 0 0
We shall use modal analysis and orthogonality to solve the above first
order system.

Bi-Orthogonality
Advanced Vibrations
Because, the matrix A is non-symmetric its eigen vectors are not

orthogonal. They are not even real.
Because of the non-symmetry of A , we must talk about the right and

left eigen values of A.
• A right eigen pair of A is a solution ( λ, V ) that satisfies
AV = λV
• A left eigen pair of A is a solution ( λ, W ) that satisfies
H H H
W A = λW where ( ) is the complex conjugate
transpose of its arguement. Note that the left eigenvectors of A
T
are the right eigenvectors of A . Please verify this.

Several Useful Observations
Advanced Vibrations
• If ( λ, V ) is a right eigen pair of A , then so is ( λ∗, V ∗ ) . Try it!

Lets order these eigen solutions by absolute value of the
imaginary part (the frequency) placing .
• Similarly, if ( λ, W ) is a left eigen pair of A , then so is

( λ∗, W ∗ ) .
• If λ is a right eigen value of A , then it is also a left eigen value of
A . The arguement is as follows. If ( A – λI )V = 0 for a non-
trivial V then A – λI is singular, which implies not only that the
columns of A – λI are linearly dependent, but that the rows of
A – λI are also linearly dependent. The linear dependence of
the rows of A – λI means there is a linear combination of them
H
which is also zero: W ( A – λI ) = 0 . Lets order these eigen
solutions identically to the corresponding right eigen solutions.
Bi-Orthogoality
Advanced Vibrations
Though the right eigen vectors are not mutually orthogonal, the left
eigenvectors are also not mutually orthogonal, there is an
orthogonallity between members of the two sets.
For the sake of simplicity, for the moment, we assume that there are no
repeated eigenvalues.
Consider one right eigen solution ( A – λ n I )V n = 0 , and contract it

H
by Wm:
H H
W m AV n = λ n W m V n
Similarly, we consider the m’th left eigen solution and contract it by V n:

H H
W m AV n = λ m W m V n

Bi-Orthogoality
Advanced Vibrations
H
Subtracting these two, we have 0 = ( λ n – λ m )W m V n
Since we had assumed that there are no repeated eigenvalues, we
H H
conclude that W m V n = 0 and W m AV n for n ≠ m .
If there are repeat eigenvalues, we select linear combinations of the

corresponding eigenvectors to achieve mutual bi-orthogonality among
those eigenvectors.
H H
We normalize the V n so that W n V n = 1 and W n AV n = λ n .

Solution of Governing Equations
Advanced Vibrations
2n
We postulate solutions y = ∑ ak ( t )Y k to the governing equation
k=1
H
ẏ – Ay = F̃ and contract by W m to obtain
H
ȧ m ( t ) – λ m a m ( t ) = W m F̃ ( t )
In principle this could be solved by Laplace transforms:
H
W m { F̃ } a m ( 0 )
{ a m } = ------------------------ + --------------- .
s – λm s – λm
Note that the initial values are found from orthogonality
H
am ( 0 ) = W m y(0) .
Advanced Vibrations
Lets let ( λ m′, V m′ ) be the eigen pair conjugate to ( λ m, V m ) :
λ m′ = λ m∗ and V m′ = V m∗
The modal coordinates associated with ( λ m′, V m′ ) are also solved by

Laplace transforms:
H H
W m′ { F̃ } W m′ y ( 0 )
{ a m′ } = ------------------------- + ---------------------
s – λ m′ s – λ m′
H ∗ H ∗
(W m ) { F̃ } (W m ) y(0)
= -------------------------------- + ----------------------------
s – λ m∗ s – λ m∗

Advanced Vibrations
Lets add the contributions of each of these terms to the solution
{ a m V m + a m′ V m′ } =
H H ∗
 m m
V W  V m W m
=  ----------------- +  ----------------- ( { F̃ } + y ( 0 ) )
 s – λ m   s – λ m′ 
Note that the right hand side of the above equation is real. The complex
modes contribute to the solution pair-wise as real values.
We reclaim the solution to the physical problem from

ẋ = y .
x

Lecture 15
Advanced Vibrations
TODAY
Last Word on Modal Analysis


A Last Word on Modal Analysis
Advanced Vibrations
Why modal analysis is attractive

• modal coordinates can be integrated independently.
• Model reduction by modal truncation:
M < modes
x(t ) = ∑ βk ( t ) { Pk }
k=1
where { P k } is the k’th column of the modal matrix simplifies the
problem further and permits larger time steps in the integration.
• Vibration modes have intrinsic meaning. Modes and modal

frequencies tell us something more understandable about
structural response than do structural matrices
• Modes and modal frequencies can be measured and compared

with prediction. Model validation and system identification

Advanced Vibrations
We have already examined problems in the frequency domain when we

assumed that the applied loads were harmonic:
iωt
f ( t ) = Re { F 0 e }.
In what follows, we will admit cases where the load is a continuous sum
of harmonic inputs:
∫ Re { F ( ω )e
iωt
f (t) = } dω
0

Advanced Vibrations
We may also write this as
∫
iωt
f (t) = F ( ω )e dω
–∞
where F ( ω ) has been defined a little differently than previously.
There is a constraint that F ( – ω ) = ( F ( ω ) )∗ .

Lets discuss why.

Advanced Vibrations
If we can express the time domain function in terms of a frequency

domain function, can we reverse the process?
Yes, of course. Following is a formulation of the relations.
∞ ∞
1 – i ωt
∫ ∫
iωt
f ( t ) = ------ F ( ω )e dω & F (ω) = f ( t )e dt
2π
–∞ –∞
Note that we rescaled F in the above by 2π and replaced ω with – ω .

There are several other conventions for the placement of the 2π term
and the minus sign. The above appears to be most common.
These relations can be derived from a limiting case of Fourier series or
from a single application of the Cauchy integral theorem.

An interesting but unimportant note
Advanced Vibrations
There is an interesting observation for fans of the Laplace transform.

Fourier transform pairs can be achieved from analytic Laplace
transforms where they exist:
∞ ∞
– i ωt
∫ ∫ [ f ( t )e
iωt iωt
F (ω) = f ( t )e dt = + f ( – t )e ] dt
–∞ 0
= (f) s = iω + (– f ) s = –i ω
This is interesting, but not used very often. The primary reason is that
one usually does not deal with analytic Fourier transforms.

Frequency Domain Analysis:
Why?
Advanced Vibrations
Frequency domain analysis is most useful when comparison is to be

made with experiment or when connection is desired between
experiment and modal analysis. We shall see examples of each.

Governing Equation in Frequency Space
Advanced Vibrations
Our familiar governing equations are M ẋ˙ + C ẋ + Kx = f ( t ) .

The Fourier transform of this equation is
2
( – ω M + iωC + K )X ( ω ) = F ( ω )
Say we are interested in the displacement x I ( t ) of degree of freedom I
due to forces at degree of freedom J . At first look this would appear to
be a difficult relationship to find, but it turns out to be reasonably do-
able.
We diagonalize the governing equation via modal decomposition:
x ( t ) = Pβ ( t ) which is X ( ω ) = PB ( ω ) in frequency space.
T 2 T
P ( – ω M + iωC + K )PX ( ω ) = P F ( ω )

Governing Equation in Frequency Space
Advanced Vibrations
T 2 T
P ( – ω M + iωC + K )PB ( ω ) = P F ( ω ) becomes
2 T
( – ω I + i2ωχ + D )B ( ω ) = P F ( ω )
The terms of B can be solved individually:
DOFs
∑
2 2
Bk = P Jk F J ( ω ) ⁄ ( ω k – ω + 2iωζ k )
J=1

Frequency Response Functions
Advanced Vibrations
The displacement at node I is

Modes
XI (ω) = ∑ P Ik B k ( ω )
k=1
Modes DOFs
∑ ∑
2 2
= P Ik P Jk F J ( ω ) ⁄ ( ω k – ω + 2iωζ k )
k=1 J=1
DOFs
= ∑ H IJ ( ω )F J ( ω )
J=1

Frequency Response Functions
Advanced Vibrations
Above,
Modes
∑
2 2
H IJ ( ω ) = P Ik P Jk ⁄ ( ω k – ω + 2iωζ k )
k=1
is the “receptance frequency response function” of DOF I with respect

to DOF J . Sometimes it is simply called the “frequency response
function”.
By construction, we see that all of the frequency response functions for

a single structure share the same peaks when plotted against
frequency.
Lets discuss what this means for experimental modal

analysis.
Modal Analysis
Advanced Vibrations
Having found the resonances of the system by identification of the

peaks in the frequency response functions, we consider the following
experiment:
We excite the structure at a single point J at a natural frequency ω k

and measure the displacements (or accelerations) at a good number of
sample of points elsewhere on the structure
2 2
H IJ ( ω ) ≅ P Ik [ P Jk ⁄ ( ω k – ω + 2iωζ k ) ]
We see that at resonance, the receptance at locations is proportional to
the mode shape of that frequency. It is possible to assess the mode
shapes experimentally.

Experimental Determination of H IJ ( ω )
Advanced Vibrations
H IJ ( ω ) = X I ( ω ) ⁄ F J ( ω ) , and we will often want to

By definition,
determine H IJ ( ω ) experimentally. What candidate forces can be
used? Most commonly used are:
• stepped sine
• swept sine
• impulse
• random
Each approach has its advantages. Lets look at them individually

By Stepped Sine
Advanced Vibrations
Stepped sine experiments are usually done by connecting a part of the

structure to a shaker via a thin wire or rod (a stringer). Accelerometers
placed at the appropriate locations on the structure provide the
kinematic information in frequency space.
Stinger or Stringer
Force Transducer
Leads to a myriad
accelerometers
Shaker
Charge Amplifier &
Signal Processor
By Stepped Sine
Advanced Vibrations
Stepped sine testing is usually done so that the structure reaches

steady state at each frequency before moving on to the next frequency.
A sine force input of unit peak amplitude is
iω 0 t – iω 0 t
e –e
f ( t ) = sin ω 0 t = ----------------------------- .
2i
∞
1 – i ωt
Since f ( t ) = ------
2π ∫ F ( ω )e dω ,
–∞
F ( ω ) = πi [ δ ( ω – ω 0 ) – δ ( ω + ω 0 ) ]

By Stepped Sine
Advanced Vibrations
Recalling that H IJ ( ω ) = X I ( ω ) ⁄ F J ( ω ) and noting that the only

information available from this experiment is at ω 0 and – ω 0 , we have
H IJ ( ω 0 ) = X ( ω 0 ) ⁄ ( πi ) and H IJ ( – ω 0 ) = X ( ω 0 )∗ ⁄ ( – π i ) .
Note that H IJ ( – ω 0 ) = H IJ ∗ ( ω 0 ) . This means that H IJ ( ω ) is the

Fourier transform of a Real function.
Can we guess which one?
The advantage of stepped sine experiments is that they provide high

resolution of both phase and amplitude of H IJ ( ω )
The disadvantage is that it takes much more time.

By Swept Sine Input
Advanced Vibrations
Swept sine input is much more quickly implemented than is the stepped
sine input.
This sort of experiment is most often used to search for resonances.
Unless the sweep is very slow, the results are hard to post process for
detailed information.

By Impulse Hammer
Advanced Vibrations
The impulse hammer contains a precision force transducer at its tip so

that the force history at the interval of impact can be recorded
accurately. The frequency content of the impulse extends roughly to the
reciprocal of the period of the pulse. The frequency content can be

concentrated at low frequencies by softening the tip and by increasing
the mass of the hammer head. Conversely, the frequency content can
be extended up to higher frequencies by dropping the mass and raising
the stiffness.
The frequency content of the structural response will cover roughly the
same interval.

By Impulse Hammer
Advanced Vibrations
In principle, the frequency response can be assessed by evaluating the

Fourier transforms of the impulse and the response and taking the
ratio.
H IJ ( ω ) = X I ( ω ) ⁄ F J ( ω )
Signals are reinforced by doing several experiments and averaging the
frequency response functions obtained from each.
Note that the magnitude information on the frequency response
function can be calculated from power spectra of the input and output:
[ H IJ ( ω )F J ( ω ) ] [ H IJ ( ω )F J ( ω ) ]∗ = X I ( ω )X I ( ω )∗ yielding
H IJ ( ω ) = S XX ( ω ) ⁄ S FF ( ω )
where S XX ( ω )= X I ( ω )X I ( ω )∗ is the power spectral density (PSD)

of displacement and S FF ( ω ) = F J ( ω )F J ( ω )∗ is the PSD of force.
Via Random Input
Advanced Vibrations
Random input experiments offer the opportunity to collect data rich in

frequency content. Further, those experiments can be run over a long
period of time to over acquiring a good signal-to-noise ratio.
We consider stationary random excitations. These are processes
whose probability distributions are constant in time. Following is a plot
of a probability density distribution of values of f J determined from a
population collected over a large time window. Stationary asserts that
the shape of this curve does not change as the sampling window is
translated in time.
ρ( f J )
fJ
Via Random Input
Advanced Vibrations
Define the auto-correlation integral for the imposed force:

∞
u ff ( τ ) = ∫ f J ( t ) f J ( t + τ ) dt . Because the distribution of f J is
–∞
stationary, u ff ( τ ) is well defined (does not change with a change of
variable s = t + α ). Similarly, because the system is linear and time-
translation invariant, the auto-correlation integral u xx ( τ ) for the system
response at location I is well defined as well.
∞
The cross-correlation integral is u xf ( τ ) = ∫ x I ( t ) f J ( t + τ ) dt .
–∞
Note that these infinite integrals should average away much of the
noise in the system. Further, if the excitation is also ergodic, we can
average the correlation integrals for multiple experiments.
Via Random Input
Advanced Vibrations
In summary, the auto-correlation and cross-correlations can be

collected and averaged in various ways to maximize the signal-to-noise.
The Fourier transforms of these integrals have interesting form:
∞ ∞ ∞ iωτ
∫ ∫ ∫e
iωτ
U xf ( ω ) = e u xf ( τ ) dτ = x I ( t ) f J ( t + τ ) dt dτ
–∞ –∞ –∞
∞ ∞ iωτ
= ∫ xI ( t ) ∫ e f J ( t + τ ) dτ dt
–∞ –∞
∞
– i ωτ
= ∫e x I ( τ ) dτ F J ( ω ) = X I ( ω )∗ F J ( ω )
–∞

Via Random Input
Advanced Vibrations
Because our integrals are actually over finite time, we use the expected
value of the right hand side.
S XF ( ω ) = E { U xf ( ω ) } = E { X I ( ω )∗ F J ( ω ) } .
where the expected value is obtained by averaging over a number of
tests. The uncertainty in the expected value is a function of the number
of averages. S XF ( ω ) is the cross-spectral density of X I and F J .
Note that S XF = S FX ∗ .
Similarly,
S XX ( ω ) = E { X I ( ω )∗ X I ( ω ) } & S FF ( ω ) = E { F J ( ω )∗ F J ( ω ) }
are power spectral densities of XI and FJ .

Via the Time Domain
Advanced Vibrations
Contracting the equation
X I ( ω ) = H IJ ( ω )F J ( ω )
on both sides by the appropriate quantity Fourier transformed quantity,
we see that we could as easily evaluate
H IJ ( ω ) = S FX ( ω ) ⁄ S FF ( ω ) and H IJ ( ω ) = S XX ( ω ) ⁄ S XF ( ω )
These equations offer two distinct estimates of H IJ ( ω ) from

experimental data. The ratio of these two estimates is the coherence
2 2
γ ( ω ) = S FX ( ω ) ⁄ ( S FF ( ω )S XX ( ω ) )
2
Unless γ is nearly 1.0, the estimate for H IJ ( ω ) is considered
unreliable.
Next Time
Advanced Vibrations
Clay Fulcher will demonstrate how modal and frequency response are
uses to assess reliability of critical structures.

Lecture 16
Advanced Vibrations
Over the next few weeks, we shall address the vibration of continuous
structure.
• We shall begin deriving the governing equations for idealized
structures such as strings, rods, Euler-Bernoulli beams,
Timoshenko beams, and plates. We shall do this exploiting
D’Alembert’s principle and Hamilton’s principle.
• Where possible, we shall derive closed form solutions to

dynamics of these structures in terms of eigenmodes.
• We shall explore approximate solutions to these structures.
• If time permits, we shall begin an examination of waves.

Strings
Advanced Vibrations
Strings are a good place to begin the process of deriving governing

equations and closed-form solutions.
Strings are one-dimensional structures too thin to exhibit bending

stiffness. Tension in the string serves to drive it toward the equilibrium
configuration - a straight line.
Lets draw a free-body diagram and derive governing equations.
The tension on the string is

F( x) p( x) F ( x + dx ) F ( x ) , the density of the
string isρ ( x ) , and the
distributed load is p ( x )
y' ( x ) ρ( x) y' ( x + dx )

Strings
Advanced Vibrations
We balance the vertical

component of force to find
ρ ( x )ẏ˙( x ) = ( F ( x )y′ ( x ) )′ F( x) p( x) F ( x + dx )
+ p ( x, t ) .
y' ( x ) ρ( x) y' ( x + dx )
In what follows, we shall assume
that F ( x ) = F = Constant
& that ρ ( x ) = ρ = Constant .
We choose easy boundary conditions: y ( 0, t ) = y ( L, t ) = 0 .
Lets now solve the unforced initial value problem:
ρẏ˙( x ) = ( Fy′ ( x ) )′
subject to y ( x, 0 ) = y 0 ( x ) and ẏ ( x, 0 ) = v 0 ( x )

Strings
Advanced Vibrations
Lets solve ρẏ˙( x ) = ( Fy′ ( x ) )′ by separation of variables

y ( x, t ) = ∑ X n ( x )T n ( t )
n
Ṫ˙n F X n ′′ 2
This yields ------ = --- ---------- = – λ n for each n .
Tn ρ Xn
2ρ
Solving X n ′′ + λ n --- X n = 0 , subject to X n ( 0 ) = X n ( L ) = 0 , we
F
find that
2ρ
2
nπx
X n = sin ---------  nπ
and ------ = λ n ---
L  L F

Strings
Advanced Vibrations
 nπ
2
2ρ nπ  F 1 ⁄ 2
------ = λ n --- => λ n = ± ------ --- .
 L F L  ρ
2
The temporal part of the solution is the solution to T n + λ n T n = 0
˙˙
which has solutions T n ( t ) = A n cos ( λ n t ) + B n sin ( λ n t )
The solution to the problem is
∑ n
nπx
y ( x, t ) = ( A cos ( λ n t ) + B n sin ( λ n t ) ) sin ---------
n
L
The velocity of the string is
∑ n n
nπx
ẏ ( x, t ) = λ ( – A sin ( λ n t ) + B n cos ( λ n t ) ) sin ---------
n
L
Strings
Advanced Vibrations
Lets match initial conditions.
∑ n --------
nπx
From prescribed initial position, y0 ( x ) = A sin -
n
L
∑ n n --------
nπx
From prescribed initial velocity, v 0 ( x ) = λ B sin -.
n
L
We use these to solve for the coefficients by exploiting the

orthogonality of sine and cosine.
L L
2 nπx 2 nπx
A n = --- ∫ y 0 ( x ) sin --------- dx & B n = --------- ∫ v 0 ( x ) sin --------- dx
L L λn L L
0 0
Please confirm these. Which is the fundamental tone?
Strings and Positive Operators
Advanced Vibrations
Note that we could have posed this in terms of kinetic energy and
potential energy operators.
∂ ∂ 
Let K be the differential operator K [ y ] = – F (y) and we
∂x ∂x 

consider its operation only on the linear space of smooth functions that
satisfy the geometric boundary conditions ( y ( 0, t ) = y ( L, t ) = 0 ).
K is a symmetric (self adjoint) operator is that

( y 1, K y 2 ) = ( K ( y 1 ), y 2 ) where ( a, b ) is the inner product of
L
functions a and b . In the case of the string, ( a, b ) = ∫ a ( x )b ( x ) dx .
0

Advanced Vibrations
We prove that K is a self-adjoint operator directly.

L
 – ∂  T ∂ (y )  dx
( y 1, K y 2 ) = ∫ 1  ∂ x ∂ x 2  
y ( x )
0
L
∂ ∂  ∂y 1 ∂y 2
= ∫ –
∂x 
y 1 ( x )T (y 2) + T
∂x  ∂x ∂x
dx
0
0 L
∂ L ∂y 1 ∂y 2
= y 1 ( x )T (y 2) + ∫T dx
∂x 0 ∂x ∂x
0
L
∂y 2 ∂y 1
= ∫T dx = ( Ky 1, y 2 )
∂x ∂x
0

Advanced Vibrations
It is also easily seen that K is a positive definite operator

L
∂y ∂y
( y, Ky ) = ∫T ∂ x∂ x
dx > 0 for all non-trivial y ( x ) .
0
We define the mass operator My ( x ) = ρy ( x ) .

It is obvious that M is symmetric: ( y 1, M y 2 ) = ( M y 1, y 2 ) ,
and that M is positive definite: ( y, My ) > 0 for all non-trivial y ( x ) .
The governing equation now has the form M ẏ˙ + Ky = 0 . Except for a
sign difference, this looks similar to the matrix equations for discrete
systems.

Eigenmodes For Continuous Systems
Advanced Vibrations
Lets define U 0 to be the set of twice differentiable functions X defined

in [ 0, L ] that satisfy the boundary conditions at 0 and L .
2 ( X , KX )
We consider the functional λ ( X ) : λ ( X ) = --------------------- operating on
( X , MX )
functions in X ∈ U0,
2 2
Let λ = min λ ( X ) and let X 1 be the function that achieves that
1
X ∈ U0
minimum.

Advanced Vibrations
2 2
Let U 1 = { X ∈ U 0 : ( X , M X 1 ) = 0 } and let λ 2 = min λ ( X )
X ∈ U1
and let X 2 be the function that achieves that minimum.
One continues in this manner:
defining U n = { X ∈ U n – 1 : ( X , M X n ) = 0 } and letting

2 2
λ n + 1 = min λ ( X ) and X n + 1 be the function that achieves that
X ∈ Un
minimum.

Advanced Vibrations
Lets examine the character of the solutions to this optimization

problem.
d ( ( X 1 + αX̃ ), K ( X 1 + αX̃ ) )
Examine ----------------------------------------------------------------- = 0 where X̃ ( x ) is
d α ( ( X 1 + αX̃ ), M ( X 1 + αX̃ ) )
α=0
any function of x satisfying the boundary conditions.
2
Appropriate differentiation shows that ( X̃ , K X 1 – λ 1 M X 1 ) = 0 for
2
all X̃ . This implies that K X 1 – λ 1 M X 1 = 0 . ( λ 1, X 1 ) are and eigen
pair for the governing equation.

Advanced Vibrations
d ( ( X n + αX̃ ), K ( X n + αX̃ ) )
Similarly, Examine ----------------------------------------------------------------- = 0 where
d α ( ( X n + αX̃ ), M ( X n + αX̃ ) )
α=0
X̃ ( x ) is any function of x satisfying the boundary conditions and
orthogonal to the previous eigen functions.
2
Appropriate differentiation shows that ( X̃ , K X n – λn M X n ) = 0 for
all X̃
orthogonal to the previous eigen-functions. This implies that
2
K X n – λ n M X n = 0 . ( λ n, X n ) are and eigen pair for the governing
equation.

Orthogonality of Eigen function
Advanced Vibrations
We examine orthogonality in the same way that we did for discrete

systems.
Say that ( λ m, X m ) and ( λ n, X n ) are eigen pairs with distinct eigen

values: λ n ≠ λ m . We examine the following inner products:
2
( X m, K X n ) – λ n ( X m, M X n ) = 0 and
2
( X n, K X m ) – λ m ( X n, M Xm) = 0.
2 2
Subtracting one from another, we find ( λ n – λ m ) ( X n, M X m ) = 0 .
Eigen functions of distinct eigenvalues are orthogonal w.r.t. M.

Advanced Vibrations
2
( X m, K X n ) – λ n ( X m, M X n ) = 0 and
2
( X n, K X m ) – λ m ( X n, M X m ) = 0
Similarly, if the eigenvalues are distinct, we multiply the first of the

2 2
equations by λ m and the second by λ n and then subtract to find
2 2
( λ n – λ m ) ( X n, K X m ) = 0 .
Eigen functions of distinct eigenvalues are orthogonal w.r.t. K.

Advanced Vibrations
What happens if there are repeated eigenvalues.

• The above proof shows that the corresponding eigen functions
can be made orthogonal to eigen functions of all other
eigenvalues
• The eigen functions of the repeated eigenvalues can be

combined in such a way as to make them orthogonal. We shall
see some examples with beam bending.

Strings and Eigenvalues
Advanced Vibrations
nπx
By the way, we see that X n = sin --------- is the n’th eigen function of
L
2
( K – λ M )X = 0 , where KX = – ( FX′ ) ′ and MX = ρX and the
2 F  nπ 2
corresponding eigen value is λ n = --- ------ .
ρ L 
Note that frequency increases with tension, decreases with density, and
decreases with length.

Strings and Energy
Advanced Vibrations
By inspection, the kinetic energy is

L
ρ 1
T ( t ) = --- ∫ ( ẏ ( x, t ) ) dx = --- ( ẏ, M ẏ ) .
2
2 2
0
Lets also note that the potential energy is

L
F 1
V ( t ) = --- ∫ ( y′ ( x, t ) ) dx = --- ( y, Ky )
2
2 2
0
The equivalence of the above integral and inner product was
established in slide Strings and Positive Operators on page 8. That this
is the strain energy is shown next.

Strings and Energy
Advanced Vibrations
Lets consider the experiment where the string is pulled through a pulley
at the boundary.
The distance that the string must be pulled through the pulleys in order
to stretch the string into its deformed shape is the arc length of the
curved string minus the distance between the boundaries
L L
1 2
∫ 1 + y′ dx – L ≅ --- ∫ y′ dx
2
d =
2
0 0
L
F 2
and the work involved is W = Fd = --- ∫ y′ dx .
2
0
Strings and the Rayleigh Quotient
Advanced Vibrations
2 ( X , KX )
The quotient that we were examining before, λ ( X ) = ---------------------
( X , MX )
actually reflects the energy in periodic motion.
iωt
Say that the deformation is y ( x, t ) = Re { e }X ( x ) . Then
iωt
ẏ = Re { iωe }X ( x ) . The maximum kinetic energy in a cycle is
L
ρ iωt 2 1
∫
2
max --- ẏ ( x, t ) dx = max ( Re { iωe } ) --- ( X , MX )
2π 2 2π 2
0 < t < ------ 0 0 < t < ------
ω ω
1 2
= --- ω ( X , MX )
2

Advanced Vibrations
1
Similarly, the maximum strain energy in each cycle is --- ( X , KX ) .
2
2 2 ( X , KX )
We now see that λ n = ω n = --------------------- is the ratio of maximum
( X , MX )
strain energy in a mode to the maximum kinetic energy in that mode.
( X , KX )
2
The function λ ( X ) = --------------------- is called the Rayleigh quotient and
( X , MX )
is often used to obtain upper bounds for the first natural frequency.

Advanced Vibrations
Lets see how well f ( x ) = x ( L – x ) approximates the first eigen

function:
∫ F ( f′ ( x ) ) dx
2
3
2 FL ⁄ 3 F
λ ( f ) = --------------------------------
0
L
- = ------------------
5
- = 10 --------
-
2
ρL ⁄ 30 ρL
∫ ρ ( f ( x ) ) dx
2
This is not to far from the exact solution

2
2 πx π F F
λ sin ------ = ---------2- = 9.87 2-
--------
 L ρL ρL

Derivation of String Equations
via Hamilton’s Principle
Advanced Vibrations
We derived the equation of motion for the string via a straightforward

application of D’Alemberts principle (balancing force and inertial
terms). Lets try it now with Hamilton’s principle.
The Weak form of Hamilton’s Principle: asserts that the actual path is
t2
one about which

∫ ( δT – δV + δW ) dt = 0 . Here, δW is the virtual
t1
work of external forces pushing the system from one path in
configuration space to another.
Lets first ignore all external forces.

Derivation of String Equations
via Hamilton’s Principle
Advanced Vibrations
t2
0 = ∫ ( δT – δV ) dt
t 2t 1 L t2 L
ρ 2  F 2 
= δ ∫ ---  ∫ ( ẏ ( x, t ) ) dx dt – δ ∫ ---  ∫ ( y′ ( x, t ) ) dx dt
2  2 
t 1 0 t 10
L t2 t2 L
∂
= ∫∫ ∂t
( ρẏδy ) – ρẏ˙δy dt dx – ∫ ∫ F [ ( y′δy )′ – y′′δy ] dx dt
0 t1 t1 0
L 0 t2 0
t2 t2
= ∫ ( ρẏδy ) t dx – ∫ ( F ( y′δy ) t ) dt
1 1
L t2
0 t1
 
– ∫  ∫ [ ρẏ˙ – Fy′′ ]δy dt dx

0 t

1
Next Time
Advanced Vibrations
A little more on Strings

Rods

Lecture 17
Rods and More Strings
Advanced Vibrations
These are 1-dimensional structures.

• With rods, only longitudinal deformation is considered.
• With strings, only lateral motions were considered.

A More Complicated String Problem
Deformations in Two Dimensions
Advanced Vibrations
We studied the string vibrating in a plane earlier. Lets now consider a

string that can move in either of two lateral directions.
y ( x, t ) = u ( x, t ) j + v ( x, t )k
Further, we consider a more complicated boundary condition.
j
k
We restrict the left hand end of the string to move within a track in the
vertical direction: y ( 0, t ) = s ( t ) j (=> v ( 0, t ) = 0 ).

The string is pinned on the right-hand-side:
y ( L, t ) = 0 (=> u ( L, t ) = v ( L, t ) = 0 ).
Advanced Vibrations
We also consider a distributed load p ( x, t ) .

Again, we employ Hamilton’s principle. We shall need each of the terms
t2
in 0 = ∫ ( δT – δV + δW ) dt .
t1
L L
1 ˙ ˙ 1
T = --- ∫ ρy ( x, t ) • y ( x, t ) dt , V = --- ∫ F y′ ( x, t ) • y′ ( x, t ) dt , and
2 2
0 0
L
δW = ∫ p ( x, t ) • δy ( x, t ) dt
0

Advanced Vibrations
t2 t2 L
 ˙ ˙ 
∫ δT dt = ∫  ∫ ρy ( x, t ) • δy ( x, t ) dx dt
t1 t 0 1
L t2
 ∂ ˙ ˙˙ 
= ∫∫ ( ρy • δy ) – ρy • δy dt dx
 ∂t 
0 t 1
L 0 L t2
˙ t2  ˙˙ 
= ∫ ( ρy • δy ) t 1 dx – ∫  ∫ ρy • δy dt dx
0

0 t

1

Advanced Vibrations
t2 t2 L
 
∫ δV dt = ∫  ∫ F y′ ( x, t ) • δy′ ( x, t ) dx dt
t1 t 0 1
t2 L
 ∂ 
= ∫ ∫ F ( y′ • δy ) – F y′′ • δy dt dx
 ∂x 
t 0 1
t2 L t2
L  
= ∫ ( F y′ • δy ) 0 dt – ∫  ∫ F y′′ • δy dt dx
 
t1 0 t1

Advanced Vibrations
L t2
 ˙˙ 
∫∫
We put this all together to find:  [ F y′′ – ρy + p ] • δy dt d x = 0

0 t 1
˙˙
for all δy from which we deduce that ρy = F y′′ + p <=>
ρu̇˙ = Fu′′ + p • j and ρv̇˙ = Fv′′ + p • k
t2
L
Also,
∫ ( F y′ • δy ) 0 dt = 0 from which we conclude that
t1
t2
∫ F y′ ( 0, t ) • δy ( 0, t ) dt = 0 => y′ ( 0, t ) • ( δs j ) => u′ ( 0, t ) = 0
t1
Advanced Vibrations
We have two independent partial differential equations with their own

boundary conditions:
ρu̇˙ = Fu′′ + p ( x, t ) • j subject to u′ ( 0, t ) = 0 & u ( L, t ) = 0

and
ρv̇˙ = Fv′′ + p ( x, t ) • k subject to v ( 0, t ) = 0 & v ( L, t ) = 0
We shall discuss how to solve these equations later in this lecture.

Governing Equations for Rods
By Hamilton’s Principle
Advanced Vibrations
F0 FL
x
u ( x, t )
We consider a rod of length L , cross sectional area A , density ρ , and

Young’s modulus E . We wish to derive a governing equation for this.
We shall use Hamilton’s Principle.

Advanced Vibrations
Recall: The Weak form of Hamilton’s Principle: asserts that the actual
t2
path is one about which

∫ ( δT – δV + δW ) dt = 0 . Here, δW is the
t1
virtual work of external forces pushing the system from one path in
configuration space to another.
Lets enumerate the components of the above equation:
L L
 ∂u 2
∂u 2
, T = --- ρA   d x ,
1 1
V = --- ∫ EA
2 ∫ ∂t 
dx
2  ∂x
0 0
and
δW = – F 0 δu ( 0, t ) + F L δu ( L, t )

Advanced Vibrations
For the rod, Hamilton’s principle becomes

t2
0 = ∫ ( δT – δV + δW ) dt
t1
t2 L t2 L
   
= ∫  ∫ ρAu̇δu̇ dx dt – ∫  ∫ EAu′δu′ dx dt

t 0
 
t 0

1 1
t2
+ ∫ ( – F 0 ( t )δu ( 0, t ) + F L ( t )δu ( L, t ) ) dt
L t2 t1 t2 L
 ∂ [ u̇δu ] – u̇˙δu dt dx – EA  ∂ [ u′δu ] – u′′δu dx dt
= ∫ ∫ ∂t
ρA

t2
∫ ∫ ∂x 
0 t1 t1 0
+ ∫ ( – F 0 ( t )δu ( 0, t ) + F L ( t )δu ( L, t ) ) dt
t1
Advanced Vibrations
After some re-organization, this becomes
L 0 L t2
t2
0 = ∫ ( ρAu̇δu ) t dx – ∫ ∫ ( ρAu̇˙ – EAu′′ )δu dt dx
1
0 0 t1
t2 t2
– ∫ ( EAu′δu ) 0 dt + ∫ [ – F 0 ( t )δu ( 0, t ) + F L ( t )δu ( L, t ) ] dt

L
t1 t1
From which we conclude that ρAu̇˙ – EAu′′ , that

[ EAu′ ( L, t ) – F L ( t ) ]δu ( L, t ) = 0 and that
[ EAu′ ( 0, t ) – F 0 ( t ) ]δu ( 0, t ) = 0 .
Hamilton’s principle provides governing equations and form of B.C.
Solutions for Rods
Advanced Vibrations
Say we have a specific problem. Our rod is fixed on one end and subject
to a prescribed load on the other.
F (t)
x
u ( x, t )
L
We shall try to solve this using a modal solution, but first, we must find
the modes.
We consider the unforced beam and solve ρAu̇˙ – EAu′′ = 0 subject

to the boundary conditions: u ( 0, t ) = 0 & u′ ( L, t ) = 0 .
From where does the second B.C. come?

Modal Solutions for Rods
Advanced Vibrations
iωt
We postulate solutions u ( x, t ) = Re ( Ce )X ( x ) . This yields the
2
2 2 ω ρ
equation for X : X′′ + λ X = 0 where λ = ---------- .
E
The only solutions to the above equation with the specified boundary
 π ( 2n – 1 ) x 
conditions are X n ( x ) = α n sin ----------------------- --- .
 2 L
Note that, as expected, ( X m, M X n ) = 0 for m ≠ n .

Mass Normalization
Advanced Vibrations
Lets choose the constants α n so that the modes are orthonormal with
respect to the mass matrix. ( X m, M X n ) = δ mn requires that
2 2  π ( 2n – 1 ) x 
αn = ------- so that X n( x) = ------- sin ----------------------- --- .
Lρ Lρ  2 L
Please confirm that this normalization is correct.

Advanced Vibrations
Lets now make a change of variables to discretize the problem.
N
Assume that u ( x, t ) = ∑ f n ( t )X n ( x )
n=1
Let now derive our governing equations in terms of our new variables
f n ( t ) using Lagrange equations. (Lagrange has been a good friend to
us.)
L N N
1   
T = --- ∫ ρA  ∑ f m ( t )X m ( x )  ∑ f n ( t )X n ( x ) dx
˙ ˙
2
0
m = 1  n = 1 
N N L N
1  1
= ---  ∑ ∑ f m ( t ) f n ( t ) ∫ ρAX m ( x )X n ( x ) dx = --- ∑ f n ( t )
2
˙ ˙ ˙
2  2
m = 1n = 1 0 n=1

Advanced Vibrations
The potential energy is a little more complicated:
L N N
1   
V = --- ∫ EA  ∑ f m ( t )X m ′ ( x )  ∑ f n ( t )X n ′ ( x ) dx
2
0
m = 1  n = 1 
N N L
1 
= --- 
2 ∑ ∑ f m ( t ) f n ( t ) ∫ EAX m ′ ( x )X n ′ ( x ) dx
m = 1n = 1 0
N
1  π EA
2
 ∑ n f n(t )
2 2
= ---  -------------
2  ρL  2
n=1

Advanced Vibrations
The generalized force associated with each mode calculated from the
work done by the physical forces against each mode:
X X
δW n = δfn X n ( L )F ( t ) = δfn Q n ( t ) => Q n(t ) = X n ( L )F ( t )
The constituents of the Lagrange equation are:
d  ∂T  ∂T ∂V  π 2 EA 2
– = f˙˙n , – = –  -------------n fn
 
d t ∂ f˙n ∂ f n ∂fn  ρL  2
The Lagrange equations for this system are:
 π 2 EA 2 2  π ( 2n – 1 )
f n = –  -------------
˙˙  n f n + F ( t ) ------- sin  -----------------------
 ρL  2 Lρ 2
Advanced Vibrations
The forcing term simplifies a little bit:
 π 2 EA 2 2 n
f˙˙n = –  ------------- n f n – F ( t ) ------
- ( – 1 )
 ρL  2 Lρ
We might solve this by Laplace transform methods or

we might solve this by numerical integration.
Having found a sufficient number of the f n ( t ) , we can reconstitute the

N
physical solution from u ( x, t ) = ∑ f n ( t )X n ( x ) .
n=1

Advanced Vibrations
To re-cap:
• Having identified the potiential and kinetic energy, we can use
Hamilton’s principle to deduce the governing equations.
• Modal solutions are found to the unforced problem
• Mode shapes are used to discretize the problem and can be used
to introduce generalized degrees of freedom in solving the forced
vibration problem.
• Using modal coordinates diagonalizes the problem for us.
Please think about how we could introduce damping into this analysis.

Next Time
Advanced Vibrations
Numerical Solutions for Strings and Rods

Galerkin Method
We saw how we could use the eigen functions to convert the partial
differential equations into systems of ordinary differential equations
that could be solved by

Lecture 18: Numerical Methods
Advanced Vibrations
Numerical Solutions for Rods

Galerkin Method
We saw how we could use the eigen functions to convert the partial
differential equations into systems of ordinary differential equations
that could be solved by expressing the displacement field a a linear
combination of eigen modes.
Application of the Lagrange equations resulted in a system of ordinary
differential equations in time - one equation per modal coefficient.
In this lecture, we discuss the solution of vibration problems when we

do not have a closed-form expression for the eigen modes. There are
two general approaches
• numerical solution to approximate the eigen functions
• solution for displacements with going through eigen analysis

Numerical Solution For Eigen Functions
Advanced Vibrations
Say that we have a problem for which we cannot calculate closed form
eigen functions. One such problem is that pictured below where the
cross section is non-constant. We shall now expand the displacement

field in terms of assumed modes
u ( x, t ) = ∑ an ( t )Pn ( x ) ,
n
and then employ this expansion in creating the Lagrange equations.
The only requirements that we place on P n ( x ) are that each satisfy
what ever geometric boundary conditions exist.

Governing Equations
Advanced Vibrations
L
T ( t ) = --- ∫ ρA ( x )  ∑ ȧ m ( t )P m ( x )  ∑ ȧ n ( t )P n ( x ) dx
1
2   
0 m n
L
V ( t ) = --- ∫ EA ( x )  ∑ a m ( t )P m ′ ( x )  ∑ a n ( t )P n ′ ( x ) dx
1
2   
0 m n
and
δW = ∑ δan ( t )Pn ( L )F ( t ) = ∑ δan ( t )Qn

n n

Governing Equations
Advanced Vibrations
We have done this derivation before. The following should be familiar.
d  ∂T  ∂V

d t ∂ ȧ m
+
∂ am
= ∑ M mn ȧ˙n + ∑ K mn an = Qm ( t )
n n
L
where K mn = ∫ EA ( x )Pm ′ ( x )Pn ′ ( x ) dx ,
0
L
M mn = ∫ ρA ( x )Pm ( x )Pn ( x ) dx , and
0
L
Qm ( t ) = ∫ Pm ( x )F ( x, t ) = P m ( L )F ( t )
0
Governing Equations
How to choose the P m
Advanced Vibrations
Criteria for choosing the assumed modes:

• linear combinations of the assumed modes must be able to take
on the admissible configurations of the problem. (They must
span the space.)
• it should not be too difficult to compute the generalized

coordinates and their time derivatives in terms of the initial
deformation and velocity of the structure.
Lets try some examples

Example with polynomial shape functions
Advanced Vibrations
Consider a tapered rod having cross section distribution
A ( x ) = A 0  1 – ------
x
 2L

Assumed Modes Case 1
Advanced Vibrations
n
Lets try the expansion defined by Pn ( x ) = ( x ⁄ L )
L
 1 – -----
x
∫ 0  2L
m n
M nm = ρ A - ( x ⁄ L ) ( x ⁄ L ) dx
0
ρ A0 L (m + n + 3)
= ------------- --------------------------------------------------------
2 (m + n + 1)(m + n + 2)
L
m+n–2
K mn
mn  x
= ------2- ∫ E A 0 1 – ------ ( x ⁄ L ) dx
 2L 
L 0
E A 0  ( mn ) ( m + n + 1 ) 
= ---------- ----------------------------------------------
2L  ( m + n – 1 ) ( m + n )

Advanced Vibrations
We have established that in this case, we can do the necessary

integrals.
The question remains: how to find initial values: a n ( 0 ) and ȧ n ( 0 ) ?.

The answer is through the use of Bernstein polynomials.
n
 k   n  x  k   x  n – k
B n ( y ( x, 0 ) ) = ∑ y L ---, 0 --- 1 – -- → y ( x, 0 )
 n   k  L   k 
0
n
 k   n  x  k   x  n – k
B n ( ẏ ( x, 0 ) ) = ∑ ẏ L ---, 0 --- 1 – -- → ẏ ( x, 0 )
 n   k  L   k 
0
We have established here that polynomials could be used to solve rod
problems. This approach is correct, but not very appealing.

Advanced Vibrations
Lets use the the chapeau functions as our assumed modes
P1 P2 P3 P4 P5 P6 P7
Each of these functions has

• a value of 1.0 on a single node,
• a non-zero value on a small region about its node (the support of

the shape function),
• and a value of zero on every other node.
The approximation for y ( x, t ) is y ( x, t ) = ∑ yn ( t )Pn ( x )

n
Advanced Vibrations
Initial conditions are obtained by several alternate methods, the most

direct of which is collocation:
y n ( 0 ) = y ( x n, 0 ) and ẏ n ( 0 ) = ẏ ( x n, 0 )
L
 1 – -----
x
The integrals M nm = ∫ 0  2L- yn ( x )ym ( x ) dx and
ρ A
0
L
K nm  x
= ∫ E A 0 1 – ------ y n ( x )y m ( x ) dx are calculated by numerical
 2L
0
integration.
The integration need be performed only where the support of ym ( x )

intersects that of yn ( x )
Lecture 19: Membranes
Formulations And Some Analytic Solutions
Advanced Vibrations
Membranes are two-dimensional systems which, at rest, will occupy the

configuration of minimum curvature consistent with the boundary
conditions. The restoring force of these membranes derives from an
isotropic surface tension.
In what follows, we assume that the boundaries lie entirely in a plane. In
this case, the membrane will also lie in a plane when at rest.
There are some cases for which it is possible to achieve closed form
solutions for the vibration of membranes. The best known of these are
those with rectangular or circular boundaries.

Governing Equations
Hamilton’s Principle Again
Advanced Vibrations
The only part of this problem which is not

immediately obvious is the calculation of
potential energy. We identify the strain
energy with the surface tension and the
surface area of the membrane. The strain
and kinetic energies and virtual work are:
1
∫ ∫A τ 1 + ∇w • ∇w d A T = --- ∫ ∫ ρẇ d A
2
V =
2 A
1
≈ ∫ ∫ τ 1 + --- ∇w • ∇w d A
A 2
δW = ∫ ∫A pδW d A
where w is the out-of-plane displacement of the membrane, F is the
surface tension in the membrane, ρ is the mass per unit area of the
membrane, and integration is over the area enclosed by the boundaries.

Governing Equations
Advanced Vibrations
t2 ∂ t2 t2
0 = ∫ ∫ ∫ ρẇ δw d A – ∫ ∫ ∫ τ∇w • ∇δw d A + ∫ ∫ ∫ pδW d A
t1 A ∂t t1 t1 A
As usual, we must do an integration by parts with respect to time to

∂
move the δw into δw and we must do an integration by parts
∂t
(Gauss’s theorem) with respect to the spacial coordinate to move the
∇δw into δw .
t2 ∂
0 = ∫t ∫ ∫A
1
ρ
∂t
( ẇδw ) – ẇ˙δw d A
t2
–∫ ∫ ∫A
2
τ [ ∇ • ( ∇wδw ) – ∇ wδw ] d A
t1
t2
+∫ ∫ ∫ pδW d A
t 1 A
Governing Equations
Advanced Vibrations
Evaluating the above and grouping like terms,
t2
∫t ∫ ∫A [ ρẇ˙ – τ∇
2
0 = w – p ]δw d A for all δw on A subject to
1
2
δw = 0 on ∂A . From the above, ρẇ˙ – τ∇ w – p = 0 on the
interior of A .
t2
Also,
∫t ∫∂A τ [ ∇w • n ]δw dΓ
1
= 0 where
the integral is over the boundary ∂A of A .

Here, n is a unit outward normal. This asserts
that at locations on the boundary for which w
∇w • n = 0 . In the following, we will consider only those
is not fixed,
case for which w is fixed on the boundaries.

Membrane Solutions on a Rectangle
Advanced Vibrations
In rectangular coordinates,
y
 ∂2 ∂
2
L2 τ  2 + 2  w + p ( x, y, t ) = ρẇ˙
∂x ∂y 
. Boundary conditions are
L1
w ( 0, y, t ) = w ( L 1, y, t ) = 0 and
x
w ( x, 0, t ) = w ( x, L 2, t ) = 0 .
Lets first try solving the homogeneous equation via separation of
variables (implying eigen analysis).
w ( x, y, t ) = ∑ f λ ( t )W λ ( x, y )
λ
2
∇ Wλ ρ fλ
˙˙
2
This leads to --------------- = --- ------ = – λ
Wλ τ fλ
Rectangular Membrane Eigen Analysis
Advanced Vibrations
2 2 πmx πny
∇ W λ + λ W λ = 0 is satisfied by W λ = sin ----------- sin ---------
L1 L2
 πm 2  πn 2
2
We see that this works if λ = ------- + ------ . (we could have
 L1   L2 
derived the above by doing another separtation of variables:
W λ ( x, y ) = ∑ X λ ( x )Y λ ( y ) .)
Substituting back into our governing equation, we see that our eigen
πmx πny τ
solutions, Wλ = sin ----------- sin --------- , have frequency ω λ = --- λ
L1 L2 ρ
We shall exploit the mutual orthogonality of the eigenmodes. As usual,

we orthonormalize the modes: ( ρW λ , W λ ) = δ λ , λ .
1 2 1 2
Forced Oscillation of Rectangular
Membranes
Advanced Vibrations
2
The governing equation is ρẇ˙ – τ∇ w – p ( x, y ) = 0 . Just as we
have done in the past, we express w ( x, y, t ) = ∑ f λ ( t )W λ ( x, y ) ,
λ
and derive Lagrange equations using the f λ(t ) as generalized
coordinates. After exploitation of the orthonormality, we have
independent equations for the modal coefficients:
τ 2
f˙˙λ + --- λ f = ( p ( x, y, t ), W λ )
ρ

Eigenmodes on triangles
This would be a good assignment
Advanced Vibrations
Analytic solutions on other polygons are

harder. Lets consider the eigen problem on s = x+y–L
the right isosceles triangle shown.
y L
We might approximate the deformed shape of

the membrane on the triangle by the following
L
polynomials: x
W = xy ( x + y – L ) [ A 1 + A 2 x + A 3 y ]
= A 1 P 1 ( x, y ) + A 2 P 2 ( x, y ) + A 3 P 3 ( x, y )
We could use these basis functions to approximate the eigenmodes of a
membrane on this triangle.

Approximate Eigenmodes on triangles
Advanced Vibrations
W = A 1 P 1 ( x, y ) + A 2 P 2 ( x, y ) + A 3 P 3 ( x, y )
Lets set up the formulation to estimate the eigen shapes:
• The Lagrange equations are [ M ] { Ȧ˙} + [ K ] { A } = 0 where

M mn = ∫ ∫ ρP m P n d A and K mn = ∫ ∫ τ∇P m • ∇ P n d A
iωt
• Postulate a harmonic motion { A } = Re ( Ce { A0 } )
2
• Solve the resulting eigen problem ( – ω M + K ) { A 0 } = 0 for
1 2 3
three arrays { A }, { A }, { A }
• Construct the eigen shapes from
k k k
W k ( x, y ) = A 1 P 1 ( x, y ) + A 2 P 2 ( x, y) + A 3 P 3 ( x, y)

Vibration of Circular Membranes
Advanced Vibrations
2 2
2 ∂ 1∂ 1 ∂
In a circular domain, ∇ = + --- + ----2 2
∂r
2 r ∂r r ∂θ
We assume that there are no holes in the membrane, so the boundary
conditions are w ( R, θ, t ) = 0 .
Lets first try solving the homogeneous equation via separation of
variables (implying eigen analysis).
w ( r, θ, t ) = ∑ f λ ( t )W λ ( r, θ )
λ
2
∇ Wλ ρ fλ
˙˙
2
This leads to --------------- = --- ------ = – λ
Wλ τ fλ
(I hope that this looks very familiar to you.)

Advanced Vibrations
2 2
∇ W λ + λ W λ = 0 is itself solved via a separation of variables:
W λ = R λ ( r )Θ λ ( θ ) . On substitution back into the differential
2
∂
2
1 ∂R λ  ∂ Θ
R + ---  2
2 λ r ∂r
∂r 1 ∂θ  2
equation, we find ----------------------------------- + ---- --------------- + λ = 0
r Θ
Rλ 2
2
 ∂ Θ
from which we conclude that   ⁄Θ = C where C is a constant.
∂θ 
2
 ∂2 1 ∂R λ C 2
We also conclude that  R
2 λ
+ --
-  + ---- R λ + λ Rλ = 0 .
∂r r ∂r  r 2

Advanced Vibrations
In order for the solution to make sense, Θ must be periodic in θ , so

2
2  ∂ Θ 2
C = – m and  2  + m Θ = 0 where m is any nonzero integer.
∂θ 
The solution to this equation, denoted as Θ m , is
Θ m ( θ ) = C 1m sin ( mθ ) + C 2m cos ( mθ ) .
The equation for the radial function is
 ∂2 1 ∂R λ m
2
2
 2 R λ, m + ---  – ------ R λ, m + λ R λ, m = 0 . This has solutions
∂r r ∂ r  r 2
R λ, m ( r ) = C 3m J m ( λr ) + C 4m Y m ( λr ) where J m is the Bessel

function of the first kind and order m and Y m is the Bessel function of
the second kind and order m .

Advanced Vibrations
Note thatY m has a singularity about zero. If we have a membrane

without holes, then we must set C 4m = 0 .
We still must satisfy the boundary conditions that W ( R, θ ) = 0 . This

requires that J m ( λR ) = 0 . This condition is used to find λ .
This offers a large number of solutions. For each order, m , there is a

Bessel function of the first kind. That Bessel function has an infinite
number of zeros. Lets call the n th zero of the m th Bessel function
x m, n .
1 2
Then W ( r, θ ) = [ C m, n sin ( mθ ) + C m, n cos ( mθ ) ]J m ( λ m, n r ) is
an eigen mode where λ m, n = x m, n ⁄ R . Frequency is
ω m, n = τ ⁄ ρ λ m, n
Advanced Vibrations
Of course, we can use orthogonality in the usual manner to construct

solutions for given initial conditions and loads on the membrane.
It is interesting to see what these modes look like. Following is a plot of
the first and second Bessel functions of the first kind
1
0.5
3.832
3.832 7.106
7.106
2.405 5.520 8.654
−0.5
0 1 2 3 4 5 6 7 8 9 10
Lets look at the first eigen modes.

First Eigen Mode
J 0 ( 2.405r ) (first zero of J 0 )
Advanced Vibrations
0.5
−0.5
1
0.8
0.6
1
0.4
0.2
0.5
0
−0.2 0
−0.4
−0.6 −0.5
−0.8
−1 −1

Second Eigen Mode
Advanced Vibrations
0.5
−0.5
−1
1
0.8
0.6
1
0.4
0.2
0.5
0
−0.2 0
−0.4
−0.6 −0.5
−0.8
−1 −1

Third Eigen Mode
Advanced Vibrations
0.5
−0.5
1
0.8
0.6
1
0.4
0.2
0.5
0
−0.2 0
−0.4
−0.6 −0.5
−0.8
−1 −1

Fourth Eigen Mode
J 0 ( 5.520r ) (second zero of J 0 )
Advanced Vibrations
0.5
−0.5
1
0.8
0.6
1
0.4
0.2
0.5
0
−0.2 0
−0.4
−0.6 −0.5
−0.8
−1 −1

Fifth Eigen Mode
Advanced Vibrations
0.5
−0.5
1
0.8
0.6
1
0.4
0.2
0.5
0
−0.2 0
−0.4
−0.6 −0.5
−0.8
−1 −1

Sixth Eigen Mode
J 1 ( 7.016r ) (second zero of J 1 )
Advanced Vibrations
0.5
−0.5
−1
1
0.8
0.6
1
0.4
0.2
0.5
0
−0.2 0
−0.4
−0.6 −0.5
−0.8
−1 −1

Seventh Eigen Mode
Advanced Vibrations
0.4
0.2
−0.2
−0.4
1
0.8
0.6
1
0.4
0.2
0.5
0
−0.2 0
−0.4
−0.6 −0.5
−0.8
−1 −1

Problem 1
Part 1
Advanced Vibrations
Part 1: For the system shown,

• Derive the governing equations of the system
• Linearize those equations about the state of stable equilibrium
R R R
M1 M1 M1
I1 I1 I1
L θ2 L θ3 L
θ1
k k
M2 M2 M2

11/30/98 1 /home/djsegal/UNM/VibCourse/slides/ProblemD.frm
Problem 1
Part 2
Advanced Vibrations
Calculate the eigen modes and frequencies of the system, constrained

by two links, as shown below
R R R
M1 M1 M1
I1 I1 I1
L θ2 L θ3 L
θ1
k
M2 M2 M2
The top linkage is has a length equal to the distance between the
centers of the disks.

Problem 2
Advanced Vibrations
Consider the configuration

F(t) = F0 sin(ωt)
shown, assuming the beams
ΕΙ, m y 1 ( x, t ) deform as
2
y 1 ( x, t ) = A ( t ) ( x ⁄ L )
K and
y 2 ( x, t ) 2
ΕΙ, m y 2 ( x, t ) = B ( t ) ( x ⁄ L ) .
In this figure, EI is the
µ bending stiffness, m is mass
L
per unit length, K is spring
stiffness, and µ is the
coefficient of the damper shown.
In steady state harmonic excitation, we expect that
iωt iωt
A ( t ) = Re { A 0 e } and B ( t ) = Re { B 0 e } .
Calculate A 0 ⁄ F 0 and B 0 ⁄ F 0
Problem E
A Damping Homework
Advanced Vibrations
We consider a structure such as that shown below

K
K
M M
A, h
u2 u1
We have two masses, two springs, and a shear damper.The cross

sectional area of the damper is A and the thickness of the damping
material is h . The damping material is that discussed earlier. For
6
M = 3 kg; K = 1.0 ×10 N/m; and A = 0.01 m2,
1. calculate the damping of both modes for h = 0.01 m
2. determine empirically the appropriate thickness, h, to make the
small modal damping as large as possible.

12/8/98 1 /home/djsegal/UNM/VibCourse/slides/ProblemE.frm
The Damping Material
Advanced Vibrations
The following plot for polyisobutylene is taken from “Vibration

Damping” by Nashif et. al.
E’=3G’
tan(δ)

Interpolating Functions
Advanced Vibrations
ω 5
2π
ω 3.3
  2π 
  ω   3.3
5.72 + 0.0112 Log ------
  2π 
G' ( ω ) = 10 and
 ω   ω  2
 2π   2π 
G'' ( ω ) = G' ( ω ) tan δ ( ω )

 ω   ω  2   ω   3.3
4.72 + 0.56Log ------ – 0.069 Log ------ + 0.0112 Log ------
 2π   2π    2π 
= 10
Solution
Using a lot of Matlab
Advanced Vibrations
The governing equation in frequency space is
= F (ω)
m 0
2 U1 K –K U1
–ω +
0 m U2 – K ( 2K + G∗ ( ω ) ( A ⁄ h ) ) U 2 0
where we are mimicking a driving force on mass 1.
1. We employ matlab to invert the above system to solve for the
driving point frequency response H 1(ω) = U 1(ω) ⁄ F (ω) .
2. We plot H 1(ω) versus ω (or f , it makes no difference. why?)

3. Find the peaks
∆ω
4. Determine fraction of critical damping for each mode via ζ = ---------
2ω n

Results for h = 0.01 == A ⁄ h = 1.0
Advanced Vibrations
−5 Frequency Response of 2−Mass System with Damping

x 10
1.2
A/h = 1
0.8 zeta = 0.069727

|FRF| Mass 1
0.6
|H1|
0.4
0.2 |H2|
zeta = 0.098924
0
0 20 40 60 80 100 120 140 160 180 200
Frequency Hz

Results for h = 0.1 == A ⁄ h = 0.1
Advanced Vibrations

x 10
8
6 A/h = 0.1
|H1|
5 zeta = 0.013697
|FRF| Mass 1
|H2|
1
zeta = 0.0085975
0
0 20 40 60 80 100 120 140 160 180 200
Frequency Hz

Results for h = 0.002 == A ⁄ h = 5
Advanced Vibrations

x 10
1.2
A/h = 5
0.8 zeta = 0.055152

|FRF| Mass 1
|H1|
0.6
0.4
0.2 |H2|
0
0 20 40 60 80 100 120 140 160 180 200
Frequency Hz

Results from many calculations
Advanced Vibrations
0.12
Mode 2
0.1
0.08
ζ
0.06 Mode 1
0.04
0.02
0
0 0.5 1 1.5 2 2.5 3 3.5 4 4.5 5
A/h
What do we observe from the above?
Advanced Vibrations
We note the following

• Increasing thickness drops the resonant frequencies. Decreasing
thickness raises natural frequencies.
• The second mode, which is associated with toward each other, is

suppressed as the thickness is decreased. What is happening.
• The first mode, which is associated with masses moving in the

same direction, does have a configuration of maximum damping.
• Optimal damping appears to happen in the vicinity of h = 0.005 m

(5.0 mm)
What do we imagine to the limiting mode as h → 0?

Final Note
Advanced Vibrations
This would be only a marginally good candidate for the modal strain
energy method.
The reason is that the viscoelastic represents an appreciable part of the

stiffness of the structure. That means that the stiffnesses are
appreciably different at different frequencies.

Take Home Final Exam
A Damping Problem
Advanced Vibrations
You may use any references, but do not ask anyone for help.
Please put you finished exams in an envelope and place that in my mail
slot no later than noon Wednesday 16 December 1998.

K
K
M M
A, h
u2 u1

8
M = 300 kg; K = 1.0 ×10 N/m; A = 0.01 m2, and h = 0.01 m
12/8/98 1 /home/djsegal/UNM/VibCourse/slides/TakeHomeFinal.frm
Problem
Advanced Vibrations
Use Matlab in any parts of the following that you wish

1. Solving the governing equation in frequency space, plot the
magnitude of the driving point frequency response function versus
frequency.
2. From the above, calculate the fraction of critical damping for each
mode using half-power points.
3. Using the modal strain energy method, calculate the fraction of

critical damping for each mode. These should be similar to the
values calculated above.
4. Find the elastic eigenmodes and frequencies. You may ignore the
slider for this, but you must tell me why.
5. Assuming modal damping, solve the problem for the following

initial conditions:
u 1 ( 0 ) = 0 , u 2 ( 0 ) = 0 , u̇ 1 ( 0 ) = 1 , u 2 ( 0 ) = 0 ,
and plot u̇ 1 ( t ) and u̇ 2 ( t ) for several cycles of the lower frequency.
Advanced Vibrations

E’=3G’
tan(δ)

Advanced Vibrations
ω 5
2π
ω 3.3
  2π 
  ω   3.3
5.72 + 0.0112 Log ------
  2π 
G' ( ω ) = 10 and
 ω   ω  2
 2π   2π 
G'' ( ω ) = G' ( ω ) tan δ ( ω )

 ω   ω  2   ω   3.3
4.72 + 0.56Log ------ – 0.069 Log ------ + 0.0112 Log ------
 2π   2π    2π 
= 10
Take Home Final Exam
A Damping Problem
Advanced Vibrations
You may use any references, but do not ask anyone for help.
Please put you finished exams in an envelope and place that in my mail
slot no later than noon Wednesday 16 December 1998.

K
K
M M
A, h
u2 u1

8
M = 300 kg; K = 1.0 ×10 N/m; A = 0.01 m2, and h = 0.01 m
TakeHomeFinalSolution.frm
Problem
Advanced Vibrations
Use Matlab in any parts of the following that you wish

frequency.
3. Find the elastic eigenmodes and frequencies. You may ignore the
slider for this, but you must tell me why.
4. Using the modal strain energy method, calculate the fraction of

critical damping for each mode. These should be similar to the
values calculated above.
5. Assuming modal damping, solve the problem for the following

initial conditions: u 1 ( 0 ) = 0 , u 2 ( 0 ) = 0 , u̇ 1 ( 0 ) = 1 ,
u 2 ( 0 ) = 0 ,and plot u̇ 1 ( t ) and u̇ 2 ( t ) for several cycles of the
lower frequency.
Advanced Vibrations

E’=3G’
tan(δ)

Advanced Vibrations
ω 5
2π
ω 3.3
  2π 
  ω   3.3
5.72 + 0.0112 Log ------
  2π 
G' ( ω ) = 10 and
 ω   ω  2
 2π   2π 
G'' ( ω ) = G' ( ω ) tan δ ( ω )

 ω   ω  2   ω   3.3
4.72 + 0.56Log ------ – 0.069 Log ------ + 0.0112 Log ------
 2π   2π    2π 
= 10
Solution Part 1
Advanced Vibrations

frequency.
In frequency space, the governing equation is
U1 k –k U1 F1(ω)
–ω M
2
+ A =
M U2 – k 2k + G∗ ( ω ) --- U 2 F2(ω)
h
U1 F1(ω)
We write this as A(ω) = . The frequency response
U2 F2(ω)
is found by inverting A(ω) .

Solution: Part 1
Advanced Vibrations
U1 –1 F1(ω) F1(ω)
= A(ω) = H (ω)
U2 F2(ω) F2(ω)
Say we apply we apply the driving force to mass 1, then the driving
point frequency response is H 11 ( ω ) . If we drive mass 2, the driving
point frequency response is H 22 ( ω ) .
Both driving point frequency response functions are plotted in the
following figure. The fractions of critical damping are calculated from
H 11 ( ω ) , but the damping numbers are identical when computed from
H 22 ( ω ) .
A very fine discretization in frequency space is necessary to converge
on the fractions of critical damping. Thirty intervals/Hz were used to
calculate the following.

Solution: Part 2
Advanced Vibrations

x 10
8
6
A/h = 1
5 zeta = 0.0014567
|FRF| Mass 1
zeta = 0.00083563
0
0 20 40 60 80 100 120 140 160 180 200
Frequency Hz
Computed damping is 0.15% and 0.084% for the first and second mode,
respectively.
Solution Part 2
Advanced Vibrations
The reason why computed damping does not depend on which mass is
driven becomes clear when we examine the modal expression for nodal
displacement:
T
Pk ( Pk F ( ω ) )
X ( ω ) = ∑ ---------------------------------------------------
2 2
-
k ( ω k – ω ) + 2iζ k ωω k
When the system is driven at frequency ω , only those modes with

frequency in that vicinity are excited. We see that all of the physical
nodes will move proportionally in response to that excitation. Hence the
half-power calculation for damping will be independent of which node is
excited.

Solution: Part 3
Advanced Vibrations
Compute elastic eigen modes and frequencies.

From the analysis of part 1, the eigen problem is
U1 k –k U1
–ω M = 0
2
+ A
M U2 ∗
– k 2k + G ( ω ) --- U 2 0
h
In the frequency range of interest,
A A 5 8
--- G′′ ( ω ) < --- G′ ( ω ) < 8.0 ×10 N ⁄ m « k = 1.0 ×10 N ⁄ m
h h
and it is not anticipated that the damper will significantly affect the
natural frequencies or modes.

Solution: Part 3
Advanced Vibrations
The eigen problem can be made non-dimensional
2
1–α –1 U1
= 0 where
2 2  M
α = ω ----- .
2 U2 0  K
–1 2–α
 3 – 5 3 + 5
2
The eigen values are α = ----------------, ---------------- = ( 0.382, 2.62 )
 2 2 
In terms of angular frequency,
2 2 K  5 5
ω = α ----- = ( 1.27 ×10 , 8.73 ×10 ) .
 M
The natural frequencies are ( ω 1, ω 2 ) = ( 357, 934 ) and

1
( f 1, f 2 ) = ------ ( ω 1, ω 2 ) = ( 56.8, 149 )Hz
2π
Solution: Part 3
Advanced Vibrations
The corresponding eigenvectors are
 1 1 
   1 1 
 5–1 , 5 + 1  =  , .
 ---------------- – ----------------   0.618 – 1.62 
 2 2 
These have not been normalized. They do not have to be for this
application, but we shall normalize them later on.

Solution: Part 4
Modal Strain Energy for First Mode
Advanced Vibrations
The first mode looks like

1 and occurs at a
0.618
ω 1 = 357, f 1 = 56.8Hz .
The maximum strain energy in a cycle of this mode is
k 2 k 2 6
E S = --- ( 1 – 0.618 ) + --- ( 0.618 ) = 26.4 ×10 NM
2 2
The maximum energy dissipated in a cycle in of this mode is
A 2 6 2 6
E D = π --- G'' ( ω ) ( 0.618 ) = π0.365 ×10 ( 0.618 ) = 0.438 ×10
h
η ED
ζ = --- = ------------- = 0.0013
2 4πE S
Solution: Part 4
Modal Strain Energy for Second Mode
Advanced Vibrations
The first mode looks like

1 and occurs at a
– 1.62
ω 2 = 934, f 2 = 149Hz .
The maximum strain energy in a cycle of this mode is
k 2 k 2 8
E S = --- ( 1 + 1.62 ) + --- ( 1.62 ) = 4.74 ×10 Nm
2 2
The maximum energy dissipated in a cycle in of this mode is
A 2 5 2 6
E D = π --- G'' ( ω ) ( 1.62 ) = π ( 1 )5.7 ×10 ( 1.62 ) = 4.7 ×10 Nm
h
η ED
ζ = --- = ------------- = 0.00079
2 4πE S
Solution: Part 5
Solution with initial conditions
Advanced Vibrations
From this point on, we shall need to normalize the eigen modes.
Recall that the physical degrees of freedom and the modal coordinates
x1 ( t ) β1 ( t ) T
are related by = P where P MP = I and
x2 ( t ) β2 ( t )
2
T ω1 0
P KP = .
2
0 ω2
Normalizing the eigen solutions calculated earlier,
P = 4.91 – 3.04 ×10

–2
3.04 4.91
Solution: Part 5
Advanced Vibrations
There are no forcing terms, so the modal equations are of the form
2
β̇˙n + 2ζω n β̇ n + ω n β n = 0 .
Since this is an “under damped” system, the solution is
–ζn ωn t 2 2
βn ( t ) = e [ C n cos ( tω n 1 – ζ n ) + D n sin ( tω n 1 – ζn ) ] .
The time derivative of β n is
β̇ n ( t ) =
–ζn ωn t 2 2
–ζn ωn e [ C n cos ( tω n 1 – ζn ) + D n sin ( tω n 1 – ζn ) ]
2 –ζn ωn t 2 2
+ ωn 1 – ζn ( e [ D n cos ( tω n 1 – ζ n ) – C n sin ( tω n 1 – ζ n ) ] )

Solution: Part 5
Advanced Vibrations
Substituting into our equation for physical displacements,
x1 ( t )
= 0.0491 β 1 ( t ) + – 0.0304 β 2 ( t )
x2 ( t ) 0.0304 0.0491
x1 ( 0 ) ẋ 1 ( 0 )
Our initial conditions are = 0 and = 1 .
x2 ( 0 ) 0 ẋ 2 ( 0 ) 0
In terms of our modal coordinates,
x1 ( 0 )
= 0.0491 C 1 ( 0 ) + – 0.0304 C 2 ( 0 )
x2 ( 0 ) 0.0304 0.0491
and we conclude that C1 = C2 = 0 .

Solution: Part 5
Advanced Vibrations
ẋ 1 ( 0 ) 0.0491 2
= ω1 1 – ζ1 D1
ẋ 2 ( 0 ) 0.0304
+ – 0.0304 2
ω2 1 – ζ2 D2
0.0491
= 17.6 D 1 + – 28.4 D 2 = 1
10.9 46 0
which is solved D 1 = 0.041, D 2 = – 0.010

Solution: Part 5
Advanced Vibrations
The calculated displacement is
x1 ( t ) 0.00202 – 0.52t
= e sin ( t358 )
x2 ( t ) 0.00125
+ 0.00030 e
– 0.78t
sin ( t935 )
– 0.00048

Solution: Part 5
Computed Displacements
Advanced Vibrations
−3
x 10
2.5
Displacement 1
2
Displacement 2
1.5
0.5
−0.5
−1
−1.5
−2
−2.5
0 0.02 0.04 0.06 0.08 0.1 0.12

Solution: Part 5
Computed Velocities
Advanced Vibrations
1.5
Velocity 1
Velocity 2
1
0.5
−0.5
−1
0 0.02 0.04 0.06 0.08 0.1 0.12


2 - Boundary Element Tearing and Interconnecting Methods

Hochgeladen von

Dokumentinformationen

Originaltitel

Copyright

Verfügbare Formate

Dieses Dokument teilen

Dokument teilen oder einbetten

Freigabeoptionen

Stufen Sie dieses Dokument als nützlich ein?

Sind diese Inhalte unangemessen?

Copyright:

Verfügbare Formate

2 - Boundary Element Tearing and Interconnecting Methods

Hochgeladen von

Copyright:

Verfügbare Formate

Advanced Vibrations

11/30/98 Copyright Dan Segalman, 1998

Frequency Domain Analysis

11/30/98 Copyright Dan Segalman, 1998

11/30/98 Copyright Dan Segalman, 1998

Applied force F ( t ) is balanced by acceleration of mass,

11/30/98 Copyright Dan Segalman, 1998

Recall: ω n is the undamped natural frequency;

and ζ is fraction of critical damping. (Note: ζ is pronounced “zeta”.)

11/30/98 Copyright Dan Segalman, 1998

and L is time-translation invarient: L ( x ( t + T ) ) = L ( x ) t+T

11/30/98 Copyright Dan Segalman, 1998

Nondimensionalization of equation for single DOF system:

Let τ = ωn t and let y ( τ ) = x ( τ ⁄ ω n ) , then

Say F ( t ) = Ak sin ( ωt ) , then

11/30/98 Copyright Dan Segalman, 1998

Consider the steady state solution of

y'' ( τ ) + 2ζy' ( τ ) + y ( τ ) = sin ( τα )

where α = ω ⁄ ω n , we anticipate a solution of the sort

If we solve forH̃ ( α ) and ψ̃ ( α ) for all α for a fixed ζ we have the

11/30/98 Copyright Dan Segalman, 1998

Please read the first chapter of

11/30/98 Copyright Dan Segalman, 1998

1. Solve the transient Single DOF problem

Hint: Let v = y' and solve v' + 2ζ 1 v = sin ( ατ )

11/30/98 Copyright Dan Segalman, 1998

• Assembly of Systems from Components

• Convolutions with Indicial and Impulse

• Beginning of Fourier Analysis and the Complex

Copyright Dan Segalman, 1998

1. Solve the transient Single DOF problem

Hint: Let v = y' and solve v' + 2ζ 1 v = sin ( ατ )

Copyright Dan Segalman, 1998

The magnitude of H̃ is determined from

Copyright Dan Segalman, 1998

Copyright Dan Segalman, 1998

Copyright Dan Segalman, 1998

• There are advantages to frequency analysis. We

• It takes a slightly damped system a very long

Copyright Dan Segalman, 1998

Recall that it was asserted that:

Does this system have two degrees of freedom?

Copyright Dan Segalman, 1998

Resolution is achieved through the elimination of massless degrees of

As shown here, F1 and F2 are applied forces: F 1 = K 1 u1 ,

Balancing forces at a virtual mass at location x1 to deduce F1 = F2.

Balancing forces at the actual mass at location x2, F M = –F 2

• The reaction forces on massless subsystems is

• Note that extra degrees of freedom are removed

• Constitutive expressions relate applied forces to

Copyright Dan Segalman, 1998

Returning to the liquid-like second order system

Recall the synthesis rules:

• The forces balance: F 1 = F 2 , and F 2 = – F M . Remember

Copyright Dan Segalman, 1998

Returning to the liquid-like second order system

Recall the constitutive rules:

Copyright Dan Segalman, 1998

Solving the system: u1 u2

Setting λ = c ⁄ K , we solve the

Copyright Dan Segalman, 1998

cu̇ 1 = K ( u 2 – u 1 ) and K ( u 2 – u 1 ) = – Mu̇˙2

Becomes: csU 1 ( s ) – cu 1 ( 0 ) = K [ U 2 ( s ) – U 1 ( s ) ] and