OSD

macl.caeds.eng.uml.
edu
Dr. Peter Avitabile

Modal Analysis & Controls Laboratory
University of Massachusetts Lowell
Presentation Topics
Intent
Things Shake and Break !
Modal Overview
TUTORIAL NOTES:
Structural Dynamics and
Experimental Modal Analysis
Analytical Modeling
SDOF Theory
MODE 1
MDOF Theory
MODE3
DSP - DQAL Windows

Measurement Definitions
Excitation Considerations
MPE Concepts
MODE 2
MODE 4
Linear Algebra
Structural Modification
Correlation/Updating
Copyright 2000 All Rights Reserved
In Trouble !!!!!
Dr. Peter Avitabile
peter_avitabile@uml.edu
Tuesday, July 06, 2004
Intent of Structural Dynamics & Modal Overview
The intent of this

Structural Dynamics and Modal Analysis Overview
is to expose undergraduate engineering students
to some of the basic concepts and ideas concerning
analytical and experimental modal analysis for solving
structural dynamic problems.
It is NOT intended to be a detailed treatment of this material.
Rather it is intended to prepare the students for some basic
material to enhance their ability to solve some structural dynamics
problems that may be encountered during this summer session.
Intent of Modal Analysis Overview
Dr. Peter Avitabile

Structural Dynamic Modeling Techniques
Could you explain
and how is it
used for solving
dynamic problems?
modal analysis
Illustration by Mike Avitabile
Overview of Structural Dynamic Modeling Techniques
Dr. Peter Avitabile

Modal Analysis and Structural Dynamics

DISK DRIVE
INDUCED VIBRATIONS
RESPONSE
OUTPUT TIME RESPONSE
INPUT TIME FORCE
FFT
IFT
INPUT FORCE
BOARD
RESPONSE
CABINET
INPUT POWER SPECTRUM
INPUT
FORCE
FAN INDUCED
VIBRATIONS
OUTPUT POWER SPECTRUM
Modal Analysis is the study of the dynamic character of a

system which is defined independently from the loads applied
to the system and the response of the system.
Structural dynamics is the study of how structures respond
when subjected to applied loads. Many times, in one form or
another, the modal characteristics of the structure is used to
determine the response of the system.
Dr. Peter Avitabile

How Do Structures Respond Dynamically ?

The raw time response of a structure may seem
complicated but it is really nothing more than the
linear combination of the effects of all the modes
that are excited by the specific input
high speed video

showing drop load
(AVI file)
response due to a vertical bump

superimposed on a random excitation
(AVI file)
Dr. Peter Avitabile

Response of a Simple Plate

Simple time-frequency response relationship
RESPONSE
increasing rate of oscillation
FORCE
time
frequency
Dr. Peter Avitabile

Measure many points on the

plate simultaneously to view
the actual response
Different deformation
patterns can be seen as the
excitation sweeps from low
frequency to high frequency
Dr. Peter Avitabile


Sine Dwell to Obtain Mode Shape Characteristics
MODE3
MODE 1
MODE 2
MODE 4
Dr. Peter Avitabile

Analytical Modal Analysis

Equation of motion
[M n ]{&x& n } + [C n ]{x& n } + [K n ]{x n } = {Fn ( t )}
Eigensolution
[[K n ] [M n ]]{x n } = {0}
Dr. Peter Avitabile

Finite Element Models
Advantages
Models used for design
development
No prototypes are
necessary
Disadvantages
Modeling assumptions
Joint design difficult to model
Component interactions are
difficult to predict
Damping generally ignored
Dr. Peter Avitabile

Finite Element Models

Analytical models are developed
to describe the system mass and
stiffness characteristics of a
component or system
The model is decomposed to
express the part in terms of its
modal characteristics - its
frequency, damping and shapes
The dynamic characteristics help
to better understand how the
structure will behave and how to
adjust or improve the component
or system design
10
Dr. Peter Avitabile


[Y]
MEASURED RESPONSE
[F]
APPLIED FORCE
fref1
fref2
[H]
FREQUENCY RESPONSE FUNCTIONS
Advantages
Modal characteristics
are defined from actual
measurements
Damping can be
evaluated
Disadvantages
Requires hardware
Actual boundary conditions
may be difficult to simulate
Different hardware
prototypes may vary
11
Dr. Peter Avitabile


Measured frequency response
functions from a modal test can
also be used to describe the
structures dynamic properties its frequency, damping and shapes
40
MODE # 1
MODE # 2
MODE # 3
DOF # 1
DOF #2
DOF # 3
COHERENCE
dB Mag
FRF
-60
0Hz
800Hz
AUTORANGING
AVERAGING
h 13
1
2
1
3
2
h 23
3
h 33
h 31
h 33
h 32
12
Dr. Peter Avitabile

Experimental Data Reduction

Measured frequency response
functions from a modal test or
operating data can be used to
develop a model of the dynamic
characteristics of the system
13
Dr. Peter Avitabile

What Are Measurements Called FRFs ?

A simple inputoutput problem
8
5
2
8
3
0
-3
8
-7
Magnitude
Real
MODE # 1
MODE # 2
MODE # 3
DOF # 1
DOF #2
DOF # 3
1.0000
Phase
-1.0000
Imaginary
14
Dr. Peter Avitabile

Digital Signal Processing Flow Diagram

ANALOG SIGNALS
OUTPUT
INPUT
ANTIALIASING FILTERS
Actual time signals

Analog anti-alias filter
AUTORANGE ANALYZER
ADC DIGITIZES SIGNALS
OUTPUT
INPUT
APPLY WINDOWS
INPUT
OUTPUT
COMPUTE FFT
LINEAR SPECTRA
LINEAR
OUTPUT
SPECTRUM
LINEAR
INPUT
SPECTRUM
Digitized time signals

Windowed time signals
Compute FFT of signal
AVERAGING OF SAMPLES
COMPUTATION OF AVERAGED
INPUT/OUTPUT/CROSS POWER SPECTRA
INPUT
POWER
SPECTRUM
OUTPUT
POWER
SPECTRUM
CROSS
POWER
SPECTRUM
COMPUTATION OF FRF AND COHERENCE
FREQUENCY RESPONSE FUNCTION
Average auto/cross spectra

Compute FRF and Coherence
COHERENC E FUNCTION
15
Dr. Peter Avitabile

Experimental Mode Shapes From FRFs

MODE 2
2
1
MODE 1
5
2
4
1
3
6
16
Dr. Peter Avitabile

Experimental Mode Shapes From FRFs
a ij1
a ij2
a ij3
HOW MANY POINTS ???
RESIDUAL
EFFECTS
RESIDUAL
EFFECTS
The task for the modal

test engineer is to
determine the parameters
that make up the pieces
of the frequency response
function
Mathematical routines
help to determine the
basic parameters that
make up the FRF
HOW MANY MODES ???
17
Dr. Peter Avitabile

Flow Diagram for Response

Why and How Do Structures Vibrate?
INPUT TIME FORCE
f(t)
y(t)
FFT
IFT
INPUT SPECTRUM
OUTPUT SPECTRUM
f(j )
h(j )
y(j )
18
Dr. Peter Avitabile

What is Operating Data ?

If an excitation is applied close to a mode, then
that mode is excited - if not, then the response
is the linear combination of all the modes excited
19
Dr. Peter Avitabile


The modes of the structure act like filters
which amplify and attenuate input excitations
on a frequency basis
OUTPUT SPECTRUM
y(j )
f(j )
INPUT SPECTRUM
20
Dr. Peter Avitabile


The raw time response of the structure may seem
complicated but it is really nothing more than the
linear combination of the effects of all the modes
that are excited by the specific input
response due to a vertical bump

superimposed on a random excitation
(AVI file)
21
Dr. Peter Avitabile

What Good is Modal Analysis ?
EXPERIMENTAL
MODAL
TESTING
FINITE
ELEMENT
MODELING
MODAL
PARAMETER
ESTIMATION
PERFORM
EIGEN
SOLUTION
Repeat
until
desired
characteristics
are
obtained
RIB
STIFFNER
MASS
DEVELOP
MODAL
MODEL
SPRING
STRUCTURAL
CHANGES
REQUIRED
Yes
USE SDM
TO EVALUATE
STRUCTURAL
CHANGES
No
DONE
DASHPOT
STRUCTURAL
DYNAMIC
The dynamic
model can be
used for studies
to determine the
effect of
structural
changes of the
mass, damping
and stiffness
MODIFICATIONS
22
Dr. Peter Avitabile

What Good is Modal Analysis ?

Simulation, Prediction, Correlation, to name a few
FREQUENCY
RESPONSE
MEASUREMENTS
FINITE
ELEMENT
MODEL
CORRECTIONS
PARAMETER
ESTIMATION
EIGENVALUE
SOLVER
MODAL
PARAMETERS
MODEL
VALIDATION
MODAL
PARAMETERS
SYNTHESIS
OF A
DYNAMIC MODAL MODEL
MASS, DAMPING,
STIFFNESS CHANGES
STRUCTURAL
DYNAMICS
MODIFICATION
FORCED
RESPONSE
SIMULATION
MODIFIED
MODAL
DATA
REAL WORLD
FORCES
STRUCTURAL
RESPONSE
23
Dr. Peter Avitabile

Correlation and Updating Models

RVAC
Analytical and
experimental models
are correlated and
adjusted to
provide
better
component
and system
models
FINITE ELEMENT MODEL
FINITE ELEMENT
CoMAC
VECTOR CORRELATION
) [U n ] , [ ]
+
g
[Tu ] = [Un ] [Ua ]
[M] , [K]
VECTOR CORRELATION
FINITE ELEMENT
MODE
SWITCHING
0.6
0.4
0.3
MAC
0.2
0.2
EXPERIMENTAL
EXPERIMENTAL MODAL MODEL

[En ] = [T u ] [E a ]
0.5
0.4
FINITE ELEMENT
COMBINING ANALYTICAL AND EXPERIMENTAL DATA
0.8
0.6
CORTHOG
Experimental Analytical
DOF CORRELATION
1.2
0.7
DOF CORRELATION
0.9
0.8
EXPERIMENTAL
DOF CORRELATION
MAC AND
ORTHOGONALITY
FRAC
0.1
GUYAN
MAC
FEM 5
FEM 4
0.6
POC
VECTOR CORRELATION
FEM 3
0.2
1
0.8
0.4
1.2
1.2
OR
FEM 2
0
EXP1 EXP 2
FEM 1
EXP 3 EXP 4
EXP 5
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
IRS
EXPERIMENTAL
SEREP
24
Dr. Peter Avitabile


FINITE ELEMENT
CoMAC
MAC
MODE
SWITCHING
MODAL
ASSURANCE
CRITERIA
MATRIX
OR
COORDINATE
MODAL
ASSURANCE
CRITERIA
CORTHOG
COORDINATE
ORTHOGONALITY
CRITERIA
OR
1
FEM 5
0.8
FEM 4
0.6
0.4
VECTOR CORRELATION
FEM 3
0.2
Experimental
FEM 2
Analytical
PSEUDO
ORTHOGONALITY
CRITERIA
MATRIX
EXP1
FEM 1
EXP 2
EXP 3
EXP 4
EXP 5
DOF CORRELATION
POC
EXPERIMENTAL
FINITE ELEMENT
EXPERIMENTAL
Vector tools
RVAC
RESPONSE
VECTOR
ASSURANCE
CRITERIA
FRAC
FREQUENCY
RESPONSE
ASSURANCE
CRITERIA
Degree of freedom tools
FINITE ELEMENT
EXPERIMENTAL
DOF CORRELATION
Frequency tools
VECTOR CORRELATION
25
Dr. Peter Avitabile


ANALYTICAL MODEL
MODEL
IMPROVEMENT
REGIONS
AMI
MODEL
IMPROVEMENT
REGIONS
SSO/MSSO
Models can be adjusted to better reflect actual measured

system characteristics
Joint stiffness can be more accurately identified
Simplistic modeling assumptions can be modified to reflect
the actual system
26
Dr. Peter Avitabile

System Models
System models are developed
from component models which
can be obtained from physical
models, reduced models, modal
models or measurement models
All of these methods may be
used to develop a system model
27
Dr. Peter Avitabile

System Assembly
Components may be
described by a variety
of different methods
depending on the
problem and results
necessary
28
Dr. Peter Avitabile

System Models
Modal Models
Reduced Models
FULL SPACE PHYSICAL MODEL
MODAL SPACE MODEL
REDUCED SPACE PHYSICAL MODEL

CONNECTION
MODAL TIE MATRIX
REDUCED SPACE PHYSICAL MODEL

MODAL SPACE MODEL

Modal/Physical Models
Impedance Models

TIE MATRIX
CONNECTION
MODAL SPACE MODEL
29
Dr. Peter Avitabile

Hybrid/Impedance Modeling
In addition to more conventional
system modeling approaches,
measured frequency response
functions can also be used to
assemble systems and provide more
realistic boundary conditions
MACHINE
CHUCK
CONNECTION IMPEDANCE
MEASURED AT MACHINE
CONNECTION IMPEDANCE
SYNTHESIZED FROM
FEM OF WORKPIECE
HYBRID MODELING
calc3_xyz
REFERENCE IMPEDANCE
SYNTHESIZED FROM
FEM OF WORKPIECE
UNIV:1974:+Z
10
10
120
-10
-10
HYBRID
-20
dB
-30
-20
-30
-40
100
-40
FEM
-50
(s2)/(kg)
-60
-50
-60
dB
-70
-70
5
100
200
255.75
Hz
Dof 15286 REFERENCE

Dof 15286 CALCULATED
(m/s2)/N
0
0
1000
2000
2550
Hz
30
Dr. Peter Avitabile

Dynamic Force Estimation

Using both measured operating data and frequency response
function, estimates of the dynamic forces driving the system can
be estimated
OPERATIONAL
DISPLACEMENTS
[Y]
Estimated force vs reference @dof7 part4

10-1
Reference
-2
10
Estimated
-3
10
-4
10
Lbf^2
-5
10
-6
10
Estimated force vs reference @dof17 part4

10 -1
Reference
-7
10
10
10
10
-2
50
100
150
Hz
200
250
Estimated
300
-3
-4
Lbf^2
10
10
10
[H]
-5
-6
-7
50
100
150
Hz
200
250
300
[F]
FREQUENCY RESPONSE
FUNCTIONS
31
Dr. Peter Avitabile

System Response
System response can be computed
for both linear and non-linear
systems by various methods.
INPUT TIME FORCE
f(t)
y(t)
FFT
IFT
INPUT SPECTRUM
OUTPUT SPECTRUM
f(j )
h(j )
y(j )
32
Dr. Peter Avitabile

Different Mathematical Models

There are basically three different types of
models that are commonly used for solving
structural dynamic problems:
Physical or Spatial Models
Modal Space Models
Response Based Models
33
Dr. Peter Avitabile

Physical or Spatial Models

These models are developed from basic physical
characteristics describing the system mass,
damping and stiffness typically from a finite
element model description:
[M ]{&x&}+[C]{x& }+[K ]{x}={F( t )}
12 6L 12 6L
6L 4L2 6L 2L2
[k] = EI3
L 12 6L 12 6L
6L 2L2 6L 4L
i
i
Fi
E, I
L
j
Fj
156 22L 54 13L

22L 4L2 13L 3L2
[m] = AL
420 54 13L 156 22L
13L 3L2 22L 4L2
34
Dr. Peter Avitabile

Modal Space Models

These models are developed from the modal
characteristics describing the frequency, damping
and mode shape:
[M ]{&x&}+[C]{x& }+[K ]{x}={F( t )}
ls
s
e
l
d
e
od
Mo
M
d
d
se
e
a
s
B
Ba
al
l
d
a
o
c
M
si
y
h
P
\
f1
p1
k1
m2
c1
MODE 1
f2
p2
m1
k2
m3
c2
MODE 2
f3
p3
k3
c3
\
{&p&} +
\
{p& } +
{p} = [U ]T {F}
MODE 3
35
Dr. Peter Avitabile


These models are developed from characteristics
of the system response typically from frequency
response measurements:
ts
n
e
on
p
m
Co
ed
t
s
Te
r
ls
so
e
l
d
e
od
Mo
M
d
ed
ase
s
B
a
B
l
se
n
a
c
o
i
p
s
Res
Phy
36
Dr. Peter Avitabile

Could you explain
and how is it
used for solving
dynamic problems?
modal analysis
37
Dr. Peter Avitabile

Analytical Topics
for
Structural Dynamic Modeling
Analytical Topics
Dr. Peter Avitabile

Structures Vibrate
All structures vibrate to some degree
Objectionable vibrations range from annoying
items such as car vibration considerations to
catastrophic failures such as the famous
Tacoma Narrows Bridge
But there are also many good vibrations some designs incorporate vibrations to achieve the
desired level of performance
Analytical Topics
Dr. Peter Avitabile

Types of Models
Models are developed to assist in the design and
understanding of system dynamics
Analytical models (such as finite element models)
are utilized in the design process
Experimental models are also used for many
systems where modeling is not practical or models
are too difficult to develop
Analytical Topics
Dr. Peter Avitabile

Finite Element Model Considerations

Finite element models are commonly used
What are we trying to do when generating a model
CONTINUOUS
SOLUTION
Analytical Topics
DISCRETIZED
SOLUTION
Dr. Peter Avitabile


Modeling Issues
continuous solutions work well with structures that are well behaved
and have no geometry that is difficult to handle
most structures don't fit this simple requirement
(except for frisbees and cymbals)
real structures have significant geometry variations that are
difficult to address for the applicable theory
a discretized model is needed in order to approximate the actual
geometry
the degree of discretization is dependent on the waveform of the
deformation in the structure
finite element modeling meets this need
Analytical Topics
Dr. Peter Avitabile


Finite element modeling involves the descretization of the structure
into elements or domains that are defined by nodes which describe
the elements.
A field quantity such as displacement is approximated using polynomial
interpolation over each of the domains.
The best values of the field quantity at nodes results from a
minimization of the total energy.
Since there are many nodes defining many elements, a set of
simultaneous equations results.
Typically, this set of equations is very large and a computer is used to
generate results.
Analytical Topics
Dr. Peter Avitabile


Nodes represent geometric locations in the structure.
Elements boundary are defined by the nodes.
The type of displacement field that exists over the domain will
determine the type of element used to characterize the
domain.
Element characteristics are determined from
Theory of Elasticity
and
Strength of Materials.
Analytical Topics
Dr. Peter Avitabile

Analytical Topics for Structural Dynamic Modeling

Structural element formulations use the same general
assumptions about their respective behavior as their respective
structural theories (such as truss, beam, plate, or shell)
Continuum element formulations (such as 2D and 3D solid
elements) comes from theory of elasticity
6L 12 6L
12
6L 4L2 6L 2L2
[k ] = EI3
L 12 6L 12 6L
6L 2L2 6L 4L
Analytical Topics
E, I
Fj
54
13L
156 22L
22L 4L2 13L 3L2
[m] = AL
13L 156
22L
420 54
13L 3L2 22L
4L2
Dr. Peter Avitabile

Analytical Topics for Structural Dynamic Modeling

The basis of the finite
element method is
summarized below
v
u
t
s
subdivide the structure into small finite elements

each element is defined by a finite number of node points
assemble all elements to form the entire structure
within each element, a simple solution to governing equations
is formulated (the solution for each element becomes a
function of unknown nodal values
general solution for all elements results in algebraic set of
simultaneous equations
Analytical Topics
Dr. Peter Avitabile


DEGREES OF FREEDOM
maximum 6 dof can be described at a point in space
finite element use a maximum of 6 dof
most elements use less than 6 dof to describe the element
TRUSS
TORSIONAL ROD
STRUCTURAL ELEMENTS
3D BEAM
PLATE
CONTINUUM ELEMENTS
Analytical Topics
10
Dr. Peter Avitabile

Advantages
Models used for design

development
No prototypes are
necessary
Disadvantages
Analytical Topics
11
Modeling assumptions
Joint design difficult to model
Component interactions are
difficult to predict
Damping generally ignored
Dr. Peter Avitabile


A TYPICAL FINITE ELEMENT USER MAY ASK
what kind of elements should be used?

how many elements should I have?
where can the mesh be coarse; where must it be fine?
what simplifying assumptions can I make?
should all of the physical structural detail be included?
can I use the same static model for dynamic analysis?
how can I determine if my answers are accurate?
how do I know if the software is used properly?
Analytical Topics
12
Dr. Peter Avitabile


ALL THESE QUESTIONS CAN BE ANSWERED, IF
the general structural behavior is well understood

the elements available are understood
the software operation is understood
(input procedures, algorithms,etc.)
BASICALLY - we need to know what we are doing !!!

IF A ROUGH BACK OF THE ENVELOP ANALYSIS
CAN NOT BE FORMULATED, THEN
MOST LIKELY THE ANALYST DOES NOT KNOW
ENOUGH ABOUT THE PROBLEM AT HAND TO
FORMULATE A FINITE ELEMENT MODEL
Analytical Topics
13
Dr. Peter Avitabile

Finite Element Modeling

Using standard finite element modeling techniques, the following steps
are usually followed in the generation of an analytical model
node generation
element generation
coordinate transformations
assembly process
application of boundary conditions
model condensation
solution of equations
recovery process
expansion of reduced model results
Analytical Topics
14
Dr. Peter Avitabile


Element Definition
Shape Functions
Each element is approximated by
Linear
{} = [N ]{x}
where
{}
[N]
{x}
- vector of displacements in element

- shape function for selected element
- nodal variable
Element shape functions can range from linear

interpolation functions to higher order polynomial
functions.
Analytical Topics
15
Quadratic
Polynomial
Dr. Peter Avitabile


Strain Displacement Relationship
The strain displacement relationship is given by
{} = [B]{x}
where
{}
- vector of strain within element
[B]
- strain displacement matrix

(proportional to derivatives of [N])
{x}
- nodal variable
Analytical Topics
16
Dr. Peter Avitabile


Mass and Stiffness Formulation
The mass and stiffness relationship is given by
[M ] = V [N][N]T V
where
[M]
[K]
[N]
{}
[B]
[C]
[K ] = V [B]T [C][B]V
- element mass matrix
- element stiffness matrix
- shape function for element
- density
- strain displacement matrix
- stress-strain (elasticity) matrix
Analytical Topics
17
Dr. Peter Avitabile


Coordinate Transformation
Generally, elements are formed in a local coordinate system which is
convenient for generation of the element.
Elemental matrices are transformed from the local elemental
coordinate system to the global coordinate system using
{x1} = [T12 ]{x 2 }

LOCAL SYSTEM
GLOBAL SYSTEM
Analytical Topics
18
Dr. Peter Avitabile


Assembly Process
Elemental matrices are then assembled into the global master matrices
using
{x k } = [c k ]{x g }
where
{xk} - element degrees of freedom
[ck] - connectivity matrix
{xg} - global degrees of freedom
The global mass and stiffness matrices are assembled and boundary
conditions applied for the structure
Analytical Topics
19
Dr. Peter Avitabile


Static Solutions
typically involve decomposition of a large matrix
matrix is usually sparsely populated
majority of terms concentrated about the diagonal
Eigenvalue Solutions
use either direct or iterative methods
direct techniques used for small matrices
iterative techniques used for a few modes from large matrices
Propagation Solutions
most common solution uses derivative methods
stability of the numerical process is of concern
at a given time step, the equations are reduced to an equivalent
static form for solution
typically many times steps are required
Analytical Topics
20
Dr. Peter Avitabile

Finite Element Modeling - Simple Example

Consider the 2 spring system shown below
u1
u2
1
u3
2
f
1
each spring element is denoted by a box with a number

each element is defined by 2 nodes denoted by the circle with a
number assigned to it
the springs have a node at each end and have a common node point
the displacement of each node is denoted by u with a subscript to
identify which node it corresponds to
there is an applied force at node 3
Analytical Topics
21
Dr. Peter Avitabile


The first step is to formulate the spring element in a general sense
ui
uj
p
f jp
f ip
j
the element label is p

the element is bounded by node i and j
assume positive displacement conditions at both nodes
define the force at node i and node j for the p element
Application of simple equilibrium gives
f ip = k p (u i u j ) = + k p u i k p u j
f jp = k p (u j u i ) = k p u i + k p u j
Analytical Topics
22
Dr. Peter Avitabile


This can be written in matrix form to give
kp
k
p
k p u i f ip
=
k p u j f jp
Now for element #1
k1 k1 u1 f11
=
k
1 k1 u 2 f 21
And for element #2
k2
k
2
k 2 u 2 f 22
=
k 2 u 3 f 32
The equilibrium requires that the sum of the internal forces equals
the applied force acting on each node
Analytical Topics
23
Dr. Peter Avitabile


The three equations can be written as
k1u1 k1u 2 = f1
k1u1 + k1u 2 + k 2 u 2 k 2 u 3 = f 2
k 2u 2 + k 2u 3 = f3
or in matrix form
k1
k1
k k + k
2
1 1
k2
Analytical Topics
u1 f1

k 2 u 2 = f 2
k 2 u 3 f 3
24
Dr. Peter Avitabile


Now applying a boundary condition of zero displacement at node 1 has
the effect of zeroing the first column of the K matrix which gives three
equations with 2 unknowns. Solving for the second and third equation
gives
k1
k1
k k + k
2
1 1
k2
u1 f1

k 2 u 2 = f 2
k 2 u 3 f 3
k1 + k 2
k
k 2 u 2 0
=
k 2 u 3 f 3
Analytical Topics
25
Dr. Peter Avitabile


Assembly of the stiffness matrix with more elements
k1
k1
k k + k + k
2
5
1 1
k2
k5
k2
k 2 + k3
k3
k3
k3 + k 4
k4
k5
k4
k 4 + k 5
Notice that the banded nature of the matrix is not preserved when
elements are arbitrarily added to the assembly
Analytical Topics
26
Dr. Peter Avitabile


Equation of Motion (n x n)
[M ]{&x&}+[C]{x& }+[K ]{x}={F( t )}

Eigensolution
[[K ][M ]]{x}=0

Frequencies (eigenvalues) and Mode Shapes (eigenvectors)
Analytical Topics
12
22
and
27
[U] = [{u1} {u 2 }
L]
Dr. Peter Avitabile


Modal transformation (n x m)
{x} = [U ]{p} = [{u1} {u 2 }
p1

L]p 2
M

Projection operation
[U]T [M ][U]{&p&} + [U]T [C][U]{p& } + [U]T [K ][U]{p} = [U]T{F}

Vector orthogonality
{u i }
mii i = j
[M ]{u j } =
0i j
Analytical Topics
{u i }
28
k ii i = j
[K ]{u j } =
0i j
Dr. Peter Avitabile

Finite Element Modeling - Modal Space

Since the mode shapes are linearly independent and
orthogonal w.r.t the mass and stiffness matrices
Modal Mass
\
[U1 ]T [M1 ][U1 ] =
Modal Damping
\
[U1 ]T [C1 ][U1 ] =
Modal Stiffness
\
[U1 ]T [K1 ][U1 ] =
Analytical Topics
29
M1
TRUE !!!
C1
???????
K1
TRUE !!!
Dr. Peter Avitabile

Finite Element Modeling - Proportional Damping

The damping matrix is only uncoupled for a special case
where the damping is assumed to be proportional to the mass
and/or stiffness matrices
\
[U1 ]T [[M ] + [K ]][U1 ] =
M + K
Many times proportional damping is assumed since we do not

know what the actual damping is
This assumption began back when computational power was
limited and matrix size was of critical concern
But even today we still struggle with the damping matrix !!!
Analytical Topics
30
Dr. Peter Avitabile

Finite Element Modeling - Non-Proportional Damping

However, if we knew the damping matrix then a solution can
be obtained after rearranging the equations.
[0]
[M ]
1
[M1 ] &x& [M1 ] [0] x& 0
=
[C1 ] x& [0] [ K1 ] x F
The equation of motion can be recast as
[B1 ]{Y& } [A1 ]{Y} = {Q}
[0] [M1 ]
[B1 ] =
[M1 ] [C1 ]
[M ] [0]
[A1 ] = 1
[0] [ K1 ]
The eigensolution and modal transformation is then
[[A1 ] [B1 ]]{Y} = {0}

Analytical Topics
{Y} = [1 ]{p1}
31
Dr. Peter Avitabile

Finite Element Modeling - Non-Proportional Damping

COMPLEX MODES
The solution to the state space formulation will result in a
set of modes that are generally complex in form.
The mode shapes will have both real and imaginary parts.
The mode shapes will become much more difficult to
describe especially as the damping becomes significantly
different than the proportional damped form.
MAKE SURE YOU REALLY WANT COMPLEX MODES !!!
DO YOU REALLY BELIEVE THE DAMPING MATRIX ???
Analytical Topics
32
Dr. Peter Avitabile

Finite Element Model Reduction & Expansion

Is there any reason to want to have a less complicated
representation of the detailed finite element model ?
or to obtain an expansion of reduced information ?
Analytical Topics
33
Dr. Peter Avitabile


Many times it is necessary to formulate a reduced model of
a structure especially for correlation and updating models
Mapping Transformation Matrix
x a
{x n }= =[T ]{x a }
x d
Reduced System Matrices
[K a ] = [T ]T [K n ][T ]
[M a ] = [T ]T [M n ][T ]
Analytical Topics
34
Dr. Peter Avitabile


Guyan Reduction
Dynamic Condensation
[ I]
[I]
[Ts ]= =
1
[
K
]
[
K
]
[
]
t
da
s
dd
[ I]
[I]
[Tf ]= =
1
[
B
]
[
B
]
[
]
t
f
dd
da
IRS Reduction
[I]
[0]
[Ti ] =
+
1
[K dd ] [ K da ] [0]
SEREP Reduction
[Tu ] = [U n ][U a ]g
Analytical Topics
[0]
1
[
][
][
]
[K a ]
M
T
M
a
n s
[K dd1 ]
(
(
)
)
[ ] [ ]T [ ] 1[ ]T
Ua
U a U a U a
1
T
T
[U d ] [U a ] [U a ] [U a ]
35
Dr. Peter Avitabile


The type of reduction utilized can have a significant impact
on the accuracy of the resulting reduced model.
Much work has been done in this area to minimize the
distortion of the reduced model.
The same reduction matrices are also used for expansion of
reduced model information (ie, test data)
Strong differing opinions prevail on this subject !!!
Analytical Topics
36
Dr. Peter Avitabile

Numerical Methods
Many times the model may be needed to perform
dynamic response studies
Some traditional methods are:
Mode Superposition
Frequency Domain Solution
Direct Integration of Equations of Motion
Analytical Topics
37
Dr. Peter Avitabile

Numerical Methods - Mode Superposition

Using the modal space formulation, a simple integration on
the SDOF system can be performed
m1
m
k
m2
c2
p& 1 k1
p& +
2
\ M
{x} = {u1 }p1
c1
{x} = [U]{p} = {u1 }p1 + {u2 }p2 + {u3 }p3
MODE 1
p
m
k
{x} = {u2 }p2
c2
k2
T
p1 {u1} {F}
p ={u }T {F}
2 2
\ M M
&p&1 c1
&p& +
2
\ M
MODE 2
p
m
k
c3
{x} = {u3 }p3
MODE 3
Analytical Topics
38
Dr. Peter Avitabile

Numerical Methods - Frequency Domain

Using the frequency domain input-output relationships,
Output Response = System Characteristic X Input Forces
the response due to many forces can be computed

No
yi ( j)= h ij ( j)f j ( j)
j=1
The frequency response function is needed for this response
rij,k
rij*,k
h ij ( j)=
+
*
j k j k
k =1
m
Analytical Topics
39
Dr. Peter Avitabile


Frequency domain input-output schematic
OUTPUT SPECTRUM
rij,k
rij*,k
+
h ij ( j)=
*
j p k j p k
k =1
y(j )
No
f(j )
y i ( j)= h ij ( j)f j ( j)
INPUT SPECTRUM
Analytical Topics
j=1
40
Dr. Peter Avitabile


Response - 5Z
Response - 3Z
No
y i ( j)= h ij ( j)f j ( j)
j=1
Response - 2Z
Applied Excitation Force - 1Z
Analytical Topics
41
Dr. Peter Avitabile

Numerical Methods - Direct Integration

The equation of motion is integrated using numerical step-by-step
procedure for a number of t steps. The term 'direct' means that the
equations of motion are formulated in a physical space at Ndof without
any transformation to another space (ie,modal space).
The equation of motion is to be satisfied not at any time t but rather an
equivalent 'static' equilibrium is sought at discrete time intervals t
apart. Note that from this statement, the static solution techniques will
be employed for the system at many different t time steps. In essence,
'effective' loads are computed from the manipulation of the velocity and
acceleration terms of the equation of motion to reduce the problem to a
simple-to-solve static equivalent problem.
Analytical Topics
42
Dr. Peter Avitabile

Numerical Methods - Direct Integration

Basically, numerical integration is a process of marching along in time
where response parameters (acceleration, velocity and displacement) at
time t are evaluated from their known historic values. Typically, three
values are needed for three unknowns. Two of these values are derived
from assumptions regarding the manner in which response parameters
vary during a time step. The third equation is the equation of motion
written at a selected point.
An important aspect of numerical integration is the selection of the time
step used in the integration process. If the time step is selected to be
too large then the computed response will suffer from the effects of
'numerical damping' that will distort the results even the scheme is
unconditionally stable.
Analytical Topics
43
Dr. Peter Avitabile

Numerical Methods - Schemes

Explicit Schemes
scheme is explicit when the equation of motion is written at time t
(which is the current time)
computationally efficient when compared to implicit schemes
these techniques are only conditionally stable
typically there is no factorization of [K] or [M] needed for most cases
Implicit Schemes
scheme is implicit when the equation of motion is written at next time
step (which is t + t)
requires more computation when compared to explicit schemes
these techniques are usually unconditionally stable
usually factorization of [K] or [M] is needed for most cases
Analytical Topics
44
Dr. Peter Avitabile

Numerical Methods - Schemes

Some common integration techniques
Central diff
Houbolt
Wilson
Newmark
Newmark
Newmark
Analytical Topics
Explicit
Implicit
Implicit
Implicit
Implicit
Implicit
=0, =0
=1/2, =1/4
=1/2, =1/6
45
(constant acceleration)
(average acceleration)
(linear acceleration)
Dr. Peter Avitabile

Numerical Methods - Central Difference
Source: UMASS Lowell Structural Dynamic Modeling Techniques Seminar Notes
Analytical Topics
46
Dr. Peter Avitabile

Numerical Methods - Central & Houbolt
Analytical Topics
47
Dr. Peter Avitabile

Numerical Methods - Wilson
Analytical Topics
48
Dr. Peter Avitabile

Numerical Methods - Comparison
Source: Bathe - Numerical Methods in Finite Element Analysis
Analytical Topics
49
Dr. Peter Avitabile

Single Degree of Freedom Overview

f(t)
x(t)
I
m
a
g
i
n
a
r
y
R
e
a
l
k
100
Fr equency
T = 2 / n
=0.1%
=1%
X1
=2%
X2
=5%
10
=10%
=20%
I
m
a
g
i
n
a
r
y
-90
=20%
/n
=10%
t1
=5%
=2%
1
h (s) =
ms 2 + cs + k
=1%
=0.1%
-180
/ n
Basic Modal Analysis Theory
t2
Real
Dr. Peter Avitabile

SDOF Definitions
Assumptions
lumped mass
f(t)
x(t)
stiffness proportional
to displacement
damping proportional to
velocity
linear time invariant
2nd order differential
equations
Dr. Peter Avitabile

SDOF Equations
Equation of Motion
d2x
dx
m 2 + c + kx = f ( t )
dt
dt
or
m &x& + cx& + kx = f ( t )
Characteristic Equation
ms 2 + cs + k = 0
Roots or poles of the characteristic equation

s1, 2
c
=
2m
c + k
m
2m
3
Dr. Peter Avitabile

SDOF Definitions
Poles expressed as
s1, 2 = n
(n )2 n 2 = jd
POLE
Damping Factor
= n
Natural Frequency
n = k
% Critical Damping
= c
m
n
cc
Critical Damping
c c = 2mn
Damped Natural
Frequency
d = n 1 2
CONJUGATE
Dr. Peter Avitabile

Poles in the S-plane

As the damping is
varied from no
damping to critical
damping and
beyond, the poles
move as shown
FRF
TIME
FRF
FRF
TIME
TIME
= 0.1
The impulse and

frequency response
are shown as the
damping is varied
=0
= 0.3
FRF
TIME
= 0.7
= 1.0
TIME
> 1.0
TIME
STABLE
UNSTABLE
Dr. Peter Avitabile

SDOF - Laplace Domain

Equation of Motion in Laplace Domain
(ms 2 +cs+k)x (s) = f (s)
with
b(s ) = (ms 2 +cs+k)
System Characteristic Equation

b(s) x (s) = f (s)
x (s) = b 1 (s)f (s) = h (s)f (s)
and
Dr. Peter Avitabile

SDOF - Transfer Function

System Transfer Function
1
h (s) =
ms 2 + cs + k
Complex valued
function defines the
surface shown
Source: Vibrant Technology
Dr. Peter Avitabile

SDOF - Transfer Function
Polynomial Form
1
h (s) =
ms 2 + cs + k
Pole-Zero Form
1/ m
h (s) =
(s p1 )(s p1* )
Partial Fraction Form
a1
a1*
h (s) =
+
(s p1 ) (s p1* )
Exponential Form
1 t
h(t) =
e sin d t
md
Dr. Peter Avitabile

Amplitude
SDOF - Damped Exponential Response
Damping Decay
Period
h(t)
1
md
e t
9
sin d t
Dr. Peter Avitabile
SDOF - Transfer Function & Residues
Residue
a1 =
h (s)(s p1 )
sp1
1
=
2 jmd
related to
mode shapes
Source: Vibrant Technology
10
Dr. Peter Avitabile

SDOF - Frequency Response Function

The Frequency Response Function is the System
Transfer Function evaluated at s = j
h ( j) = h (s)
s = j
a1
a 1*
=
+
( j p1 ) ( j p1* )
11
Dr. Peter Avitabile


Coincident-Quadrature Plot
Bode Plot
Nyquist Plot
12
Dr. Peter Avitabile


DYNAMIC COMPLIANCE
DISPLACEMENT / FORCE
MOBILITY
VELOCITY / FORCE
INERTANCE
ACCELERATION / FORCE
DYNAMIC STIFFNESS
FORCE / DISPLACEMENT
MECHANICAL IMPEDANCE
FORCE / VELOCITY
DYNAMIC MASS
FORCE / ACCELERATION
13
Dr. Peter Avitabile

SDOF - Effects of Damping & Damping Estimates

Damping Effects
100
Damping Estimates
MAG
=0.1%
=1%
=2%
=5%
10
Q=
0.707
MAG
=10%
=20%
1
n
=
2 2 1
/n
Half Power Points
T = 2 / n
-90
X1
=20%
X2
=10%
= ln
=5%
=2%
=1%
=0.1%
x1
2
x2
-180
/ n
t1
14
t2
Log Decrement
Dr. Peter Avitabile

Multiple Degree of Freedom Overview

[B(s )]1 = [H(s )] = Adj[B(s )] = [A(s )]
det[B(s )] det[B(s )]
f1
p1
k1
f2
p2
c1
f3
p3
m2
m1
m3
k2
c2
k3
c3
R1
D1
MODE 1
MODE 2
MODE 3
R2
D2
\
{&p&} +
MDOF Overview
\
{p& } +
{p} = [U ]T {F}
R3
D3
F1
F2F3
Dr. Peter Avitabile

MDOF Definitions
Assumptions
f2
lumped mass
m2
stiffness proportional
k2
to displacement
damping proportional to
velocity
linear time invariant
f1
c2
x1
m1
k1
2nd order differential
x2
c1
equations
MDOF Overview
Dr. Peter Avitabile

MDOF Equations
Equation of Motion - Force Balance
m1&x&1+(c1 + c 2 )x& 1c 2 x& 2 +(k1 + k 2 )x1k 2 x 2 =f1 (t )
m 2 &x& 2 c 2 x& 1+c 2 x& 2 k 2 x1 +k 2 x 2 =f 2 (t )
Matrix Formulation
m1
&x&1

m 2 &x& 2
(c1 + c 2 ) c 2 x& 1
+
x&
c
c
2
2 2
Matrices and
Linear Algebra
are important !!!
(k1 + k 2 ) k 2 x1 f1 ( t )
+
=
k
k
x
f
(
t
)
2
2
2 2
MDOF Overview
Dr. Peter Avitabile

MDOF Equations
Equation of Motion
[M ]{&x&}+[C]{x& }+[K ]{x}={F( t )}
Eigensolution
[[K ][M ]]{x}=0

Frequencies (eigenvalues) and
Mode Shapes (eigenvectors)
\
MDOF Overview
1
=

\
22
and [U ] = [{u1}
\
4
{u 2 } L]
Dr. Peter Avitabile

Modal Space Transformation

Modal transformation
{x} = [U ]{p} = [{u1} {u 2 }

p1

L]p 2
M

[U ]T [M ][U]{&p&} + [U]T [C][U]{p& } + [U ]T [K ][U ]{p} = [U ]T{F}

Modal equations (uncoupled)
m1
m2
&p&1 c1
&p& +
2
\ M
MDOF Overview
c2
p& 1 k1
p& +
2
\ M
5
k2
T
p1 {u1} {F}
p ={u }T {F}
2 2
\ M M
Dr. Peter Avitabile


Diagonal Matrices Modal Mass
Modal Damping
{&p&} +
Modal Stiffness
{p& } +
{p} = [U ]T {F}
Highly coupled system

transformed into
k1
k2
m3
c2
MODE 2
f3
p3
m2
c1
MODE 1
f2
p2
m1
simple system
MDOF Overview
f1
p1
k3
c3
MODE 3
Dr. Peter Avitabile


PHYSICAL MODEL
..
.
[M]{x} + [C]{x} + [K]{x} = {F(t)}
=
f1
p1
m1
{x} = [U]{p} = [{u 1 } {u 2 } {u 3 }
MODAL
]{p}
k1
SPACE
c1
MODE 1
f2
p2
m2
k2
..
.
T
[ M ]{p} + [ C ]{p} + [ K ]{p} = [U] {F(t)}
{x} = [U]{p} = {u 1 }p1 + {u 2 }p2 + {u 3 }p3
c2
MODE 2
f3
p3
m3
k3
c3
MODE 3
MDOF Overview
Dr. Peter Avitabile

MDOF - Laplace Domain

Laplace Domain Equation of Motion
[[M]s
+ [C]s + [K ]]{x (s )} = 0 [B(s )]{x (s )} = 0
System Characteristic (Homogeneous) Equation
[[M]s +[C]s+[K ]] = 0
p k = k jdk
Damping
MDOF Overview
Frequency
Dr. Peter Avitabile

MDOF - Transfer Function

System Equation
x (s )}
{
[B(s )]{x (s )} = {F(s )} [H(s )] = [B(s )] =
{F(s )}
1
System Transfer Function
[B(s )]
[
Adj[B(s )]
A(s )]
= [H(s )] =
=
det[B(s )] det[B(s )]
[A(s )]
Residue Matrix
det[B(s )]
Characteristic Equation
MDOF Overview
Mode Shapes
Poles
Dr. Peter Avitabile

MDOF - Residue Matrix and Mode Shapes

Transfer Function evaluated at one pole
[H(s )]s=s
qk
T
{u k }
= {u k }
sp k
can be expanded for all modes

m
[H(s )] =
k =1
MDOF Overview
q k {u k }{u k }
q k {u }{u }
+
(sp*k )
(sp k )
T
10
*
k
* T
k
Dr. Peter Avitabile

MDOF - Residue Matrix and Mode Shapes

Residues are related to mode shapes as
[A(s )]k
a11k
a
21k
a 31k
M
a12 k
a 22 k
a 32 k
M
MDOF Overview
a13k
a 23k
a 33k
M
= q k {u k }{u k }
L
u1k u1k
u u
L
=q k 2 k 1k
L
u 3k u1k
M
O
11
u1k u 2 k
u 2k u 2k
u 3k u 2 k
M
u1k u 3k
u 2 k u 3k
u 3k u 3k
M
L
L
L
O
Dr. Peter Avitabile

MDOF - Drive Point FRF
h ij ( j ) =
a ij1
( j p1 )
+
( j p 2 )
MDOF Overview
( j p*2 )
( j p 3 )
a *ij 3
( j p*3 )
( j p1 )
q1 u i 1u j 1
( j p*1 )
q 2u i 2u j 2
( j p 2 )
+
a *ij 2
a ij 3
q1 u i 1u j 1
( j p*1 )
a ij 2
h ij ( j ) =
a *ij1
q 2u i 2u j 2
( j p*2 )
q 3u i 3u j 3
( j p 3 )
12
q 3u i 3u j 3
( j p*3 )
Dr. Peter Avitabile

MDOF - FRF using Residues or Mode Shapes

h ij ( j ) =
R1
+
( j p1 ) ( j p*1 )
+
D1
a*ij1
a ij1
a*ij 2
a ij 2
+
+ L
( j p 2 ) ( j p*2 )
R2
D2
R3
D3
F1
F2F3
a ij1
h ij ( j ) =
q1u i1u j1
* * *
1 i 1 j1
*
1
qu u
+
( j p1 ) ( j p )
+
q 2u i 2 u j 2
* * *
2 i2 j2
*
2
qu u
( j p 2 ) ( j p )
MDOF Overview
a ij2
a ij3
2
3
+ L
13
Dr. Peter Avitabile

Time / Frequency / Modal Representation

PHYSICAL
TIME
FREQUENCY
ANALYTICAL
MODAL
f1
p1
m1
k1
MODE 1
c1
MODE 1
+
p2
f2
m2
k2
MODE 2
c2
MODE 2
p3
f3
m3
k3
MODE 3
MODE 3
MDOF Overview
c3
14
Dr. Peter Avitabile

Overview Analytical and Experimental Modal Analysis

LAPLACE
DOMAIN
[B(s)] = [M]s + [C]s + [K]
TRANSFER
FUNCTION
[B(s)] -1 = [H(s)]
qk u j {u k}
[U]
[ A(s) ]
det [B(s)]
[U]
FINITE
ELEMENT
MODEL
[MA] = [T] T[M N] [T]
[K - M]{X} = 0
MODAL
PARAMETER
ESTIMATION
H(j )
LARGE DOF
MISMATCH
H(j ) =
Xj (j )
Fi (j )
CORRELATION &
MODEL UPDATING
[EN ]' = [TU ] [EA]
X j(t)
MDOF Overview
ANALYTICAL
MODEL
REDUCTION
FFT
Fi (t)
15
MODAL
TEST
EXPERIMENTAL
MODAL MODEL
EXPANSION
Dr. Peter Avitabile

Digitization,
Quantization,
Aliasing,
Leakage
T
Digitization, Quantization, Aliasing, Leakage
Dr. Peter Avitabile


Objectives of this lecture:
Overview basic digital signal processing concepts
Discuss digitization and sampling
Discuss quantization
Discuss aliasing and anti-aliasing filters
Discuss leakage
Dr. Peter Avitabile

Time - Frequency - Laplace

Each domain casts the same information from a different view
point. Many times things that are confusing or unclear in one
domain become easier to interpret in another domain.
* Time domain represents the
physics of the system
TIME DOMAIN
TRANSFORMATION
* Frequency domain represents

the system in terms of it's
periodicities
SUBSET
FREQUENCY
DOMAIN
* Laplace domain represents

the system in terms of its
poles and residues
TRANSFORMATION
PARAMETER ESTIMATION
LAPLACE
DOMAIN
FREQUENCY
AMPLITUDE
TIME
Dr. Peter Avitabile

Transformation from Time to Frequency

Many times a transformation is performed to provide a better
or clearer understanding of a phenomena. The time
representation of a sine wave may be difficult to interpret. By
using a Fourier series representation, the original time signal
can be easily transformed and much better understood.
Transformations are also
performed to respresent the same
data with significantly less
information. Notice that the
original time signal was defined by
many discrete time points (ie,
1024, 2048, 4096 ) whereas the
equivalent Fourier representation
only requires 4 amplitudes and 4
frequencies.
Dr. Peter Avitabile

The Anatomy of the FFT Analyzer

The FFT Analyzer can be broken
down into several pieces which
involve the digitization, filtering,
transformation and processing of a
signal.
ANALOG
SIGNAL
ANALOG
FILTER
ADC
DISPLAY
DIGITAL
FILTER
FFT
DISCRETE
DATA
Several items are important here:

Digitization and Sampling
Quantization of Signal
Aliasing Effects
Leakage Distortion
Windows Weighting Functions
The Fourier Transform
Measurement Formulation
Dr. Peter Avitabile

The Anatomy of the FFT Analyzer

Actual time signals
ANALOG SIGNALS
OUTPUT
INPUT
AUTORANGE ANALYZER
OUTPUT
INPUT
APPLY WINDOWS
INPUT
OUTPUT
COMPUTE FFT
LINEAR SPECTRA
LINEAR
OUTPUT
SPECTRUM
LINEAR
INPUT
SPECTRUM

INPUT
POWER
SPECTRUM
OUTPUT
POWER
SPECTRUM
CROSS
POWER
SPECTRUM

COHERENC E FUNCTION
Dr. Peter Avitabile

Analog Filter
The analog filter removes the portion of the data that can cause
aliasing.
dB
Rolloff
Fc
Frequency
Dr. Peter Avitabile

Sinusoidal Terminology
The peak displacement, peak-to-peak displacement, average
value and rms value are shown below for a sinusoid.
PEAK
AVERAGE
RMS
PEAK TO PEAK
RMS = 0.707x PEAK

AVERAGE = 0.637 x PEAK
PEAK TO PEAK = 2 x PEAK
x AVG =
1 2
x RMS = x ( t )dt
To
1
x dt
To
Dr. Peter Avitabile

Complex Notation
Real / Imaginary Representation
a + jb
a - real part
b - imaginary part
j = 1
Magnitude / Phase Representation
= a tan (b / a )
(a + jb) = a 2 + b 2
Complex Conjugate
(a + jb)* = a jb
Complex Multiplication
(a + jb)(a jb) = a 2 + b 2
(a + jb)(c jd ) = (ac bd ) + j(bc + ad )
Dr. Peter Avitabile

Digitization of Time Signals

With analog sampling devices, only the performance of the
analog instrumentation was of concern. With the use of digital
signal processing (DSP) techniques, additional consideration must
be given to the analog to digital conversion (ADC) process.
The analog signal must be digitized and several additional items
become important in order to minimize distortion of the original
signal. These are quantization, sampling, aliasing and leakage.
10
Dr. Peter Avitabile

Digitization of Time Signals

Sampling rate of the ADC is specified as a maximum that is
possible. Basically, the digitizer is taking a series of
snapshots at a very fast rate as time progresses
Digital
Analog Signal
Representation
ADC
11
Dr. Peter Avitabile

Quantization
Sampling refers to the rate at which the signal is collected.
Quantization refers to the amplitude description of the signal.
A 4 bit ADC has 24 or 16 possible values
ADC BIT STEPS

4bit = 0000 = 23 + 2 2 + 21 + 2 0 = 16levels
12bit = 000000000000= 211 + 210 + L + 21 + 20 = 4096levels = 72dBdynamicrange
12
Dr. Peter Avitabile

Quantization
Quantization errors refer to the accuracy of the amplitude
measured. The 6 bit ADC represents the signal shown much
better than a 4 bit ADC
A
D
C
A
D
C
M
A
X
M
A
X
R
A
N
G
E
R
A
N
G
E
13
Dr. Peter Avitabile

Quantization Error
Underloading of the ADC causes amplitude errors in the signal
All of the available
dynamic range of the
analog to digital
converter is not used
effectively
10 volt
range
on
ADC
0.5 volt signal

This causes amplitude
and phase distortion of
the measured signal in
both the time and
frequency domains
14
Dr. Peter Avitabile

Quantization Error
A large DC bias can cause amplitude errors in the alternating
part of the signal. AC coupling uses a high pass filter to
remove the DC component from the signal
All of the available
dynamic range of the
analog to digital
converter is dominated
by the DC signal
10 volt
range
on
ADC
The alternating part of

the signal suffers from
quantization error
the measured signal
15
Dr. Peter Avitabile

Quantization Error
Overloading of the ADC causes severe errors also
The ADC range is set
too low for the signal
to be measured and
causes clipping of the
signal
1 volt
range
on
ADC
A
D
C
M
A
X
1.5 volt signal
R
A
N
G
E

the measured signal in
both the time and
frequency domains
16
Dr. Peter Avitabile

Sampling
Each sample is spaced delta t seconds apart. Sufficient
sampling is needed in order to assure that the entire event is
captured. The maximum observable frequency is inversely
proportional to the delta time step used
Fs = 1 / t
Digital Sample
t spacing
17
Dr. Peter Avitabile

Sampling Theory
In order to extract valid frequency information, digitization of
the analog signal must occur at a certain rate.
Shannon's Sampling Theorem states
fs > 2 fmax
That is, the sampling rate must be at least twice the desired
frequency to be measured.
For a time record of T seconds, the lowest frequency
component measurable is
f = 1 / T
With these two properties above, the sampling parameters can
be summarized as
fmax = 1 / 2 t
t = 1 / 2 fmax
18
Dr. Peter Avitabile

Sampling Parameters
Due to the Rayleigh Criteria and Shannons Sampling Theorum,
the following sampling parameters must be observed.
T=N t
19
Dr. Peter Avitabile

Sampling Parameters
Due to the Rayleigh Criteria and Shannons Sampling Theorum,
the following sampling parameters must be observed.
PICK
THEN
AND
fmax = 1 / (2 t)
T = N t
fmax
t = 1 / (2 fmax )
f = 1/(N t)
T = 1 / f
t = T / N
f =1 / T
fmax = N f / 2
If we choose f = 5 Hz and N = 1024

T = 1 / f = 1 / 5 Hz = 0.2 sec
Then
fs = N f = (1024) (5 Hz) = 5120 Hz
fmax = fs = (5120 Hz) / 2 = 2560 Hz
20
Dr. Peter Avitabile

Time vs. Frequency

An inverse relationship between time and frequency exists
T
BW
Given delta t = .0019531 and N = 1024 time points,
then T = 2 sec and BW= 256 Hz and delta f = 0.5 Hz
TIME DOMAIN
FREQUENCY DOMAIN
BW
then T = 1sec sec and BW = 512 Hz and delta f = 1 Hz
TIME DOMAIN
FREQUENCY DOMAIN
BW
then T = 1 sec and BW = 256 Hz and delta f = 1 Hz
TIME DOMAIN
FREQUENCY DOMAIN
21
Dr. Peter Avitabile

Aliasing
WRAP-AROUND
ACTUAL SIGNAL
OBSERVED
ACTUAL
ALIASED SIGNAL
f max
Aliasing results when the sampling does not occur fast enough.
Sampling must occur faster than twice the highest frequency
to be measured in the data - sampling of 10 to 20 times the
signal is sufficient for most time representations of varying
signals
However, in order to accurately represent a signal in the
frequency domain, sampling need only occur at greater than
twice the frequency of interest
22
Dr. Peter Avitabile

Anti-Aliasing Filters
Most good FFT analyzers have
anti-aliasing filters which
protect against aliasing.
WRAP-AROUND
OBSERVED
These are low pass filters that

typically have a roll off rate and
are not ideal.
Usually only 80% of the antialiasing filter range is used to
provide additional protection
against aliasing.
ACTUAL
f max
BW
CF
800
23
1024
Dr. Peter Avitabile

Fourier Transform
Forward Fourier Transform

+
Sx (f )= x ( t )e j2 ft dt
and Inverse Fourier Transform

+
x ( t )= Sx (f )e j2 ft df
24
Dr. Peter Avitabile

Discrete Fourier Transform

Even though the actual time signal is continuous, the signal is
discretized and the transformation at discrete points is
+
Sx (mf )= x ( t )e j2 mf t dt
This integral is evaluated as

Sx (mf )t
x(nt )e j2mf nt
n =
However, if only a finite sample is available (which is generally

the case), then the transformation becomes
N 1
Sx (mf )t x( nt )e j2 mf nt
n =0
25
Dr. Peter Avitabile

Fourier Transform - Periodic Signal

Actual Time
Signal
ACTUAL
DATA
Captured Time
Signal
CAPTURED
DATA
Reconstructed
Time Signal
RECONTRUCTED
DATA
Frequency
Spectrum
26
Dr. Peter Avitabile

Fourier Transform - Nonperiodic Signal

Actual Time
Signal
ACTUAL
DATA
Captured Time
Signal
CAPTURED
DATA
Reconstructed
Time Signal
RECONTRUCTED
DATA
Frequency
Spectrum
27
Dr. Peter Avitabile

Leakage
F
R
E
Q
ACTUAL
DATA
CAPTURED
DATA
T
I
M
E
Periodic Signal
RECONTRUCTED
DATA
Non-Periodic Signal
U
E
N
C
Y
ACTUAL
DATA
CAPTURED
DATA
RECONTRUCTED
DATA
Leakage due to
signal distortion
28
Dr. Peter Avitabile

Leakage
When the measured signal is not periodic in the sample
interval, incorrect estimates of the amplitude and frequency
occur. This error is referred to as leakage.
Basically, the actual energy distribution is smeared across the
frequency spectrum and energy leaks from a particular f into
adjacent f s.
Leakage is probably the most common and most serious digital
signal processing error. Unlike aliasing, the effects of leakage
can not be eliminated.
29
Dr. Peter Avitabile

Windows
0
-10
AMPLITUDE
-20
0
-30
-10
-40
-20
-50
-30
-60
dB
ROLLOFF
-40
-70
-50
- 80
-60
- 90
dB
-100
-16
-70
-14
-12
-10
-8
-6
-4
-2
10
12
14
-100
-16
-14
-12
-10
-8
15.9375
- 80
- 90
WIDTH
Windows
-6
-4
-2
10
12
14
15.9375
Dr. Peter Avitabile

Windows
Overview window concept
Discuss different windows
Discuss effects of windows
Windows
Dr. Peter Avitabile

Windows
A window is a weighting function that is applied to the measured
signal. The function of the window is to make the measured
signal appear to look more periodic in the sample interval
thereby reducing the effects of leakage
Some common windows are
* Rectangular
* Hanning
* Flat Top
* Force / Exponential
Windows
Dr. Peter Avitabile

Windows
In order to better satisfy the periodicity requirement of the
FFT process, time weighting functions, called windows, are used.
Essentially, these weighting functions attempt to heavily weight
the beginning and end of the sample record to zero - the middle
of the sample is heavily weighted towards unity
ACTUAL
DATA
CAPTURED
DATA
T
I
M
E
Periodic Signal
RECONTRUCTED
DATA
Non-Periodic Signal
ACTUAL
DATA
CAPTURED
DATA
RECONTRUCTED
DATA
Windows
Dr. Peter Avitabile

F
R
E
Q
U
E
N
C
Y
Windows - Rectangular/Hanning/Flattop
Rectangular - Unity gain applied to entire sample interval; this
window can have up to 36% amplitude error if the signal is not
periodic in the sample interval; good for signals that inherently
satisfy the periodicity requirement of the FFT process
Hanning - Cosine bell shaped weighting which heavily weights the
beginning and end of the sample interval to zero; this window
can have up to 16% amplitude error; the main frequency will
show some adjacent side band frequencies but then quickly
attenuates; good for general purpose signal applications
Flat Top - Multi-sine weighting function; this window has
excellent amplitude characteristics (0.1% error) but very poor
frequency resolution; very good for calibration purposes with
discrete sine
Windows
Dr. Peter Avitabile

Windows - Rectangular/Hanning/Flattop
Time weighting functions
are applied to minimize
the effects of leakage
AMPLITUDE
ROLLOFF
Rectangular
Hanning
WIDTH
General window
frequency characteristics
Flat Top
and many others
Windows DO NOT eliminate leakage !!!

Windows
Dr. Peter Avitabile

Windows - Rectangular
The rectangular window function is shown below. The main lobe is narrow, but the side lobes are very large
and roll off quite slowly. The main lobe is quite rounded and can introduce large measurement errors. The
rectangular window can have amplitude errors as large as 36%.
-10
-20
Amplitude
-30
-40
-50
-60
dB
-70
- 80
- 90
-100
-16
-14
-12
Windows
-10
-8
-6
-4
-2
10
12
14
15.9375
-3
-2.5
-2
-1.5
-1
-0.5
0.5
1.5
2.5
Dr. Peter Avitabile

Windows - Hanning
The hanning window function is shown below. The first few side lobes are rather large, but a 60 dB/octave
roll-off rate is helpful. This window is most useful for searching operations where good frequency
resolution is needed, but amplitude accuracy is not important; the hanning window will have amplitude errors
of as much as 16%.
0
-10
-20
Amplitude
-30
-40
-50
-60
dB
-70
- 80
- 90
-100
-16
-14
-12
Windows
-10
-8
-6
-4
-2
10
12
14
15.9375
-3
-2.5
-2
-1.5
-1
-0.5
0.5
1.5
2.5
Dr. Peter Avitabile

Windows - Flat Top

The flat top window function is shown below. The main lobe is very flat and spreads over several frequency
bins. While this window suffers from frequency resolution, the amplitude can be measured very accurately
to 0.1%.
-10
-20
Amplitude
-30
-40
-50
-60
dB
-70
- 80
- 90
-100
-16
-14
-12
-10
Windows
-8
-6
-4
-2
10
12
14
-3
15.9375
-2.5
-2
-1.5
-1
-0.5
0.5
1.5
2.5
Dr. Peter Avitabile

Windows - Hanning Window Process
Windows
10
Dr. Peter Avitabile

Windows - Flattop Window Process
Windows
11
Dr. Peter Avitabile

Windows - Periodic Signal

Rectangular
Hanning
Flat Top
-10
-10
-10
-20
-20
-20
-30
-30
-30
-40
-40
-40
-50
-50
-50
-60
-60
-60
dB
dB
dB
-70
-70
-70
- 80
- 80
- 80
- 90
- 90
- 90
-100
-16
-14
-12
-10
-8
-6
-4
Windows
-2
10
12
14
15.9375
-100
-16
-14
-12
-10
-8
-6
-4
-2
12
10
12
14
15.9375
-100
-16
-14
-12
-10
-8
-6
-4
-2
Dr. Peter Avitabile

10
12
14
15.9375
Window Effects
THEORETICAL WINDOW SHAPE
ACTUAL SIGNAL
0
-1 DELTA F
0 DELTA F
1 DELTA F
RESULTING FREQUENCY SPECTRUM
Windows
X
7
CONVOLUTION OF THE
THEORETICAL WINDOW
AND THE ACTUAL SIGNAL
IN THE FREQUENCY DOMAIN
13
Dr. Peter Avitabile

Windows - Force/Exponential for Impact Testing

Special windows are used for impact testing
Force
window
Windows
14
Dr. Peter Avitabile

Windows - Force/Exponential for Impact Testing

Special windows are used for impact testing
Exponential
window
Windows
15
Dr. Peter Avitabile

INPUT
SYSTEM
u(t)
n(t)
x(t)
OUTPUT
v(t)
H
m(t)
ACTUAL
NOISE
MEASURED
y(t)
-10
-20
-30
-40
1.0000
-50
-60
-1.0000
dB
-70
- 80
- 90
-100
-16
-14
-12
-10
-8
-6
-4
-2
10
12
14
15.9375
Dr. Peter Avitabile

Define the basic measurements needed for
experimental modal analysis
Define linear spectra and power spectra

Include the effects of noise
Dr. Peter Avitabile

Actual time signals
ANALOG SIGNALS
OUTPUT
INPUT
AUTORANGE ANALYZER
OUTPUT
INPUT
APPLY WINDOWS
INPUT
OUTPUT
COMPUTE FFT
LINEAR SPECTRA
LINEAR
OUTPUT
SPECTRUM
LINEAR
INPUT
SPECTRUM
INPUT
POWER
SPECTRUM
OUTPUT
POWER
SPECTRUM
CROSS
POWER
SPECTRUM
COHERENC E FUNCTION
Dr. Peter Avitabile

Measurements - Linear Spectra

x(t)
INPUT
Sx(f)
h(t)
y(t)
TIME
SYSTEM
OUTPUT
FFT & IFT
H(f)
Sy(f)
FREQUENCY
x(t)
- time domain input to the system
y(t)
- time domain output to the system
Sx(f)
- linear Fourier spectrum of x(t)
Sy(f)
- linear Fourier spectrum of y(t)
H(f)
- system transfer function
h(t)
- system impulse response
Dr. Peter Avitabile


+
x ( t )= Sx (f )e j2 ft df
Sx (f )= x ( t )e j2 ft dt
y( t )= S y (f )e
j2 ft
S y (f )= y( t )e j2 ft dt
df
h ( t )= H (f )e
j2 ft
H (f )= h ( t )e j2 ft dt
df
Note: Sx and Sy are complex valued functions
Dr. Peter Avitabile

Measurements - Power Spectra

Rxx(t)
INPUT
Gxx(f)
Ryx(t)
Ryy(t)
TIME
SYSTEM
OUTPUT
FFT & IFT
Gxy(f)
Gyy(f)
FREQUENCY
Rxx(t) - autocorrelation of the input signal x(t)

Ryy(t) - autocorrelation of the output signal y(t)
Ryx(t) - cross correlation of y(t) and x(t)
Gxx(f) - autopower spectrum of x(t)
G xx ( f ) = S x ( f ) S*x ( f )
Gyy(f) - autopower spectrum of y(t)
G yy ( f ) = S y ( f ) S*y ( f )
Gyx(f) - cross power spectrum of y(t) and x(t)
G yx ( f ) = S y ( f ) S*x ( f )
Dr. Peter Avitabile


lim 1
R xx ()=E[ x ( t ), x ( t + )]=
x ( t )x ( t + )dt
T TT
+
G xx (f )= R xx ()e j2 ft d=Sx (f )S*x (f )
lim 1
R yy ()=E[ y( t ), y( t + )]=
y( t )y( t + )dt
T TT
+
G yy (f )= R yy ()e j2 ft d=S y (f )S*y (f )
lim 1
R yx ()=E[ y( t ), x ( t + )]=
y( t )x ( t + )dt
T TT
+
G yx (f )= R yx ()e j2 ft d=S y (f )S*x (f )
Dr. Peter Avitabile

Measurements - Derived Relationships

S y =HSx
H1 formulation
- susceptible to noise on the input
- underestimates the actual H of the system
S y S*x G yx
H=
=
*
Sx Sx G xx
S y S*x =HSx S*x
H2 formulation
- susceptible to noise on the output
- overestimates the actual H of the system
Other
formulations
for H exist
S y S*y G yy
H=
=
*
Sx S y G xy
S y S*y =HSx S*y

COHERENCE
2
xy
(S y S*x )(Sx S*y )

(Sx S*x )(S y S*y )
8
G yx / G xx
G yy / G xy
H1
H2
Dr. Peter Avitabile

Measurements - Noise
H=G uv /G uu
1
H1 =H
1+G nn
G uu
G mm
H 2 =H1+
vv
INPUT
SYSTEM
u(t)
n(t)
OUTPUT
v(t)
H
m(t)
x(t)
ACTUAL
y(t)
NOISE
MEASURED
Dr. Peter Avitabile

H1 FORMULATION - OUTPUT NOISE ONLY
Using the basic input-output model and adding noise Sm on
the output, gives
Sm + Sv = H Su
Post-multiplying by the conjugate of the input spectrum Su* ,
gives
( Sm + Sv ) Su* = H1 Su Su*
Sm Su* + Sv Su* = H1 Su Su*
If the output noise is incoherent with input signal
(uncorrelated), then SmSu* = 0 as more averages are taken.
Then the following can be written
INPUT
SYSTEM
u(t)
H1 = Sv Su* / Su Su* = Guv / Guu

n(t)
x(t)
10
OUTPUT
v(t)
H
m(t)
ACTUAL
y(t)
Dr. Peter Avitabile

NOISE
MEASURED
H2 FORMULATION - OUTPUT NOISE ONLY
Using the basic input-output model and adding noise Sm on the
output, gives
Sm + Sv = H Su
Post-multiplying by the conjugate of the output spectrum
( Sm* + Sv* ) , gives
( Sm + Sv ) ( Sm* + Sv* ) = H2 Su ( Sm* + Sv* )
Sm Sm* + Sv Sv* + Sv Sm* + Sm Sv* = H2 Su Sm* + H2 Su Sv*
If the output noise is incoherent with input and output signal
(uncorrelated), then as more averages are taken, the following can
be written
INPUT
SYSTEM
Sm Sm* + Sv Sv* = H2 Su Sv*
u(t)
Gmm + Gvv = H2 Guv
n(t)
OUTPUT
v(t)
H
m(t)
ACTUAL
NOISE
H2 = ( Gmm + Gvv ) / Guv = H + Gmm / Guv

x(t)
H2 = H ( 1 + Gmm / Gvv )
11
y(t)
Dr. Peter Avitabile

MEASURED
H1 FORMULATION - INPUT NOISE ONLY
Using the basic input-output model and adding noise Sn on the input,
gives
Sv = H ( Su + Sn )
Post-multiplying by the conjugate of the input spectrum ( Su* + Sn* ) ,
gives
Sv ( Su* + Sn* ) = H1 ( Su + Sn ) ( Su* + Sn* )
Sv Su* + Sv Sn* = H1 ( SuSu* + SnSn* + SnSu* + SuSn* )
If the input noise is incoherent with input and output signal
(uncorrelated), then as more averages are taken, the following can be
written
INPUT
SYSTEM
u(t)
Sv Su* = H1 ( SuSu* + SnSn* )

Gvu = H1 ( Guu + Gnn )
n(t)
OUTPUT
v(t)
H
m(t)
ACTUAL
NOISE
H1 = Guv / ( Guu + Gnn ) = ( Guv / Guu ) / ( 1 + Gnn / Guu )

x(t)
H1 = H / ( 1 + Gnn / Guu )
12
y(t)
Dr. Peter Avitabile

MEASURED
H2 FORMULATION - INPUT NOISE ONLY
Using the basic input-output model and adding noise Sn on
the input, gives
Sv = H ( Su + Sn )
Post-multiplying by the conjugate of the output spectrum
Sv* , gives
Sv Sv* = H2 ( Su + Sn ) Sv*
Sv Sv* = H2 ( Su Sv* + Sn Sv* )
If the input noise is incoherent with input and output signal
(uncorrelated), then as more averages are taken, the
following can be written
INPUT
SYSTEM
u(t)
H2 = Gvv / Guv
n(t)
x(t)
13
OUTPUT
v(t)
H
m(t)
ACTUAL
y(t)
Dr. Peter Avitabile

NOISE
MEASURED
COHERENCE - OUTPUT NOISE
Using the basic input-output coherence model and adding
noise Sm on the output, gives
2 = ( Gyx 2 ) / Gxx Gyy
2 = ( H Guu )2 / [( Su Su* ) ( Sv + Sm ) ( Sv* + Sm* )
2 = ( H Guu )2 / [( Guu ) ( SvSv* + SmSm* + SmSv* + SvSm* )
As more averages are taken, the following can be written

2 = ( H 2 Guu 2 ) / [( Guu ) ( Gvv + Gmm )]
Recalling that
H2 =
Gvv / Guu
(since Sv = H Su)
INPUT
SYSTEM
u(t)
the following can be written

2 = Gvv / ( Gvv + Gmm )
n(t)
OUTPUT
v(t)
H
m(t)
ACTUAL
NOISE
2 = 1 / ( 1 + Gmm/Gvv )
x(t)
14
y(t)
Dr. Peter Avitabile

MEASURED
Frequency Response Function - Hv

When considering both noise on the input and
output simultaneously, another frequency reponse
function can be computed from the total least
squares solution
{S y } {Sm }= [H V ]{{Sx } {Sn }}

This formulation is a better approximation of the
true frequency response of the system in the
presence of noise on both input and output
simultaneously.
15
Dr. Peter Avitabile

H1, H2, HV - Comparison

H 1 reduces noise on the output only
H 2 reduces noise on the input only
H v reduces noise on the input and output sim ultaneously
HV
OUTPUT
H1
H2
INPUT
16
Dr. Peter Avitabile

Measurements - Auto Power Spectrum
x(t)
y(t)
OUTPUT RESPONSE
INPUT FORCE
G xx (f)
G yy (f)
AVERAGED INPUT
AVERAGED OUTPUT
POWER SPECTRUM
POWER SPECTRUM
17
Dr. Peter Avitabile

Measurements - Cross Power Spectrum
AVERAGED INPUT
AVERAGED OUTPUT
POWER SPECTRUM
POWER SPECTRUM
G yy (f)
G xx (f)
AVERAGED CROSS
POWER SPECTRUM
G yx (f)
18
Dr. Peter Avitabile

Measurements - Frequency Response Function
AVERAGED INPUT
AVERAGED CROSS
AVERAGED OUTPUT
POWER SPECTRUM
POWER SPECTRUM
POWER SPECTRUM
G yx (f)
G yy (f)
G xx (f)
H(f)
19
Dr. Peter Avitabile

Measurements - FRF & Coherence

Coherence
1
Real
0
0Hz
AVG:
200Hz
COHERENCE
Freq Resp
40
dB Mag
-60
0Hz
AVG:
200Hz
20
Dr. Peter Avitabile

1
h 13
1
2
1
h 23
3
h 33
h 31
h 33
h 32
Dr. Peter Avitabile

Impact Excitation
Dr. Peter Avitabile

Impact Excitation
Overview impact excitation techniques
Review hammer/tip characteristics
Review special DSP considerations
Dr. Peter Avitabile

Impact Excitation
An impulsive excitation which is very short in the time window
usually lasting less than 5% of the sample interval.
ADVANTAGES
- easy setup
- fast measurement time
- minimum of equipment
- low cost
DISADVANTAGES
- poor rms to peak levels
- poor for nonlinear structures
- force/response windows needed
- pretrigger delay needed
- double impacts may occur
- high potential for signal overload and underload of ADC
Dr. Peter Avitabile

Impact Excitation - Hammer Tip Selection

The force spectrum can be customized to some extent
through the use of hammer tips with various hardnesses.
A hard tip has a very short pulse and will excite a wide
frequency range. A soft tip has a long pulse and will excite
a narrow frequency range.
However, the hammer tip alone does not totally determine
the frequency range excited. The local flexibility of the
structure must also be considered.
Dr. Peter Avitabile

Impact Excitation - Hammer Tip Selection

METAL TIP
HARD PLASTIC TIP
Real
Real
-976.5625us
TIME PULSE
123.9624ms
-976.5625us
dB Mag
TIME PULSE
123.9624ms
dB Mag
FREQUENCY SPECTRUM
0Hz
6.4kHz
0Hz
FREQUENCY SPECTRUM
6.4kHz
RUBBER TIP
SOFT PLASTIC TIP

Real
Real
-976.5625us
TIME PULSE
-976.5625us
123.9624ms
TIME PULSE
123.9624ms
dB Mag
dB Mag
0Hz
FREQUENCY SPECTRUM
6.4kHz
0Hz
FREQUENCY SPECTRUM
6.4kHz
Dr. Peter Avitabile

Impact Excitation - Pretrigger Delay

Pretrigger delay is often used to minimize the distortion of
the triggering effect from the impact pulse
t=0
NO PRETRIGGER
USED
t=0
PRETRIGGER
SPECIFIED
Dr. Peter Avitabile

Impact Excitation - Double Impact

Double impacts can occur due to a sloppy hammer swing or
many times due to the responsive nature of many structures.
They should be avoided wherever possible.
DOUBLE IMPACT
Real
DOUBLE IMPACT
-976.5625us
998.53516ms
TIME PULSE
Real
dB Mag
-976.5625us
0Hz
FREQUENCY SPECTRUM
TIME PULSE
998.53516ms
800Hz
dB Mag
FREQUENCY SPECTRUM
0Hz
800Hz
Dr. Peter Avitabile

Impact Excitation - Right Hammer for the Test

40
COHERENCE
dB Mag
FRF
-60
0Hz
800Hz
40
COHERENCE
FRF
dB Mag
-60
0Hz
200Hz
Dr. Peter Avitabile

Impact Excitation - Windows May Be Necessary
ACTUAL TIME SIGNAL
SAMPLED SIGNAL
WINDOW WEIGHTING
WINDOWED TIME SIGNAL
10
Dr. Peter Avitabile

Impact Excitation - Exponential Window

If the signal does not
naturally decay within the
sample interval, then an
exponentially decaying
window may be necessary.
T=N t
However, many times

changing the signal
processing parameters such
as bandwidth and number of
spectral lines may produce a
signal which requires less
window weighting
T=N t
11
Dr. Peter Avitabile

Multiple Reference Impact Test

Mount a few accelerometers at key points on the structure
where the majority of the modes can be observed.
Impact ALL points in ALL directions.
Multiple reference data is then obtained.
Ref#1
Ref#2
Ref#3
12
Dr. Peter Avitabile

Multiple Reference Impact Test

Mount ALL the
accelerometers at ALL
points in ALL of the
required directions.
Impact a few key points
where most of the
desired modes can be
observed.
Multiple reference data
is then obtained.
Ref#1
13
Ref#2
Ref#3
Dr. Peter Avitabile

Shaker Excitation
14
Dr. Peter Avitabile

Shaker Excitation
RESPONSE TRANSDUCER
Excitation device is
attached to the
structure using a long
rod called a stinger
or quill
STRUCTURE UNDER TEST

FORCE TRANSDUCER
STINGER
SHAKER
Excitation device is attached to the structure using a long rod called a

stinger or quill
Its purpose is to provide input along the shaker excitation axis with
essentially no excitation of the other directions
It is also intended to be flexible enough to not provide any stiffness
to the other directions
The force gage is always mounted on the structure side of the quill
NOT ON THE SHAKER SIDE
15
Dr. Peter Avitabile

Signal Types
Excitation techniques can be broken down into two categories:
Deterministic Signals
Non-Deterministic (Random) Signals
Deterministic Signals
conform to a particular mathematical relationship
can be described exactly at any instant in time
response of the system can also be exactly defined if the
system character is known
swept sine, sine chirp, digital stepped sine are examples
Non-Deterministic (Random) Signals
do not conform to a particular mathematical relationship
can not be described exactly at any instant in time
described by some statistical character of the signal
generally have varying amplitude, phase and frequency
content at any point in time
pure random, periodic random, burst random are examples
16
Dr. Peter Avitabile

Swept Sine Excitation

INPUT EXCITATION
Slowly changing sine signal sweeping from one frequency to another

frequency
17
Dr. Peter Avitabile

Swept Sine Excitation

A slowly changing sine output sweeping from one frequency to
another frequency
ADVANTAGES
best peak to RMS level
best signal to noise ratio
good for nonlinear characterization
widely accepted and understood
DISADVANTAGES
slowest of all test methods
leakage is a problem
does not take advantage of speed of FFT process
18
Dr. Peter Avitabile

Random Excitation with Hanning Window

AUTORANGING
AVERAGING WITH WINDOW
An ergodic, stationary signal with Gaussian probability distribution.

Typically, has frequency content at all frequencies.
19
Dr. Peter Avitabile


An ergodic, stationary signal with Gaussian probability distribution.
Typically, has frequency content at all frequencies.
ADVANTAGES
gives a good linear approximation for a system with slight nonlinearities
relatively fast
overlap processing can be used
relatively good general purpose excitation
DISADVANTAGES
even with windows applied to the measurement leakage
is a very serious problem
FRFs are generally distorted due to leakage with
(significant distortion at the peaks)
excessive averaging necessary to reduce variance on data
20
Dr. Peter Avitabile


Time signal
Frequency Signal
0s
1.999s
0Hz
0s
1.999s
0Hz
21
400Hz
AVG: 10
Dr. Peter Avitabile

400Hz
Random Excitation with Overlap Processing

OVERLAP PROCESSING
10
22
used to reduce test time with

pure random excitations
Hanning window tends to weight
the first and last quarter of
the time block to zero and this
data is not effectively used in
the normal averaging process
effectively uses the portion of
the block that has been heavily
weighted to zero
overlap processing allows for
almost twice as many averages
with the same data when fifty
percent overlap is used
Dr. Peter Avitabile

Burst Random Excitation

AUTORANGING
AVERAGING
A random excitation that exists over only a portion of the data block
(typically 50% to 70%).
23
Dr. Peter Avitabile


A random excitation that exists over only a portion of the data
block (typically 50% to 70%)
ADVANTAGES
has all the advantages of random excitation
the function is self-windowing
no leakage
DISADVANTAGES
if response does not die out within on sample interval, then
leakage is a problem
24
Dr. Peter Avitabile


Time signal
Frequency Signal
End of burst
0s
Shaker off
0Hz
1.999s
400Hz
Response decays exponentially

0s
0Hz
1.999s
25
AVG: 10
Dr. Peter Avitabile

400Hz
Sine Chirp Excitation

AUTORANGING
AVERAGING
A very fast swept sine signal that starts and stops within one sample
interval of the FFT analyzer
26
Dr. Peter Avitabile


A very fast swept sine signal that starts and stops within one
sample interval of the FFT analyzer
ADVANTAGES
has all the same advantages as swept sine
self windowing function
DISADVANTAGES
nonlinearities will not be averaged out
27
Dr. Peter Avitabile


Time signal
Frequency Signal
0s
1.999s
0Hz
0s
1.999s
0Hz
28
400Hz
AVG: 10
Dr. Peter Avitabile

400Hz
Digital Stepped Sine Excitation

AUTORANGING
AVERAGE
AUTORANGING
IFT
IFT
AVERAGE
Sine waves are generated at discrete frequencies which correspond to

the digital values of the FFT analyzer for the frequency resolution
available. The system is excited with a single sine wave and steady
state response measured. Once one spectral line is obtained, the next
digital frequency is acquired until all frequencies have been measured.
29
Dr. Peter Avitabile

Digital Stepped Sine Excitation

Sine waves are generated at discrete frequencies which correspond
to the digital values of the FFT analyzer for the frequency
resolution available. The system is excited with a single sine wave
and the steady state response is measured. Once one spectral line is
obtained, the next digital frequency is acquired until all frequencies
have been measured.
ADVANTAGES
excellent peak to RMS level
excellent signal to noise ratio
leakage free measurements obtained
DISADVANTAGES
slowest of all test methods
30
Dr. Peter Avitabile

Comparison - Random/Hann, Burst Random, Chirp
RANDOM
BURST RANDOM
SINE CHIRP
31
Dr. Peter Avitabile

Random with Hanning Window vs Burst Random

Frequency Response Function
Coherence
RANDOM
RANDOM
BURST RANDOM
BURST RANDOM
When comparing the measurement with random and burst random,

notice that the random excitation peaks are lower and appear to be
more heavily damped when compared to the burst random. - also notice
the coherence improvement at the resonant peaks.
32
Dr. Peter Avitabile

Random with Hanning Window vs Burst Random
RANDOM
COH
FRF
BURST RANDOM
RANDOM & HANNING
33
BURST RANDOM
Dr. Peter Avitabile

Excitation Considerations - Shaker Excitation

AUTORANGING
Random
with
Hanning
Burst
Random
AVERAGING WITH WINDOW
AUTORANGING
Accurate FRFs are necessary
AVERAGING
AUTORANGING
Leakage is a serious concern
Special excitation
techniques can be
used which will result
in leakage free
measurements without
the use of a window
AVERAGING
Sine
Chirp
1
as well as other techniques

34
Dr. Peter Avitabile

Excitation Considerations - MIMO

Multiple referenced FRFs are
obtained from MIMO test
Energy is distributed
better throughout the
structure making
better measurements
possible
Ref#1
35
Ref#2
Ref#3
Dr. Peter Avitabile

Large or
complicated
structures
require
special
attention
36
Dr. Peter Avitabile

[G XF ]=[H][G FF ]
H11
H
[H]= 21
M
H
No ,1
H12
H 22
M
H No , 2
[H ]=[G XF ][G FF ]1
H1, Ni
L H 2, Ni
M
L H No , Ni
L
Measurements are
developed in a
similar fashion to
the single input
single output case
but using a matrix
formulation
where
No - number of outputs
Ni - number of inputs
37
Dr. Peter Avitabile


Measurements on the same structure can show
tremendously different modal densities depending
on the location of the measurement
38
Dr. Peter Avitabile

Modal Parameter Estimation Concepts

j
[
[
A k ] [A*k ]
A k ] [A*k ] upper [A k ] [A *k ]
[H(s )]=
+
+
+
+
+
*
*
*
terms (ss k ) (ss k )
k =i (ss k ) (ss k ) terms (ss k ) (ss k )
lower
SYSTEM EXCITATION/RESPONSE
SDOF POLYNOMIAL
PEAK PICK
MULTIPLE REFERENCE FRF MATRIX DEVELOPMENT
INPUT FORCE
RESIDUAL COMPENSATION
INPUT FORCE
INPUT FORCE
LOCAL CURVEFITTING
IFT
GLOBAL CURVEFITTING
POLYREFERENCE CUVREFITTING
COMPLEX EXPONENTIAL
MDOF POLYNOMIAL
Dr. Peter Avitabile

Parameter Estimation Concepts

NO COMPENSATION
Y
y=mx
X
COMPENSATION
Y
y=mx+b
WHICH DATA ???

Y
Dr. Peter Avitabile

Parameter Extraction Considerations

HOW MANY POINTS ???
RESIDUAL
EFFECTS
RESIDUAL
EFFECTS
ORDER OF THE MODEL

AMOUNT OF DATA TO
BE USED
COMPENSATION FOR
RESIDUALS
HOW MANY MODES ???
The test engineer identifies these items

NOT THE SOFTWARE !!!
Dr. Peter Avitabile


HOW MANY POINTS ???
lower
[Ak ]
[A ]
terms
*
k
*
k
[H( s) ] = ( s s ) + s s
( )
j
[Ak ]
[A ]
*
k
*
k
( s s ) + (s s )
k=i
RESIDUAL
EFFECTS
upper
[Ak ]
[A ]
terms
*
k
RESIDUAL
EFFECTS
*
k
( s s ) + (s s )
HOW MANY MODES ???
[
[
Ak ]
A*k ]
[H(s )] = lower residuals +
+
+ upper residuals
*
(ss k )
k =i (ss k )
j
Dr. Peter Avitabile

Classification of Modes
Well separated - lightly damped
Closely spaced - lightly damped
Well separated - heavily damped
Closely spaced - heavily damped
Dr. Peter Avitabile

Time and Frequency Domain Representations

The basic equations can be cast in either the
time or frequency domain
a1
a1*
h (s)=
+
(s p1 ) (s p1* )
1 t
h ( t )=
e sin d t
md
There is essentially no difference in these

equations from a theoretical standpoint
provided there is an infinite amount of
amplitude and frequency resolution
Dr. Peter Avitabile

Different Forms of the Same Equation
Polynomial Form
1
h (s) =
ms 2 + cs + k
Pole-Zero Form
1/ m
h (s) =
(s p1 )(s p1* )
Partial Fraction Form
a1
a1*
h (s) =
+
(s p1 ) (s p1* )
Exponential Form
1 t
h(t) =
e sin d t
md
Dr. Peter Avitabile

Why Have Different Methods
Time Domain
Frequency Domain
Time or frequency - which to use ???
It really depends on which domain

has the most data
Dr. Peter Avitabile

MDOF - Summation of Many SDOF Systems

h ij ( j ) =
R1
+
( j p1 ) ( j p*1 )
+
D1
a*ij1
a ij1
a*ij 2
a ij 2
+
+ L
( j p 2 ) ( j p*2 )
R2
D2
R3
D3
F1
F2F3
a ij1
h ij ( j ) =
q1u i1u j1
* * *
1 i 1 j1
*
1
qu u
+
( j p1 ) ( j p )
+
q 2u i 2 u j 2
* * *
2 i2 j2
*
2
qu u
( j p 2 ) ( j p )
a ij2
a ij3
2
3
+ L
9
Dr. Peter Avitabile


The FRF is made up
from each individual
mode contribution
which is determined
from the
a ij1
a ij2
a ij3
frequency,
damping,
residue
10
Dr. Peter Avitabile

MDOF
SDOF
The task for the modal test engineer is to

determine the parameters that make up the pieces
of the frequency response function
11
Dr. Peter Avitabile


The FRF matrix contains
redundant information
regarding the system
frequency, damping and
mode shapes
Multiple referenced data
can be used to obtain
better estimates of
modal parameters
12
MULTIPLE REFERENCE FRF MATRIX DEVELOPMENT
LOCAL CURVEFITTING
GLOBAL CURVEFITTING
POLYREFERENCE CUVREFITTING
Dr. Peter Avitabile


Local Curvefitting
- Each measurement is curvefit to estimate
the frequency, damping and residue for
each FRF
- The frequency and damping is allowed to vary for each measurement
and may not be the same for every measurement
ADVANTAGES
- Good for systems where the poles are not global
DISADVANTAGES
- Frequency and damping is different for the system
- Local modes/node points are not characterized well
13
Dr. Peter Avitabile


Global Curvefitting
- A set of measurements are curvefit to
estimate the frequency and damping
- The residue is estimated in a second pass
ADVANTAGES
- Good for systems where the poles are global
- Better estimate of the frequency and damping
- Local modes are better characterized
DISADVANTAGES
- Frequency and damping must be global in FRFs
14
Dr. Peter Avitabile


Polyreference Curvefitting
- A set of measurements are curvefit to
estimate the frequency and damping
- The residue is estimated in a second pass
and is based on redundant FRF matrix information
ADVANTAGES
- Good for systems where the poles are global
- Better estimate of the frequency and damping
- Repeated roots can be identified
DISADVANTAGES
- Frequency and damping must be global in FRFs
15
Dr. Peter Avitabile

Selection of Bands
Select bands for possible SDOF or MDOF
extraction for frequency domain technique.
Residuals ???
Complex ???
16
Dr. Peter Avitabile

Mode Determination Tools

Summation
MIF
A variety of tools assist in the determination

and selection of modes in the structure
1 Point Each From Panels 1,2, and 3 (37,49,241)
4
10
10
10
10
CMIF
10
-1
10
-2
10
0
50
100
150
CMIF
200
250
Frequency (Hz)
300
350
400
450
500
Stability Diagram
17
Dr. Peter Avitabile

Modal Extraction Methods

Circle Fitting
Peak Picking
SDOF Polynomial
I
m
a
g
i
n
a
r
y
Real
A multitude of techniques exist
IFT
Complex Exponential
MDOF Polynomial Methods

18
Dr. Peter Avitabile

Model Validation
Synthesis
Validation tools exist

to assure that an
accurate model has
been extracted from
measured data
1.2
1
0.8
0.6
0.4
0.2
0
MAC
S1
S2
S3
S4
S5
S6
19
Dr. Peter Avitabile

Single Degree of Freedom

Modal Parameter Estimation
Circle Fitting
Peak Picking
SDOF Polynomial
I
m
a
g
i
n
a
r
y
Real
20
Dr. Peter Avitabile

Peak Pick - First Step in Reviewing Test Data

MODE # 1
Simple Peak Picking
MODE # 2
MODE # 3
DOF # 1
DOF #2
DOF # 3
21
Dr. Peter Avitabile

Peak Pick - First Step in Reviewing Test Data

Simple Peak Picking
MODE 2
2
1
MODE 1
5
2
4
1
3
6
22
Dr. Peter Avitabile

Simple Peak Pick - SDOF System

Substitute the pole into the SDOF FRF equation
h ( j)
a1
a1*
=
+
( jn + jd ) ( jn + + jd )
a1 =h ( j)
23
Dr. Peter Avitabile

Simple Peak Pick - Consider Additional Modes

Substitute the first pole into the FRF equation
h ( j)
a1
a1*
a2
a *2
=
+
+
+
( j1 + 1 jd ) ( j1 + 1 + jd ) ( jn + 2 jd 2 ) ( jn + 2 + jd 2 )
MODE 1 CONTRIBUTION
MODE 2 CONTRIBUTION
24
Dr. Peter Avitabile

Simple Peak Pick - Considerations and Use

Peak pick is a quick and simple check
Modes must be well spaced
(ie: no significant modal overlap)
Approximate since peak is determined from the
frequency resolution
Good quick check before attempting major modal
parameter estimation
Quickly identify erroneous measurement points
(ie: phase reversal, incorrect direction, etc)
25
Dr. Peter Avitabile

SDOF - Complex Exponential Response
response can be
used to extract
parameters
Amplitude
Time domain
Damping Decay
This time domain
technique is
generally used on
multiple mode time
response data
Period
Generally not used
for single mode

response extraction
h(t)
26
1
md
e t
sin d t
Dr. Peter Avitabile

SDOF Approximation - Effects of Other Modes

Simple symmetric
characteristics of
SDOF system
distorted by
adjacent modes
Real part is shifted

Nyquist circle is
displaced and
rotated
Remove effects of
adjacent modes or
add compensation to
basic equations
Source: Ewins - Modal Testing, 2nd Edition
27
Dr. Peter Avitabile

SDOF Circle Fit - Kennedy-Pancu

Simple equation of a
circle is fit to the data

in the Nyquist
Noise and leakage have
a pronounced effect on
circle
Use of windows tends
to make the circle look

egg-shaped
Circle fit method not
used as much today due

to the availability of
many MDOF methods
Source: Heylen, Modal Analysis - Theory and Testing
h ( j)=
=
28
U + jV
+ R + jI
r + j( r )
U2 +V2
; tan() = U V
Dr. Peter Avitabile

SDOF Polynomial Form

Simple equation of a
polynomial for one mode is

used to fit the function
h (s)=
1
ms 2 + cs + k
Additional terms are
typically added to account

for effects of adjacent or
out-of-band modes
Fast, simple, easy to use

Inappropriate for use with
compensation terms can be added to

account for out of band effects
very closely spaced modes
29
Dr. Peter Avitabile

SDOF Rational Fraction Polynomial

Simply the ratio of two
polynomials
While useful for estimating
SDOF type characteristics,

its real benefits are for
multiple modes
Orthogonal polynomials used
to greatly simplify the

numerical processing
More discussion on this in
the advanced curvefitting
Source: Richardson, Rational fraction Polynomial
30
Dr. Peter Avitabile

Multiple Degree of Freedom

IFT
Complex Exponential
MDOF Polynomial Methods
31
Dr. Peter Avitabile

Complex Exponential
One of the first mdof estimators was the complex exponential which uses
the Prony Algorithm to solve the set of equations. The Toeplitz equations
are used to form the characteristic polynomial followed by the mode
shape extraction using Vandemonde Equation formulation.
m
1
h ( t )=
e kt sin dk t
k =1 m k dk
ADVANTAGE
numerically fast and stable
handles many modes
IFT
DISADVANTAGE
time domain leakage is a concern
must overspecify modes to handle residuals
32
Dr. Peter Avitabile

MDOF Polynomial
This method uses a Rational Fraction polynomial form of the FRF in order
to extract modal parameters. Both the numerator and denominator
polynomials are used in a least squares fit to extract the polynomial
coefficients.
h ij ( j) =
a ij2
a *ij2
( j p 2 ) ( j p*2 )
+
a ij3
a *ij3
( j p 3 ) ( j p*3 )
A key advantage of the frequency domain representation of the FRF is

that the effects of out-of-band modes can be easily accounted for by
adding extra terms to the numerator polynomial.
33
Dr. Peter Avitabile

Other Time Domain Techniques

Other time domain techniques exist which extend the Complex
Exponential technique described above.
Techniques such as Ibrahim Time Domain and Polyreference LSCE
utilize some variant of the equation below to formulate the problem
O
[L]
[h ( t )] = [V] e t
global parameters are extracted for poles and shapes

uses MIMO time data for the estimation process
34
Dr. Peter Avitabile

Other Frequency Domain Techniques

Other frequency domain techniques exist which extend the
polynomial technique described above.
Techniques such as Least Squares Frequency Domain, Orthogonal
Polynomial, Frequency Domain Parameter Identification utilize some
variant of the rational fraction, partial fraction or reduced equation
of motion to formulate the problem
LR ij
u ik L kj
[h ij ( j)] =
+ * + UR ij + 2
k =1 ( j p k )
global poles, MPF and shapes extracted

LSFD nonlinear problem solved iteratively
RFP - ill-conditioning possible for higher order polys
use of orthogonal poly to minimize numerical problems
35
Dr. Peter Avitabile


MODAL PARAMETER ESTIMATION MODELS
Time representation
h ij ( n ) ( t ) + a1h ij ( n 1) ( t ) + L + a 2 n h ij ( n 2 N ) ( t ) = 0
Frequency representation
[( j)
2N
+ a1 ( j) 2 N 1 + L + a 2 N ]h ij ( j) =
[( j)
2M
+ b1 ( j) 2 M 1 + L + b 2 M ]
36
Dr. Peter Avitabile

Modal Parameter Estimation Considerations

Numerical Considerations
Generally the time domain is numerically more
stable than the frequency domain approaches
Time domain techniques are generally better for
handling noisy data acquired
The frequency domain is more advantageous for
averaging noise in the data acquisition phase
37
Dr. Peter Avitabile


Bandwidth Considerations
The time domain is best suited for wide
bandwidths with many modes included in the

estimation process
The frequency domain is best suited for narrow
frequency bands with limited number of modes in

the band
38
Dr. Peter Avitabile


Out-of-Bandwidth Considerations
The frequency domain is best suited for
compensation effects through the use of

residuals in the mathematical formulation
The time domain can only account for out of
band effects of other modes through the use of

more poles in the estimation process
39
Dr. Peter Avitabile


Damping Considerations
The time domain is generally well suited for
lightly damped systems - there is an abundance

of data available in the time domain
The frequency domain is generally well suited for
heavily damped systems - there is sufficient data

represented in the frequency domain
40
Dr. Peter Avitabile


Best Combination
The MPE process can be broken down into two
stages
- poles extraction
- residue estimation
First, estimate polyreference poles (or global
poles) using either a time or frequency domain

technique
Second, estimate residues using a frequency
domain technique where residuals are easier to

include in the estimation process
41
Dr. Peter Avitabile

Linear Algebra Concepts

[A] = [{u1} {u 2 } {u 3}
x . . . x
0 x . . .
[U ]= . 0 x . .
. . 0 x .
0 . . 0 x
s3
T
{v1}
T
{v 2 }
{v 3 }T
O M
[A]nm{X}m ={B}n
[A]nm =[V]nn [S]nm [U]Tmm
{X}m =[A ]gnm{B}n =[[V ]nn [S]nm [U ]Tmm ] {B}n

{X}m =[[U ]mm [S]gnm [V]Tnn ]{B}n
[A ]1=[Adjo int[A]]
Det[A ]
s1
s2
L]
Dr. Peter Avitabile

Linear Algebra
The analytical treatment of structural dynamic
systems naturally results in algebraic equations
that are best suited to be represented through
the use of matrices
Some common matrix representations and linear
algebra concepts are presented in this section
Dr. Peter Avitabile

Linear Algebra
Common analytical and experimental equations
needing linear algebra techniques
[G ] = [H][G ]
yf
[H ] = [G yf ][G ff ]
ff
[[K ][M ]]{x}=0
[M ]{&x&}+[C]{x& }+[K ]{x}={F( t )}
[B(s )]1 = [H(s )] = Adj[B(s )]

det[B(s )]
[B(s )]{x (s )} = {F(s )}

or
O
[H(s )] = [U] S [L]T
O
Dr. Peter Avitabile

Matrix Notation
A matrix [A] can be described using row,column as
a 11
a
21
[A ] = a 31
a
41
a 51
a 12
a 22
a 13
a 23
a 14
a 24
a 32
a 42
a 52
a 33
a 43
a 53
a 34
a 44
a 54
( row , column )
[A]T -Transpose - interchange rows & columns

[A]H - Hermitian - conjugate transpose
Dr. Peter Avitabile

Matrix Notation
A matrix [A] can have some special forms
Diagonal
Square
a 11
a
21
[A ] = a 31
a
41
a 51
a 12
a 22
a 13
a 23
a 14
a 24
a 32
a 42
a 52
a 33
a 43
a 53
a 34
a 44
a 54
a 15
a 25
a 35
a 45
a 55
a 11
a 22
[A ] =
a 33
a 44
a 55
Toeplitz
Symmetric
a 11
a
12
[A ] = a13
a
14
a 15
Triangular
a 12
a 22
a 13
a 23
a 14
a 24
a 23
a 24
a 25
a 33
a 34
a 35
a 34
a 44
a 45
a 15
a 25
a 35
a 45
a 55
a 5
a
4
[A ] = a 3
a
2
a 1
a11
0
[A ] = 0
0
a 12
a 22
a 13
a 23
a 14
a 24
0
0
0
a 33
0
0
a 34
a 44
0
Vandermonde
a6
a5
a7
a6
a8
a7
a4
a3
a2
a5
a4
a3
a6
a5
a4
a9
a8
a7
a6
a 5
1
[A ] =
1
a1
a2
a3
a4
a 12
a 22
a 32
a 24
Dr. Peter Avitabile

a 15
a 25
a 35
a 45
a 55
Matrix Manipulation
A matrix [C] can be computed from [A] & [B] as
a 11
a
21
a 31
a 12
a 22
a 32
a 13
a 23
a 33
a 14
a 24
a 34
b11
a 15 b 21
a 25 b 31
a 35 b 41
b 51
b12
b 22 c11
b 32 = c 21
c
b 42
31
b 52
c12
c 22
c 32
c 21 = a 21b11 + a 22 b 21 + a 23 b 31 + a 24 b 41 + a 25 b 51
c ij = a ik b kj
k
Dr. Peter Avitabile

Simple Set of Equations

A common form of a set of equations is
[A ] {x} = [b]
Underdetermined
# rows < # columns
more unknowns than equations
(optimization solution)
Determined
# rows = # columns
equal number of rows and columns
Overdetermined
# rows > # columns
more equations than unknowns
(least squares or generalized inverse solution)
Dr. Peter Avitabile

Simple Set of Equations

This set of equations has a unique solution
2x y = 1
x + 2 y 1z = 2
y+z =3
2 1 0 x 1
1 2 1 y = 2

0 1 1 z 3
whereas this set of equations does not

2x y = 1
x + 2 y 1z = 2
4x 2 y = 2
2 1 0 x 1
1 2 1 y = 2

4 2 0 z 2
Dr. Peter Avitabile

Static Decomposition
A matrix [A] can be decomposed and written as
[A ] = [L][U ]
Where [L] and [U] are the lower and upper
diagonal matrices that make up the matrix [A]
x
x
[L] = x
x
0
x
x
x
x
0 0
0 0
x 0
x x
x x
0
0
0
0
x
0
[U] = 0
0
0
9
x
x
0
0
0
x
x
x
0
0
x
x
x
x
0
x
x
x
x
Dr. Peter Avitabile

Static Decomposition
Once the matrix [A] is written in this form then
the solution for {x} can easily be obtained as
[A ] = [L][U ]
[U ] {X} = [L]1 [B]
Applications for static decomposition and inverse
of a matrix are plentiful. Common methods are
Gaussian elimination
Crout reduction
Gauss-Doolittle reduction
Cholesky reduction
10
Dr. Peter Avitabile

Eigenvalue Problems
Many problems require that two
matrices [A] & [B] need to be reduced
[[B] [A ]] {x} = 0
[A]{&x&} + [B]{x} = {Q( t )}
Applications for solution of eigenproblems are

plentiful. Common methods are
Jacobi
Givens
Subspace Iteration
Householder
Lanczos
11
Dr. Peter Avitabile

Singular Valued Decomposition

Any matrix can be decomposed using SVD
[A ] = [U ][S][V ]T
[U] - matrix containing left hand eigenvectors
[S] - diagonal matrix of singular values
[V] - matrix containing right hand eigenvectors
12
Dr. Peter Avitabile


SVD allows this equation to be written as
[A] = [{u1} {u 2 } {u 3}
s1
s2
L]
{v1}
T
{v 2 }
{v 3 }T
O M
T
s3
which implies that the matrix [A] can be written in

terms of linearly independent pieces which form
the matrix [A]
[A ] = {u1}s1{v1}T + {u 2 }s 2 {v 2 }T + {u 3 }s 3 {v3 }T +
13
Dr. Peter Avitabile


Assume a vector and singular value to be
1

u1= 2
3

and
s1 = 1
Then the matrix [A1] can be formed to be
[A1 ] = {u1} s1 {u1}T
1
1 2 3

= 2 [1] {1 2 3} = 2 4 6
3

3 6 9
The size of matrix [A1] is (3x3) but its rank is 1

There is only one linearly independent
piece of information in the matrix
14
Dr. Peter Avitabile


Consider another vector and singular value to be
1

u 2= 1
1

and
s2 = 1
Then the matrix [A2] can be formed to be
[A 2 ] = {u 2 } s 2 {u 2 }T
1
1 1 1

= 1 [1] {1 1 1} = 1 1 1
1
1 1 1

The size and rank are the same as previous case

Clearly the rows and columns
are linearly related
15
Dr. Peter Avitabile


Now consider a general matrix [A3] to be
2 3 2
[A 3 ] = 3 5 5 = [A1 ] + [A 2 ]
2 5 10
The characteristics of this matrix are not

obvious at first glance.
Singular valued decomposition can be used to
determine the characteristics of this matrix
16
Dr. Peter Avitabile


The SVD of matrix [A3] is
1
[A] = 2
3
or
1

1
1

0 1
{1 2 3}

{1 1 1}
0
1

0
0 {0 0 0}

1
0
1
[A] = 21{1 2 3}T + 1 1{1 1 1}T + 00{0 0 0}T
3
1
0

These are the independent quantities that

make up the matrix which has a rank of 2
17
Dr. Peter Avitabile

Linear Algebra Applications

The basic solid mechanics formulations as well as
the individual elements used to generate a finite
element model are described by matrices
6L 12 6L
12
6L 4L2 6L 2L2
[k ] = EI3
L 12 6L 12 6L
6L 2L2 6L 4L
E, I
Fj
x C11
C
y 21
C
{} = [C]{} z = 31
xy C 41
xz C51

yz C 61
54
13L
156 22L
22L 4L2 13L 3L2
[m] = AL
420 54
22L
13L 156
13L 3L2 22L
4L2
C12
C 22
C13
C 23
C14
C 24
C15
C 25
C32
C33
C34
C35
C 42
C52
C 43
C53
C 44
C54
C 45
C55
C 62
C 63
C 64
C 65
18
C16 x
C 26 y

C36 z

C 46 xy
C56 xz

C 66 yz
Dr. Peter Avitabile


Finite element model development uses individual
elements that are assembled into system matrices
19
Dr. Peter Avitabile


Structural system equations - coupled
[M ]{&x&}+[C]{x& }+[K ]{x}={F( t )}

Eigensolution - eigenvalues & eigenvectors
[[K ][M ]]{x}=0

Modal space representation
of equations - uncoupled
\
\
{&p&} +
\
{p& } +
20
{p} = [U ]T {F}
Dr. Peter Avitabile


Multiple Input Multiple Output Data Reduction
[G ] = [H][G ]
yx
[H ] = [G yx ][G xx ]1
xx
[Gyx]
[H]
[Gxx]
RESPONSE
(MEASURED)

(UNKNOWN)
FORCE
(MEASURED)
Matrix inversion can only be performed if the

matrix [Gxx] has linearly independent inputs
21
Dr. Peter Avitabile


Principal Component Analysis using SVD
[G xx ] = [{u1} {u 2 } {0}
[Gxx]
s1
s2
L]
{v1}
T
{v 2 }
0
{0}T
O M
T
SVD of the input excitation matrix identifies the

rank of the matrix - that is an indication of how
many linearly independent inputs exist
22
Dr. Peter Avitabile


SVD of Multiple Reference FRF Data
[H] = [{u1} {u 2 } {u 3}
[H]
{v1}
T
{v 2 }
{v 3 }T
O M
T
s1
s2
L]
s3
50
100
150
200
250
Frequency (Hz)
300
350
400
450
500
SVD of the [H] matrix gives an indication

of how many modes exist in the data
23
Dr. Peter Avitabile


Least Squares or Generalized Inverse for
Modal Parameter Estimation Techniques
[
[
Ak ]
A*k ]
[H(s )] =
+
*
(
s
s
(
)
s
s
k =i
k
k)
j
Least squares error minimization of

measured data to an analytical function
24
Dr. Peter Avitabile


Extended analysis and evaluation of systems
[K ][U] = [M I ][U][`2 ]
[U ][K ][U] = [U] [M ][U][` ]

[K ] = [K ] + [V] [` + K ][V]
T
] [
[K S ][U ][U ] [M I ] [K S ][U ][U ] [M I ]

T
[K I ] = [K S ] + [V ]T [`2 + K S ][V ] [[K S ][U ][V ]] [[K S ][U ][V ]]T

generally require matrix manipulation of some type
25
Dr. Peter Avitabile


Many other applications exist
Correlation
Model Updating
Advanced Data Manipulation

Operating Data
Nonlinearities
Rotating Equipment
and the list goes on and on
26
Dr. Peter Avitabile

EXPERIMENTAL
MODAL
TESTING
FINITE
ELEMENT
MODELING
MODAL
PARAMETER
ESTIMATION
PERFORM
EIGEN
SOLUTION
TWENTY YEARS OF
STRUCTURAL DYNAMIC
MODIFICATION
RIB
STIFFNER
MASS
DEVELOP
MODAL
MODEL
MODAL SPACE MODEL

Repeat
until
desired
characteristics
are
obtained
STRUCTURAL
CHANGES
REQUIRED
Yes
MODAL TIE MATRIX
SPRING
No
DONE
DASHPOT
MODAL SPACE MODEL
USE SDM
TO EVALUATE
STRUCTURAL
CHANGES
Peter Avitabile
Modal Analysis and Controls Laboratory
(Excerpt of slides used for presentation at IMAC20 in Los Angeles, California February 2002)
Twenty Years of Structural Dynamic Modification
IMAC20 Los Angeles, California February 2002
Dr. Peter Avitabile


Objectives of this lecture -->> 20 years in 20 minutes:
Describe the basic types of mathematical models used in
modification studies
Real Normal Modes -or- Complex Modes

Modal Space Models -or- Impedance Models
Simplistic Modifications -or- Realistic Modifications
Structural Dynamic Modification -or- System Modeling
Truncation Effects ! ! !
Rotational Degrees of Freedom ! ! !

Dr. Peter Avitabile


Structural Dynamic Modification:
Introduced in the late 70s
Desktop computers were lacking in computational power
Local Eigenvalue Modification Technique allowed for
efficient computation of structural changes but only one

change at a time was possible
Simple changes in mass, damping and stiffness could be
quickly evaluated
Proportional mode assumption used in early approaches

Structural changes or resonant specification allowed
Tuned absorber studies allowed
Dr. Peter Avitabile


Structural Dynamic Modification - Complex Modes:
The proportional mode approximation works well with mass
and stiffness changes for a proportional mode system
However, modifications involving damping tends to disrupt
any proportionality that may exist requiring the formulation

to address non-proportional modes
State space formulation allowed for the use of non-
proportional damping models
Eigenvalue modification approach or Local Eigenvalue
Modification Procedure used with the complex mode

approach

Dr. Peter Avitabile


Local Eigenvalue Modification Procedure:
Due to the significant computational advantage and lack of
significant computational resources, the Local Eigenvalue

Modification Procedure (LEMP) was the method used in
early approaches
As computational power increased, the need for the LEMP
decreased
Many implementations reverted to the Eigenvalue
Modification Technique as computational resources became

more readily available

Dr. Peter Avitabile


Realistic Structural Changes:
Early implementations only allowed for simple changes in
mass, damping and stiffness
3D beam elements and rib stiffeners using finite element
approaches were developed but experimental data typically

only contained translational DOF - no rotational DOF
Approaches for beam using Guyan reduction and 3 point
beam bending were developed to approximate beam

characteristics for structural changes
System models were also developed using the unconstrained
modal approach for individual components

Dr. Peter Avitabile


Estimation of Rotational DOF:
The lack of rotational DOF spurred the development of
different approaches to estimate these necessary DOF
Spline fitting approaches were developed

FEM shape expansion approaches were developed
Experimental approaches were attempted to attempt to
estimate rotational DOF

Dr. Peter Avitabile


Modal Truncation Effects:
The most serious of all effects is related to the lack of all
of the modes to completely describe the entire space of

the model
Modal truncation is by far one of the most serious concerns
in defining the modal data base
Attempts to approximate residual approximations for the
truncated modes help to improve the structural modification

process

Dr. Peter Avitabile


Impedance Modeling as an Alternative:
At approximately the same time, structural modifications
using response functions were developed
This impedance modeling approach provided a good
alternative to the modal approach
FRF functions contain the effects of all modes in the
measured functions - modal truncation is not a issue
However, rotational DOF are difficult to obtain for
measured functions
System modeling using the impedance modeling approach
were also developed

Dr. Peter Avitabile


The first 10 years of IMAC were seen to be the birth
and development of Structural Dynamic Modification.
The Eigenvalue Modification and LEMP. This was followed
by the development of more realistic structural elements.
Issues pertaining to RDOF and truncation were addressed.
The next 10 years leaned towards the utilization of the
SDM Technique and development of System Modeling
tools. There was also a trend towards using frequency
based modification and system modeling techniques.
Even today, the two most important issues still pose
problems for the modal and frequency based techniques.
Rotational DOF
Truncation
10
Dr. Peter Avitabile

Modal Space Models

These models are developed from the modal
characteristics describing the frequency, damping
and mode shape:
[M ]{&x&}+[C]{x& }+[K ]{x}={F( t )}
ls
s
e
l
d
e
od
Mo
M
d
d
se
e
a
s
B
Ba
al
l
d
a
o
c
M
si
y
h
P
\
f1
p1
k1
m2
c1
MODE 1
f2
p2
m1
k2
m3
c2
MODE 2
f3
p3
k3
c3
\
{&p&} +
\
{p& } +
{p} = [U ]T {F}
MODE 3

11
Dr. Peter Avitabile

MDOF Equations
Equation of Motion (n x n)
[M ]{&x&}+[C]{x& }+[K ]{x}={F( t )}

Eigensolution
[[K ][M ]]{x}=0

Frequencies (eigenvalues) and Mode Shapes (eigenvectors)
\
2
1
=

\
22
and
Modal transformation (n x m)
{x} = [U ]{p} = [{u1} {u 2}

12
[U] = [{u1} {u 2}
L]
p1

L]p 2
M

Dr. Peter Avitabile

[U ]T [M ][U ]{&p&} + [U ]T [C][U ]{p& } + [U ]T [K ][U ]{p} = [U ]T{F}

Modal equations (uncoupled)
m1
m2
&p&1 c1
&p& +
c2
2
\ M
Modal Mass
\
[U1 ]T [M1 ][U1 ] =
p& 1 k1
p& +
k2
2
\ M
T
p1 {u1} {F}
p ={u }T {F}
2 2
M
\ M
Modal Damping
M1
\
[U1 ]T [C1 ][U1 ] =

13
Modal Stiffness
C1
\
[U1 ]T [K1 ][U1 ] =
Dr. Peter Avitabile

K1

Diagonal Matrices (m x m) Modal Mass
Modal Damping
{&p&} +
Modal Stiffness
{p& } +
{p} = [U ]T{F}
Highly coupled system
transformed into
simple system
f1
p1
k1

14
k2
m3
c2
MODE 2
f3
p3
m2
c1
MODE 1
f2
p2
m1
k3
c3
MODE 3
Dr. Peter Avitabile

Structural Changes to the System

Mass and Stiffness Changes (n x n)
[K 2 ] = [K1 ] + [K12 ]
[M 2 ] = [M1 ] + [M12 ]
Another eigensolution is needed to uncouple the system
[[[K1 ] + [K12 ]] [[M1 ] + [M12 ]]]{x} = {0}

Some advantages can be obtained if the existing
modal space solution is used to estimate the changes
to the system

15
Dr. Peter Avitabile

Eigenvalue Modification Technique

The modal projection is used to recast the equations as
O
+ [K ]{p } = [0]
+ [M ]{&p& } +
M
K
1
12
1
1
12 1

O
O
where
[M12 ] = [U1 ]T [M12 ][U1 ]
[K12 ] = [U1 ]T [K12 ][U1 ]
An eigensolution of an (m x m) system is required to

uncouple the set of equations
O
+ [U ]T [K ][U ]
+ [U ]T [M ][U ] {p } = {0}
K
M
1
1
12
1
1
1
12
1 1

O
O

16
Dr. Peter Avitabile


The mode shapes are updated using
{p1} = [U12 ]{p 2 }
resulting in
[U 2 ] = [U1 ][U12 ]
which implies that the final modes are developed as

linear combinations of the original modes of the
unmodified system
For the ith mode of the system, the following
describes the modified mode
{u 2 }(i ) = {u1}(1) U12(1i) + {u1}( 2) U12( 2i ) + {u1}(3) U12(3i ) +L

17
Dr. Peter Avitabile


The process is best shown in the schematic below
ORIGINAL
STATE
PHYSICAL
SPACE
[M ],[K ]
1
1
MODAL
TRANSFORMATION
{ x } = [ U 1 ] { p 1}
MODAL
SPACE
2
[ ],[U ]
1
1
MOD IFIC ATION
[ M
12
] , [ K 12 ]
{ p } = [U
]{ p }
1
12
2

18
MODIFIED
STATE
[M ],[K ]
2
2
'N'
PHYSICAL
DOF
{ x } = [ U 2 ] { p 2}
M<<N
2
[ ],[U ]
2
2
'M'
MODAL
DOF
Dr. Peter Avitabile

Local Eigenvalue Modification Technique

Additional computation advantages are obtained through
singular value decomposition of the changes to the system
O
r
T
[M12 ] = [Tm ] m [Tm ] = {t mi } mi {t mi }T

i =1
O
r
T
[K12 ] = [Tk ] k [Tk ] = {t ki } ki {t ki }T

i =1
O
which can be projected to modal space as
O
r
T
T
[M12 ] = [ U1] [Tm ] m [Tm ] [ U] = {v mi } mi {vmi }T

i =1
O
O
r
T
T
[K12 ] = [ U1] [Tk ] k [Tk ] [ U1] = {v ki } ki {v ki }T

i =1
O

19
Dr. Peter Avitabile

Local Eigenvalue Modification Technique

If only one change of mass or stiffness is considered then
these equations can be reduced to
O
+ {v } {v }T
{p } = {0}
K
M
1
1k 1k 1k
1

O
O
O
O

+ {v } {v }T {p } = {0}
K
M
1
1
1m 1m 1m 1

O
O
The solution then reduces to a second order equation for

each of the m modes of the system
{ }
u (i ) T {t }
m
k
1
=
k i =1 22 i2

1
2 m
20
{ }
u (i ) T {t }
m
m
=
22 i2
i =1
Dr. Peter Avitabile

System Modeling Formulation

A system model can be developed using the same approach
for individual modal components
UA
MA
&p& A
T
[
]
[
][
]
+
U
M
U
&p&B
M
KA

O
+
O
KB

{ }
[U ] = [
[U ]
B
{ }
21
pA
T
[
]
[
][
]
+
U
K
U
pB
= {0}
{ }
{ }
Dr. Peter Avitabile

Complex Mode Solution

The Structural Dynamic Modification process can also be
applied to systems with nonproportional modes.
[0]
[M ]
1
[M1 ] &x& [M1 ] [0] x& 0
=
[C1 ] x& [0] [ K1 ] x F
The equation of motion can be recast as
[B1 ]{Y& } [A1 ]{Y} = {Q}
[0] [M1 ]
[B1 ] =
[M1 ] [C1 ]
[M ] [0]
[A1 ] = 1
[0] [ K1 ]
The eigensolution and modal transformation is then
[[A1 ] [B1 ]]{Y} = {0}

{Y} = [1 ]{p1}
22
Dr. Peter Avitabile


Considering changes to the system as
[M 2 ] = [M1 ] + [M12 ]
[C2 ] = [C1 ] + [C12 ]
[K 2 ] = [K1 ] + [K12 ]
[M 2 ]{&x&} + [C2 ]{x& } + [K 2 ]{x} = {F( t )}
And expressing them in state space
[A 2 ] = [A1 ] + [A12 ]
[B2 ] = [B1 ] + [B12 ]
[B2 ]{Y& } [A 2 ]{Y} = {Q( t )}

[0]
[B12 ] =
[M12 ]
[M12 ]
[0]
[M12 ]
[
]
=
A
12
[0]
[C12 ]
[ K12 ]

23
Dr. Peter Avitabile


The eigensolution can be obtained
O
+ [B ]{p& }
+ [A ]{p} = {Q ( t )}
I
1
12
1
12
O
O
[A12 ] = [1 ]T [A12 ][1 ]
[B12 ] = [1 ]T [B12 ][1 ]
The Local Eigenvalue Modification Procedure can also be

applied to the complex mode solution

24
Dr. Peter Avitabile


These models are developed from characteristics
of the system response typically from frequency
response measurements:
ts
n
e
on
p
m
Co
ed
t
s
Te
r
ls
so
e
l
d
e
od
Mo
M
d
ed
ase
s
B
a
B
l
se
n
a
c
o
i
p
s
Res
Phy

25
Dr. Peter Avitabile

Impedance Modeling
Frequency Response Functions can also be used to investigate
structural modifications. The FRF can be written as
m
Hij ( j) =
q k u ik u jk
k =1 ( j p k )
q k u ik*u jk*
( j p k * )
Using force balance and compatibility equations, the effects

of a modification can be written in terms of the unmodified
system as
x a = H ab Fb + H aa Fa
Fa =
1
H aa
x
~
1
H cb = c = H cb H ca H aa
H ab
Fb
H ab Fb
x c = H ca Fa + H cb Fb

26
Dr. Peter Avitabile

Impedance Modeling Extended to System Modeling

A system model can be developed using the impedance
modeling approach
h Cij = h Aij H A iS ([H A ]SS + [H B ]SS )1{H A }Sj
FRFs
describing
connection
points
FRFs
describing
output response
points
FRFs
describing
input force
points
COMPONENT A
CONNECTION POINTS
COMPONENT B

27
Dr. Peter Avitabile

Truncation
Generally, the lower order modes are sufficient to describe
a structural dynamic problem
However, the SDM process
may require modes that are
not included in the frequency
range of response interest

28
Dr. Peter Avitabile

Truncation
A single structural change can have the effect of recoupling
all the uncoupled modal DOF. Truncation effects are quite
different for the modified and unmodified models

29
Dr. Peter Avitabile

Truncation
A free-free beam is subject to 2 changes in stiffness to
develop a simple support beam and a cantilever beam.
While one modification has accurate predictions with only 5
modes, the other modification does not

30
Dr. Peter Avitabile

Truncation
Simple support and cantilever modification
Modal transformation from modal space 1 to modal space 2

31
Dr. Peter Avitabile

Truncation
Simple support modification - Modal truncation is not a
problem since the available unmodified modes are adequate
to span the space of the problem
{u 2 }(i ) = {u1}(1) U12(1i) + {u1}( 2) U12( 2i ) + {u1}(3) U12(3i ) +L

32
Dr. Peter Avitabile

Truncation
Cantilever modification - Modal truncation is a problem since
the available unmodified modes are not adequate to span the
space of the problem
{u 2 }(i ) = {u1}(1) U12(1i) + {u1}( 2) U12( 2i ) + {u1}(3) U12(3i ) +L

33
Dr. Peter Avitabile

Truncation
All final modified system modes are NOT affected the same
Each mode is made from a different recipe
{u 2 }(i ) = {u1}(1) U12(1i) + {u1}( 2) U12( 2i) + {u1}(3) U12(3i ) +L

Just because one ingredient is missing doesnt
mean that you cant make any other recipes
34
Dr. Peter Avitabile

Rotational DOF
Adding additional Translation DOF (residual terms) does not
improve the modification - Extra Rotational DOF are needed
CANTILEVER BEAM
1
2
3
Ref.
Freq.
(Hz)
5 Modes
(1-5 TDOF)
(1-5 RDOF)
(Hz)
10-5 Modes
(1-10 TDOF)
(1-5 RDOF)
(Hz)
10 Modes
(1-10 TDOF)
(1-10 TDOF)
(Hz)
21.6
139.6
398.6
24.8
162.8
476.0
24.8
162.6
473.7
22.2
144.9
411.4

35
Dr. Peter Avitabile

Rotational DOF
Rotational DOF are needed for Impedance Methods also
CANTILEVER BEAM
TDOF
RDOF

36
Dr. Peter Avitabile

Rigid Body Modes

Many experimental modal test databases do not contain the
rigid body modes of the free-free system.
In any modifications that tie an unconstrained component to
ground or to another substructure, the rigid body modes
must be available as part of the modal data base.
If they are not available, then some other approximation
of them must be included or the modification process will
be missing a key ingredient.

37
Dr. Peter Avitabile

Complex vs Proportional Modes

A proportional based approximation is generally acceptable
if the original modal database is proportional and the
modifications to be studied do not significantly disrupt the
proportionality of the system.
However, if the starting modal database is complex or the
changes to be investigated will disrupt the proportionality
of the system, then a complex mode formulation is
recommended.

38
Dr. Peter Avitabile

TOP TEN SDM BLUNDERS

10 Geometry not accurate (rib modifications)
9
Calibration not performed
Weight vs mass density
Inconsistent units used (inch - feet)
Add stiffness between 2 dof that have no shape value
Add mass - modes shift up in frequency
Add stiffness - modes shift down in frequency
Forget to include rigid body modes for free-free models
Forget that only the connection DOF matter for SDM
and the #1 SDM blunder

TRUNCATION
39
Dr. Peter Avitabile

Summary
A brief review of the
Structural Dynamic
Modification process
was given to summarize
the past twenty years.
The paper has a
significant amount of
material that cannot be
covered in this short
presentation.
EXPERIMENTAL
MODAL
TESTING
FINITE
ELEMENT
MODELING
MODAL
PARAMETER
ESTIMATION
PERFORM
EIGEN
SOLUTION
MASS
DEVELOP
MODAL
MODEL
Repeat
until
desired
characteristics
are
obtained
RIB
STIFFNER
SPRING
STRUCTURAL
CHANGES
REQUIRED
Yes
USE SDM
TO EVALUATE
STRUCTURAL
CHANGES
No
DONE
DASHPOT
STRUCTURAL
DYNAMIC
MODIFICATIONS

40
Dr. Peter Avitabile

Test-Analysis
Correlation-Updating
Considerations
Peter Avitabile
Modal Analysis and Controls Laboratory
Test/Analysis Correlation/Updating Considerations
Dr. Peter Avitabile

The Overall Correlation and Updating Process

MAC AND ORTHOGONALITY
FINITE ELEMENT MODEL

1
0.9
1.2
0.8
0.7
0.6
0.8
0.5
0.4
0.6
0.3
0.4
0.2
0.2
0.1
MODEL
IMPROVEMENT
REGIONS
MAC
GUYAN
1.2
[U n ] , [ ]
+
g
[Tu ] = [Un ] [U a ]
[M] , [K]
RVAC
1.2
0.8
0.8
0.6
0.6
0.4
0.4
0.2
0.2
IRS
FRAC
SEREP
COMBINING ANALYTICAL AND EXPERIMENTAL DATA
FINITE ELEMENT
EXPERIMENTAL
DOF CORRELATION
FINITE ELEMENT
VECTOR CORRELATION
MAC
DOF CORRELATION
CoMAC
EXPERIMENTAL MODAL MODEL

[En ] = [T u ] [E a ]
MODE
SWITCHING
VECTOR CORRELATION
CORTHOG
Experimental Analytical
OR
1
FEM 5
0.8
FEM 4
0.6
0.4
0.2
POC
VECTOR CORRELATION
FEM 3
FEM 2
0
EXP1 EXP 2
FEM 1
EXP 3 EXP 4
EXP 5
DOF CORRELATION
FINITE ELEMENT
EXPERIMENTAL
EXPERIMENTAL
Dr. Peter Avitabile

Test-Analysis Correlation-Updating Considerations

Briefly describe the different correlation tools
available
Conceptually, overview the correlation process

Briefly overview the model updating process
A significant amount of effort is required to
completely describe all the techniques and tools

available
Dr. Peter Avitabile

Correlation Techniques
Correlation between analytical and experimental
data is an important part of the structural
dynamic characterization and updating of systems
FINITE ELEMENT
CoMAC
MAC
COORDINATE
MODAL
ASSURANCE
CRITERIA
MODE
SWITCHING
MODAL
ASSURANCE
CRITERIA
MATRIX
CORTHOG
COORDINATE
ORTHOGONALITY
CRITERIA
OR
OR
FEM 5
0.8
FEM 4
0.6
0.4
VECTOR CORRELATION
Experimental
FEM 3
0.2
Analytical
FEM 2
PSEUDO
ORTHOGONALITY
CRITERIA
MATRIX
EXP1
FEM 1
EXP 2
EXP 3
EXP 4
EXP 5
DOF CORRELATION
POC
RVAC
FINITE ELEMENT
EXPERIMENTAL
RESPONSE
VECTOR
ASSURANCE
CRITERIA
EXPERIMENTAL
FRAC
FREQUENCY
RESPONSE
ASSURANCE
CRITERIA
FINITE ELEMENT
EXPERIMENTAL
DOF CORRELATION
VECTOR CORRELATION
Dr. Peter Avitabile

Overview of Correlation Techniques

Vector correlation provides global indicator:
Modal Assurance Criteria
Orthogonality Checks
DOF correlation provides spatial indicator:

Coordinate Modal Assurance Criteria
Coordinate Orthogonality Check
Frequency Response Assurance Criteria
Other tools:
MAC Contribution
Force Unbalance
Dr. Peter Avitabile


Two basic levels of correlation are considered:
Modal vector correlation provides a
global indicator of the level of

correlation achieved
Degree of freedom (dof) correlation
provides an indicator as to how the

individual dofs contribute to the overall
modal vector correlation
Dr. Peter Avitabile


Vector Correlation Techniques:
Modal Assurance Criteria (MAC):
Simple dot product
independent of mass weighting
Orthogonality Checks (POC):
Performed at n space or a space

mass reduced for a space calc
shape expanded for n space calc
reduction/expansion has an effect
Dr. Peter Avitabile


DOF Correlation Techniques:
Coordinate Modal Assurance Criteria (CoMAC):
Simple dot product
correlation on dof basis for correlated
mode pairs
independent of mass weighting
Enhanced Coordinate Modal Assurance Criteria:

Extension of CoMAC
Dr. Peter Avitabile


DOF Correlation Techniques:
Frequency Response Assurance Criteria (FRAC):
simple dot product
correlation of FEM and Test FRFs
Coordinate Orthogonality Check (CORTHOG)

Identified correlation on a dof basis
mass matrix used for weighting
similar to CoMAC in concept except
correlated mode pairs not required
Dr. Peter Avitabile

Modal Assurance Criteria - MAC

Originally formulated for the test engineer to
determine the degree of correlation between
vectors from different tests, MAC between two
vectors is defined as:
(
{V } {V })
MAC =
({V } {V })({V } {V })
2
ij
values range between 0 and 1

approaching zero indicates no similarity
approaching one indicates high similarity
10
Dr. Peter Avitabile

Modal Assurance Criteria - MAC

MAC was extended to allow an assessment
between analytical and experimental modal
vectors:
MACij
MAC
[
{u } {e }]
=
[{u } {u }][{e } {e }]
2
MODAL
ASSURANCE
CRITERIA
MATRIX
FINITE ELEMENT
MODE
SWITCHING
VECTOR CORRELATION
low values - not similar

1
high values - very similar
FEM 5
0.8
FEM 4
0.6
0.4
FEM 3
0.2
FEM 2
0
EXP1
FEM 1
EXP 2
EXP 3
EXP 4
EXP 5
EXPERIMENTAL
11
Dr. Peter Avitabile

Orthogonality Check
For modal vectors scaled to unit modal mass, the
vectors must satisfy the orthogonality condition:
[ U ]T [M ] [ U ] = [I]
[ U ]T [K ] [ U] = [ 2 ]
12
Dr. Peter Avitabile

Pseudo Orthogonality Check - POC

The Pseudo Orthogonality Check relating the
correlation between the analytical and
experimental modal vectors with the analytical
mass matrix is
?
POC = [E ] [M ][U ] = [I]

T
1
0.9
0.8
0.7
1.2
1.2
0.8
0.8
0.8
0.6
0.6
0.6
0.4
0.4
0.4
0.2
0.2
0.2
1.2
0.6
0.5
0.4
0.3
0.2
0.1
0
MAC
GUYAN
IRS
SEREP
Typically, most people feel the smaller the POC

off-diagonal terms the better correlation that
exists. However, these terms may be small
and vectors may still be relatively uncorrelated
13
Dr. Peter Avitabile


The Pseudo Orthogonality Check is an assessment
as to how close the experimental vectors are
aligned with the analytical vectors
?
[E ] [M ] [ U ] = [I]
T
[E ] [K ] [ U ] = [ 2 ]
T
These equations can be evaluated at:

FEM Space - requires expansion
Reduced Space - requires reduction
Intermediate space - requires both
Substantial numerical advantages using SEREP!

14
Dr. Peter Avitabile

Expansion to
Full Space
may smear
and distort
mode shapes
Reduction to
Test Space
may result
in distorted
mass and
stiffness
matrices
15
Dr. Peter Avitabile

Cross Orthogonality Check

The Cross Orthogonality Check is also used for
correlation purposes
?
[E ] [M ] [E ] = [I]
T
[E ] [K ] [E ] = [ 2 ]
T
These equations can be evaluated at:

FEM Space - requires expansion
Reduced Space - requires reduction
Similar to POC (off-diagonal terms are squared)
16
Dr. Peter Avitabile

Coordinate Modal Assurance Criteria - CoMAC

The CoMAC gives an indication of the contribution
of each dof to the MAC for a given mode pair
(c)
(c)
u k e k
c=1
CoMAC(k ) =
(u ) (e )
c=1
Low values of CoMAC

indicate little correlation
whereas high values of
CoMAC indicate very
high correlation
(c) 2
k
c=1
CoMAC
COORDINATE
MODAL
ASSURANCE
CRITERIA
17
Experimental
Analytical
DOF CORRELATION
FINITE ELEMENT
(c) 2
k
EXPERIMENTAL
Dr. Peter Avitabile

Modulus Difference
The Modulus Difference was developed to
supplement the results from CoMAC
ModulusDifference( k ) = u (kc ) e (kc )
Assists in identifying
discrepancies
between analytical
and experimental
vectors
DOF CORRELATION
FINITE ELEMENT
18
EXPERIMENTAL
Dr. Peter Avitabile

Enhanced CoMAC - ECoMAC

The CoMAC gives an indication of the contribution
of each dof to the MAC for a given mode pair
m (c) (c)
u k e k
ECoMAC(k ) = c=1
2m
Low values of ECoMAC indicate high correlation

whereas high values of CoMAC indicate very low
correlation
Very sensitive to phasing of vectors - which
makes it more sensitive
19
Dr. Peter Avitabile

Frequency Response Assurance Criteria - FRAC

The FRAC is used to identify similarity between a
measured and analytical FRF - formed like MAC
{H( ) } {H( ) }
FRAC( j) =
({H( ) } {H( ) } ) ({H( ) } {H( ) } )
a
i j
a
i j
2
x *
i j
a *
i j
x
i j
Low values of FRAC

indicate little
correlation whereas
high values of FRAC
indicate very high
correlation
x *
i j
FRAC
FREQUENCY
RESPONSE
ASSURANCE
CRITERIA
FINITE ELEMENT
EXPERIMENTAL
DOF CORRELATION
20
Dr. Peter Avitabile

Response Vector Assurance Criteria - RVAC

The RVAC is used to identify the degree of
similarity that exists at a particular frequency
RVAC() = MAC({E test ()}, {U fem ()})
Low values of RVAC

indicate little
correlation whereas
high values of RVAC
indicate very high
correlation
RVAC
RESPONSE
VECTOR
ASSURANCE
CRITERIA
FINITE ELEMENT
EXPERIMENTAL
VECTOR CORRELATION
21
Dr. Peter Avitabile

Coordinate Orthogonality Check - CORTHOG

The Coordinate Orthogonality Check helps to
identify the contribution of individual dofs to
each of the off-diagonal terms of the POC
matrix
Identifies which dof are most discrepant between
the analytical and experimental vectors on a mass
weighted basis
POC
Orthogonality
ORTHOG ijk = u ki m kp u pj
POC ijk = e ki m kp u pj
22
Dr. Peter Avitabile

Coordinate Orthogonality Check - CORTHOG

The Coordinate Orthogonality Check is simply the
comparison of what should have been obtained
analytically for each dof in an orthogonality check
to what was actually obtained for each dof in a
pseudo-orthogonality check from test
SD = CORTHOG ijk = e ki m kp u pj u ki m kp u pj
p
Variety of different
formulations with
different scaling
approaches
-4
-3
-2
dof 3
23
emu
umu
dof 1
dof 2
-1
emu
Experimental
Analytical
umu
umu
emu
Dr. Peter Avitabile

MAC Contribution
The MAC Contribution is a relatively simple and
straightforward technique to determine the degree
of contribution of each dof to the MAC value
achieved
pick a mode pair of interest
select a target MAC value
delete dof until target MAC value achieved
24
Dr. Peter Avitabile

Force Unbalance
The Force Balance is a simple calculation to
determine the inequality that exists in the
equation of motion
?
[[K ] [M ]]{x}={0}
uses the FEM mass and stiffness matrices

uses experimental frequencies and mode shapes
compute the inequality that exists
25
Dr. Peter Avitabile

Model Updating Topics

Model Updating techniques can be broken down
into two categories:
Direct Techniques
Indirect Techniques (Sensitivity based)
Modal Based Techniques

Response Based Techniques
Some basic theory of analytical model
improvement and localization of model change are
described
26
Dr. Peter Avitabile

Model Improvement Terminology
27
Dr. Peter Avitabile

Analytical Model Improvement - AMI
28
Dr. Peter Avitabile

29
Dr. Peter Avitabile

30
Dr. Peter Avitabile

31
Dr. Peter Avitabile

Analytical Model Improvement - SSO
32
Dr. Peter Avitabile

Analytical Model Improvement - MSSO
33
Dr. Peter Avitabile

Model Updating - Sensitivity Approaches
34
Dr. Peter Avitabile

Comments on Direct Techniques

Direct Techniques
Usually a one step process that does not require iteration
to obtain a solution
Usually based on equation of motion and orthogonality
conditions
Exact results obtained (in the sense that the target modes
are reproduced
Generally updated matrices are difficult to interpret and
smearing of results occurs
Skyline approaches attempt to retain the original topology

of the system assembly
Reduction and expansion have a dramatic effect on results
35
Dr. Peter Avitabile

Comments on Direct Techniques

Skyline containment
Matrix smearing
36
Dr. Peter Avitabile

Model Updating - Sensitivity Approaches

Differences that are typically minimized:
Frequency differences
Mode shape differences
Frequency response differences
Parameters that may be updated:
mass/stiffness of individual elements

mass/stiffness of groups of elements
parameters associated with individual elements
parameters associated with groups of elements
37
Dr. Peter Avitabile

Comments on Indirect Techniques

Indirect Techniques - Sensitivity approach
Modal Based Techniques
Shape differences
Response differences
38
Dr. Peter Avitabile


Indirect Techniques -Sensitivity -Modal Approach
Likely to be the most accurate parameter
measured
No spatial information needed
Relatively simple calculations
No reduction/expansion problems
39
Dr. Peter Avitabile


Indirect Techniques -Sensitivity -Modal Approach
Shape differences
Less accurate on a dof basis

Spatial information included
Mode pairing necessary
Calculations more complicated
Reduction/expansion is a problem
40
Dr. Peter Avitabile


Indirect Techniques - Sensitivity approach
Response Based Techniques
Contains complete information in frequency range
No need to estimate modal parameters
FRFs are more accurate than modal parameters
Response may be item of interest
Damping may be difficult to determine
Selection of certain spectral lines may cause
numerical difficulties
Using only a few FRFs may distort the results
Difficult to identify parameters for change
Measured FRFs must be acquired with high accuracy
41
Dr. Peter Avitabile

General Comments
42
Dr. Peter Avitabile

General Comments
Use of all the correlation tools necessary to
interpret the data available
Both modal and response based techniques should be
used together for the updating
One technique alone may not be sufficient to
adequately update the model
Once updated, the model should be perturbed both

analytically and experimentally and the correlation
process repeated to assure that meaningful
parameters have been obtained from the updating
process
43
Dr. Peter Avitabile

General Comments
Model Updating requires extreme care in order to

obtain reliable results
A firm understanding of the modeling techniques
employed are necessary in order to adequately adjust
the finite element model
A thorough understanding of the experimental data
used for the updating process is critical
A clear definition of what is meant by an improved
model is necessary
The analyst has a tremendous responsibility in
identifying which areas of the model are to be
updated and which sets of modes are the best modes
to use in the updating process
44
Dr. Peter Avitabile

macl.caeds.eng.uml.edu
Dr. Peter Avitabile

Presentation Topics
Intent
Things Shake and Break !
Modal Overview
TUTORIAL NOTES:
Structural Dynamics and
Analytical Modeling
SDOF Theory
MODE 1
MDOF Theory
MODE3
DSP - DQAL Windows

MPE Concepts
MODE 2
MODE 4
Linear Algebra
Structural Modification
Correlation/Updating
Copyright 2000 All Rights Reserved
In Trouble !!!!!
Dr. Peter Avitabile
peter_avitabile@uml.edu
Tuesday, July 06, 2004

OSD

Caricato da

Informazioni sul documento

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

OSD

Caricato da

Copyright:

Formati disponibili

macl.caeds.eng.uml.

Dr. Peter Avitabile

DSP - DQAL Windows

Copyright 2000 All Rights Reserved

Tuesday, July 06, 2004

Intent of Structural Dynamics & Modal Overview

The intent of this

Intent of Modal Analysis Overview

Dr. Peter Avitabile

Structural Dynamic Modeling Techniques

Could you explain

Illustration by Mike Avitabile

Overview of Structural Dynamic Modeling Techniques

Illustration by Mike Avitabile

Illustration by Mike Avitabile

Dr. Peter Avitabile

Modal Analysis and Structural Dynamics

INPUT TIME FORCE

INPUT POWER SPECTRUM

Modal Analysis is the study of the dynamic character of a

Dr. Peter Avitabile

How Do Structures Respond Dynamically ?

high speed video

response due to a vertical bump

Dr. Peter Avitabile

Response of a Simple Plate

increasing rate of oscillation

Overview of Structural Dynamic Modeling Techniques

Dr. Peter Avitabile

Response of a Simple Plate

Measure many points on the

Overview of Structural Dynamic Modeling Techniques

Dr. Peter Avitabile

Response of a Simple Plate

Overview of Structural Dynamic Modeling Techniques

Dr. Peter Avitabile

Analytical Modal Analysis

[M n ]{&x& n } + [C n ]{x& n } + [K n ]{x n } = {Fn ( t )}

Overview of Structural Dynamic Modeling Techniques

[[K n ] [M n ]]{x n } = {0}

Dr. Peter Avitabile

Finite Element Models

Overview of Structural Dynamic Modeling Techniques

Dr. Peter Avitabile

Finite Element Models

Dr. Peter Avitabile

Experimental Modal Analysis

Overview of Structural Dynamic Modeling Techniques

Dr. Peter Avitabile

Experimental Modal Analysis

Overview of Structural Dynamic Modeling Techniques

Dr. Peter Avitabile

Experimental Data Reduction

Overview of Structural Dynamic Modeling Techniques

Dr. Peter Avitabile

What Are Measurements Called FRFs ?

Overview of Structural Dynamic Modeling Techniques

Dr. Peter Avitabile

Digital Signal Processing Flow Diagram

Actual time signals

Digitized time signals

COMPUTATION OF FRF AND COHERENCE

FREQUENCY RESPONSE FUNCTION

Average auto/cross spectra