The Fundamentals of

Modal Testing
Application Note 243 - 3

n φi φj /m
Η(ω) = Σ
r=1
(ω n - ω2)2 + (2ξωωn ) 2
2
Preface

Modal analysis is defined as the study

of the dynamic characteristics of a
mechanical structure. This applica-
tion note emphasizes experimental
modal techniques, specifically the
method known as frequency response
function testing. Other areas are
treated in a general sense to intro-
duce their elementary concepts and
relationships to one another.

Although modal techniques are math-

ematical in nature, the discussion is
inclined toward practical application.
Theory is presented as needed to
enhance the logical development of
ideas. The reader will gain a sound
physical understanding of modal
analysis and be able to carry out
an effective modal survey with
confidence.

Chapter 1 provides a brief overview

of structural dynamics theory.
Chapter 2 and 3 which is the bulk
of the note – describes the measure-
ment process for acquiring frequency
response data. Chapter 4 describes
the parameter estimation methods
for extracting modal properties.
Chapter 5 provides an overview
of analytical techniques of structural
analysis and their relation to
experimental modal testing.

2
Table of Contents

Preface 2

Chapter 1 — Structural Dynamics Background 4

Introduction 4
Structural Dynamics of a Single Degree of Freedom (SDOF) System 5
Presentation and Characteristics of Frequency Response Functions 6
Structural Dynamics for a Multiple Degree of Freedom (MDOF) System 9
Damping Mechanism and Damping Model 11
Frequency Response Function and Transfer Function Relationship 12
System Assumptions 13

Chapter 2 — Frequency Response Measurements 14

Introduction 14
General Test System Configurations 15
Supporting the Structure 16
Exciting the Structure 18
Shaker Testing 19
Impact Testing 22
Transduction 25
Measurement Interpretation 29

Chapter 3 — Improving Measurement Accuracy 30

Measurement Averaging 30
Windowing Time Data 31
Increasing Measurement Resolution 32
Complete Survey 34

Chapter 4 — Modal Parameter Estimation 38

Introduction 38
Modal Parameters 39
Curve Fitting Methods 40
Single Mode Methods 41
Concept of Residual Terms 43
Multiple Mode-Methods 45
Concept of Real and Complex Modes 47

Chapter 5 — Structural Analysis Methods 48

Introduction 48
Structural Modification 49
Finite Element Correlation 50
Substructure Coupling Analysis 52
Forced Response Simulation 53

Bibliography 54

3
Chapter 1
Structural Dynamics Background

Introduction Figure 1.1

Phases of a Test Structure
A basic understanding of structural modal test
dynamics is necessary for successful
modal testing. Specifically, it is
important to have a good grasp of the
relationships between frequency
response functions and their individ-
ual modal parameters illustrated in
Figure 1.1. This understanding is of
value in both the measurement and Frequency Response Measurements
analysis phases of the survey. Know-
ing the various forms and trends of
frequency response functions will
lead to more accuracy during the
measurement phase. During the
analysis phase, knowing how equa-
tions relate to frequency responses
leads to more accurate estimation of
modal parameters.
Curve Fit Representation
The basic equations and their various
forms will be presented conceptually φir φjr
n
to give insight into the relationships Ηij(ω) = Σ
r = 1 mr (ω - ω 2 + j2ζωωr)
2
between the dynamic characteristics r
of the structure and the correspond-
ing frequency response function
measurements. Although practical
systems are multiple degree of free-
dom (MDOF) and have some degree Modal Parameters
of nonlinearity, they can generally
be represented as a superposition ω — Frequency
of single degree of freedom (SDOF) ζ — Damping
{φ} — Mode Shape
linear models and will be developed
in this manner.

First, the basics of an SDOF linear

the trends and their usefulness in the
dynamic system are presented to gain
measurement process. Finally, these
insight into the single mode concepts
concepts are extended into MDOF
that are the basis of some parameter
systems, since this is the type of
estimation techniques. Second, the
behavior most physical structures
presentation and properties of vari-
exhibit. Also, useful concepts
ous forms of the frequency response
associated with damping mechanisms
function are examined to understand
and linear system assumptions
are discussed.

4
Structural Dynamics of a Single Figure 1.2
Degree of Freedom (SDOF) System SDOF discrete
parameter model
Although most physical structures are
continuous, their behavior can usual-
ly be represented by a discrete
Damper c
parameter model as illustrated in Spring k
Figure 1.2. The idealized elements
are called mass, spring, damper and
excitation. The first three elements
describe the physical system. Energy
is stored by the system in the mass Response Excitation
and the spring in the form of kinetic Displacement (x) Mass m Force (f)
and potential energy, respectively.
Energy enters the system through
excitation and is dissipated through
damping.

The idealized elements of the physi- Figure 1.3

cal system can be described by the Equation of .. .
mx + cx + kx = f(t)
equation of motion shown in Figure motion —
modal definitions c or ζ = c
1.3. This equation relates the effects ωn2 = k , 2ζωn =
m m 2km
of the mass, stiffness and damping in
a way that leads to the calculation of
natural frequency and damping factor
of the system. This computation is
often facilitated by the use of the def-
initions shown in Figure 1.3 that lead Figure 1.4
directly to the natural frequency and Complex roots s1,2 = -σ + jωd
damping factor. of SDOF
equation
σ — Damping Rate
The natural frequency, ω, is in units ωζ — Damped Natural Frequency
of radians per second (rad/s). The
typical units displayed on a digital
signal analyzer, however, are in Hertz
(Hz). The damping factor can also be
represented as a percent of critical Figure 1.5
damping – the damping level at which SDOF
the system experiences no oscillation. impulse
response/
e -σ t
This is the more common understand- free decay
ing of modal damping. Although there
are three distinct damping cases,
only the underdamped case (ζ< 1)
is important for structural dynamics
applications.

ωd

5
When there is no excitation, the roots Figure 1.6
of the equation are as shown in Frequency
Figure 1.4. Each root has two parts: response —
the real part or decay rate, which polar
coordinates
defines damping in the system and
the imaginary part, or oscillatory
rate, which defines the damped
natural frequency, wd. This free

Magnitude
vibration response is illustrated
in Figure 1.5. 1/m
H(ω) =
(ωn -ω ) + (2ζωωn ) 2
2 2 2

When excitation is applied, the equa-

tion of motion leads to the frequency
response of the system. The frequen-
cy response is a complex quantity
and contains both real and imaginary
parts (rectangular coordinates). It can
be presented in polar coordinates as
magnitude and phase, as well. ωd

Presentation and Characteristics of

Frequency Response Functions
Because it is a complex quantity, the
frequency response function cannot
be fully displayed on a single two-
dimensional plot. It can, however, be
presented in several formats, each of
Phase

which has its own uses. Although the

response variable for the previous 2ξωωn
θ(ω) = tan-1
discussion was displacement, it could ωn2 -ω2
also be velocity or acceleration.
Acceleration is currently the accepted
method of measuring modal
response.

One method of presenting the data

is to plot the polar coordinates, mag-
ωd
nitude and phase versus frequency
as illustrated in Figure 1.6. At reso-
nance, when ω = ωn, the magnitude
is a maximum and is limited only by
the amount of damping in the system.
The phase ranges from 0° to 180°
and the response lags the input by
90° at resonance.

6
Another method of presenting Figure 1.7
the data is to plot the rectangular Frequency
coordinates, the real part and the response —
rectangular
imaginary part versus frequency. coordinates
For a proportionally damped system,
the imaginary part is maximum at
resonance and the real part is 0, as
shown in Figure 1.7.

Real
A third method of presenting the
frequency response is to plot the real
ω2n - ω2
part versus the imaginary part. This is H(ω) =
often called a Nyquist plot or a vector (ωn -ω2)2 + (2ζωωn ) 2
2

response plot. This display empha-

sizes the area of frequency response
at resonance and traces out a circle,
as shown in Figure 1.8.

By plotting the magnitude in decibels

vs logarithmic (log) frequency, it is
possible to cover a wider frequency
range and conveniently display the
range of amplitude. This type of plot,
often known as a Bode plot, also
has some useful parameter character-
istics which are described in the
following plots.
Imaginary

When ω << ωn the frequency

response is approximately equal to
the asymptote shown in Figure 1.9. -2ξωωn
H(ω) =
This asymptote is called the stiffness (ωn -ω2)2 + (2ζωωn ) 2
2

line and has a slope of 0, 1 or 2 for

displacement, velocity and accelera-
tion responses, respectively. When
ω >> ωn the frequency response is
approximately equal to the asymptote
also shown in Figure 1.9. This asymp-
tote is called the mass line and has a
slope of -2, -1 or 0 for displacement Figure 1.8
velocity or acceleration responses, Nyquist plot
respectively. of frequency
response

θ(ω)

H(ω)

7
The various forms of frequency Figure 1.9
response function based on the Different forms
type of response variable are also of frequency
response
defined from a mechanical engineer-
ing viewpoint. They are somewhat
intuitive and do not necessarily corre-
spond to electrical analogies. These 1
≈
k

Displacement
forms are summarized in Table 1.1.

1
≈
ω2m

Table 1.1
Different forms
of frequency response

Definition Response Variable Frequency

Compliance X Displacement
F Force

Mobility V Velocity
F Force

Accelerance A Acceleration
F Force
Velocity

1
≈
ωm
ω
≈
k

Frequency

1
≈−
Acceleration

ω2
≈
k

Frequency

8
Structural Dynamics for a Multiple Figure 1.10
Degree of Freedom (MDOF) System MDOF discrete
parameter model
The extension of SDOF concepts to
a more general MDOF system, with k1 k2
c1 c2
n degrees of freedom, is a straightfor-
ward process. The physical system is
simply comprised of an interconnec- m1 m2
tion of idealized SDOF models, as
illustrated in Figure 1.10, and is
described by the matrix equations k3 k4
of motion as illustrated in Figure 1.11. c3 c4

The solution of the equation with no

m
excitation again leads to the modal 3
parameters (roots of the equation)
of the system. For the MDOF case,
however, a unique displacement
vector called the mode shape exists
for each distinct frequency and damp-
ing as illustrated in Figure 1.11. The
free vibration response is illustrated Figure 1.11
in Figure 1.12. Equations of .. .
[m]{x} + [c]{x} + [k]{x} = {f(t)}
motion —
modal definitions
The equations of motion for the {φ}r , r = 1, n modes
forced vibration case also lead to
frequency response of the system.
It can be written as a weighted
summation of SDOF systems shown
in Figure 1.13.

The weighting, often called the modal

participation factor, is a function of Figure 1.12
excitation and mode shape coeffi- MDOF impulse
cients at the input and output degrees response/
free decay
of freedom.
Amplitude

0.0 Sec 6.0

9
The participation factor identifies the Figure 1.13
amount each mode contributes to the MDOF frequency
total response at a particular point. response
An example with 3 degrees of free-
dom showing the individual modal
contributions is shown in Figure 1.14.

dB Magnitude
The frequency response of an MDOF
system can be presented in the same
forms as the SDOF case. There are
other definitional forms and proper-
ties of frequency response functions, n φi φj /m
such as a driving point measurement, Η(ω) = Σ
that are presented in the next chap-
r=1
(ω n - ω2)2 + (2ξωωn ) 2
2

ter. These are related to specific

0.0
locations of frequency response
ω1 ω2 ω3 Frequency
measurements and are introduced
when appropriate.

Figure 1.14
SDOF modal Mode 1
contributions Mode 3
Mode 2
dB Magnitude

0.0
ω1 ω2 ω3 Frequency

10
Damping Mechanism and Figure 1.15
Viscous damping .
Damping Model cx
energy dissipation
Damping exists in all vibratory
systems whenever there is energy
dissipation. This is true for mechani-
cal structures even though most are
inherently lightly damped. For free
vibration, the loss of energy from
x
damping in the system results in the
decay of the amplitude of motion.
In forced vibration, loss of energy is
balanced by the energy supplied by
excitation. In either situation, the X
effect of damping is to remove energy ∆E = πωceq X2
from the system.
Force vs Displacement

In previous mathematical formula-

tions the damping force was called
viscous, since it was proportional to
velocity. However, this does not
imply that the physical damping Figure 1.16
System Input Output
mechanism is viscous in nature. It is G(s)
block diagram Excitation Response
simply a modeling method and it is
important to note that the physical System
damping mechanism and the mathe-
matical model of that mechanism are
two distinctly different concepts.

Most structures exhibit one or more

forms of damping mechanisms, such
as coulomb or structural, which
result from looseness of joints, inter- Figure 1.17
nal strain and other complex causes. Definition of Output
Transfer Function =
transfer function Input
However, these mechanisms can be
modeled by an equivalent viscous G(s) = Y(s)
damping component. It can be X(s)
shown that only the viscous compo-
nent actually accounts for energy loss
from the system and the remaining
portion of the damping is due to non-
linearities that do not cause energy
dissipation. Therefore, only the
viscous term needs to be measured to
hysteresis loop in Figure 1.15. E is
characterize the system when using a
the energy dissipated per cycle of
linear model.
vibration, ceq is the equivalent vis-
cous damping coefficient and X is the
The equivalent viscous damping
amplitude of vibration. Note that the
coefficient is obtained from energy
criteria for equivalence are equal
considerations as illustrated in the
energy distribution per cycle and the
same relative amplitude.

11
Frequency Response Function and Figure 1.18
S-plane iω
Transfer Function Relationship
representation ωd
The transfer function is a mathemati-
ωn
cal model defining the input-output
relationship of a physical system.
Figure 1.16 shows a block diagram σ
σ
of a single input-output system.
System response (output) is caused
by system excitation (input). The s-plane
casual relationship is loosely defined
as shown in Figure 1.17. Mathemati-
cally, the transfer function is defined
as the Laplace transform of the out-
put divided by the Laplace transform This relationship can be further
of the input. explained by the modal test process.
The measurements taken during a
The frequency response function is modal test are frequency response
defined in a similar manner and is function measurements. The parame-
related to the transfer function. ter estimation routines are, in gener-
Mathematically, the frequency al, curve fits in the Laplace domain
response function is defined as the and result in the transfer functions.
Fourier transform of the output divid- The curve fit simply infers the loca-
ed by the Fourier transform of the tion of system poles in the s-plane
input. These terms are often used from the frequency response func-
interchangeably and are occasionally tions as illustrated in Figure 1.18. The
a source of confusion. frequency response is simply the
transfer function measured along the
jω axis as illustrated in Figure 1.19.

12
Figure 1.19 System Assumptions
3-D Laplace
representation
jω The structural dynamics background
theory and the modal parameter
Real Part
estimation theory are based on two
major assumptions:
σ
● The system is linear.
● The system is stationary.

There are, of course, a number of

other system assumptions such as
jω observability, stability, and physical
Imaginary Part realizability. However, these assump-
tions tend to be addressed in the
inherent properties of mechanical
σ systems. As such, they do not pres-
ent practical limitations when making
frequency response measurements as
do the assumptions of linearity
and stationarity.

jω
Magnitude

Phase

Transfer Function – surface

Frequency Response – dashed

13
Chapter 2
Frequency Response Measurements

Introduction Figure 2.1

System block
This chapter investigates the current Excitation H(ω) Response
diagram X(ω) Y(ω)
instrumentation and techniques
available for acquiring frequency
response measurements. The discus-
sion begins with the use of a dynamic
signal analyzer and associated periph-
erals for making these measurements.
The type of modal testing known as
Figure 2.2
the frequency response function Structure
method, which measures the input under test
excitation and output response simul-
taneously, as shown in the block dia-
gram in Figure 2.1, is examined. The
focus is on the use of one input force,
a technique commonly known as sin-
gle-point excitation, illustrated in Structure
Figure 2.2. By understanding this
technique, it is easy to expand to the
multiple input technique.

With a dynamic signal analyzer,

which is a Fourier transform-based Force Transducer
instrument, many types of excitation Shaker
sources can be implemented to meas-
ure a structure’s frequency response
function. In fact, virtually any physi-
ods, are also included. Techniques
cally realizable signal can be input
for improving the quality and
or measured. The selection and
accuracy of measurements are then
implementation of the more common
explored. These include processes
and useful types of signals for modal
such as averaging, windowing and
testing are discussed.
zooming, all of which reduce mea-
surement errors. Finally, a section
Transducer selection and mounting
on measurement interpretation is
methods for measuring these signals
included to aid in understanding the
along with system calibration meth-
complete measurement process.

14
General Test System Configurations Figure 2.3
General test
The basic test setup required for configuration Controller
making frequency response measure-
ments depends on a few major Analyzer
factors. These include the type of
structure to be tested and the level
of results desired. Other factors,
including the support fixture and Structure
the excitation mechanism, also affect
the amount of hardware needed to
perform the test. Figure 2.3 shows
a diagram of a basic test system
configuration.
Transducers
The heart of the test system is the Exciter
controller, or computer, which is the
operator’s communication link to the
analyzer. It can be configured with
various levels of memory, displays
and data storage. The modal analysis
software usually resides here, as well
as any additional analysis capabilities
such as structural modification and For making measurements on simple Transducers, along with a power
forced response. structures, the exciter mechanism supply for signal conditioning, are
can be as basic as an instrumented used to measure the desired force
The analyzer provides the data hammer. This mechanism requires and responses. The piezoelectric
acquisition and signal processing a minimum amount of hardware. types, which measure force and
operations. It can be configured with An electrodynamic shaker may be acceleration, are the most widely
several input channels, for force and needed for exciting more complicated used for modal testing. The power
response measurements, and with structures. This shaker system re- supply for signal conditioning may be
one or more excitation sources for quires a signal source, a power ampli- voltage or charge mode and is some-
driving shakers. Measurement func- fier and an attachment device. The times provided as a component of the
tions such as windowing, averaging signal source, as mentioned earlier, analyzer, so care should be taken in
and Fast Fourier Transforms (FFT) may be a component of the analyzer. setting up and matching this part of
computation are usually processed the test system.
within the analyzer.

15
Supporting The Structure Figure 2.4a
Example of
The first step in setting up a free support
structure for frequency response situation
measurements is to consider the fix-
turing mechanism necessary to obtain
the desired constraints (boundary
conditions). This is a key step in the Free
process as it affects the overall struc- Boundary
tural characteristics, particularly for
subsequent analyses such as structur-
al modification, finite element corre-
lation and substructure coupling.

Analytically, boundary conditions Figure 2.4b

can be specified in a completely free Example of
or completely constrained sense. In constrained
support
testing practice, however, it is gener- situation
ally not possible to fully achieve
these conditions. The free condition
means that the structure is, in effect,
floating in space with no attachments
to ground and exhibits rigid body
behavior at zero frequency. The Constrained
Boundary
airplane shown in Figure 2.4a is an
example of this free condition.
Physically, this is not realizable,
so the structure must be supported
in some manner. The constrained
condition implies that the motion,
(displacements/rotations) is set to
zero. However, in reality most struc-
tures exhibit some degree of flexibili- system is used, the rigid body fre- The implementation of a constrained
ty at the grounded connections. The quencies will be much lower than the system is much more difficult to
satellite dish in Figure 2.4b is an frequencies of the flexible modes and achieve in a test environment. To
example of this condition. thus have negligible effect. The rule begin with, the base to which the
of thumb for free supports is that the structure is attached will tend to have
In order to approximate the free sys- highest rigid body mode frequency some motion of its own. Therefore, it
tem, the structure can be suspended must be less than one tenth that of is not going to be purely grounded.
from very soft elastic cords or placed the first flexible mode. If this criteri- Also, the attachment points will have
on a very soft cushion. By doing this, on is met, rigid body modes will have some degree of flexibility due to the
the structure will be constrained to negligible effect on flexible modes. bolted, riveted or welded connec-
a degree and the rigid body modes Figure 2.5 shows a typical frequency tions. One possible remedy for these
will no longer have zero frequency. response measurement of this type problems is to measure the frequency
However, if a sufficiently soft support with nonzero rigid body modes.

16
response of the base at the attach- Figure 2.5 FREQ RESP
ment points over the frequency range Frequency 50.0
of interest. Then, verify that this response 1st Flexible Mode
of freely
response is significantly lower than suspended 10.0
/Div
the corresponding response of the system
structure, in which case it will have
a negligible effect. However, the
Rigid Body Mode
frequency response may not be mea- dB
surable, but can still influence the
test results.

There is not a best practical or

appropriate method for supporting
a structure for frequency response -30.0
testing. Each situation has its own FxdXY 0 Hz 1.5625k
characteristics. From a practical
standpoint, it would not be feasible
to support a large factory machine
weighing several tons in a free test
state. On the other hand, there may
be no convenient way to ground a
very small, lightweight device for the
constrained test state. A situation
could occur, with a satellite for exam-
ple, where the results of both tests
are desired. The free test is required
to analyze the satellite’s operating
environment in space. However, the
constrained test is also needed to
assess the launch environment
attached to the boost vehicle.
Another reason for choosing the
appropriate boundary conditions is
for finite element model correlation
or substructure coupling analyses. At
any rate, it is certainly important dur-
ing this phase of the test to ascertain
all the conditions in which the results
may be used.

17
Exciting the Structure random function requires a shaker processing equipment, characteristics
system for implementation. In gener- of the structure, general measure-
The next step in the measurement al, the reverse is also true. Choosing ment considerations and, of course,
process involves selecting an a hammer for the excitation system the excitation system.
excitation function (e.g., random dictates an impulsive type excitation
noise) along with an excitation sys- function. A full function dynamic signal analyz-
tem (e.g., a shaker) that best suits the er will have a signal source with a
application. The choice of excitation Excitation functions fall into four sufficient number of functions for
can make the difference between a general categories: steady-state, exciting the structure. With lower
good measurement and a poor one. random, periodic and transient. quality analyzers, it may be necessary
Excitation selection should be There are several papers that go into to obtain a signal source as a sepa-
approached from both the type of great detail examining the applica- rate part of the signal processing
function desired and the type of exci- tions of the most common excitation equipment. These sources often
tation system available because they functions. Table 2.1 summarizes the provide fixed sine and true random
are interrelated. The excitation func- basic characteristics of the ones that functions as signals; however, these
tion is the mathematical signal used are most useful for modal testing. may not be acceptable in applications
for the input. The excitation system True random, burst random and where high levels of accuracy are
is the physical mechanism used to impulse types are considered in the desired. The types of functions
prove the signal. Generally, the context of this note since they are the available have a significant influence
choice of the excitation function most widely implemented. The best on measurement quality.
dictates the choice of the excitation choice of excitation function depends
system, a true random or burst on several factors: available signal

Table 2.1
Excitation Periodic* Transient
functions in analyzer window in analyzer window
Sine True Pseudo Random Fast Impact Burst Burst
steady random random sine sine random
state
Minimze leakage No No Yes Yes Yes Yes Yes Yes
Signal to noise Very Fair Fair Fair High Low High Fair
high
RMS to peak ratio High Fair Fair Fair High Low High Fair
Test measurement time Very Good Very Fair Fair Very Very Very
long good good good good
Controlled frequency content Yes Yes* Yes* Yes* Yes* No Yes* Yes*
Controlled amplitude content Yes No Yes* No Yes* No Yes* No
Removes distortion No Yes No Yes No No No Yes
Characterize nonlinearity Yes No No No Yes No Yes No

* Requires additional equipment or special hardware

18
The dynamics of the structure cable and measuring the transients. Shaker Testing
are also important in choosing the Self-operating involves exciting the
excitation function. The level of structure through an actual operating The most useful shakers for modal
nonlinearities can be measured and load. This input cannot be measured testing are the electromagnetic
characterized effectively with sine in many cases, thus limiting its useful- shown in Fig. 2.6 (often called
sweeps or chirps, but a random func- ness. Shakers and impactors are the electrodynamic) and the electro
tion may be needed to estimate the most common and are discussed in hydraulic (or, hydraulic) types. With
best linearized model of a nonlinear more detail in the following sections. the electromagnetic shaker, (the more
system. The amount of damping and Another method of excitation mecha- common of the two), force is generat-
the density of the modes within the nism classification is to divide them ed by an alternating current that
structure can also dictate the use of into attached and nonattached drives a magnetic coil. The maximum
specific excitation functions. If devices. A shaker is an attached frequency limit varies from approxi-
modes are closely coupled and/or device, while an impactor is not, mately 5 kHz to 20 kHz depending
lightly damped, an excitation function (although it does make contact for a on the size; the smaller shakers
that can be implemented in a leakage- short period of time). having the higher operating range.
free manner (burst random for exam- The maximum force rating is also a
ple) is usually the most appropriate. function of the size of the shaker and
varies from approximately 2 lbf to
Excitation mechanisms fall into four 1000 lbf; the smaller the shaker, the
categories: shaker, impactor, step lower the force rating.
relaxation and self-operating. Step
relaxation involves preloading the With hydraulic shakers, force
structure with a measured force is generated through the use of
through a cable then releasing the hydraulics, which can provide much
higher force levels – some up to
several thousand pounds. The maxi-
mum frequency range is much lower
though – about 1 kHz and below. An
advantage of the hydraulic shaker is
its ability to apply a large static pre-
load to the structure. This is useful
for massive structures such as grind-
ing machines that operate under
relatively high preloads which may
alter their structural characteristics.

19
There are several potential problem Figure 2.6
areas to consider when using a Electrodynamic
shaker system for excitation. To shaker with
power amplifier
begin with, the shaker is physically and signal source
mounted to the structure via the
force transducer, thus creating the
possibility of altering the dynamics Power Amplifier
of the structure. With lightweight
structures, the mechanism used to
mount the load cell may add appre-
ciable mass to the structure. This
causes the force measured by the
load cell to be greater than the force
actually applied to the structure.
Figure 2.7 describes how this mass
loading alters the input force. Since
the extra mass is between the load
cell and the structure the load cell
senses this extra mass as part of
the structure.

Since the frequency response is a

single input function, the shaker
should transmit only one component
of force in line with the main axis of
the load cell. In practical situations, Figure 2.7
when a structure is displaced along Mass loading
a linear axis it also tends to rotate from shaker Ax
setup
about the other two axes. To mini-
mize the problem of forces being Structure
applied in other directions, the shaker
should be connected to the load
cell through a slender rod, called Fs
a stinger, to allow the structure to
move freely in the other directions. Loading
This rod, shown in Figure 2.8, has Mass
a strong axial stiffness, but weak
bending and shear stiffnesses. In Fm
effect, it acts like a truss member,
carrying only axial loads but no Load
moments or shear loads. Cell

F1

Fs = Fm - MmAx

20
The method of supporting the shaker Figure 2.8
is another factor that can affect the Stinger
force imparted to the structure. The attachment
to structure
main body of the shaker must be
isolated from the structure to prevent
any reaction forces from being trans-
mitted through the base of the shaker
back to the structure. This can be
accomplished by mounting the shaker
on a solid floor and suspending the
structure from above. The shaker
could also be supported on a mechan-
ically isolated foundation. Another
method is to suspend the shaker, in
which case an inertial mass usually
needs to be attached to the shaker
body in order to generate a measura-
ble force, particularly at lower
frequencies. Figure 2.9 illustrates Figure 2.9
the different types of shaker setups. Test support
mechanisms
Another potential problem associated
with electromagnetic shakers is the
impedance mismatch that can exist
between the structure and the shaker
coil. The electrical impedance of the
shaker varies with the amplitude of
motion of the coil. At a resonance
with a small effective mass, very
little force is required to produce a
response. This can result in a drop
in the force spectrum in the vicinity
of the resonance, causing the force
measurement to be susceptible to
noise. Figure 2.10 illustrates an
example of this phenomenon. The
problem can usually be corrected by Figure 2.10 FREQ RESP 30 Avg 50%Ovlp Hann
using shakers with different size coils Shaker/structure 40.0
or driving the shaker with a constant- impedance
current type amplifier. The shaker mismatch dB

could also be moved to a point with a

larger effective mass. FRF
-40.0
Fxd Y 1.1k Hz 2.1k
POWER SPEC1 30 Avg 50%Ovlp Hann
-20.0

dB
Force Spectrum
rms
V2

-100
Fxd Y 1.1k Hz 2.1k

21
Impact Testing Figure 2.11
Impact devices
Another common excitation mecha- for testing
nism in modal testing is an impact
device. Although it is a relatively
simple technique to implement, it’s
difficult to obtain consistent results.
The convenience of this technique
is attractive because it requires very
little hardware and provides shorter
measurement times. The method of
applying the impulse, shown in Figure
2.11, includes a hammer, an electric
gun or a suspended mass. The ham-
mer, the most common of these, is
used in the following discussion.
However, this information also
applies to the other types of
impact devices.
Figure 2.12 0%Ovlp
FILT TIIME1
Frequency
Since the force is an impulse, the content of
350
Hard
m
amplitude level of the energy applied various pulses
50.0
to the structure is a function of the m
mass and the velocity of the hammer. /Div
Medium
This is due to the concept of linear
momentum, which is defined as mass
Real
times velocity. The linear impulse is
equal to the incremental change in Soft
v
the linear momentum. It is difficult
though to control the velocity of the
hammer, so the force level is usually Pulse t
controlled by varying the mass. -50.0
m
Impact hammers are available in
Fxd Y -46.9µ Sec 15.9m
weights varying from a few ounces to
1Avg 0%Ovlp Fr/Ex
several pounds. Also, mass can be POWER SPEC1
-50.0
added to or removed from most ham-
mers, making them useful for testing
10.0
objects of varying sizes and weights. /Div

Hard
The frequency content of the energy
applied to the structure is a function Medium
dB
of the stiffness of the contacting sur-
Soft
faces and, to a lesser extent, the mass rms
of the hammer. The stiffness of the v2

contacting surfaces affects the shape

of the force pulse, which in turn
determines the frequency content.
-130
Fxd X 0 Hz 2.5k

22
It is not feasible to change the Figure 2.13 1Avg 0%Ovlp Fr/Ex
POWER SPEC1
stiffness of the test object, therefore Useful frequency -50.0
the frequency content is controlled by range of pulse
spectrum
varying the stiffness of the hammer 10.0
/Div -72 dB
tip. The harder the tip, the shorter
the pulse duration and thus the high-
er the frequency content. Figure 2.12
-88 dB
illustrates this effect on the force dB
spectrum. The rule of thumb is to
choose a tip so that the amplitude rms
v2
of the force spectrum is no more
than 10 dB to 20 dB down at the
maximum frequency of interest as
shown in Figure 2.13. A disadvantage -130
to note here is that the force spec- Fxd Y 0 Hz 2.5k
trum of an impact excitation cannot
be band-limited at lower frequencies
when making zoom measurements,
so the lower out-of-band modes will
still be excited.

Impact testing has two potential Figure 2.14 0%Ovlp

FILT TIIME1
Force pulse
signal processing problems associat- with force
120
m
ed with it. The first – noise – can window applied
20.0
be present in either the force or m
response signal as a result of a long /Div

time record. The second – leakage –

can be present in the response signal
Real
as a result of a short time record.
Compensation for both these prob- v
lems can be accomplished with
windowing techniques.

Since the force pulse is usually very -40.0

m
short relative to the length of the time
record, the portion of the signal after
Fxd X 250µ τ Sec 100m
Force Pulse
the pulse is noise and can be elimi-
Ampl
nated without affecting the pulse
itself. The window designed to 1.0
accomplish this, called a force
window, is shown in Figure 2.14.
The small amount of oscillation that
occurs at the end of the pulse is
actually part of the pulse. It is a
result of signal processing and
should not be truncated.

Time
Force Window

23
The response signal is an exponential Figure 2.15 No Data
FILT TIIME2
decaying function and may decay out Decaying 250
before or after the end of the meas- response with m
exponential
urement. If the structure is heavily window applied 62.5 Response
m
damped, the response may decay out /Div
before the end of the time record. In
this case, the response window can
be used to eliminate the remaining Real
noise in the time record. If the struc-
ture is lightly damped, the response v

may continue beyond the end of the

time record. In this case, it must be
artificially forced to decay out to min-
-250
imize leakage. The window designed m
to accomplish either result, called Fxd Y 0.0 Sec 512m
the exponential window, is shown in Exponential Decaying Response
Figure 2.15. The rule of thumb for Ampl
setting the time constant, (the time
required for the amplitude to be
reduced by a factor of 1/e), is about
one-fourth the time record length, T.
The result of this is shown in Exponential Window
Figure 2.15.

Unlike the force window, the expo-

nential window can alter the resulting
frequency response because it has
the effect of adding artificial damping
to the system. The added damping
coefficient can usually be backed Time
out of the measurement after signal 0% Ovlp
WIND TIIME2
processing, but numerical problems 250
may arise with lightly damped struc- m

tures. This can happen when the 62.5

m
added damping from the exponential /Div
window is significantly more than Windowed Response
the true damping in the structure. A
better measurement procedure in this Real
case would be to zoom, thus utilizing
a longer time record in order to v

capture the entire response, instead

of relying on the exponential window.
-250
m
Fxd Y 0.0 Sec 512m

24
Transduction Figure 2.16
Frequency
Now that an excitation system has response of
been set up to force the structure into transducer
motion, the transducers for sensing
force and motion need to be selected.
Although there are various types of
transducers, the piezoelectric type
is the most widely used for modal
testing. It has wide frequency and
dynamic ranges, good linearity and is
relatively durable. The piezoelectric
transducer is an electromechanical
sensor that generates an electrical
output when subjected to vibration.
This is accomplished with a crystal
element that creates an electrical Response Transducer (Accelerometer)
charge when mechanically strained.

The mechanism of the force transduc-

er, called a load cell, functions in a
fairly simple manner. When the crys-
tal element is strained, either by ten-
sion or compression, it generates a Useable Frequency Range
charge proportional to the applied
dB Mag

force. In this case, the applied force

is from the shaker. However, due to
mounting methods discussed earlier,
this is not necessarily the force trans-
mitted to the structure.

The mechanism of the response

transducer, called an accelerometer,
functions in a similar manner. When
the accelerometer vibrates, an inter- Frequency
nal mass in the assembly applies a
force to the crystal element which is
proportional to the acceleration. This
relationship is simply Newton’s Law:
force equals mass times acceleration. Figure 2.17
Stud mounted
The properties to consider in load cell
selecting a load cell include both the
type of force sensor and its perfor-
mance characteristics. The type of
force sensing for which load cells are
designed include compression, ten-
sion, impact or some combination

25
thereof. Most shaker tests require at Figure 2.18
least a compression and tension type. Frequency 20k-25k
A hammer test, for example, would response of
transducer
require an impact transducer.

Some of the operating specifications

Useable Frequency Range
to consider are sensitivity, resonant
frequency, temperature range and
shock rating. Sensitivity is measured
in terms of voltage/force with units
of mV/1b or mV/N. Analyzers have
a range of input voltage settings;
therefore, sensitivity should be
chosen along with a power supply
amplification level to generate a
measurable voltage. Stud

The resonant frequency of a load cell

is simply a function of its physical
mass and stiffness characteristics.
The frequency range of the test
should fall within the linear range 15k-20k
below the resonant peak of the fre-
quency response of the load cell, as
shown in Figure 2.16. The rule of Useable Frequency Range
thumb for shock rating is that the
maximum vibration level expected
during the test should not exceed one
third the shock rating.

The load cell should be mounted to

the structure with a threaded stud for
best results as shown in Figure 2.17.
If this is not feasible, then an alterna- Cement, Wax
tive method of first fixing a spacer to
the structure with some type of adhe-
sive (such as dental cement) and then
stud mounting the load cell to this 5k-7k
spacer will usually suffice for low
force levels.

The properties to consider in select- Useable Frequency Range

ing an accelerometer are very similar
to those of the load cell, although
they are related to acceleration rather
than force. The type of response is
limited to acceleration as the term
implies, since displacement and
velocity transducers are not available

Magnet

26
Figure 2.19 3Avg 0%Ovlp FR/Ex
FREQ RESP
Mass loading
50.0
from Frequency
accelerometer
10.0 Amplitude
/Div

dB

-30.0
Fxd Y 0 Hz 2.5k

in the piezoelectric type. However, if Another important consideration is

displacement or velocity responses the effect of mass loading from the
are desired, the acceleration response accelerometer. This occurs as a
can be artificially integrated once or result of the mass of the accelerome-
twice to give velocity and displace- ter being a significant fraction of the
ment responses, respectively. effective mass of a particular mode.
A simple procedure to determine if
In general, the optimum acceler- this loading is significant can be
ometer has high sensitivity, wide done as follows:
frequency range and small mass.
Trade-offs are usually made since ● Measure a typical frequency
high sensitivity usually dictates a response function of the test object
larger mass for all but the most using the desired accelerometer.
expensive accelerometers. The ● Mount another accelerometer
sensitivity, measured in mV/G, and the (in addition to the first) with the
shock rating should be selected in the same mass at the same point and
same manner as with the load cell. repeat the measurement.
● Compare the two measurements and
Although the resonant frequency of
look for frequency shifts and ampli-
the accelerometer (freely suspended)
tude changes.
is a function of its mass and stiffness
characteristics, the actual natural If the two measurements differ signif-
frequency (when mounted) is general- icantly, as illustrated in Figure 2.19,
ly dictated by the stiffness of the then mass loading is a problem and
mounting method used. The effect of an accelerometer with less mass
various mounting methods is shown should be used. On very small struc-
in Figure 2.18. The rule of thumb is tures, it may be necessary to measure
to set the maximum frequency of the the response with a non-contacting
test at no more than one-tenth the transducer, such as an acoustical or
mounted natural frequency of the optical sensor, in order to eliminate
accelerometer. This is within the any mass loading.
linear range of the mounted frequen-
cy response of the accelerometer.

27
Figure 2.20
Example of
the input
half ranging
INST TIME2 COHERENCE 50 Avg 50% Ovlp Hann
80.0 1.1
m
Mag
20.0
m
/Div

0.0

Real Fxd Y 1.1k Hz 2.1k

FREQ RESP 50 Avg 50% Ovlp Hann
40.0
v
dB

-80.0
m -40.0
Fxd XY 0.0 Sec 8.0m Fxd XY 1.1k Hz 2.1k

Figure 2.21
Example of
the input
under ranging
INST TIME2 COHERENCE 50 Avg 50% Ovlp Hann
130 1.1
m
Mag
32.5
m
/Div

0.0

Real Fxd Y 1.1k Hz 2.1k

FREQ RESP 50 Avg 50% Ovlp Hann
40.0
v
dB

-130
m -40.0
Fxd XY 0.0 Sec 8.0m Fxd XY 1.1k Hz 2.1k

Figure 2.22
Example of
the input
over ranging
INST TIME2 Ov2 COHERENCE 50 Avg 50% Ovlp Hann Ov2
80.0 1.1
m
Mag
20.0
m
/Div

0.0

Real Fxd Y 1.1k Hz 2.1k

FREQ RESP 50 Avg 50% Ovlp Hann Ov2
40.0
v
dB

-80.0
m -40.0
Fxd X 0.0 Sec 8.0m Fxd XY 1.1k Hz 2.1k

28
Measurement Interpretation Figure 2.23 FILT TIME 1 50%Ovlp
Random 10.0
Having discussed the mechanics of test signals m Random Noise
Real
setting up a modal test, it is appropri-
ate at this point to make some trial v
measurements and examine their
trends before proceeding with data -10.0
m
collection. Taking the time to investi- Fxd Y 0.0 Sec 799m
gate preliminaries of the test, such HIST 1 90 Avg 50%Ovlp
10.0
as exciter or response locations, k Histogram
various types of excitation functions Real

and different signal processing

parameters will lead to higher quality
measurements. This section includes 0.0
preliminary checks such as adequate Fxd XY -10.0m V 10.0m
signal levels, minimum leakage
measurements and linearity and
reciprocity checks. The concept and
trends of the driving point measure- Figure 2.24 FILT TIME 1 0%Ovlp
ment and the combinations of mea- Transient 400
test signals m
Typical Force
surements that constitute a complete Real
modal survey are discussed.
v

After the structure has been support-

-400
ed and instrumented for the test, m
Fxd Y 0.0 Sec 512m
the time domain signals should be FILT TIME 2 0%Ovlp
examined before making measure- 700
Exponential Decaying
m
ments. The input range settings on Real
the analyzer should be set at no more
than two times the maximum signal v
level as shown in Figure 2.20. Often -700 Response
called half-ranging, this takes advan- m
Fxd Y 0.0 Sec 512m
tage of the dynamic range of the
analog-to-digital converter without
underranging or overranging the
signals.
It is also advisable to verify that the
The effect resulting from under-
signals are indeed the type expected,
ranging a signal, where the response
(e.g., random noise). With a random
input level is severely low relative to
signal, it is advisable to measure the
the analyzer setting, is illustrated in
histogram to verify that it is not
Figure 2.21. Notice the apparent
contaminated with other signal
noise between the peaks in the fre-
components, i.e., it has a Gaussian
quency response and the resulting
distribution as shown in Figure 2.23.
poor coherence function. In
This can be visually checked as
Figure 2.22, the response is severely
illustrated with the transient signals
overloading the analyzer input sec-
in Figure 2.24.
tion and is being clipped. This results
in poor frequency response and,
consequently, poor coherence since
the actual response is not being
measured correctly.

29
Chapter 3
Improving Measurement Accuracy

Measurement Averaging the frequency response. The circle

appears very distorted for a measure-
In order to reduce the statistical ment with few averages, but begins
variance of a measurement with a to smooth out with more and more
random excitation function (such as averages. This process can be seen
random noise) and also reduce the in Figure 3.1. With each data record
effects of nonlinearities, it is neces- acquired, the frequency spectrum has
sary to employ an averaging process. a different magnitude and phase
By averaging several time records distribution. As these spectra are
together, statistical reliability can averaged, the nonlinear terms tend
be increased and random noise asso- to cancel, thus resulting in the best
ciated with nonlinearities can be linear estimate.
reduced. One method to gain insight
into the variance of a measurement
is to observe the Nyquist display of

Figure 3.1
Measurement
averaging
frequency
response

COHERENCE 5 Avg 50% Ovlp Hann FREQ RESP 5 Avg 50% Ovlp Hann
1.1 100

Mag

0.0
Fxd Y 1.1k Hz 2.1k Imag
FREQ RESP 5 Avg 50% Ovlp Hann
40.0

dB

-40.0 -100
Fxd Y 1.1k Hz 2.1k Fxd Y -128 Real 128
5 Averages
COHERENCE 30 Avg 50% Ovlp Hann FREQ RESP 30 Avg 50% Ovlp Hann
1.1 100

Mag

0.0
Fxd Y 1.1k Hz 2.1k Imag
FREQ RESP 30 Avg 50% Ovlp Hann
40.0

dB

-40.0 -100
Fxd Y 1.1k Hz 2.1k Fxd XY -128 Real 128

30 Averages

30
Windowing Time Data Figure 3.2 FILT TIME2 50%Ovlp
Frequency 50.0
There is a property of the Fast response with m
Fourier transform (FFT) that affects true random
12.5
the energy distribution in the frequen- signal and m
no windows /Div
cy spectrum. It is the result of the
physical limitation of measuring a
finite length time record along with
Real
the periodicity assumption required
of the time record by the FFT. This v
does not present a problem when the
signal is exactly periodic in the time
record or when a transient signal is
completely captured within the time -50.0
m
record. However, in the case of true Fxd Y 0.0 Sec 799m
random excitation or in the transient True Random Noise
case when the entire response is not COHERENCE 50 Avg 50% Ovlp Unif
captured, a phenomenon called leak- 1.1
age results. This has the effect of Mag
smearing or leaking energy into adja-
cent frequency lines of the spectrum,
thus distorting it. Figure 3.2 illus-
trates an example of the effects of 0.0
Fxd Y 1.1k Hz 2.1k
severe leakage problems with true FREQ RESP 50 Avg 50% Ovlp Unif
random excitation. The effect is to 40.0
underestimate the amplitude and
dB
overestimate the damping factor.

One of the most common techniques

for reducing the effects of leakage -40.0

with a non-periodic signal is to artifi- Fxd Y 1.1k Hz 2.1k

cially force the signal to 0 at the

beginning and end of the time record
to make it appear periodic to the
analyzer. This is accomplished by Figure 3.3
COHERENCE 50 Avg 50% Ovlp Hann
multiplying the time record by a Frequency 1.1
mathematical curve, known as a win- response with
dow function, before processing the burst random Mag
signal
FFT. Another measurement is taken
with a Hann window applied to the
true random excitation signal, shown 0.0
in Figure 3.3. This measurement is Fxd Y 1.1k Hz 2.1k
FREQ RESP 50 Avg 50% Ovlp Hann
more accurate, but notice that the 40.0
coherence is still less than unity at
dB
the resonance. The window does not

-40.0
Fxd Y 1.1k Hz 2.1k

31
eliminate leakage completely and it Figure 3.4 FILT TIME2 0%Ovlp
also distorts the measurement as a Frequency 50.0
result of eliminating some data. A response with m
true random
better measurement technique is to signal and
12.5
m
use an excitation that is periodic Hann window /Div
within the time record such as burst
random, in order to eliminate the
leakage problem as illustrated in Real
Figure 3.4.
v

Increasing Measurement Resolution

Another measurement capability
that is often needed, particularly -50.0
m
for lightly damped structures, is to Fxd Y 0.0 Sec 799m
obtain more frequency resolution in Burst Random Signal
the vicinity of resonance peaks. It
COHERENCE 15 Avg 0% Ovlp Unif
may not be possible in a baseband 1.1
measurement to extract valid modal
Mag
parameters with inadequate informa-
tion. Normally, the Fourier transform
is calculated over a frequency range
from 0 to some maximum frequency. 0.0

Zoom processing is a technique in Fxd Y 1.1k Hz 2.1k

FREQ RESP 15 Avg 0% Ovlp Unif
which the lower and upper frequency 40.0
limits are independently selectable
dB
over fixed ranges within the analyzer.
The capability to zoom allows closely
spaced modes to be more accurately
identified by concentrating the mea- -40.0

surement points over a narrower Fxd Y 1.1k Hz 2.1k

band. The result of this increased

measurement accuracy is shown in
Figure 3.5. Another result is that
distortion due to leakage is reduced,
because the smearing of energy
is now within a narrower bandwidth,
but not eliminated. Another related
process associated with zooming is
the ability to band-limit the excitation
to concentrate the available energy
within the given frequency
range of the test.

32
 Figure 3.5
Effects of
increasing
frequency
resolution

COHERENCE 50 Avg 0% Ovlp Hann FREQ RESP 50 Avg 0% Ovlp Hann

1.1 200

Mag

0.0
Fxd Y 0 Hz 4k Imag
FREQ RESP 50 Avg 0% Ovlp Hann
50.0

dB

-30.0 -200
Fxd Y 0 Hz 4k Fxd Y -257 Real 257
Baseband Measurement
COHERENCE 50 Avg 5% Ovlp Hann FREQ RESP 50 Avg 5% Ovlp Hann
1.1 100

Mag

0.0
Fxd Y 745 Hz 2.345k Imag
FREQ RESP 50 Avg 5% Ovlp Hann
40.0

dB

-40.0 -100
Fxd Y 745 Hz 2.345k Fxd XY -128 Real 128

1st Zoom Span

COHERENCE 50 Avg 50% Ovlp Hann FREQ RESP 50 Avg 50% Ovlp Hann
1.1 100

Mag

0.0
Fxd Y 1.2k Hz 1.6k Imag
FREQ RESP 50 Avg 50% Ovlp Hann
40.0

dB

-40.0 -100
Fxd Y 1.2k Hz 1.6k Fxd Y -128 Real 128

2nd Zoom Span

33
Complete Survey Figure 3.6
Frequency
As frequency response functions are response
being acquired and stored for subse- matrix H11 H12 H1N
quent modal parameter estimation,
an adequate set of measurements
must be collected in order to arrive
at a complete set of modal parame- H21 H22
ters. This section describes the
number and type of measurements
that constitute complete modal sur-
vey. Definitions and concepts, such
as driving point measurement and
a row or column of the frequency
response matrix are discussed.
Optimal shaker and accelerometer
locations are also included. HN1 HNN

A complete, although redundant, set

of frequency response measurements
would form a square matrix of size N,
where the row corresponds to Figure 3.7
Driving point
response points and the columns measurement
correspond to excitation points, as setup
illustrated in Figure 3.6. It can be
shown, however, that any particular
row or column contains sufficient
information to compute the complete
set of frequencies, damping, and
mode shapes. In other words, if
the excitation is at point 3, and the
response is measured at all the
points, including point 3, then column
of the frequency response matrix will
be measured. This situation would be
the result of a shaker test. On the
other hand, if an accelerometer is
attached to point 7, and a hammer is
used to excite the structure at all
points, including point 7, then row 7
on the matrix will be measured. This
would be the result of an impact test. form the diagonal of the frequency and stiffness as well. It is not neces-
response matrix shown above. They sary to make a driving point measure-
The measurement where the also exhibit unique characteristics ment to obtain only frequencies,
response point and direction are the that are not only useful for checking damping factors and unscaled mode
same as the excitation point and measurement quality, but necessary shapes. However, a set of scaled
direction is called a driving point for accomplishing a comprehensive mode shapes and consequently,
measurement. The beam in Figure modal analysis, which includes not modal mass and stiffness cannot be
3.7 illustrates a measurement of this only frequencies, damping factors and extracted from a set of measurements
type. Driving point measurements scaled mode shapes, but modal mass that does not contain a driving point.

34
Recall from Chapter 1, Structural Figure 3.8 FREQ RESP 3 Avg 0% Ovlp Fr/Ex
Dynamics Background, that the Driving point 56.0
response of a MDOF system is simply frequency
response Real
the weighted sum of a number of
SDOF systems. The characteristics
of the driving point measurement
can be easily explained and presented -56.0

as a consequence of this property. Fxd Y 0 Hz 1.25k

FREQ RESP 3 Avg 0% Ovlp Fr/Ex
Figure 3.8 shows a typical driving 100
point measurement displayed in
Imag
rectangular and polar coordinates.
As seen in the imaginary part of the
rectangular coordinates, all of the
resonant peaks lie in the same direc- -100

tion. In other words, they are in Fxd Y 0 Hz 1.25k

phase with each other. This charac-

teristic becomes more intuitive when FREQ RESP 3 Avg 0% Ovlp Fr/Ex
illustrated with the beam modes in 40.0

Figure 3.9. The response point moves dB

in the same direction as the excita-
tion point at all the modes, since it
is measured at the same physical
-40.0
location as the excitation.
Fxd Y 0 Hz 1.25k
FREQ RESP 3 Avg 0% Ovlp Fr/Ex
By observing the trends of this 180

measurement in polar coordinates Phase

in Figure 3.9, a further understanding
of its characteristics can be gained.
When the magnitude is displayed in
-180
log format (dB), anti-resonances
Fxd Y 0 Hz 1.25k
occur between every resonance
throughout the frequency range. The
individual SDOF systems sum to 0 at
the frequencies where the mass and
stiffness lines of adjacent modes
intersect since all the modes are in
phase with each other. This results Figure 3.9
in the near 0 magnitude of an anti- Typical
resonance. Also notice the phase free beam
lead as the magnitude passes through mode shapes
an anti-resonance and the opposite
phase lag as the magnitude passes
through a resonance. These trends
of the driving point measurement
should be observed and monitored
throughout the measurement process
as a check for maintaining a
consistent set of data.

35
The remaining measurements, where Figure 3.10 FREQ RESP 3 Avg 0% Ovlp Fr/Ex
the response coordinates are differ- Cross-point 56.0
ent from the excitation coordinates, frequency
response Real
are called cross-point measurements.
Figure 3.10 illustrates a typical cross-
point measurement. All the modes
are not necessarily in phase with -56.0

each other, as seen in the imaginary Fxd Y 0 Hz 1.25k

FREQ RESP 3 Avg 0% Ovlp Fr/Ex
display. Since the response points 100
are not at the same location as the
Imag
excitation point, the response can
move either in phase or out of phase
with the excitation. This motion,
which defines the mode shape, -100

is a function of the measurement Fxd Y 0 Hz 1.25k

location, and will vary from measure-

ment to measurement. In the dB FREQ RESP 3 Avg 0% Ovlp Fr/Ex
display, if any two adjacent modes 40.0

are in phase at a particular point, dB

then an anti-resonance will exist
between them. If any two adjacent
modes are out of phase, then their
-40.0
mass and stiffness lines will not
Fxd Y 0 Hz 1.25k
cancel at the intersection and a FREQ RESP 3 Avg 0% Ovlp Fr/Ex
smooth curve will appear instead, 180
as seen in Figure 3.10. Phase

In order to excite all the modes

within the frequency range of inter-
-180
est, several shaker or accelerometer
Fxd Y 0 Hz 1.25k
locations should be examined. A
point or line on the structure that
remains stationary is called a node
point or node line. The node points
of a cantilever beam are illustrated in
Figure 3.11. The number and location
of these nodes are a function of the Figure 3.11
particular mode of vibration and Typical
increase as mode number increases. cantilever
beam mode
shapes

3 2 1

36
If the response is measured at the Figure 3.12 FREQ RESP 3 Avg 0% Ovlp Fr/Ex
end of the beam at point 1 and Frequency 100
excitation is applied at point 3, all response with
excitation at Imag
modes will be excited and the result- node of
ing frequency response will contain vibration
all the modes as shown in Figure
3.10. However, if the response is at -100
Fxd Y 0 Hz 1.25k
point 1 but the excitation is moved to FREQ RESP 3 Avg 0% Ovlp Fr/Ex
point 2, the second mode will not be 40.0
excited and the resulting frequency
dB
response will appear as shown in
Figure 3.12. Referring to Figure 3.11,
note that point 2 is a node point for
mode 2 and very near a node point -40.0

for mode 3. Mode 2 does not appear Fxd Y 0 Hz 1.25k

in the imaginary display and mode 3

is barely discernible. In dB coordi- FREQ RESP 3 Avg 0% Ovlp Fr/Ex
nates, the mode 3 appears to exist but 56.0

it is still difficult to observe mode 2. Imag

It may be advisable or necessary at
times to gather more than one set of
data at different excitation locations
-56.0
in order to measure all the modes. It
Fxd Y 0 Hz 1.25k
should also be noted that the same FREQ RESP 3 Avg 0% Ovlp Fr/Ex
observations can be made in an 40.0

impact test where the response dB

point would also be moved to
various locations.
-40.0
Another concept associated with a
Fxd Y 0 Hz 1.25k
linear structure concerns a property
of the frequency response matrix.
The frequency response matrix
for a linear system can be shown
to be symmetric due to Maxwell’s Figure 3.13
Reciprocity Theorem. Simply stated, Symmetric
frequency
a measurement with the excitation response H11 H12 H1N
at point i and the response at point j matrix
is equal to the measurement with
the excitation at point j and the
response at point i. This is illustrated
H21 H22
in Figure 3.13. A check can be made
on the measurement process by Hij
comparing these two reciprocal
measurements at various pairs of
points and observing any differences
between them. This can be helpful
for noting nonlinearities when Hji
applying different force levels.
HN1 HNN

Hij = Hji

37
Chapter 4
Modal Parameter Estimation

Introduction and increase the confidence level of

the estimated parameters. The fre-
The previous chapter presented quency and damping for each mode
several techniques for making can be estimated from any combina-
frequency response measurements tion of these measurements. The
for modal analysis. Having acquired residues and, consequently, the modal
this data, the next major step of the coefficients are then computed for
process is the use of parameter esti- each measurement point. The mode
mation techniques – “curve fitting” – shapes are then scaled and sorted for
to identify the modal parameters. A each resonant frequency. Finally,
vast amount of literature exists on the modal mass and stiffness can be
the subject of curve fitting measured determined from these scaled param-
data to estimate the modal properties eters as illustrated in Figure 4.1.
of a structure. However, this
information tends to be mathemati-
cally vigorous and is generally biased
toward a particular type of algorithm.
It is the intent of this chapter to cate-
gorize, in a conceptual manner, the
Figure 4.1
different types of curve fttters and Typical flow
discuss the applications and prob- Measure
of modal test
complete
lems associated with those most
set of
commonly implemented. frequency
responses
It was discussed earlier that a
minimum of one row or column of
the frequency response matrix, or its
equivalent, must be measured in
order to identify a complete set of
modal parameters. Although addi- Identify Modal
l Frequency
tional data is, in principle, redundant
l Damping
information, it can be used to verify
l Mode Shape

Scale Mode
Shape
l Modal Mass
l Modal
Stiffness

38
Modal Parameters Figure 4.2
Concepts of Vibrating Beam Mode = 1
One of the most fundamental modal parameters Mode = 2
assumptions of modal testing is that Mode = 3
a mode of vibration can be excited
at any point on the structure, except
at nodes of vibration where it has no Measurement
Points
motion. This is why a single row or
column of the frequency response
matrix provides sufficient informa-
tion to estimate modal parameters.
As a result, the frequency and damp-
ing of any mode in a structure are
constants that can be estimated Dampin
g
from any one of the measurements
as shown in Figure 4.2. In other
Freque
words, the frequency and damping ncy
of any mode are global properties of
the structure. Damping, frequency — same at each measurement point
Mode shape — obtained at same frequency from all measurement points
In practical applications, it is impor-
tant to include sufficient points in the
test to completely describe all the
modes of interest. If the excitation
point has not been chosen carefully
or if enough response points are
Figure 4.3
not measured, then a particular mode Modal
may not be adequately represented. parameters ω — Frequency
At times it may become necessary to ζ — Damping
include more than one excitation
location in order to adequately
describe all of the modes of interest. Global
Frequency responses can be meas-
ured independently with single-point
excitation or simultanously with
multiple-point excitations.
{φ} — Mode Shape
The mode shapes as a whole are also
global properties of the structure, but
have relative values depending on the Local
point of excitation and scaling and
sorting factors. On the other hand,
each individual modal coefficient that
makes up the mode shape is a local
property in the sense that it is esti-
mated from the particular measure-
ment associated with that point as
shown in Figure 4.3.

39
Curve Fitting Methods Users generally fall into one of three modal model. For example, they
major groups. The first group is pri- require more accuracy and are willing
Due to the large amount of literature marily concerned with troubleshoot- to spend more time obtaining results.
and algorithms currently available for ing existing mechanical equipment. The final group is pushing the state of
curve fitting structural data, it has They are usually concerned with time the art and is involved with develop-
become difficult to determine the and require a fast, medium quality ment work. Accuracy, rather than
exact need for each method and curve fitter. The second group is time, is of paramount importance.
which method is best. There is no more serious about quantitative
ideal solution and the common parameter estimates for use in a
methods are only approximations.
Also, many of the methods are very
similar to each other and, in some
cases, simply extensions of a few
basic techniques.

Although there are several ways in

which curve fitting methods can be Figure 4.4
categorized, the most straightforward Increasing
is single-mode versus multiple- accuracy in Single-Mode
curve fit Methods
mode classification. Besides the
methods l Easy, fast
intuitive reasoning for single- and l Limited
multiple-mode approximations, there
are some practical reasons for this
classification. The major difference
in the level of sophistication, or l
evel of accuracy, among curve fitters
is between a single-mode and a multi-
Multi-Mode
ple-mode method. Also, the comput- Methods
ing resources needed (computation l More
speed, memory size and I/0 capabili- Accurate
ty) for multiple-mode methods l Slower
can increase tremendously. Other
sub-catagories and extensions that
fall mostly within multiple-mode
methods are shown in Figure 4.4.
Multi-Measurement
l Global Accuracy
l Large Amount of
Data Processing

40
 Single-Mode Methods Figure 4.5
Frequency
As stated earlier, the general response
procedure for estimating modal
parameters is to estimate frequencies Magnitude
and damping factors, then estimate
modal coefficients. For most single-
mode parameter estimation tech-
niques, however, this is not always
the case. In fact, it is not absolutely
necessary to estimate damping in
order to obtain modal coefficients.
This is typical in a troubleshooting Phase
environment where frequencies and
mode shapes are of primary concern.

The basic assumption for single-mode

approximations is that in the vicinity ωn — Frequency
of a resonance, the response is due
primarily to that single mode. The
resonant frequency can be estimated
from the frequency response data
(illustrated in Figure 4.5) by observ-
ing the frequency at which any of the Real
following trends occur:

● The magnitude of the frequency

response is a maximum.
● The imaginary part of the
frequency response is a
maximum or minimum. Imaginary
● The real part of the frequency
response is zero.
● The response lags the input by
90° phase.
ωn — Frequency

It was discussed earlier that the

height of the resonant peak is a
function of damping. The damping
factor can be estimated by the half-
power method or other related math-
ematical or graphical method. In the
half-power method, the damping is
estimated by determining the sharp-
ness of the resonant peak. It can be
shown from Figure 4.6 that damping
can be related to the width of the

41
peak between the half-power points: Figure 4.6
points below and above the resonant Damping factor 640
m
peak at which the response magni- from half power
tude is .7071 times the resonant
80.0
magnitude. m
/Div

One of the simplest single-mode

Damping
modal coefficient estimation tech-
Mag
niques is the quadrature method,
often called “peak picking”. Modal
coefficients are estimated from the
imaginary (quadrature) part of the
frequency response, so the method is
not a curve fit in the strict sense of
the term. As mentioned earlier, the 0.0
imaginary part reaches a maximum at 0 Hz 100
the resonant frequency and is 90°
out of phase with respect to the
input. The magnitude of the modal Figure 4.7
640
coefficient is simply taken as the Quadrature m
value of the imaginary part at peak pick
resonance as illustrated in Figure 4.7.
160
The sign (phase) is taken from the m
/Div
direction that the peak lies along
the imaginary axis, either positive or
negative. This implies that the phase
Imag
angle is either 0° or 180°.

The quadrature response method is

Modal Coefficient
one of the more popular techniques
for estimating modal parameters
because it is easy to use, very fast
and requires minimum computing -640
m
resources. It is, however, sensitive to
noise on the measurement and effects
from adjacent modes. This method is
best suited for structures with light Figure 4.8
damping and well separated modes Circle fit Imaginary
where modal coefficients are essen- method
tially real valued. It is most useful for
troubleshooting problems, however,
where it is not necessary to create a
modal model and time is limited.
Real

Modal Coefficient

42
Another single-mode technique, A SDOF method related to the circle Concept of Residual Terms
called the circle fit, was originally fit is a frequency domain curve fit to
developed for structural damping but a single-mode analytical expression Before proceeding to multiple-mode
can be extended to the viscous damp- of the frequency response. This methods, it is appropriate to discuss
ing case. Recall from Chapter 1 that expression is generally formulated the residual effects that out-of-band
the frequency response of a mode as a second order polynomial with modes have on estimated parameters.
traces out a circle in the imaginary residual terms to take into account In general, structures possess an
plane. The method fits a circle to the effects of out of band modes. infinite number of modes. However,
the real and imaginary part of the Because of its similarities to the there are only a limited number that
frequency response data by minimiz- circle fit, it possesses the same basic are usually of concern. Figure 4.10
ing the error between the radius of advantages and disadvantages. illustrates the analytical expression
the fitted circle and the measured
data. The modal coefficient is then
determined from the diameter of the
circle as illustrated in Figure 4.8. The
phase is determined from the positive
or negative half of the imaginary axis
in which the circle lies.
Figure 4.9
Frequency and damping can be Comparison of Imaginary
estimated by one of the methods quadrature and
Real
circle fit methods
discussed earlier or by some of the
MDOF methods to be discussed later.
Damping can also be estimated from
φ φ Circle Fit
the spacing of points along the φ
Peak
Nyquist plot from the circle. Pick φ Circle Fit Peak
Pick
The circle fit method is fairly fast
and requires minimum computer
resources. It usually results in better
parameter estimates than obtained
by the quadrature method because
Figure 4.10 n
it uses more of the measurement Analytical φi φj
information and is not as sensitive H(ω) = Σ
expression of r=1 ( ω2
n - ω2
) + j(2 ζωωn)
to effects from adjacent modes as frequency
illustrated in Figure 4.9. It is also response
less sensitive to noise and distortion
on the measurement. However, it
requires much more user interaction
than the quadrature method; conse-
quently, it is prone to errors, particu-
larly when fitting closely spaced
modes.

43
for the frequency response of a Figure 4.11
structure taking into account the Single mode
total number of realizable modes. summation of Mass Line
Unfortunately, the measured frequen- frequency
response
cy response is limited to some fre-
quency range of interest depending

dB Magnitude
on the capabilities of the analyzer
and the frequency resolution desired.
This range may not necessarily
include several lower frequency
modes and most certainly will not
include some higher frequency Stiffness Line
modes. However, the residual effects
of these out-of-band modes will be
present in the measurement and,
0.0
consequently, affect the accuracy
of parameter estimation. ω1 ω2 ω3 Frequency

Although parameters of the

out-of-band modes cannot be identi-
fied, their effects can be represented
by two relatively simple terms. It can
be seen from Figure 4.11 that the
effects of the lower modes tend to
have mass-like behavior and the
dB Magnitude

effects of the higher modes tend to

have stiffness-like behavior. The
analytical expression for the residual
terms can then be written as shown
in Figure 4.12. Notice that the
residual terms are equivalent to the
asymptotic behavior of the mass and
stiffness of a SDOF system discussed
in the chapter on structural dynamics.
0.0
Useful information can often be ω1 ω2 ω3 Frequency
gained from the residual terms that
has some physical significance. First,
if the structure is freely supported
during the test, then the low frequen-
cy residual term can be a direct
measure of rigid body mass proper- Figure 4.12
φi φj
ties of the structure. The high fre- Residual terms H(ω) = Lower Modes
of frequency -w2m
quency term, on the other hand, can response φi φj
be a measure of the local flexibility H(ω) = Higher Modes
k
of the driving point.

44
Multiple-Mode Methods Figure 4.13
Damping and 64.0
The single-mode methods discussed modal coupling
earlier perform reasonably well for
structures with lightly damped and 80
well separated modes. These meth- /Div
ods are also satisfactory in situations
where accuracy is of secondary con-
cern. However, for structures with Mag

closely spaced modes, particularly

when heavily damped, (as shown in
Figure 4.13) the effects of adjacent
modes can cause significant approxi-
mations. In general, it will be neces-
sary to implement a multiple-mode
0.0
method to more accurately identify
433.7 Hz 2.4437k
the modal parameters of these types Light Damping and Coupling
of structures.
64.0
The basic task of all multiple-mode
methods is to estimate the coeffi-
cients in a multiple- mode analytical 2.0

expression for the frequency /Div

response function. This is done by
curve fitting a multiple-mode form
of the frequency response function Mag

to frequency domain measurements.

An equivalent method is to curve fit
a multiple response form of the
impulse response function to time
domain data. In either process, all
the modal parameters (frequency,
0.0
damping and modal coefficient)
443.7 Hz 2.4437k
for all the modes are estimated Heavy Damping and Coupling
simultaneously.

There are a number of multiple-mode

methods currently available for curve
fitting measured data to estimate
modal parameters. However, there
are essentially two different forms
of the frequency response function Figure 4.14
which are used for curve fitting. Frequency n rk rk*
H(ω) = Σ + Partial Fraction
These are the partial fraction form response k=1 jω - pk jω - pk*
and the polynomial form which are representation
a0sm + a, s m - 1 + . . .
shown in Figure 4.14. They are H(ω) = Polynomial
equivalent analytical forms and can b0sn + b, sn - 1 + . . .
n
h(t) = Σ rk e -σk t + rk*e-σ*k t Complex Exponential
k=1

45
be shown to be related to the struc- The complex exponential method faster than the corresponding time
tural frequency response developed is a time domain method that fits domain correlation functions. It is
earlier in Chapter 1. The impulse decaying exponentials to impulse true that the time domain can be used
response function, obtained by response data. The equations are to select modes having different
inverse Fourier transforming the nonlinear, so an iterative procedure is damping values, but this is usually
frequency response function into the necessary to obtain a solution. The not as important as the ability to
time domain, is also shown in Figure method is relatively insensitive to select a frequency range of interest.
4.14. noise on the data, but suffers from
sensitivity to time domain aliasing, as Each method has its advantages and
When the partial fraction form of the a result of truncation in the frequency disadvantages, but the fundamental
frequency response is used, the domain from inverse Fourier trans- problems of noise, distortion and
modal parameters can be estimated forming the frequency responses. interference from adjacent modes
directly from the curve fitting remain. As a result, none of the
process. A least squares error In principle, it should not matter methods work well in all situations.
approach yields a set of linear whether frequency domain data or It is also unlikely that some “magic”
equations that must be solved for time domain data is used for curve method will be discovered that elimi-
the modal coefficients and a set of fitting since the same information is nates all of these problems. All of the
nonlinear equations that must be contained in both domains. However, methods work well with ideal data,
solved for frequency and damping. there are some practical reasons, but cannot be evaluated by analytical
Because an iterative solution is based on frequency domain and time means alone. The important factor is
required to solve these equations, domain operations, that seem to favor how well they work, or gracefully fail,
there is potential for convergence the frequency domain. One, the mea- with real experimental data complete
problems and long computation surement data can be restricted to with noise and distortion.
times. some desired frequency range and
any noise or distortion outside this
If the polynomial form of the frequen- range can effectively be ignored.
cy response is used, the coefficients Another, the cross spectra and
of the polynomials are identified dur- autospectra needed to compute fre-
ing the curve fitting process. A root quency responses can be formed
finding solution must then be used to
determine the modal parameters. The
advantage of the polynomial form is
that the equations are linear and the
coefficients can be solved by a nonit-
erative process. Therefore, conver-
gence problems are minimal and
computing time is more reasonable.

46
Concept of Real and Complex Modes When damping is light, as is the case coupling is a second-order effect. It
in most mechanical structures, the should be noted that closely-spaced
The structural model discussed so far proportional damping assumption is modes often appear complex as a
is based on the concept of propor- generally an accurate approximation. result of the effects from adjacent
tional viscous damping which implies Although damping is not proportional modes as illustrated in Figure 4.18.
the existence of real, or normal, to the mass and stiffness, the nonpro- In reality, they may actually be
modes. Mathematically, this implies portional coupling effects may be more real than they appear.
that the physical damping matrix small enough not to cause serious
can be defined as linear combination errors. Physically, this means that the
of the physical mass and stiffness damping is sufficiently small so that
matrices as shown in Figure 4.15.
The mode shapes, are, in effect real
valued, meaning the phase angles Figure 4.15
differ by 0° or 180°. Physically, all Proportional [C] = a [M] + β [K]
the points reach their maximum damping
excursion at the same time as in a representation
standing wave pattern. One of the
consequences of this assumption,
discussed earlier, is that the imagi-
Figure 4.16
nary part of the frequency response
Imaginary
reaches a maximum at resonance
and the real part is 0 valued as illus- Real
trated in Figure 4.16. Note also that
the Nyquist circle lies along the
imaginary axis.
2σk
However, physical structures exhibit
a more complicated form of damping
which results in non-proportional ω = ωk
damping. The mode shapes are,
generally, complex valued, meaning
the phase angles can have values
other than 0° or 180°. Physically, Figure 4.17
the points reach their maximum Imaginary
excursions at various times as in a
traveling wave pattern. With non-
r1k2 +r2k 2
proportional damping, the imaginary
2σk
part of the frequency response no
longer reaches a maximum at reso- ak
nance nor is the real part nonzero
valued as illustrated in Figure 4.17.
Note also that the Nyquist circle is ω = ωk
rotated at an angle in the complex
plane.

Figure 4.18
Two closely Imaginary
spaced real
Real
modes

47
Chapter 5
Structural Analysis Methods

Introduction Figure 5.1

Alternative Physical Structure Model
The basic techniques for performing a types of
modal test to identify the dynamic structural
properties of a structure have been models
described in the previous chapters.
An introduction to the applications
of the resulting frequency responses
and modal parameters is the focus FFT
of this chapter. The discussion is spe-
cifically concerned with the uses of a
Response Model (measurements)
response model or a modal model
with structural analysis methods as
shown in Figure 5.1. The intent is to
bring together the experimental and
analytical tools for solving noise,
vibration and failure problems.

A response model is simply the set of

Curve Fit
frequency response measurements
acquired during the modal test.
These measurements contain all the Modal Model (parameters)
dynamics of the structure needed for
ω — Frequency
subsequent analyses. A modal model
ζ — Damping
is derived from the response model {φ} — Mode Shape
and is a function of the parameter
estimation technique used. It not
only includes frequencies, damping
factors, and mode shapes, but also
modal mass and modal stiffness.
These masses and stiffnesses depend
on the method that was used to scale
the mode shapes. A subset of the
modal model consisting of only the
frequencies and unscaled mode
shapes can be useful for some substructure coupling, a complete
troubleshooting applications where modal model is required. This
frequencies and mode shapes are the definition of a complete modal model
primary concern. However, for appli- should not be confused with the
cations involving analysis methods, concept of a truncated mode set in
such as structural modification and which all the modes are not included.

48
Structural Modification Figure 5.2
Structural Engine Mount
When troubleshooting a vibration modification
problem or investigating simple simulation
design changes, an analysis method
known as structural modification,
illustrated in Figure 5.2, can be very
useful. Basically, the method deter-
mines the effects of mass, stiffness ∆ Stiffness, Damping
and damping changes on the dynamic
characteristics of the structure. It is SM
a straightforward technique and gives
reasonable solutions for simple
design studies. Some of the benefits Troublesome Mode
of using structural modification are
SYNTHESIS Pole Residue
reduced time and cost for implement-
40.0
ing design changes and elimination
of the trial-and-error approach to
solving existing vibration problems.
7.5
The technique can be extended to
an iterative process, often called /Div

sensitivity analysis, in order to

categorize the sensitivity of specific
dB
amounts of mass, stiffness or
damping changes.

In general, structural modification

involves two interrelated design
investigations. In the first, a physical
mass, stiffness or damping change
-20
can be specified with the analysis FxdXY 1.375k Hz 2.975k
determining the modified set of
modal parameters. The second
involves specifying a frequency and
having the analysis determine the Structural modification can be imple- the support system. A modal model
amount of mass, stiffness or damping mented with either a response model requires only one driving point
needed to shift a resonance to this or a modal model. The advantages of measurement whereas a response
new frequency. For the user’s response model are that it contains model requires a driving point
convenience, specific applications of the effects of modes outside the measurement at every location to be
these basic methods, such as tuned analysis range and the measurements modified. The modal model may be
absorber, design are usually included. can be limited to the selected set of more intuitive since it contains direct
modified points. It is sensitive, mass, stiffness and damping informa-
though, to the quality of frequency tion directly. However, it does suffer
response measurements and the from being tedious and time consum-
effects of rigid body modes from ing to derive and is sensitive to the
number and type of modes extracted.

49
Finite Element Correlation Table 5.1
FE (Hz) Test (Hz)
Tabular
Finite element analysis is a numerical comparisons 17.5 15.7
procedure useful for solving structur- of frequency 21.3 19.4
al mechanics problems. More specifi-
26.4 25.5
cally, it is an analytical method for
30.0 28.3
determining the modal properties of
a structure. It is often necessary to 31.2 30.5
validate the results from this theoreti-
cal prediction with measured data
from a modal test. This correlation
method is generally an iterative
process and involves two major steps.
First, the modal parameters, both fre-
quencies and mode shapes, are com- Figure 5.3
pared and the differences quantified. Graphical
Second, adjustments and modifica- comparison
of frequencies
tions are made, usually to the finite
element model, to achieve more com-
parable results. The finite element Fe
model can then be used to simulate
responses to actual operating
environments.

The correlation task is usually begun

by comparing the measured and pre- Test
dicted frequencies. This is often done
by making a table to compare each
mode frequency by frequency as
shown in Table 5.1. It is more useful,
however, to graphically compare the
entire set of frequencies by plotting
measured versus the predicted an inappropriate element type or ensure a one-to-one correspondence
results as shown in Figure 5.3. This a poor element mesh in the finite between the frequency and the mode
shows not only the relative differenc- element model. It could also result shape. Remember that a distinct
es between the frequencies, but also from incorrect boundary conditions mode shape is associated with each
the global trends and suggests possi- in either the test or the analysis. If distinct frequency. One technique
ble causes of these differences. If the points lie on a straight line, but for performing this comparison is
there is a direct correlation the points with a slope other than 1, then the to simply overlay the plotted mode
will lie on a straight line with a slope problem may be a mass loading prob- shapes from the test and analysis and
of 1.0. If a random scatter arises, lem in the modal test or an incorrect observe their general trends. This
then the finite element model may material property, such as elastic can become rather difficult, though,
not be an accurate representation of modulus or material density, in the for structures with complicated
the structure. This could result from finite element model. geometry because the plots tend to
get cluttered.
The parameter comparison is not
actually this simple, nor is it com-
plete, because the mode shapes must
also be compared at the same time to

50
Numerical techniques have been Figure 5.4
developed to perform statistical Numerical MSF — Proportionality Constant
comparisons between any two mode comparison of How were the modes scaled?
mode shapes
shapes, illustrated in Figure 5.4. The
[M] [ I ]
first results in the modal scale factor
(MSF) – a proportionality constant Unit Modal Mass
between the two shapes. If the
MAC — Correlation Coefficient
constant is equal to 1.0, this means
the shapes were scaled in the same Are the modes the same mode?
manner such as unity modal mass. If
Mode 1
the constant is any value other than
1.0, then the shapes were scaled
differently. The second, and more
important method, results in the
modal assurance criterion (MAC), a
Mode 2
correlation coefficient between the
two mode shapes. If the coefficient is
equal to 1.0, then the two shapes are
perfectly correlated. In practice,
any value between 0.9 and 1.0 is
considered good correlation. If the
coefficient is any value less than 1.0,
then there is some degree of
inconsistency, proportional to the There are other numerical methods
value of the factor, between the for comparing the measured and
shapes. This can be caused by an predicted modal parameters of a
inaccurate finite element model, as structure. One such technique, called
described earlier, or the presence of direct system parameter identifica-
noise and nonlinearities in the mea- tion, is the derivation of a physical
sured data. It should be noted that in model of a structure from measured
order for these comparisons to have force and response data. However,
a reasonable degree of accuracy, it techniques such as this are beyond
is very important that coordinate the scope of this text and can be
locations in the modal test coincide found in technical articles about
with coordinates in the finite modal analysis.
element mesh.

51
Substructure Coupling Analysis A modal model of a component for tive modes of the structure. In the
substructure coupling must contain special case where the mode shapes
In analyses involving large structures the modal mass, stiffness and damp- have been scaled to unity modal
or structures with many components ing factors along with the modal mass, the modal model reduces to
it may not be feasible to assemble a matrix. The modal matrix of a the frequencies, damping and mode
finite element model of the entire structure is simply a matrix whose shapes.
structure. The time involved in build- columns are comprised of the respec-
ing the model may be unacceptable
and the model may contain more de-
grees of freedom than the computer
can handle. As a result, it may be
necessary to employ a modeling
reduction method known as substruc- Figure 5.5
Substructure Airplane
ture coupling or component mode coupling
synthesis illustrated in Figure 5.5. analysis

Substructure coupling involves the

division of the structure into various
components, modeling these compo-
nents for their individual dynamics
and then combining these individual
results into one model to analyze the
dynamics of the complete structure.
These component models can take on
several different mathematical forms Component Analysis
each of which has a particular useful-
ness. Common models include modal
models and physical models from a
finite element analysis, modal models
from a modal test, rigid body models
and physical springs and dampers. Fuselage
The component models are combined
through a transformation that relates
their dynamics at the interfaces. The Wing
results from the analysis of the com- Engine
plete structure can then be correlated
with equivalent modal test results in
the same manner as described earlier.

Wing

Fuselage

Engine

52
Forced Response Simulation Figure 5.6 Force Inputs
Forced
One of the major design goals for response
most engineering analyses is to be simulation
able to predict system responses to
actual operating forces. This can
enable engineers to ultimately find
optimal solutions to troublesome
noise or vibration problems. This
technique, illustrated in Figure 5.6,
is commonly called forced response Engine RPM Road Surface
simulation or forced response predic-
tion. Forces can be specified for any
degree of freedom in the modal
model and displacements, velocities System Model
or accelerations can be predicted for
any degree of freedom.

ω, ζ, φ

System Response

Ride Motion

53
Bibliography

Robert D.Cook R. J. Allemang, D. L. Brown, D. L. Brown, R. J. Allemang,

Concepts and Applications of R. W. Rost Ray Zimmerman, M. Mergeay
Finite Element Analysis Multiple Input Estimation of Parameter Estimation Techniques
John Wiley & Sons Frequency Response Functions for Modal Analysis
1981 Proceedings of the 2nd International Society of Automotive Engineers,
Modal Analysis Conference #790221
Roy R. Craig, Jr. 1984 1979
Structural Dynamics:
An Introduction to Computer Dave Corelli, David L. Brown Richard Jones, Yuji Kobayashi
Methods Impact Testing Considerations Global Parameter Estimation
John Wiley & Sons Proceedings of the 2nd International Using Rational Fraction Polynomials
1981 Modal Analysis Conference Proceedings of the 4th International
1984 Modal Analysis Conference
Clare D. McGillem, George R. Cooper 1986
Continuous and Discrete Signal Norm Olsen
and System Analysis Excitation Functions for Structural Richard Jones, Kathleen Iberle
Holt, Rinehart, and Winston, Inc. Frequency Response Measurements Structural Modification:
1974 Proceedings of the 2nd International A Comparison of Techniques
Modal Analysis Conference Proceedings of the 4th International
Francis S. Tse, Ivan E. Morse, 1984 Modal Analysis Conference
Rolland T. Hinkle 1986
Mechanical Vibrations: John R. Crowley, G. Thomas Rocklin,
Theory and Applications Albert L. Klosterman, Havard Vold
Allyn and Bacon, Inc. Direct Structural Modification
1978 Using Frequency Response Functions
Proceedings of the 2nd International
D. Brown, G.Carbon, K. Ramsey Modal Analysis Conference
Survey of Excitation Techniques 1984
Applicable to Automotive Structures
Society of Automotive Engineers,
#770029
1977

R. J. Allemang, D. L. Brown
A Correlation Coefficient for Modal
Vector Analysis
Proceedings of the 1st International
Modal Analysis Conference
1982

54
55
Agilent Technologies' Test and Measurement
Support, Services, and Assistance
Agilent Technologies aims to maximize the
value you receive, while minimizing your risk
and problems. We strive to ensure that you get the
test and measurement capabilities you paid for
and obtain the support you need. Our extensive
support resources and services can help
you choose the right Agilent products for your
applications and apply them successfully. Every
instrument and system we sell has a global war-
ranty. Support is available for at least five years
beyond the production life of the product. Two
concepts underlie Agilent's overall support
policy: "Our Promise" and "Your Advantage."

Our Promise
Our Promise means your Agilent test and
measurement equipment will meet its advertised
performance and functionality. When you are
choosing new equipment, we will help you with
product information, including realistic perform-
ance specifications and practical recommenda-
tions from experienced test engineers. When
you use Agilent equipment, we can verify that it
works properly, help with product operation, and
provide basic measurement assistance for the use
of specified capabilities, at no extra cost upon
request. Many self-help tools are available.

Your Advantage
Your Advantage means that Agilent offers a wide
range of additional expert test and measurement
services, which you can purchase according to
your unique technical and business needs. Solve
problems efficiently and gain a competitive edge
by contracting with us for calibration, extra-cost
upgrades, out-of-warranty repairs, and on-site
education and training, as well as design, system
integration, project management, and other
professional engineering services. Experienced
Agilent engineers and technicians worldwide can
help you maximize your productivity, optimize the
return on investment of your Agilent instruments
and systems, and obtain dependable measurement
accuracy for the life of those products.

For More Assistance with Your

Test & Measurement Needs go to
www.agilent.com/find/assist

Or contact the test and measurement experts

at Agilent Technologies
(During normal business hours)
United States: Latin America:
(tel) 1 800 452 4844 (tel) (305) 267 4245
(fax) (305) 267 4286

Canada: Australia:
(tel) 1 877 894 4414 (tel) 1 800 629 485
(fax) (905) 206 4120 (fax) (61 3) 9272 0749

Europe: New Zealand:

(tel) (31 20) 547 2323 (tel) 0 800 738 378
(fax) (31 20) 547 2390 (fax) 64 4 495 8950

Japan: Asia Pacific:

(tel) (81) 426 56 7832 (tel) (852) 3197 7777
(fax) (81) 426 56 7840 (fax) (852) 2506 9284

Product specifications and descriptions in this

document subject to change without notice.
Copyright © 1997, 2000 Agilent Technologies
Printed in U.S.A. 5/00
5954-7957E