11th IMAC Conference, Kissimmee, FL February, 1993

OPERATING DEFLECTION SHAPES

FROM
TIME VERSUS FREQUENCY DOMAIN MEASUREMENTS

Mark H. Richardson Patrick L. McHargue

Vibrant Technology, Inc. Vibrant Technology, Inc.
Jamestown, California 95327 Jamestown, California 95327

ABSTRACT [H(jω)] = (n by n) Frequency Response Function (FRF)

The vibration parameters of a structure are typically de- matrix.
rived from acquired time domain signals, or from frequency
domain functions that are computed from acquired time do- {x (t)} = forced response n-vector.
f
main signals. For example, the modal parameters of a
{uk} = complex mode shape (n-vector) for the kth mode.
structure can be obtained by curve fitting a set of frequency
response functions (FRFs), or by curve fitting a set of (time pk = pole location for the kth mode = - σk + jωk
domain) impulse response functions. Similarly, the operat-
ing deflection shapes of a structure can be obtained either σk = damping of the kth mode (radians/second).
from a set of time domain responses, or from a set of fre-
quency domain responses. ωk = frequency of the kth mode (radians/second).
Two of the most commonly asked questions about vibration Ak = a non-zero scaling constant for the kth mode.
are:
[Rk] = the (n by n) residue matrix for the kth mode
1. What is the deformation (deflection shape) of a ma-
chine or structure under a particular operating condi- tr
= Ak{uk}{uk}
tion?
tr - denotes the transpose.
2. How much is the machine or structure actually moving
at certain points? * - denotes the complex conjugate.

Time domain responses can be used to answer both of these INTRODUCTION

questions, for linear as well as non-linear vibration. On the The vibration parameters of a structure are typically derived
other hand, frequency domain responses can be used to from acquired time domain signals, or from frequency do-
answer these questions for specific frequencies. main functions that are computed from acquired time sig-
nals. Using a modern multichannel data acquisition system,
NOMENCLATURE the vibration response of a structure is measured for multi-
t = time variable (seconds). ple points and directions (DOFs) with motion sensing trans-
ducers. Signals from the sensors are then amplified, digit-
ω = frequency variable (radians/second).
ized, and stored in the system's memory as blocks of data,
n = number of measured DOFs. one data block for each measured DOF of the structure. A
m = number of modes. key requirement of the multichannel system is that it be able
[M] = (n by n) mass matrix (force/unit of acceleration). to simultaneously sample or digitize the vibration signals as
it converts them from analog signals (voltages) to digital
{x''(t)} = acceleration response n-vector.
data (numbers).
[C] = (n by n) damping matrix (force/unit of velocity).
If the acquisition system is also an FFT-based system (an
{x'(t)} = velocity response n-vector.
FFT Analyzer), then one additional requirement must be
[K] = (n by n) stiffness matrix (force/unit of displacement). met in order to compute valid frequency domain functions.
{x(t)} = displacement response n-vector. To prevent aliasing (false frequencies in their frequency
{f(t)} = excitation force n-vector. spectra) the frequency content of the time domain signals
must be bounded to satisfy the Nyquist criterion. That is,
{X(jω)} = discrete Fourier transform of the displacement the maximum frequency in the analog signals cannot ex-
response n-vector. ceed one half of the sampling frequency used to digitize
{F(jω)} = discrete Fourier transform of the excitation force them.
n-vector.

Page 1 of 8
11th IMAC Conference, Kissimmee, FL February, 1993

There are two ways to meet this criterion; digitize the time of a vibration problem by assuming that the structure is be-
domain signals at a very high sampling rate (twice the high- having in a linear (or near linear) manner. Then, both time
est expected frequency in the signals), or band limit the time and frequency domain techniques can be used.
domain signals with filters. Since the time signals must be
The equation of motion for a vibrating structure is common-
frequency limited before they are digitized, they must be
ly derived by applying Newton's second law to all of the
filtered as analog signals using analog filters. In a multi-
DOFs of interest on the structure. In an experimental situa-
channel system, all of the anti-aliasing filters must filter all
tion, this results in a countable set of equations, one for each
channels in the same way, and this usually adds to the cost
measured DOF:
of the system.
[M] {x''(t)} + [C] {x'(t)} + [K] {x(t)} = {f(t)} (1)
Having acquired either a set of sampled time domain re-
sponses, or computed (via the FFT) a set of frequency do- Notice that the excitation forces and responses are functions
main responses, an operating deflection shape is defined as: of time (t), and that the coefficient matrices [M], [C], and
[K] are constants. This is a dynamic model for describing
Operating Deflection Shape: The values of a set of simul-
the vibration of a linear, time invariant structure.
taneously sampled time domain responses at a specific time,
or the values of a set of frequency domain responses at a
specific frequency. The Fourier Transform
The Fourier transform is defined for continuous signals.
The analogous discrete Fourier transform (DFT) is defined
Operating Deflection Shapes versus Mode Shapes
for discrete (sampled) signals. The FFT algorithm performs
The vibration response of a structure depends on both the
the DFT on a finite number of samples of time domain data.
amount and the location of its excitation. Therefore its op-
More specifically, the FFT transforms (N) samples of real
erating deflection shapes always depend on its excitation
valued time domain data into (N/2) samples of complex
source(s). On the other hand, the mode shapes of a struc-
valued frequency domain data. Likewise, the Inverse FFT
ture do not depend on either the amount or location of its
converts (N/2) samples of complex valued frequency do-
excitation. That is, they are natural properties of the struc-
main into (N) samples of real time domain data. (For prac-
ture, and will not change unless its physical properties
tical reasons, the number of samples (N) is usually restrict-
(mass, stiffness, and damping), or its boundary conditions
ed to an integer power of 2.)
are changed.
When initial conditions are ignored, the equivalent frequen-
Mode shapes, by themselves, do not have fixed values (they
cy domain form of the dynamic model for a structure can be
have no units), and therefore cannot tell you how much a
represented in terms of discrete Fourier transforms:
structure is actually moving. They can only show the rela-
tive motions between two DOFs of a structure. Using mod- {X(jω)} = [H(jω)] {F(jω)} (2)
al parameters, the answer to the question, "How much ...?"
can only be answered once the amount and location of all of This equation is valid for all discrete frequency values for
the excitation forces is specified. Then, modal parameters which the discrete Fourier transforms are computed. Taking
can be used to synthesize an operating deflection shape. -1
the inverse FFT (FFT ) of the above equation yields the
forced response equation for the structure. (For a fixed
But, what if the amount and location of all of the excitations
value of time (t), the forced response vector {x (t)} is the
cannot be identified, or are too complex to identify? Then, f
direct measurement of the operating deflection shape is the operating deflection shape.)
only way to answer the question, "How much ...?"
-1
{x (t)} = FFT {[H(jω)] {F(jω)}} (3)
Since many vibration problems involve the excitation of f
modes (or resonances), operating deflection shapes and
mode shapes must be closely related to one another, and
indeed they are. In the following section, this relationship
is examined in more detail.

THEORETICAL BACKGROUND
Since an operating deflection shape is defined simply as the
response of a structure at a specific time or frequency, no
assumption is made regarding the linearity of the structure's
response. However, many vibration problems in structures
involve the excitation of modes, which are only defined for
linear systems. Furthermore, since the FFT is a linear trans-
formation, it is much more profitable to begin the analysis

Page 2 of 8
11th IMAC Conference, Kissimmee, FL February, 1993

tion, the FRF matrix can be represented as a summation of

terms, each term due to the contribution of a single mode of
vibration:
[H(jω)] = [H1(jω)] + [H2(jω)] +...+ [Hk(jω)] +...+

[Hm(jω)]

where:
tr
[Hk(jω)] = Ak{uk}{uk} /(jω - pk) (4)

tr
+ A*k{u*k}{u*k} /(jω - p*k)

Figure 1. An FRF Measurement. Notice that each term of the FRF matrix is represented in
terms of a pole location and a mode shape. Notice also that
Several conclusions can be drawn from the forced response all the numerators are simply constants, and that only the
equation: denominators are functions of frequency. The numerators
• The Fourier transform of the forced response is made are also called residues. Each term of the FRF matrix can
up of a summation of the transforms of all of the exci- also be represented in terms of poles and residues:
tation forces times the columns of the FRF correspond- [Hk(jω)] = [Rk] / (jω - pk) + [R*k] / (jω - p*k) (5)
ing to the excitation DOFs.
where:
• The unit Impulse Response of the structure is obtained
by setting one of the elements of the transformed exci- [Rk] = the (n by n) residue matrix for the kth mode
tation vector {F(jω)} equal to one (1), since the Fourier
transform of a unit impulse is one for all frequencies. tr
= Ak{uk}{uk}
From the forced response equation, an n-vector of im-
pulse responses is obtained by inverse Fourier trans- Again, it's worth noting that the numerators (residues) of an
forming one of the columns of the FRF matrix, the col- FRF are merely constants, fixed in value. The mode shapes
umn corresponding to the DOF where the impulse is are eigenvectors; that is, they can change in value, but not
applied. This column is also called the reference col- in shape. The denominators are functions of frequency, and
umn. cause the peaks in an FRF. The locations of the peaks are
dictated by the pole locations (pk). Each peak in the FRF is
• The impulse response of the structure depends on
where the impulsive force is applied, which certainly evidence of at least one pole, or mode, or resonant condi-
serves our intuition. tion. A typical FRF measurement (one element of the FRF
matrix) is shown in Figure 1.
• The Sinusoidal Response of the structure to a single
sinusoidal excitation is obtained by inverse Fourier Forced Response in Terms of Modes
transforming the values of the reference (excitation) Several more conclusions can be drawn by substituting the
column of the FRF matrix, at the sinusoidal frequency. modal parametric form of the FRF matrix into the forced
This is so because the Fourier transform of a sine wave response equation:
signal is non-zero at the sine frequency, and zero for all
other frequencies. The forced response equation also • Every element of the FRF matrix (between any pair of
shows that for sine excitation, the operating deflection DOFs) is a summation of contributions from all of the
shape is merely the values of the FRFs from the refer- modes of the structure. Therefore, the forced response
ence column, at the excitation frequency. potentially contains contributions from all of the
modes.
• The forced response, (and hence the operating deflec-
tion shape), is completely arbitrary, depending on the • The operating deflection shape depends not only on the
combination of excitation forces acting on the structure. excitation forces, but also on the locations of the poles
(resonant peak frequencies) and the structure's mode
Modal Parameters shapes.
If it is further assumed that reciprocity is valid for the test
structure, (the [M], [C], and [K] matrices are symmetric), • If an excitation force puts energy into a structure near a
then the FRF matrix can be represented completely in terms resonant peak frequency, the operating deflection shape
of the modal parameters of the structure. Using superposi- could be very large, depending on the value of the

Page 3 of 8
11th IMAC Conference, Kissimmee, FL February, 1993

modal residue between the excitation and response

where:
DOFs.
• The modal residue between an excitation DOF and a {1i} = n-vector with one (1) in the ith element, zero else-
response DOF is the product of the two mode shape where.
components corresponding to the two DOFs. If either σ
- t
of the mode shape components is zero (on a nodal line {Imi,k(t)} = {Ri,k e k sine( ωkt + αi,k)}
of the shape), the mode will not contribute to the oper-
ating deflection shape. = response of the kth mode, due to
an impulse applied at DOF(i).
The mathematics predicts what every vibration engineer
knows from experience; namely, if either the excitation or {Ri,k} = ith column of the residue matrix for the
response DOF is on a nodal line of a mode shape, that
mode will not contribute to the operating deflection shape. kth mode.
α
t
Impulse Response in Terms of Modes = {Ri,k e i,k } in polar form.
Since the Fourier transform of the unit impulse response is σ
- t
one (1) for all frequencies, the impulse response n-vector e k = exponential decay envelope due to damping
due to an impulse applied at DOF(i) can also be written in
terms of modal parameters: of the kth mode.

-1 sine( ωkt + αi,k) = sinusoidal response at the

{Imi(t)} = FFT {[H(jω)] {1i}}
frequency of the kth mode.
= {Imi,1(t)} + {Imi,2(t)} +..+ {Imi,k(t)}
+..+{Imi,m(t)} This parametric form shows that the overall response of the
structure to an impulse is the summation of the impulse
responses of each of its modes. Each modal contribution is
a damped sinusoidal response, with the oscillation frequen-
cy equal to the mode's natural frequency, and the decay en-
velope controlled by the mode's damping.

Inverse
FFT

Figure 2. FRF Measurements and Corresponding Impulse Responses.

OPERATING DEFLECTION SHAPES FROM TIME Impulse Response

DOMAIN FUNCTIONS
A set of impulse responses can be measured directly from a
The operating deflection shape can be obtained directly
structure by simply impacting it and simultaneously sam-
from measured time domain responses of a structure, or
pling its responses at multiple DOFs. This is fast and di-
from frequency domain measurements that have been in-
rect testing method but requires a lot of simultaneously
verse Fourier transformed.
sampling acquisition channels if responses for a lot of
DOFs are desired.

Page 4 of 8
11th IMAC Conference, Kissimmee, FL February, 1993

Alternatively, a set of impulse responses can be obtained

by inverse Fourier transforming a set of FRF measure-
ments. This second approach requires less equipment than
the first approach because the FRF measurements don't
have to be measured simultaneously. Only a 2 channel
analyzer is required to make an FRF measurement. Since
an FRF is a "normalized" measurement, computed, in ef-
fect, by dividing the transformed response by the trans-
formed excitation, a set of FRFs can be measured one at a
time. Hence, the requirement that all of the response sig-
nals be simultaneously sampled is relaxed. FRFs are typi-
cally measured one at a time during impact testing of a
structure to obtain its modal properties.

Units of the Impulse Response

The units of an impulse response depend on whether it was
acquired directly from the structure, or was computed by
inverse Fourier transforming an FRF. If the impulse re-
sponse was measured directly, and was calibrated, the re-
sponse units are those of the vibration sensor; acceleration,
velocity or displacement.
If the impulse response was obtained by inverse Fourier
transforming an FRF, its units are acceleration, velocity, or
displacement per unit of excitation force. For instance, if
the FRF was measured using an accelerometer for response
and a load cell for excitation, the impulse response units
are acceleration per unit of impact force at the impacting
DOF.
Figure 2 shows some FRF measurements, and their corre-
sponding impulse responses, computed with the Inverse
FFT. Figure 3 shows the mode shapes of the fundamental
(lowest frequency) modes of a plate structure. Figure 4
shows 16 operating deflection shapes taken from the im-
pulse responses, for 16.1 milliseconds following the im-
pulse. These pictures show that all of the fundamental
modes are excited from the impact point (reference DOF)
on the corner of the plate, which is expected.

Figure 3. Mode Shapes of the Fundamental Modes.

Page 5 of 8
11th IMAC Conference, Kissimmee, FL February, 1993

reference
DOF

Figure 4. Operating Deflection Shape from Impulse Responses.

Sinusoidal Response OPERATING DEFLECTION SHAPES FROM FRE-

If the structure is excited by a single sinusoidal force, its QUENCY DOMAIN FUNCTIONS
steady state response will also be sinusoidal, regardless of Most modern FFT-based Analyzers can compute a variety
the frequency of excitation. However, the amplitude of the of frequency domain functions that can be used for deriving
response will depend on whether or not a mode (resonance) operating deflection shapes, including the Linear Spec-
is also excited. trum (FFT), Auto Power Spectrum, FRF, and Transmis-
sibility. Each of these measurement functions has certain
From the Theoretical Background section, we saw that in
advantages, depending on the test situation.
order to excite a mode, two conditions must be met:
When any set of frequency domain functions is used to de-
• The excitation force must be applied at a DOF which is rive the operating deflection shape, the underlying assump-
not on a nodal line of the mode shape. tion is that the deflection shape is the response of the struc-
• The excitation frequency must be "close" to the reso- ture if it were excited by a single sine wave at the reference
nant peak frequency of the mode. DOF of the set of measurements. Frequency domain func-
tions are therefore useful for examining how a structure
If both of these conditions are met, the mode will act as a would deflect if excited at the reference DOF, at any fre-
"mechanical amplifier" and greatly increase the amplitude quency within the bandwidth of the measurements.
of response of the entire structure. This is commonly
called a resonance condition. Conversely, if either condi- In the Theoretical Background section, it was shown that
tion is not met, the mode will not participate in the forced sinusoidal excitation is equivalent to selecting the FRF val-
response of the structure. ues at the excitation frequency, from the column of FRFs
corresponding to the reference DOF.

This is also true for the other types of frequency domain

functions listed above. In other words, any set of vibration

Page 6 of 8
11th IMAC Conference, Kissimmee, FL February 1993

data taken from a structure is the result of applied excitation

forces. Whether it be operating data, caused by self excita- Transmissibility
tion, or data taken during a modal test, under tightly con- Transmissibility measurements are made when the excita-
trolled excitation conditions, the operating deflection shapes tion force(s) cannot be measured. Transmissibility is a 2-
are always subject to both the amount and location of the channel measurement like the FRF. It is estimated in the
excitation. same way as the FRF, but the response is "normalized" by a
reference response signal instead of an excitation signal.
Linear Spectrum Phase is also preserved in Transmissibility's, and a set of
This frequency domain function is simply the FFT of a them need not be obtained by simultaneously sampling all
sampled time domain function. Phase is preserved in the of the time domain responses. Each (response, reference
Linear Spectrum, so in order to obtain operating deflection response) pair must be simultaneously sampled, however.
shapes from a set of Linear Spectra, the time domain signals
As with FRFs, a set of Transmissibility's contain both mag-
must be simultaneously sampled. Since the Linear Spec-
nitude and phase at each frequency, so operating deflection
trum is complex valued (contains both magnitude and phase
shapes derived from a set of Transmissibility's will also
information), the resulting operating deflection shapes will
contain magnitude and phase information. The units of the
also contain magnitude and phase information.
operating deflection shapes are response units per unit of
response at the reference DOF.
Auto Power Spectrum
This frequency domain function is derived by taking the Figure 5 shows the operating deflection shape taken from a
FFT of a sampled time domain function and multiplying the set of FRFs for the plate structure at 780 Hz, in between the
resulting Linear Spectrum by the complex conjugate of the 3rd and 4th modes. Figure 6 shows the operating deflection
Linear Spectrum at each frequency. Phase is not preserved shape for 760 Hz, close to the resonant frequency of the 3rd
in the Auto Power Spectrum, so a set of these measurements mode. Since, the motion at this frequency is dominated by
need not be obtained by simultaneously sampling all of the the mode shape of the 753 Hz mode, the operating deflec-
time domain responses. Since phase is not retained in these tion shape "looks like" the 753 Hz mode shape, but its val-
measurements, operating deflection shapes derived from ues will be different. Figure 7 shows the operating deflec-
them will contain only magnitude, and no phase infor- tion shape for 800 Hz, close to the resonant frequency of the
mation. 4th mode.

FRF
The FRF is a 2-channel measurement, involving a response
and an excitation signal. It can be estimated in several
ways, depending on whether the excitation or the response
has more measurement noise associated with it. The most
common calculation involves dividing the Cross Power
Spectrum between the response and excitation signals by
the Auto Power Spectrum of the excitation, at each frequen-
cy. Averaging of several Cross and Auto Power Spectra
together is also commonly done, to reduce noise. Phase is
preserved in the FRF, but a set of FRFs need not be ob-
tained by simultaneously sampling all of the time domain Figure 5. Operating Deflection Shape from FRFs at 780 Hz.
responses. Each (response, excitation) pair must be simul-
taneously sampled, however.
Since a set of FRFs contains both magnitude and phase at
each frequency, the operating deflection shapes derived
from a set of FRFs will also contain both magnitude and
phase information. The units of the operating deflection
shapes are acceleration, velocity, or displacement per unit
of excitation force at the reference DOF.

Figure 6. Operating Deflection Shape from FRFs at 765 Hz.

Page 7 of 8
11th IMAC Conference, Kissimmee, FL February 1993

REFERENCES
[1] Potter, R. and Richardson, M.H. "Identification of the
Modal Properties of an Elastic Structure from Measured
Transfer Function Data" 20th International Instrumentation
Symposium, Albuquerque, New Mexico, May 21-23, 1974.
[2] Døssing, Ole "Structural Stroboscopy-Measurement of
Operational Deflection Shapes" Sound and Vibration Maga-
zine, August 1988.

Figure 7. Operating Deflection Shape from FRFs at 785 Hz.

CONCLUSIONS
Operating deflection shapes were defined for both time and
frequency domain functions. Operating deflection shapes
and mode shapes were compared, and it was shown that
they are quite different, yet related to one another.
• Operating deflection shapes depend on both the amount
and the location of excitation forces on a structure,
whether the forces are known or not. Modes shapes do
not depend on excitation forces, but are natural proper-
ties of a structure.
• Operating deflection shapes show how much a structure
is really moving. Mode shapes have no unique value,
so they cannot be used directly to determine how much
a structure is moving.
• Operating deflection shapes are functions of the modal
properties (frequencies, damping, and mode shapes) of
a structure. If the amounts and locations of excitation
forces are known, then modal properties can be used to
synthesize operating deflection shapes. For excitations
close to the modal frequencies of a structure, its mode
shapes will closely approximate its operating deflection
shapes, but only in "shape", not in value.
There are many other details associated with manipulation
of time and frequency domain measurement data, in order
to animate and compare the operating deflection shapes and
mode shapes of a structure. These will be presented in
forthcoming papers.

Page 8 of 8