IMPACT TESTING AND ITS APPLICATIONS

By
Ming Xu and Frank A. Findley
ENTEK IRD International Corporation
ABSTRACT

The effects of structural resonances on rotating machinery can dramatically

reduce machine life and machine performance. Impact testing aids in
understanding how forces are transmitted throughout a structure or mechanical
system. In this paper, the setup of impact testing and frequency response function
are described. The signal processing, data integrity check and three types of bad
impacts are discussed in detail. The measurement considerations, including
transducer mounting, FRF display and dynamic stiffness, are also discussed. Two
case studies of impact testing on machine tool spindles are presented. The case
studies show that impact testing can effectively troubleshoot vibration problems
in machine tool applications. For new spindles of the same model, the test results
show that the frequencies at which the lowest dynamic stiffness occurred were
more consistent than the magnitudes of the measured dynamic stiffness.

INTRODUCTION

The impact testing is an effective way to troubleshoot resonance related

problems. Resonance is the condition where the frequency of the disturbing force
coincides with the natural frequency of a structure or mechanical system. In
rotating machinery, the disturbing force can be induced by unbalance,
misalignment or other mechanical faults. Impact testing aids in understanding
how forces are transmitted throughout a structure or mechanical system. The
natural frequencies of a structure, machine or component are determined by its
mass, stiffness and damping characteristics. Thus, the natural frequencies are
inherent properties of a mechanical system. Sometimes, the natural frequency is
also called resonance frequency. The system response is determined by both the
input and the characteristics of the system, as illustrated in Figure 1.

Figure 1. System block diagram

There are basically two types of vibration measurement:
(1). only response output is measured;
(2). both input and response output are measured.
It is obvious that the system characteristics, such as natural frequencies of the
system, can be determined completely when both the input and response are
measured. In the traditional “bump test”, only system response is measured and,
at times, it is difficult to say whether a particular large response level is due to a
strong excitation or to a resonance of the system. In the newly developed
dataPAC™ 1500 FRF (Frequency Response Function) module, the capability of
simultaneously measuring both input and response is provided. The system
natural frequencies and other properties can be determined directly from the
measured data.

In the machining industry, the importance of dynamic stiffness measurement is

emphasized in recent years because it reflects the machine component’s vibration
characteristics under the cutting condition. Dynamic stiffness is the ratio of the
force amplitude to the displacement amplitude at a certain vibration frequency. In
general, dynamic stiffness is a function of frequency. The vibration problems in a
machining system, which includes machine tool structure, workpiece, tooling and
fixture, are among the major obstacles to greater productivity and quality of the
machined parts. The vibratory motion between a cutting tool and a workpiece is
recorded on the workpiece surface. Depending upon the severity of the
oscillatory motion and the surface finish specifications, the surface quality of the
parts may be unsatisfactory. The high frequency vibrational condition known as
chatter may seriously decrease tool life. In addition, chatter also affects the life of
bearings and other machine components. [1] It is necessary to solve a vibration
problem if unsatisfactory tool life, surface finish, or noise result. The minimum
value of the dynamic stiffness and its frequency is of primary interest since it
corresponds to a resonance condition and the machine component will have a
tendency to naturally vibrate at that frequency. This lowest dynamic stiffness
represents a dynamic weak point of that structure or machine component.
Dynamic stiffness is one of the frequency response functions and can be
determined by impact testing.

In this paper, the impact testing setup and selection of impact hammer and
transducer are discussed. The characteristics of various hammer tip materials are
compared and the selection of sensitivities of both hammer and transducer are
discussed. The definition and six representations of frequency response function
are described. The signal processing techniques, including pre-triggering, force
2
and exponential windows, are discussed. To ensure the quality of the acquired
data, the use of coherence function and the detection of three types of bad
impacts, namely overload, underload and double hit, are discussed in detail. The
measurement techniques, such as accelerometer mounting methods, the different
graphic displays of testing results, as well as static and dynamic stiffness are also
discussed. Two case studies of impact testing on machine tool spindles are
presented. The case studies illustrate how to troubleshoot tool breakage problem
and the impact testing data of several new spindles.

IMPACT TESTING SETUP

To perform impact testing, a calibrated force hammer is used to strike or excite

the structure being tested and an accelerometer to record how the structure
responds to the excitation. The measurement setup is shown in Figure 2. An
instrumented force hammer is connected to a power unit and then to the
reference channel (REF) of dataPAC™ 1500. The response signal is measured by
Deleted: ¶
an accelerometer. The accelerometer is connected to the vibration input channel
Deleted: dataPAC 1500
of dataPAC™ 1500.

Figure 2. Measurement setup

Hammer Selection

A typical impulse-force hammer is illustrated in Figure 3. It consists of a rigid

mass with a force transducer (load cell) attached to one end and an impactor tip
attached to the opposite side of the force transducer. A set of impactor tips made
from different materials, such as steel, plastic (vinyl) and rubber, and
interchangeable mass can be selected to obtain a desired force pulse for testing
different structures, machines or components.

3
 Figure 3. Typical impulse-force hammer

The hammer impulse consists of a nearly constant force over a certain frequency
range and is capable of exciting all resonances in that range. The hammer weight
and tip hardness determine the amplitude and frequency range of the force
impulse. The frequency range is directly proportional to the hardness of the tip
and inversely proportional to the weight of the hammer. Since the weight of the
hammer determines the magnitude of the force pulse, the hammer is usually
chosen for its weight and then the tip hardness is varied to achieve the desired
pulse time duration or frequency range. The effects of tip hardness on the
frequency content are illustrated in Figure 4 (a) through (d).

(a). Hard (steel) tip (b). Medium (plastic) tip

4
 (c). Soft (rubber) tip (d). Supersoft (rubber) tip

Figure 4. Force time wavefroms and spectra with various tip materials

The measurement conditions for Figure 4 (a) through (d) are identical except the
tip material. For comparison purpose, the maximum frequency (Fmax) was set at
60,000 cpm or
1 kHz for all four cases. A steel tip can provide a short duration force which
excites higher frequency modes. As shown in Figure 4 (a), the magnitude of the
force frequency spectrum maintained at almost the same level in the measured
frequency range. A rubber tip, on the other hand, concentrates more energy at
lower frequencies. As shown in Figure 4 (c) and (d), the magnitudes of the force
spectra drop as frequency increases, especially in the case of supersoft tip. A
plastic tip falls somewhere between steel and rubber. As shown in Figure 4 (b),
the force level maintained fairly constant before 40,000 cpm and started to drop
down after that. In general, the softer the impactor tip, the lower the frequency
range of excitation. Another consideration in tip selection is to use appropriate
tip material so that there is no permanent damage on either the tip or the test
structure.

The types and sensitivities of the force hammer can be selected and entered in
dataPAC™ 1500 FRF measurement setup menu. Typical operating characteristics
of force hammers are listed in Table 1. The force range of hammers can be
chosen from 100 lbf (440 N) up to 5000 lbf (22000 N). The corresponding
frequency can be 8 kHz for small and medium size hammers or 5 kHz for large
hammer. Generally, the maximum voltage output of a hammer force is 5 Volts.
The selection of sensitivity is inversely proportional to the force range, i.e.,
smaller sensitivity should be selected for larger force range. The calibrated
sensitivity of each hammer will deviate slightly from the nominal values shown
in Table 1 due to different hammer construction.
5
 Table 1. Typical force hammer properties
Force range, lbf (N) 100 (440) 500 (2200) 1000 (4400) 5000 (22000)
Frequency range, Hz 8000 8000 8000 5000
Sensitivity, mV/lbf 50 (12) 10 (2.3) 5 (1.2) 1 (0.23)
(mV/N)

Transducer Selection

Accelerometers are the widely used transducers for the impact testing. In the
selection of accelerometers, the weight and sensitivity of an accelerometer are
two important factors to be considered. If the weight of an accelerometer is
heavy relative to the test structure, it may cause “dynamic mass loading” and
alter the dynamic characteristics of the structure under test. Dynamic mass
loading is the change in physical properties of the structure caused by adding the
mass of the accelerometer. In impact testing, low weight accelerometers should
be used to minimize the effect of dynamic mass loading.

The selection of accelerometer sensitivity depends upon the response amplitude

of the structure under test. In general, the maximum amplitude range is inversely
proportional to the sensitivity of the transducer, as shown in Table 2.

Table 2. Accelerometer sensitivity versus amplitude range

Sensitivity, mV/g 10 50 100 500
Amplitude range, g peak 500 100 50 10

It is not unusual to generate large acceleration (g) readings in impact testing.

Thus, accelerometers with lower sensitivity, such as 10 mV/g or 50 mV/g, are
generally used to avoid saturation or clipping of transducers. In some machine or
structure, low frequency response is of primary interest. In this particular case, high
sensitivity accelerometers (500 mV/g) may be used with great care to prevent
from saturation problem. Also, the soft or supersoft tip should be used to
concentrate energy in the low frequency range.

FREQUENCY RESPONSE FUNCTIONS

6
Impact testing is performed to determine the frequency response function (FRF)
or transfer function. For a typical single input/single output system shown in
Figure 5, the time waveform of an impact force to the system is u(t) and the time
waveform of the system response is v(t). The Fourier transforms (frequency
spectra) for input u(t) and output v(t) are U(f) and V(f), respectively.

Figure 5. Single input/single output system

The frequency response function H(f) is defined as the ratio of the Fourier
transforms of the system output V(f) to the system input U(f) and given by:

OUTPUT V ( f )
H( f ) = =
INPUT U( f )

It is required for the frequency response function H(f) that the input force
spectrum U(f) is non-zero at all frequencies of interest. Depending upon the
selected parameters for the system response output V(f) and inverse relationship,
there are six representations of frequency response functions as follows:

Accelerance (g/lbf or g/N) = Acceleration / Force

Effective Mass (lbf/g or N/g) = Force / Acceleration
Mobility (in/lbf⋅sec or m/N⋅sec) = Velocity / Force
Mechanical Impedance (lbf⋅sec/in or N⋅sec/m) = Force / Velocity
Dynamic Compliance (in/lbf or m/N) = Displacement / Force
Dynamic Stiffness (lbf/in or N/m) = Force / Displacement

where g is acceleration due to gravity, and g is equal to 386 (in/sec2) or

9.81(m/sec2).
The dataPAC™ 1500 FRF module can calculate and display any one of the six
FRF functions. Table 3 summaries all six forms of the FRF’s.

Table 3. Six Frequency Response Functions (FRF’s)

Response Parameter Response/Impact Impact Force/Response
7
 Force
Acceleration Accelerance Effective Mass
Velocity Mobility Mechanical Impedance
Displacement Dynamic Compliance Dynamic Stiffness

If the system output is measured by an accelerometer in g’s and hammer input

force in pounds, the conversion relationship of different FRF’s is shown in
Figure 6.

Figure 6. Conversion relationship between different FRF’s

Depending upon the source of measurement noise, three different algorithms,

namely H1, H2 and Hv [2], are used in dataPAC™ 1500 FRF module to estimate
the FRF. If noise in a measurement is assumed in the output signal, H1 algorithm
is a better choice because H1 attempts to minimize the error on the output. On the
other hand, H2 algorithm is suitable for noise in the input signal since H2
minimizes the error on the input. If noise is assumed in both input and output
signals, Hv gives the best estimation of FRF. Generally, H1 technique
underestimates the amplitude at resonances and causes damping to be
overestimated. In contrast, H2 overestimates the amplitude at resonances and
causes damping to be underestimated. At anti-resonances, H1 overestimates the
amplitude and H2 underestimated the amplitude. At both resonance and anti-
resonance, Hv yields the best estimation of FRF amplitude. The FRF estimators
can be selected from the display setup menu of dataPAC™ 1500 FRF module, as
shown in Figure 7.
8
 Figure 7. Selection of FRF estimators

Frequency response functions are properties of linear dynamic systems which do

not depend on the type of excitation function. Excitation can be harmonic,
random, or transient functions of time. The test results obtained with one type of
excitation can thus be used for predicting the response of the system to any other
type of excitation. [3] The impact testing falls into the class of transient
excitation. This technique is the simplest and fastest technique for obtaining good
estimates of the required frequency response information. However, the impact
testing is ill-suited for nonlinear systems because of its very high ratio of peak
level to total energy. In practice, linearity of the system is a condition which will
be met only approximately, depending on the type of system and on the
magnitude of the input. Care has to be taken to avoid non-linear effects when
applying impulse excitation. Typical examples of structural non-linearities
include clearance or looseness between parts, nonlinear damping associated with
mechanical joints, and load-sensitive stiffness. Structures which are known to be
non-linear should not be tested with impulse excitation.

SIGNAL PROCESSING

The frequency response functions are measured by simultaneously sampling both

impact force and system response signals. The captured time waveforms of both
9
impact force and system response contain amplitude and relative phase
information. The frequency response function and coherence function can be
computed from the Fourier transforms of both impact and response signals. The
block diagram of signal processing is shown in Figure 8.

Figure 8. Block diagram of signal processing

Pre-Trigger
Pre-trigger is most appropriate for impact testing to capture the impact signal
from a hammer strike because the waveform occurring before the trigger point is
captured and displayed. In the pre-trigger mode, a negative trigger delay is used
to acquire and display the measured data a period of time before the trigger point,
as illustrated in Figure 9 (a). Triggering is used to initiate data acquisition when a
signal satisfies the trigger condition. The trigger point is determined by trigger
force level and slope, as shown in Figure 9 (b). The trigger level and slope can be
specified in the FRF setup menu of dataPAC™ 1500.

(a). Pre-trigger (b). Trigger force level and slope

Figure 9. Pre-trigger and trigger setting

Force Window

In general, it is desirable to have a forcing function that is short in time duration

to provide a good estimate of the FRF for the system under test because a short
10
duration signal in the time domain corresponds to a broad-band frequency
content in the frequency domain. Since the impact force signal represents a very
small portion of the digital record analyzed, even very low level noise may
produce significant errors in the autospectrum of the force signal and FRF
estimation. This is especially true if the duration of the force signal is short
relative to the total sample time. Such noise is not reduced by averaging with the
usual FRF estimate because the noise is additive in the force autospectrum.
Random noise can be greatly reduced by multiplying a force window to the force
signal.

The shape of a typical force window is shown in Figure 10. The force window
has unity gain for a portion of the record which includes the force signal, and sets
the remaining samples in the record to zero prior to Fourier processing. This
procedure introduces no distortion as long as none of the force data is attenuated
and the background noise is broad-band in nature. The data record length is T
and the window start time is To which is related to the pre-trigger setting. Two
cosine tapers (curves AB and CD) are added to both rising edge and trailing edge
of the narrow rectangular window. The tapers have the effect of smoothing the
data and sharpening the spectral window.

Figure 10. Force window

Exponential Window

Noise problems may also be encountered in the response signal, particularly

when dealing with heavily damped structures. In this case, the duration of the
response signal may be short relative to the total sample time, so that noise may
comprise a significant portion of the total energy in the time-sample even with
relatively high signal-to-noise ratios. On the other hand, leakage error in
response signals may be encountered when testing lightly damped structures. In
this particular case, the response signal does not significantly decay during the
11
sampling period. It may require a very large data record length to avoid
truncation errors. In order to reduce the errors in both situations, a time-domain
exponential decay window is used for the response signal. The shape of a typical
exponential window is shown in Figure 11. The data record length T and the
window start time To are the same as those of the force window. A cosine taper
(curve AB) is also applied to the rising edge of the window to avoid abrupt
starting and smooth the data.

Figure 11. Exponential window

CHECK DATA QUALITY - COHERENCE FUNCTION

To determine the quality of the FRF estimation, it is not sufficient to know only
the relationship between input and output. The question is whether the system
output is caused by the system input or by other factors. For example, noise
and/or non-linear effects can cause large outputs at various frequencies, inducing
errors in estimating the frequency response function. The influence of noise
and/or non-linearities, and thus the degree of noise contamination in the FRF is
measured by calculating the coherence function.

The coherence function can be calculated from the averaged autospectrum and
cross spectrum of the impact force and the system response, as shown in Figure
8. The coherence function is dimensionless and its value is always between 0 and
1. If the coherence is equal to 1 at any specific frequency, the system is said to
have perfect causality at that frequency. In other words, the measured response
power is caused totally by the measured input power and the system is linear. If
the value of coherence function is equal to 0, the two signals are totally
uncorrelated. A coherence value less than 1 at a given frequency indicates that
the contamination exists in the two signals in terms of noise and nonlinear
effects. Thus, coherence values help to make a judgment on whether a
measurement is acceptable or should be rejected. Generally, the measured data is
good or acceptable if the coherence value is greater than or equal to 0.75.
12
Figure 12 shows a typical coherence function (top trace) and corresponding
dynamic stiffness (bottom trace). In this example, good coherence values are
obtained over the frequency range of interest except at two locations, those being
at the points of high dynamic stiffness or anti-resonance frequencies. In
applications, low coherence at anti-resonance frequencies is not generally a cause
for concern about data quality. At such frequencies the machine or structural
response is very small and may be near the noise floor of the instrumentation.
Thus, the low coherence merely indicates the reduced signal-to-noise ratio. On
Deleted: it
the other hand, one should be aware that the coherence would always be unity for
Deleted: a
single impact estimation. In most cases, four or five impacts are needed to
average impact testing data to assure the FRF measurements are valid. If there is
no big change in both coherence and any one of the six FRF’s, the measurements
are stable and can be used.

Figure 12. Display of coherence and dynamic stiffness

CHECK BAD IMPACTS

In addition to the coherence measurement, bad impacts should also be checked

on every impact measurement to ensure data integrity. Three types of bad
impacts are common problems in impact testing, namely overload, underload,
and more than one impact (also referred to as double hit). Any measurement data
from bad impacts should be rejected and excluded from the average process.

Overloading is the most difficult problem to overcome in impact testing. When

an overload occurs, the impact force signal is clipped, causing distortions in the
measurement. In some cases, anti-aliasing filter may smooth a clipped signal to
13
make distorted measurement look good. To detect overload problem, the
unfiltered force signal should be checked before it passes through anti-aliasing
filter. In the FRF module of dataPAC™ 1500, the overload force level can be
specified in the measurement setup menu. The overload limit can be set in
pounds or Newtons based on the experience of testing similar types of machines.
If the impact force exceeds the limit, the dataPAC™ 1500 gives an overload
warning, as shown in Figure 13. In this particular example, the overload limit is
set at 200 pounds.

The opposite problem is underload due to low impact force level or improper
gain setting. If the impact force is not adequate, the measured frequency response
function does not represent the dynamic characteristics of the measured system.
Therefore, underload needs to be checked before signal processing. In the FRF
module of dataPAC™ 1500, the underload force level is specified as a percentage
of the overload limit. As shown in Figure 14, the underload limit is set at 25% of
overload or 50 pounds. In this particular case, the impact force level is below the
limit and the dataPAC™ 1500 gives a underload warning.

Figure 13. Overload display of a unfiltered impact force

14
 Figure 14. Underload display of a unfiltered impact force

It is important to avoid more than one impact or impactor rebounds. If more than
one impact occurs within the data record, the Fourier transforms of the pulses
tend to cancel at certain frequencies, creating sharp notches in the force spectrum
[4], as shown in Figure 15. This can cause significant errors in the FRF
measurement at these frequencies due to a low signal-to-noise ratio in the force
spectrum. Furthermore, the force window can not eliminate secondary impacts
from the force record prior to Fourier processing. In fact, the response will still
include the effects of the multiple impacts, thus resulting in an erroneous
estimate of the frequency response function.

Figure 15. Time waveform and frequency spectrum of double hit

Multiple impacts sometimes occur, especially when testing weak, lightly damped
structures. In such cases, the structures bounce back against the impactor before
it can be drawn away after the initial impact. The problem of more than one
15
impact or double hit can be detected by checking any secondary peak(s) within a
data record. If the amplitude of any secondary impact is greater than a preset
percentage level of the primary impact, the dataPAC™ 1500 FRF module gives a
double hit warning. An example of more than one impact is shown in Figure 16.
The double hit limit is set at 20% of the primary impact as indicated by the dash
line. The warning of double hit is showed up due to clear multiple impacts.

Figure 16. Unfiltered force signal with more than one impact

Deleted: It is important to avoid

impact hammer rebounds. If more
MEASUREMENT CONSIDERATIONS than a single impact occurs within
the data record, the Fourier
transforms of the pulses tend to
In addition to the data integrity check, the accuracy of the measurement using a cancel at certain frequencies,
creating sharp notches in the force
hand-held impact hammer depends heavily on the ability of the operator to spectrum. This can cause
significant errors in the FRF
maintain the correct location and direction of impact. The hammer should strike measurement at these frequencies,
due to a low signal-to-noise ratio
the structure under test in the same plane as the axis of sensitivity of the response in the force spectrum. Furthermore,
the force window should not be
accelerometer. The setup of FRF measurement parameters, such as frequency used to eliminate secondary
impacts from the force record prior
range (Fmax), number of lines and average type, are also important. As discussed to Fourier processing. In this case,
the response will still include the
earlier, the frequency range depends on the tip stiffness and the hammer mass. effects of the multiple impacts,
thus resulting in an erroneous
The frequency range of a given hammer can be reduced either by decreasing the estimate of the frequency response
function. An option in the menu is
tip stiffness or by increasing the hammer mass. The frequency range also provided to let the instrument beep
whenever overloading,
depends on the effective stiffness and mass of the test structure at the point of underloading, or hammer rebounds
occur.¶
impact. Low structure stiffness limits the increase in frequency range achievable
by increasing the tip stiffness. In this case, a more effective way of increasing the
frequency range of the excitation is to reduce the mass of the hammer. In FRF
measurements, a number of averages are necessary to obtain valid data. In some
cases, no average is needed, such as time waveform of impact force or response,
and real or imaginary display of the FRF. The dataPAC™ 1500 FRF module
16
allows the operator to setup desired Fmax, number of lines and average type, as
shown in Figure 17 (a). To setup response channel, the dataPAC™ 1500 allows
the operator to make a few trial impacts and record the highest response level, as
shown in Figure 17 (b). The suggested range will be used in the FRF
measurements.

(a). Measurement setup (b). Response channel setup

Figure 17. FRF measurement setup menu

Transducer Mounting

There are various means of mounting the accelerometer to the surface of the test
structure, such as stud mounting, magnet mounting or using a thin layer of wax
[5]. In order to minimize the loss of high frequency signals in the transmission
path, the contact surface should be flat, smooth, clean and free of rust and paint.
The stud mounting is the most

reliable method and it allows more transmission of high frequency signal.

However, it is not always possible to tap a mounting hole on the structure under
test because it requires the appropriate modification of the structure. In such
cases, the use of a magnetic mounting base or a thin layer of wax can be
alternative methods of attachment.

Magnet mounting offers a convenient, temporary attachment to magnetic

surfaces, but the first resonance of magnet mounting is much lower than that of
the stud mounting. Therefore, the operational frequency range of an
accelerometer is decreased. Petro wax is also a convenient, temporary mounting
method. Test results show that frequency response equivalent to stud mounting
can be achieved with a thin coating of wax on good flat surfaces. [6] As the
thickness of wax at the interface of the accelerometer and test object increases,
17
the first resonance of the system decreases, which limits the frequency response
of the test structure. To obtain the best frequency response on flat surfaces, one
should use as little wax as possible at the interface and press the accelerometer
firmly against the test structure using a circular motion to secure mounting.

FRF Display

FRF is a complex function and three quantities are involved, namely frequency,
magnitude and phase. This is equivalent to frequency, real part and imaginary
part. These three quantities can not be fully displayed on a standard two-
dimensional (2-D) graph because any 2-D plot can only shows two of the three
quantities. The interpretation of impact testing data can be greatly enhanced by
the use of a suitable graphic display. The display formats of the dataPAC™ 1500
FRF module include the following: (a). time waveforms and frequency spectra of
impact force and response; (b). magnitude, phase, real part and imaginary part of
six FRF’s (see Table 3); and (c). coherence function. A linear or logarithmic type
display for FRF’s may be selected to display data. However, only linear
amplitude type display is used for time waveform, phase, and coherence
function.

The time waveforms and frequency spectra of impact force and system response
are shown in Figure 18 (a) and (b). The improper setting of Fmax can be detected
by inspecting the frequency spectrum of the force signal. The hammer impact
excites the machine component or structure over a wide range of frequencies and
the frequency spectrum is nearly flat over this frequency range. In the useful
frequency range, the force level should be high enough to excite all the
resonances. If the impact force drops to a certain level before it reaches the Fmax,
it presents a problem of improper setting of Fmax. In this case, either the impactor
tip or mass needs to be changed or the Fmax should be lowered to ensure
sufficient energy level in the interested frequency range. Figure 18 (a) shows a
typical impact force in both time and frequency domain. As a rule-of -thumb, the
Fmax setting is acceptable if the force level at this frequency has not dropped
significantly below 20% of the level at Fmin. In addition, if the force level has
virtually no drop in amplitude at the desired Fmax, the selected hammer tip
may be too stiff and a softer tip may be more suitable. In the example shown in
Figure 18 (a), the hammer and tip were ideally suited for a Fmax in the range of a
few hundred Hertz up to 2500 Hz.

18
 (a). Impact force (b). System response
Figure 18. Time waveforms and frequency spectra

The Bode type plot is commonly used form of FRF presentation. It consists of
two graphs: magnitude vs. frequency and phase vs. frequency. A typical dynamic
stiffness plot is illustrated in Figure 19. In this example, logarithmic magnitude
and linear frequency are used for the plot. At the resonant frequency (2656 Hz),
low dynamic stiffness and a phase shift of approximately 1800 are shown in the
plot. The coherence function shown in Figure 12 is also used to examine the
quality of the measurement data. In this particular case, the coherence is near 1.0
at the frequency 2656 Hz indicating a direct relationship between the impact and
response.

Figure 19. Magnitude and phase of dynamic stiffness

Depending upon measurement parameters, either the imaginary part or real part
of FRF can be used to plot the mode shape. A mode shape can be thought of as a
photograph of the structure deflection curve at the instant of maximum strain at a
given natural frequency. The natural frequencies and mode shapes can be
measured by impact testing or calculated from the free vibration equations. The
19
mode shape is different from the operational deflection shape (ODS). ODS is the
structure deflection under a forcing frequency, such as unbalance or other
external forces. A mode shape is structure deflection at a natural frequency. An
example of measuring mode shape is shown in Figure 20 and Table 4.

(a). Impact at location A (b). Impact at location B

(c). Second bending mode

Figure 20. Mode shape - the imaginary part of accelerance

In this particular case, the imaginary part of the accelerance is measured by

fixing the accelerometer at one location and moving the impact hammer along
the mesh layout of the

Table 4. Peak values of the imaginary part of accelerance vs. coordinates of a

cantilever beam
Coordinates, 0.5 1.0 1.5 2.0 2.5 3.0 3.5 4.0 4.5
in
Peak Values, 0.00 0.36 0.91 1.28 1.81 2.11 2.40 2.51 2.56
20
 g/N
Coordinates, 5.0 5.5 6.0 6.5 7.0 7.5 8.0 8.5 9.0
in
Peak Values, 2.43 2.08 1.71 1.12 0.64 - - - -
g/N 0.25 0.79 1.67 2.31

structure. According to Maxwell’s reciprocity, the same results will be obtained

if the impact location remains fixed and the accelerometer is moved. At the
frequency 2656 Hz, the amplitude when impact occurred at location A is 2.469
g/N and location B 3.663 g/N, as shown in Figure 20 (a) and (b). A set of peak
values of the imaginary part of accelerance and the coordinates of the
corresponding data points can be obtained by using this approach, as illustrated
in Table 4. The measurement results can be used to plot the mode shape, as
shown in Figure 20 (c). In this example, the second bending mode of a cantilever
beam is plotted and one nodal (zero displacement) point is located near 7.5 in. In
applications, the number of data points should be sufficient to avoid spatial
aliasing problem. [7] The real part of the mobility or imaginary part of the
dynamic compliance can also be used to measure the mode shape. The measured
mode shape is determined to within a constant multiplier.

Static and Dynamic Stiffness

One of the FRF applications is to measure dynamic stiffness of a machine or

component. The difference between the static stiffness and dynamic stiffness is
whether the stiffness value changes as the frequency changes. The dynamic
stiffness is a function of frequency, but the static stiffness remains a constant.
The static stiffness or stiffness of a structure can be defined as the amount of
force required to produce a unit of displacement. The force is measured in
pounds (lbf) or Newtons (N) and the displacement is measured by inches (in) or
meters (m). The stiffness can be calculated by the following equation:

F
K=
X
where:
K = stiffness, lbf/in or N/m
F = force, lbf or N
X = displacement, in or m

21
For example, if a machine deflects 0.001 inches under a steady load of 900
pounds, the static stiffness is:
900
K= = 900,000 lbf / in
0.001

The analysis of the behavior of structures under dynamic loads is far more
complicated than that of structural behavior under steady loads. The basic
principle is that at very low frequencies the structure responds to dynamic loads
the same as it does to static loads. [1] However, as the forcing frequency
increases, the stiffness may change from that of the steady state value. In general,
dynamic stiffness is a function of frequency f and can be expressed by the
following equation:
F( f )
Kd ( f ) =
A( f )
where:
Kd(f) = dynamic stiffness at frequency f, lbf/in or N/m
F(f) = force amplitude at frequency f, lbf or N
A(f) = displacement amplitude at frequency f, in or m

At a certain vibration frequency, say f = fo, the calculation of dynamic stiffness is

the same as that of static stiffness. In other words, it is simply the ratio of the
forcing function amplitude to the amplitude of the resulting displacement
measured at the frequency fo. Examples of dynamic stiffness measurements will
be given in the case studies.

CASE STUDIES

Two case studies of dynamic stiffness testing on machine tool spindles are
presented. The first case study deals with troubleshooting a tool breakage
problem. The second case study discusses the impact testing data of several new
spindles.

Case 1
22
 The first case study concerns a machining system which includes a motorized
spindle, tooling, part and part-clamping apparatus, as shown in Figure 21. The
spindle was mounted at 450 angle to the part flow direction and was fixed at the
rear end. The spindle was used for finish drilling a hole on aluminum parts of
engine water pumps.

This spindle had frequent tool breakage problem. To find out the possible cause
of the problem, both dynamic stiffness and vibration data were measured on the
spindle. The dynamic stiffness at the spindle nose was 78.764 lbf/mils at 360 Hz,
at the end of the tool holder was 14.728 lbf/mils at 360 Hz, and at the part was
243.930 lbf/mils at 1290 Hz. The measured dynamic stiffness and coherence at
the end of the tool holder and the part are shown in Figure 22 and Figure 23.
Four averages were used in the coherence calculation and the coherence values at
these measurement frequencies were greater than 0.994, indicating good
measurements. The vibration data were also obtained and the displacement
spectrum measured at the rear bearing of the spindle (point 2) is shown in Figure
24. The frequency spectrum showed clearly the harmonics of the spindle running
speed.

The measurement data indicated that the possible cause of frequent tool breakage
was due to improper spindle mounting arrangement. In this particular case, the
spindle was a cantilever

Figure 21. Schematic of a machining system

23
 Figure 22. Dynamic stiffness and coherence at the end of tool holder

beam. The dynamic stiffness at the end of tool holder was substantially lower
than the dynamic stiffness at the part, i.e., 14.728 lbf/mils vs. 243.930 lbf/mils.
Also, the frequency at which the lowest dynamic stiffness occurred was much
higher at the part (1290 Hz) as compared to the frequency at the tool holder (360
Hz). Therefore, the impact and other disturbances during the cut can induce
resonance causing vibratory motion of the tooling, but

Figure 23. Dynamic stiffness and coherence at the part

24
Figure 24. Displacement spectrum measurement at the rear bearing of the spindle
(point2)

the part was constrained with sufficient dynamic stiffness. In other words, the
part was much less susceptible to the resonance. The large difference in the
lowest dynamic stiffness and the corresponding frequencies may be the cause of
frequent tool breakage. The measured displacement spectrum also showed
typical symptom of looseness due to improper mounting of the spindle.

Case 2

In the second case study, the dynamic stiffness data were taken from nine
identical box spindles, as shown in Figure 25. These spindles were newly
manufactured and to be used for boring operation. The front and rear bearings of
the spindles were duplex set precision angular contact bearings. The operating
speed of the spindles was 1400 RPM. The overall vibrations, including
acceleration, velocity and displacement, were also measured at the front bearings
in the horizontal direction, i.e., point 1 in Figure 25.

Figure 25. Schematic of a box spindle

The impact testing was performed on these spindles. During the test, the spindles
were suspended. An impact hammer with plastics tip was used to give a force
impulse at the spindle nose. The response was measured by using a small size
accelerometer with sensitivity of 10 mV/g. The accelerometer was attached to the
inside of the spindle nose at the driving or impact point by using a thin layer of

25
wax. The dynamic stiffness and coherence of spindle No. 4 are shown in Figure
26. Four averages were used in the data collection.

In this example, the measured frequency range was from 0 to 4000 Hz and the
lowest dynamic stiffness was 165.22 lbf/mils at 1620 Hz. The coherence at this
particular frequency was 0.9996 indicating a direct relationship between the
impact and response. Similar results were obtained for all the spindles and listed
in Table 5. In addition, the static deflection under 200 pound axial load as well as
the overall vibration data at the spindle operating speed were also given in Table
5. The coherence values for all the tests were greater than 0.998.

Figure 26. Dynamic stiffness and coherence of spindle No. 4

Table 5. Testing data for the box spindles

Spindle Static Accel Vel Displacement Dynamic Frequency
Number Deflection, g in/s Mils Stiff- Hz
in ness, lbf/mils
1 0.00010 0.086 0.003 0.006 127.00 1630
2 0.00012 0.096 0.003 0.006 130.60 1510
3 0.00011 0.118 0.003 0.004 145.79 1570
4 0.00008 0.113 0.003 0.005 165.22 1620
5 0.00010 0.126 0.003 0.003 169.37 1600
6 0.00012 0.151 0.013 0.012 172.00 1490
7 0.00010 0.178 0.013 0.010 175.40 1590
8 0.00010 0.130 0.008 0.005 222.41 1610
9 0.00010 0.207 0.008 0.007 321.00 1550
26
The static deflection data were fairly consistent at 0.00010 ± 0.00002 in. It is
interesting to notice that the lowest dynamic stiffness for the spindles of the same
model under the same measurement conditions had wide variations. The largest
value in Table 5 was 321 lbf/mils and it was 253% higher than the smallest one
(127 lbf/mils). On the other hand, the frequencies at which the lowest dynamic
stiffness occurred had limited variations (± 4.5%) and were within 1560 ± 70 Hz
range. Thus, the frequency corresponding to the lowest dynamic stiffness may be
a more consistent indicator of the spindle resonance as compared to the
magnitude itself.

The measurement results also show that the overall acceleration of the spindles
increases as the lowest dynamic stiffness value increases. These two
measurement parameters had direct correlation, as shown in Figure 27. Similar to
coherence function, the correlation coefficient is used to measure the linear
dependence of two variables. If correlation coefficient is equal to 1, two variables
have perfect correlation. If it is equal to 0, two variables have no correlation. In
this particular case, the correlation coefficient was equal to 0.816, indicating a
strong correlation.

Figure 27. Correlation between overall acceleration and dynamic stiffness

CONCLUDING REMARKS

The impact testing can be used effectively to troubleshoot spindle resonance

related problems. The case study showed that the possible cause of frequent tool
breakage was due to the large difference in dynamic stiffness between the end of
tool holder and the part. The substantially low dynamic stiffness and associated
frequency made the tool holder more susceptible to vibrations. The problem may
be caused by the improper spindle mounting.

27
The impact testing results on nine new spindles of the same model showed large
variations in the magnitude of the lowest dynamic stiffness but fairly consistent
in the frequency range. This suggests that the frequency at which the lowest
dynamic stiffness occurs may be a better criterion for new spindle acceptance
tests with regards to the resonance. The measurement results also showed that the
overall acceleration increases as the lowest dynamic stiffness value increases for
the new spindles of the same model. Although the data sample size was limited,
the correlation between the two variables was fairly strong. Further studies are
needed to fully understand the causes of machine tool - workpiece vibration
problems and alternative approaches for their elimination.

ACKNOWLEDGMENT

The authors wish to thank John R. Birchmeier of ENTEK IRD International

Corporation for his contribution during this study. The authors also wish to
acknowledge Joseph M. Shea and Doug Lawrence of SETCO Company for their
contributions to the case studies.

REFERENCES

1. Machining Data Handbook, Section 22: “Machine Chatter and Vibration,”

Institute for Advanced Manufacturing Science (IAMS), Machine Ability Data
Center, Cincinnati (1992).

2. Brown, D. L., Ewins, D. J., and Lorenz, R. D., Modal Analysis: Theory and
Application, lecture notes, Society for Experimental Mechanics, Inc., Nashville,
Tennessee (February 9-11, 1995).

3. Halvorsen, W. G. and Brown, D. L., “Impulse Technique for Structural

Frequency Response Testing,” Sound and Vibration, Vol. 11, No. 11, pp. 8-21
(November 1977).

4. ISO 7626: Vibration and shock - Experimental determination of mechanical

mobility, Part 1 - Basic definitions and transducers (1986), Part 5 -
Measurements using impact excitation with an exciter which is not attached to
the structure (1994).

28
5. Ewins, D. J., Modal Testing: Theory and Practice, Research Studies Press Ltd.,
John Wiley & Sons Inc., New York (1984).

6. Quartz Sensors, Section 8: “Accessories,” PCB Piezotronics, Inc. Catalog G-

500, p. 165, Depew, New York (1991).

7. Singleton, K., “Tutorial on Experimental Modal Analysis,” Vibrations, Vol.

11, No. 2, pp. 3-9 (June, 1995).

8. Xu, M. and Birchmeier, J. R., “Dynamic Stiffness Testing and Its Applications
in Machine Tools,” Proceedings of the 8th Annual Predictive Maintenance
Technology National Conference, Indianapolis, Indiana, pp. TP 4-11 (December
3-6, 1996).

29