WINDOWS USED FOR IMPACT TESTING

Bill Fladung
Structural Dynamics Research Lab
University of Cincinnati
Cincinnati, Ohio USA

1. Introduction

Windowing is a common signal processing technique available on

all modem data acquisition systems used for modal testing. ‘Ihe
force and expauential windows arc tlmc domain weighting
functions that have been developed for the tramlent signals of
impact testing.“’ These windows are de&cd to reduce the
effects of leakage and improve the signal-to-noise ratio of the
measured signals. ln gaeral, to improve impact test@ FRF
measurements, the force and exponcotial w i n d o w s s h o u l d almost
always be applied to the time signals, with vev few exceptions.

.To correctly measure a frequency response function (FRF), the

exponential window must also be applied to the input signal, in
addition to the force window, if it is applied to the response
signals. The effect of the exponential window is to increase the
apparent damp@ of the measured system. This elfrct, which is
governed by the shit? property of the Laplace transform, is prr-
dictable and easily corrected in the estimated modal pammcters.

2. Causer and Symptoms of Leakage in Impact Testing

I.r&age is a signal processing hias ~ITTOT caused hy the violatlug

au assumption of the discrete Fourier trmsfonn. The assumption
IS that the signal to he transformed is periodic within the sample
period. For cases in which both the input and output are hannouic
func(ions of the sampling period rx arc completely observed
transirnts, there will be no leakage errors. Since real systems do
oat generally respond as multiples of some arbitrary sampling
frequnxy, windows are used to coustrain the signal to more
closely meet the requirement of a completely ohserved transient.
The rffkcts of leakage on measured FRFs is an underestimation of
the magnitude at the peaks ad distortion of the phase, and a drop
in coherence at the peaks,“’

For a signal to be a completely ohsrrvcd transient, it must start at

zero and end at zero wthiu the sampling period. For the (ransirnt
signals of impact testing, the input is always completely observ-
able, but the response may not he. If the response does not decay
to near zero by the end of the sampling period, the exponential

1662
window is applied to reduce leakage. Note that windowing can
only reduce the effects of leakage, not entirely eliminate theln.

3 . Use of the Exponential Window

The exponential window is simply an exponential function as

defined by either of the two forms in equation (I), where the
parameter 7 is the time constant of the exponential fmlction. The
time variable for the exponential function struts at zero, 0 T
regwdless if a pretrigger delay is used in the measurement.

The principal purpose of the exponential window is to reduce the The decay of the expouential window is typically detinrd in
commercially available data acquwtion software by one of the
effects of leakage on lightly damped response signals. The
transieut response of a lightly damped system will typically not several ways listed helow:
de+ to near zero by the end of the time record, as show ,n
Figure I(a). For lightly damped systems, the exponential wmdow l specifying the time constant (7) in sr&onds
should reduce the measured response signal at the end of the time l specifying the reciprocal parameter(B) in radJsrc
record to approximately one percent, as showo i n Figurc I(b); t o l specifying the value of the exponential function at the end of the
&ctively reduce the effects of leakage. time period as a percentage of unity
l specifying the value of the exponential function, as a percentage
of unity, at some point on the time axis, as a percentage of the
time record length
l graphically shaping the cxponentzd curve with the mouse

A common suggestion is that the time constant should bc one

quarter of the time record length, which creates an exponential
function that decays to about two percent at the end 01 the time
record. Ilov~evcr, this would not he appropriate for hravil)
damped systems, as explained ahow Also, note that if the
frequency span; number of spectral hnes, or any other measure-
ment parameter that affects the time record length is changed,
then T or ,3 must also be updatcd so that the shape of the exponen-
tial window is prcscrved.

4. Use of the Force Window

, /, The purpose of the force window is to improve the signal-to-noise

I ratio of the measured input by eliminating the noise on the signal
@I 1 following the duration of the impact. After the impact, the
T impactor is no longer in contact with the surfdce and can not
impart any excitation into the system The data in the trailing
segment of the unfiltered force signal consists of only electrical
uoise on the input channel. However, due to the short duration of
the impulse, the total energy of the noise may be on the same
The exponential window can also improve the signal-to-noise order as the energy of the input force. The force window passes
ratio of heavily damped response signals. It may seem that an the initial segment of the time record containiug the impact sign4
exponential window is not needed smce the trausirnt response and suppresses the noise in the remainder of the time record. The
decays very quickly, within the measured time record, and the tmiling segment of the filtered force signal also contains respouse
signal is a completely observed transxnt. IIowever, an exponen- to the anti-&sing filter, which should not he truncated by the
tial window is used in this case to attenuate the noise on the force window because it conldins energy of the input.
measured output after the response has decayed due to system
damping. For heavily damped systems, the exponential window The force window is unity over the leading five to ten percent of
should follow the damping of the system, as shown in Figure 2. the time record, has a steep cosine taper to zero, and is zero for
the remainder of the time record. ln some cases, the trailing
segment of the time record is set to the average value of the noise

1663
instead of zero. As will be discussed in a following section, the eliminated, then the periodic noise components can be removed
exponential window must also be applied to the force signal, in by signal processing techniques described Ref. 131.
addition to the force window. The force window and the
combination for the force and exponential windows are s h o w n i n
Figure 3.

ww
I+zquency Axs Frequency Ax,s

6 . Applying the Exponential Window to the Force Signal

Proper use of the exponential window requires that it must be

applied to the both response signals and the input signal.
\I Although an FRF is defined as a division of the output spectrum
0.1 T 0.2 T T by the input spectrum, applying the exponential window to both
the input and output does not cancel the effect of the window
Figure 3. The Jiircr window. from the FRF

The force window is defined by the duration of the leading unity The shit? propertves of the Laplace transform”’ govern the effects
portion, which is commonly refered to as the “length“ of the of the exponential window and illustrate why it must be applied
window In commercially available data acquisition soltware, the to both the input and output time signals. In the discussion below,
force window length is typically specified in either absolute time the Laplace variable (s) is used, but the Fourier tmnsfonn is
or as a percentage of the time record length. A duration of 5.10% equivalent to the Laplace transform evaluated at the imaginary
of the time record length, a&r the trigger, is usually (jo) axis.
recommended for the length of the force window. Note that the
force window length may or may not by default take into Multiplying a time signal y(t) b y an exponential function shifts
consideration the use of a pretrigger. the independent variable of the associated Laplace transform Y(S).

I f y(t) 0 Y(s), then e”‘y(l) = Y((s a ) (2)

5. Exception to the Rule
Shifting the independent variable of a time signal y(t)by an
The force and exponential windows should normally be applied to amount I, multiplies the associated Laplace tmnsfonn l’(s) by an
the time signals when using impact testing. The exception t o t h i s exponrnlial function in the s-domain.
rule is when the measured signals contain significant components
of periodic noise. In this case, the windows will smear the If Y(f) - w then y(f - 1,) 0 e-Y(s) (3)
periodic noise components, contaminating the adjacent spectral
lines. The periodic noise components are undamped and, if not Equation (2) govenx the effect of the exponential window and
windowed, appear in the spectrum as nal~ow peaks. However, the equation (3) governs the effect 01.the pretigger delay.
exponential window increases the apparent damping of the
measured signals, which results in broadening of the periodic The unwindowcd, equation (4), and exponentially windowed,
now~ peaks. The line shapes of the force and exponential equation (5), input .fit) and output x(t) time signals are trans.
windows arc show in Figure 4. The line shape of a time domain formed to the s-domain
window determines the properties of the window in the frequency
domain and X(f) Q X(s)
.X0 Q m) (4)
The typrs of periodic noise commonly acountered with impact CPlf(f) Q F(s + p) and e-KY(f) c, X(s + p) (5)
testmg include the DC-component, electrical line noise, and
permdlc excitation sources. Because of the frequency domain The transfer function, H(s), is defined as the ratio of the Laplace
effects of the windows, the periodic noise must be removed from
transforms of the output and input. For the case in which neither
the data before applying the windows in the time domain. If
the output or input time signals are exponentially windowed,
possible, the noise sources should be eliminated by appropriate
measurement practices. If the source of the noise can not be

1664
 residues could he corrected for this effect, this is an uunecessay
(6) complication that can he avoided by applying the exponential
window to the force signal.
For the case in which both the output aud input signals are
exponentially windowed,
7. Correction for the Exponential Window

(7) .The exponential window increases the apparent damping of the

measured system, and the amount of added damping is
Since the transfer function i s computed from the shifted input ad determined by the exponential time constant. This effect is
output functions. it is a functiou of the same iudependent variable. predictable and the correctiou for estimated modal paremetcrs i s
For the case in which the output signal is exponentially windowed developed hclow.‘61 Stating with equation (7), the l-W i s i n v e r s e
but the iuput signal is not, trzmsfonned to the impulse response fuuction (IRF). where t,(t) i s
the lRF of the true (i.e., unwindowed) system.

I/(& + p) 0 e-b(t) (9)

Equatiou (7) clearly shows that the exponential window must be The IRF cau he written as a summation of damped exponential
applied to both the response and force time signals so that the terms, where h, is the complex cigcnvalue of mode I, A, is the
indepeudent variable of the tmusfer functiou is unemhiguously residue for mode r, and the caret notation (“) denotes a parameter
defined. If the exponential window is applied to the response, its associated with the measured (i.e~, windowed) system
transform is a function of s+b, but if the window is not applied
the force, its transform is B function of just S, as indicated in
equatiou (8). So the question is, what is the independent variable
of II for this case’? Ihe practical consequences of applying the Equating the like tenns in equation (IO) and iuscrting the real and
exponential window to an impact signal are investigated in Ref. imaginary parts of the eigenvalues yields the correction for the
151 b y cousideriug the dependence of F on p. Also considered are exponential window. where CO, and CT, are the damped natural
the more subtle, comhiued ekl’ects of the exponential window and frequency and damping factor of mode r. respectively.
a pretrigger delays The conclusions can be summarized 8s follows.

The exponeutial windo\\ and pretrigger delay at?%3 the measured

FRF aud estimated modal parameters in a predictable manner,
The cffccts are hcst uuderstood by studying the transfer function
of the system and the Laplace transform of a pulse input fimction.
It is the combination of the Laplace transform shin properties
which gown, the effects of the exponential window aud the
prctrigger delay that urcessitate the expouential window always
be applied to the force signal when measuring frequency response
tu”ctl”“s, .io I

‘The exponential tindow effect 01. incrcasiug the apparent

damping of the measured response is widely acknowledged and
readily corrected in the modal parameters. The exponenhal
window essentially shitts the plaue 01. the s-domain input and
o u t p u t functious that are actually measured, and to measure the
conect FRF, the sane planes of the input and output functions
must he measured. That is, both the input and output must by B
fimctiou of s+p. The errors introduced by not windowing the force
signal will not affect the estimated frequency or damping hut can
effect the scaling of the residue due to au incorrect FRF mag-
nitude, especially if a pretrigger delay is used in the
*“eaSuTeme”tS.

Prrtriggering is commouly used in impact testing to ohservc the Equations I1 and I3 indicate that the damped natural frequencies
leading edge of the measured time signals, and its effects are also and residues of the measured system tre identical to those of the
governed by a shiti property of the Laplace transform. The errors true system and the difference of the dampiug factors between the
introduced by not windowing the force signal are amplified when tie and measured systems is a function of the exponential
a pretrigger delay is used in the measurements, which can lead to wiudow time constant. .The effects ot. the exponential wiudow
a significant underestimation of the residue. Although the described m the above equations are illustrated on the complex

1665
plane in Figure 5~ Since the damping factors of natural system are [7] R.C. Taber, H. Void, D.L. Brown, G.T. Rocklin, Exponential
expected to be negative, the damping factors of the measured Window fir Burst Random Excitorion, Proceedings of the
system will have larger negative values than that sf the true Third International Modal Analysis Conference, 1985, p.
system, which causes the apparent increased damping in the 989.995.
measurements. Note that since the measured FRFs are computed 181 R.L. Clark, A.L. Wicks, W.J. Becker, Eficrs ofan Exponori-
from data modified by the exponential window, residue and lid Window on Ihe Damping f~‘o@cicnf, Proceedings of the
modal scaling calculations and FRF synthesis should use the Seventh International Modal Analysis Conference, 1989, p.
uncorrected poles. X3-86.
191 W.P. Jennings, NIL. Olsen, N.J. Walter, Trmsient Excitation
A damping correction for the exponential window historically and Data Processing Techniques Employing fhc Fasf
given in the literature”.8~9’ that deals with the damping ratio of a Fourier Transfinn for Aeroelastlc Testing, Proceedings of
m o d e , C,r Flutter Testing Techniques Conference, NASA SP-145,
1975.p.77.113.
[IO) W.A. Fladung, R.W. Rest, Application and Correction offbe
Expono&?I Window for Frequency Response Functions,
Journal of Mechanical Systems and Signal Processing,
is actually an approximate method. “ ” ‘The approximation is that accepted Jan. 1996.
the estimated undamped natural frequency, cl,, is equal to the
true undamped natural frequency, cl,, when in fact il. > Q, since
j&,1 > lo,l. The approximate correction underestimates the
damping ratio by a factor of Q/h,, but approaches the true value
asymptotically as the damped natural frequency increases, for a I I
given damping ratlo and exponential time constant.

Another possible effect of the exponential window which should

be noted is that it may complicate separation of closely spaced
modes due to the increased damping. Figure 6 shows the response
to an impact of an steel T-plate (approx. lSx29x29cm, 0.635cm
thick) and the applied exponential windows. The windows decay
to 1% at lOO%, 50%, 20% and 10% of the time record length.
which corresponds to n time constat of 0.2171, 0.1086, 0.0434
and 0.0217 sec., respectively. Figure 7 shows the FRF for each of
these four cases, note the two closely spaced modes at 580 and
588 Hz. As the added damping from the rxpo- nential window
increases, these two peaks merge such that two distinct modes
cannot be distinguished.

8. References

iI j D.L Brown, W.G. Halvorsen, lm&e Techniqwfor Srruc-

rural Frequency Response Tesrirrg, Sound and Vibration,
Nov. 1977,~. 8.21.
21 K.J. Allcmang, Vihratiom: Expuimvrfal hfodal Analysis,
llC-SDRIrCN-20-263.663/664, April 1994, p. 4.14.
31 W.A. Fladung, R.W~ Rost, 11.L. Brown, Fwther Develop-
nxnfs of Multiple Re$wnce Inryacl Testirtg, Proceedings of
the Twelfth International Modal Analysis Conference, 1994,
p. 1026.1033.
41 M.R. Spiegel, Theory and Problems of Lap/ace Tmnsfomrs,
New York: Schaum Publishing Co., 1965, p, 3 .
51 WA. Fladung, R.W. Rest, (huse and Eficr qfApply;y the
-701
Expor~enr;a/ Window to ON fmpoct Force Signal, Proceedings 250 Frequency (Hz) 65
of the Fourteenth International Modal Analysis Conference,
1996, p. 1553-1559.
6) W.A. Fladung, T&! Dcvelopnw~t and Implemrtafion of
~<uitiple R&wm Impact Tesrirrp, M a s t e r s T h e s i s ,
University of CincUnnati, 1994.

1666