Often times, people ask some simple questions regarding
modal analysis and how structures vibrate. Most times, it is impossible to describe this simply and some of the basic un- derlying theory needs to be addressed in order to fully explain some of these concepts. However, many times the theory is just a little too much to handle and some of the concepts can be described without a rigorous mathematical treatment. This article will attempt to explain some concepts about how struc- tures vibrate and the use of some of the tools to solve struc- tural dynamic problems. The intent of this article is to sim- ply identify how structures vibrate from a nonmathematical perspective. With this being said, lets start with the first ques- tion that is usually asked: In a nutshel l , we coul d say that modal anal ysi s i s a process whereby we descri be a structure i n terms of i ts natural char- acteristics which are the frequency, damping and mode shapes its dynamic properties. Well thats a mouthful so lets explain what that means. Without getting too technical, I often explain modal anal ysi s i n terms of the modes of vi brati on of a si mpl e pl ate. Thi s expl anati on i s usual l y useful for engi neers who are new to vi brati ons and modal anal ysi s. Lets consi der a freel y supported fl at pl ate (Fi gure 1). Lets appl y a constant force to one corner of the pl ate. We usual l y think of a force in a static sense which would cause some static deformati on i n the pl ate. But here what I woul d l i ke to do i s to appl y a force that vari es i n a si nusoi dal fashi on. Lets consi der a fi xed frequency of osci l l ati on of the constant force. We wi l l change the rate of oscillation of the frequency but the peak force wi l l al ways be the same val ue onl y the rate of osci l l ati on of the force will change. We will also measure the response of the pl ate due to the exci tati on wi th an accel erometer attached to one corner of the pl ate. Now i f we measure the response on the pl ate we wi l l noti ce that the ampl i tude changes as we change the rate of osci l l a- tion of the input force (Figure 2). There will be increases as well as decreases i n ampl i tude at di fferent poi nts as we sweep up i n ti me. This seems very odd si nce we are appl yi ng a constant force to the system yet the ampl i tude vari es dependi ng on the rate of osci l l ati on of the i nput force. But thi s i s exactl y what happens the response ampl i fi es as we appl y a force wi th a rate of osci l l ati on that gets cl oser and cl oser to the natural fre- quency (or resonant frequency) of the system and reaches a maxi mum when the rate of osci l l ati on i s at the resonant fre- quency of the system. When you thi nk about i t, thats pretty amazi ng si nce I am appl yi ng the same peak force al l the ti me onl y the rate of osci l l ati on i s changi ng! Thi s ti me data provi des very useful i nformati on. But i f we take the ti me data and transform i t to the frequency domai n usi ng the Fast Fouri er Transform then we can compute some- thi ng cal l ed the frequency response functi on (Fi gure 3). Now there are some very i nteresti ng i tems to note. We see that there are peaks in this function which occur at the resonant frequen- ci es of the system. And we noti ce that these peaks occur at fre- quenci es where the ti me response was observed to have maxi - mum response correspondi ng to the rate of osci l l ati on of the i nput exci tati on. Now i f we overl ay the ti me trace wi th the frequency trace what we wi l l noti ce i s that the frequency of osci l l ati on at the ti me at whi ch the ti me trace reaches i ts maxi mum val ue cor- responds to the frequency where peaks i n the frequency re- sponse functi on reach a maxi mum (Fi gure 4). So you can see that we can use either the time trace to determine the frequency at which maximum amplitude increases occur or the frequency response functi on to determi ne where these natural frequen- ci es occur. Cl earl y the frequency response functi on i s easi er to eval uate. Most people are amazed at how the structure has these natu- ral characteri sti cs. Wel l , whats more amazi ng i s that the de- formati on patterns at these natural frequenci es al so take on a vari ety of di fferent shapes dependi ng on whi ch frequency i s used for the exci tati on force. Now l ets see what happens to the deformati on pattern on the structure at each one of these natural frequenci es. Lets pl ace 45 evenl y di stri buted accel erometers on the pl ate and measure the ampl i tude of the response of the pl ate wi th di f- ferent exci tati on frequenci es. If we were to dwel l at each one of the frequenci es each one of the natural frequenci es we woul d see a deformati on pattern that exi sts i n the structure (Figure 5). The figure shows the deformation patterns that will resul t when the exci tati on coi nci des wi th one of the natural frequenci es of the system. We see that when we dwel l at the fi rst natural frequency, there i s a fi rst bendi ng deformati on pattern i n the pl ate shown i n bl ue (mode 1). When we dwel l at the second natural frequency, there i s a fi rst twi sti ng defor- mati on pattern i n the pl ate shown i n red (mode 2). When we dwel l at the thi rd and fourth natural frequenci es, the second bendi ng and second twi sti ng deformati on patterns are seen i n green (mode 3) and magenta (mode 4), respecti vel y. These de- formati on patterns are referred to as the mode shapes of the structure. (Thats not actual l y perfectl y correct from a pure mathemati cal standpoi nt but for the si mpl e di scussi on here, these deformati on patterns are very cl ose to the mode shapes, from a practi cal standpoi nt.) These natural frequenci es and mode shapes occur i n al l structures that we desi gn. Basi cal l y, there are characteri sti cs that depend on the wei ght and sti ffness of my structure whi ch determi ne where these natural frequenci es and mode shapes wi l l exi st. As a desi gn engi neer, I need to i denti fy these fre- quenci es and know how they mi ght affect the response of my structure when a force excites the structure. Understanding the mode shape and how the structure wi l l vi brate when exci ted helps the design engineer to design better structures. Now there i s much more to i t al l but thi s i s just a very si mpl e expl anati on of modal anal ysi s. So, basically, modal analysis is the study of the natural char- acteri sti cs of structures. Understandi ng both the natural fre- quency and mode shape hel ps to desi gn my structural system for noi se and vi brati on appl i cati ons. We use modal anal ysi s to hel p desi gn al l types of structures i ncl udi ng automoti ve struc- tures, aircraft structures, spacecraft, computers, tennis rackets, gol f cl ubs . . . the l i st just goes on and on. Now we have i ntroduced thi s measurement cal l ed a fre- quency response functi on but exactl y what i s i t? The frequency response functi on i s very si mpl y the rati o of the output response of a structure due to an appl i ed force. We measure both the appl i ed force and the response of the struc- ture due to the applied force simultaneously. (The response can be measured as di spl acement, vel oci ty or accel erati on.) Now Experimental Modal Analysis A Simple Non-Mathematical Presentation Peter Avitabile, University of Massachusetts Lowell, Lowell, Massachusetts Could you explain modal analysis for me? Just what are these measurements that are called FRFs? 2 SOUND AND VIBRATION/JANUARY 2001 the measured ti me data i s transformed from the ti me domai n to the frequency domai n usi ng a Fast Fouri er Transform al go- ri thm found i n any si gnal processi ng anal yzer and computer software packages. Due to thi s transformati on, the functi ons end up bei ng com- plex valued numbers; the functions contain real and imaginary components or magni tude and phase components to descri be the functi on. So l ets take a l ook at what some of the functi ons mi ght l ook l i ke and try to determi ne how modal data can be extracted from these measured functi ons. Lets fi rst eval uate a si mpl e beam wi th onl y 3 measurement l ocati ons (Fi gure 6). Noti ce that the beam has 3 measurement l ocati ons and 3 mode shapes. There are 3 possi bl e pl aces that forces can be applied and 3 possible places where the responses can be measured. Thi s means that there are a total of 9 pos- si bl e complex-valued frequency response functions that could be acqui red; the frequency response functi ons are usual l y de- scri bed wi th subscri pts to denote the i nput and output l oca- ti ons as h out,in (or wi th respect to typi cal matri x notati on thi s woul d be h row,col umn ). Fi gure 7 shows the magni tude, phase, real and i magi nary parts of the frequency response functi on matri x. (Of course, I am assuming that we remember that a complex number is made up of a real and i magi nary part whi ch can be easi l y converted to magni tude and phase. Si nce the frequency response i s a compl ex number, we can l ook at any and al l of the parts that can descri be the frequency response functi on.) Now l ets take a l ook at each of the measurements and make some remarks on some of the i ndi vi dual measurements that coul d be made. Fi rst l ets dri ve the beam wi th a force from an i mpact at the ti p of the beam at poi nt 3 and measure the re- sponse of the beam at the same l ocati on (Fi gure 8). Thi s mea- surement i s referred to as h 33 . Thi s i s a speci al measurement referred to as a drive point measurement. Some important char- acteri sti cs of a dri ve poi nt measurement are: a Al l resonances (peaks) are separated by anti -resonances. a The phase l oses 180 degrees of phase as we pass over a reso- nance and gains 180 degrees of phase as we pass over an anti- resonance. a The peaks i n the i magi nary part of the frequency response functi on must al l poi nt i n the same di recti on. So as I conti nue and take a measurement by movi ng the i mpact force to poi nt 2 and measuri ng the response at poi nt 3 and then movi ng the i mpact force on to poi nt 1 to acqui re two more measurements as shown. (And of course I could continue on to col l ect any or al l of the addi ti onal i nput-output combi - nati ons.) So now we have some i dea about the measurements that we coul d possi bl y acqui re. One i mportant i tem to note i s that the frequency response functi on matri x i s symmetri c. Thi s i s due to the fact that the mass, dampi ng and sti ffness matri ces that descri be the system are symmetri c. So we can see that h ij =h ji thi s i s cal l ed reci proci ty. So we dont need to actual l y mea- sure al l the terms of the frequency response functi on matri x. Figure 1. Simple plate excitation/response model. Figure 2. Simple plate response. Figure 3. Simple plate frequency response function. Output Response Input Force Frequency M a g n i t u d e Figure 4. Overlay of time and frequency response functions. Figure 5. Simple plate sine dwell responses. Figure 6. A 3 DOF model of a beam. Mode 1 Mode 2 Mode 3 Mode 4 Increasing Rate of Oscillation 1 2 3 8 5 2 8 0 3 6 8 7 3 SOUND AND VIBRATION/JANUARY 2001 One questi on that al ways seems to ari se i s whether or not i t i s necessary to measure al l of the possi bl e i nput-output com- bi nati ons and why i s i t possi bl e to obtai n mode shapes from onl y one row or col umn of the frequency response functi on matri x. It i s very i mportant for us to understand how we arri ve at mode shapes from the various measurements that are available i n the frequency response functi on matri x. Wi thout getti ng mathemati cal , l ets di scuss thi s. Lets just take a l ook at the thi rd row of the frequency re- sponse functi on matri x and concentrate on the fi rst mode. If I l ook at the peak ampl i tude of the i magi nary part of the fre- quency response functi on, I can easi l y see that the fi rst mode shape for mode 1 can be seen (Fi gure 9a). So i t seems fai rl y straightforward to extract the mode shape from measured data. A qui ck and di rty approach i s just to measure the peak ampl i - tude of the frequency response functi on for a number of di f- ferent measurement poi nts. Now l ook at the second row of the frequency response func- ti on matri x and concentrate on the fi rst mode (Fi gure 9b). If I l ook at the peak ampl i tude of the i magi nary part of the fre- quency response functi on, I can easi l y see that the fi rst mode shape for mode 1 can be seen from thi s row al so. We could also look at the first row of the frequency response functi on matri x and see the same shape. Thi s i s a very si mpl e pi ctori al representati on of what the theory i ndi cates. We can use any row to descri be the mode shape of the system. So i t i s very obvi ous that the measurements contai n i nformati on per- tai ni ng to the mode shapes of the system. Lets now take a l ook at the thi rd row agai n and concentrate on mode 2 (Fi gure 9c). Agai n i f I l ook at the peak ampl i tude of the i magi nary part of the frequency response functi on, I can easi l y see the second mode shape for mode 2. And i f I l ook at the second row of the frequency response functi on matri x and concentrate on the second mode, I wi l l be a l i ttl e surpri sed because there i s no ampl i tude for the second mode (Fi gure 9d). I wasnt expecti ng thi s but i f we l ook at the mode shape for the second mode then we can qui ckl y see that thi s i s a node poi nt for mode 2. The reference poi nt i s l ocated at the node of the mode. So this points out one very important aspect of modal analy- si s and experi mental measurements. The reference poi nt can- not be l ocated at the node of a mode otherwi se that mode wi l l not be seen i n the frequency response functi on measurements and the mode cannot be obtai ned. Now we have onl y used 3 measurement poi nts to descri be the modes for thi s si mpl e beam. If we add more i nput-output measurement locations then the mode shapes can be seen more cl earl y as shown i n Fi gure 10. The fi gure shows 15 measured frequency response functi ons and the 3 measurement poi nts used in the discussion above are highlighted. This figure shows the 15 frequency response functi ons i n a waterfal l styl e pl ot. Usi ng thi s type of pl ot, i t i s much easi er to see that the mode shapes can be determi ned by l ooki ng at the peaks of the i magi - nary part of the frequency response functi on. Now the measurements that we have discussed thus far have been obtai ned from an i mpact testi ng consi derati on. What i f the measured frequency response functions come from a shaker test? From a theoreti cal standpoi nt, i t doesnt matter whether the measured frequency response functi ons come from a shaker test or an i mpact test. Fi gures 11a and 11b show the measure- ments that are obtai ned from an i mpact test and a shaker test. An impact test generally results in measuring one of the rows Figure 7. Response measurements of a 3 DOF model of a beam: a) mag- nitudes; b) phases; c) real components; and d) imaginary components. 80 0 M a g n i t u d e ,
d B 80 0 M a g n i t u d e ,
d B 80 0 M a g n i t u d e ,
d B 0 40 Frequency, Hz 0 40 Frequency, Hz 0 40 Frequency, Hz 180 180 P h a s e 180 180 P h a s e 180 180 P h a s e 0.5 0.5 R e a l 0.5 0.5 R e a l 0.5 0.5 R e a l I m a g i n a r y 1 1 I m a g i n a r y 1 1 I m a g i n a r y 1 1 a b c d Why is only one row or column of the FRF matrix needed? Whats the difference between a shaker test and an impact test? 4 SOUND AND VIBRATION/JANUARY 2001 of the frequency response function matrix whereas the shaker test generally results in measuring one of the columns of the frequency response function matrix. Si nce the system matri - ces describing the system are square symmetric, then reciproc- ity is true. For the case shown, the third row is exactly the same as the thi rd col umn, for i nstance. Theoreti cal l y, there i s no di fference between a shaker test and an i mpact test. That i s, from a theoreti cal standpoi nt! If I can appl y pure forces to a structure wi thout any i nteracti on between the appl i ed force and the structure and I can measure response wi th a massl ess transducer that has no effect on the structure then thi s i s true. But what i f thi s i s not the case? Now l ets thi nk about performi ng the test from a practi cal standpoi nt. The poi nt i s that shakers and response transduc- ers general l y do have an effect on the structure duri ng the modal test. The mai n i tem to remember i s that the structure under test i s not just the structure for whi ch you woul d l i ke to obtai n modal data. It i s the structure pl us everythi ng i nvol ved i n the acqui si ti on of the data the structure suspensi on, the mass of the mounted transducers, the potenti al sti ffeni ng ef- fects of the shaker/sti nger arrangement, etc. So whi l e theory tel l s me that there shoul dnt be any di fference between the i mpact test resul ts and the shaker test resul ts, often there wi l l Figure 8. a) Drive point FRFs for reference 3 of the beam model. b&c) Cross FRFs for reference 3 of the beam model. Figure 9. a) Mode 1 from third row of FRF matrix. b) Mode 1 from sec- ond row of FRF matrix. c) Mode 2 from third row of FRF matrix. d) Mode 2 from second row of FRF matrix. Figure 10. Waterfall plot of beam frequency response functions. ? ? a b c d Mode #1 DOF #1 DOF #2 DOF #3 Mode #2 Mode #3 1 1 2 3 2 3 h 33 1 1 2 3 2 3 h 32 1 1 2 3 2 3 h 31 a b c 5 SOUND AND VIBRATION/JANUARY 2001 be di fferences due to the practi cal aspects of col l ecti ng data. The most obvi ous di fference wi l l occur from the rovi ng of accel erometers duri ng a shaker test. The wei ght of the accel - erometers may be extremel y smal l rel ati ve to the total wei ght of the whole structure, but their weight may be quite large rela- ti ve to the effecti ve wei ght of di fferent parts of the structure. Thi s i s accentuated i n mul ti -channel systems where many ac- celerometers are moved around the structure in order to acquire al l the measurements. Thi s can be a probl em especi al l y on l i ghtwei ght structures. One way to correct thi s probl em i s to mount al l of the accel erometers on the structure even though onl y a few are measured at a ti me. Another way i s to add dummy accelerometer masses at locations not being measured; thi s wi l l el i mi nate the rovi ng mass effect. Another difference that can result is due to the shaker/stinger effects. Basi cal l y, the modes of the structure may be affected by the mass and sti ffness effects of the shaker attachment. Whi l e we try to mi ni mi ze these effects, they may exi st. The purpose of the sti nger i s to uncoupl e the effects of the shaker from the structure. However, on many structures, the effects of the shaker attachment may be si gni fi cant. Si nce an i mpact test does not suffer from these probl ems, di fferent resul ts may be obtai ned. So whi l e theory says that there i s no di fference between a shaker test and an i mpact test, there are some very basi c practi cal aspects that may cause some di fferences. The most i mportant measurement that i s needed for experi - mental modal analysis is the frequency response function. Very si mpl y stated, thi s i s the rati o of the output response to the input excitation force. This measurement is typically acquired usi ng a dedi cated i nstrument such as an FFT (Fast Fouri er Transform) analyzer or a data acquisition system with software that performs the FFT. Lets bri efl y di scuss some of the basi c steps that occur i n the acqui si ti on of data to obtai n the FRF. Fi rst, there are anal og si gnal s that are obtai ned from our measuri ng devi ces. These analog signals must be filtered to assure that there is no aliasing of hi gher frequenci es i nto the anal ysi s frequency range. Thi s i s usual l y done through the use of a set of anal og fi l ters on the front-end of the analyzer called anti-aliasing filters. Their func- ti on i s to remove any hi gh frequency si gnal s that may exi st i n the si gnal . The next step i s to di gi ti ze the anal og si gnal to form a di gi - tal representati on of the actual si gnal . Thi s i s done by the ana- l og to di gi tal converter that i s cal l ed the ADC. Typi cal l y thi s di gi ti zati on process wi l l use 10, 12 or 16 bi t converters; the more bi ts avai l abl e, the better the resol uti on possi bl e i n the digitized signal. Some of the major concerns lie in the sampling and quanti zati on errors that coul d potenti al l y creep i nto the digitized approximation. Sampling rate controls the resolution i n the ti me and frequency representati on of the si gnal s. Quan- ti zati on i s associ ated wi th the accuracy of magni tude of the captured si gnal . Both sampl i ng and quanti zati on can cause some errors i n the measured data but are not nearl y as si gni fi - cant and devastati ng as the worst of al l the si gnal processi ng errors l eakage! Leakage occurs from the transformati on of time data to the frequency domain using the Fast Fourier Trans- form (FFT). The Fouri er Transform process requi res that the sampl ed data consist of a complete representation of the data for all time or contai n a peri odi c repeti ti on of the measured data. When thi s i s sati sfi ed, then the Fouri er Transform produces a proper representati on of the data i n the frequency domai n. However, when thi s i s not the case, then l eakage wi l l cause a seri ous di storti on of the data i n the frequency domai n. In order to mi ni mi ze the di storti on due to l eakage, wei ghti ng functi ons cal l ed wi ndows are used to cause the sampl ed data to appear to better sati sfy the peri odi ci ty requi rement of the FFT. Whi l e wi ndows greatl y reduce the l eakage effect, i t cannot be com- pl etel y removed. Once the data are sampled, then the FFT is computed to form l i near spectra of the i nput exci tati on and output response. Typi cal l y, averagi ng i s performed on power spectra obtai ned from the l i near spectra. The mai n averaged spectra computed are the i nput power spectrum, the output power spectrum and the cross spectrum between the output and i nput si gnal s. These functi ons are averaged and used to compute two i m- portant functions that are used for modal data acquisition the frequency response functi on (FRF) and the coherence. The coherence functi on i s used as a data qual i ty assessment tool whi ch i denti fi es how much of the output si gnal i s rel ated to the measured i nput si gnal . The FRF contai ns i nformati on re- gardi ng the system frequency and dampi ng and a col l ecti on of FRFs contai n i nformati on regardi ng the mode shape of the system at the measured l ocati ons. Thi s i s the most i mportant measurement related to experimental modal analysis. An over- vi ew of these steps descri bed i s shown i n Fi gure 12. Of course, there are many important aspects of measurement acqui si ti on, averagi ng techni ques to reduce noi se and so on, that are beyond the scope of thi s presentati on. Any good refer- Figure 11. a) Roving impact test scenario. b) Roving response scenario. h 32 h 33 h 31 h 23 h 33 h 13 1 2 3 1 2 3 1 2 3 3 a b What measurements do I actually make to compute the FRF? 6 SOUND AND VIBRATION/JANUARY 2001 ence on digital signal processing will provide assistance in this area. Now the i nput exci tati on needs to be di scussed next. Basi cal l y, there are two commonl y used types of exci tati on for experi mental modal anal ysi s i mpact exci tati on and shaker exci tati on. Now l ets consi der some of the testi ng consi derati ons when performi ng an i mpact test. There are many i mportant consi derati ons when performi ng i mpact testi ng. Onl y two of the most cri ti cal i tems wi l l be menti oned here; a detai l ed expl anati on of al l the aspects per- tai ni ng to i mpact testi ng i s far beyond the scope of thi s arti cl e. Fi rst, the sel ecti on of the hammer ti p can have a si gni fi cant effect on the measurement acqui red. The i nput exci tati on fre- quency range i s control l ed mai nl y by the hardness of the ti p selected. The harder the tip, the wider the frequency range that i s exci ted by the exci tati on force. The ti p needs to be sel ected such that al l the modes of i nterest are exci ted by the i mpact force over the frequency range to be consi dered. If too soft a ti p i s sel ected, then al l the modes wi l l not be exci ted ad- equatel y i n order to obtai n a good measurement as seen i n Fi g- ure 13a. The i nput power spectrum does not exci te al l of the frequency ranges shown as evi denced by the rol l off of the power spectrum; the coherence i s al so seen to deteri orate as wel l as the frequency response functi on over the second hal f of the frequency range. Typi cal l y, we stri ve to have a fai rl y good and rel ati vel y fl at input excitation forcing function as seen in Figure 13b. The fre- quency response functi on i s measured much better as evi - denced by the much i mproved coherence functi on. When per- formi ng i mpact testi ng, care must be exerci sed to sel ect the proper ti p so that al l the modes are wel l exci ted and a good frequency response measurement i s obtai ned. The second most i mportant aspect of i mpact testi ng rel ates to the use of an exponenti al wi ndow for the response trans- ducer. General l y for l i ghtl y damped structures, the response of the structure due to the i mpact exci tati on wi l l not di e down to zero by the end of the sample interval. When this is the case, the transformed data wi l l suffer si gni fi cantl y from a di gi tal si gnal processi ng effect referred to as l eakage. In order to mi ni mi ze l eakage, a wei ghti ng functi on referred to as a wi ndow i s appl i ed to the measured data. Thi s wi ndow is used to force the data to better satisfy the periodicity require- ments of the Fourier transform process, thereby minimizing the di storti on effects of l eakage. For i mpact exci tati on, the most common wi ndow used on the response transducer measure- ment i s the exponenti al l y decayi ng wi ndow. The i mpl emen- tati on of the wi ndow to mi ni mi ze l eakage i s shown i n Fi gure 14. Wi ndows cause some di storti on of the data themsel ves and shoul d be avoi ded whenever possi bl e. For i mpact measure- ments, two possi bl e i tems to al ways consi der are the sel ecti on of a narrower bandwidth for measurements and to increase the number of spectral l i nes of resol uti on. Both of these si gnal processing parameters have the effect of increasing the amount of ti me requi red to acqui re a measurement. These wi l l both tend to reduce the need for the use of an exponenti al wi ndow and shoul d al ways be consi dered to reduce the effects of l eak- age. Now l ets consi der some of the testi ng consi derati ons when performi ng a shaker test. Agai n, there are many i mportant i tems to consi der when performing shaker testing but the most important of those cen- ter around the effects of exci tati on si gnal s that mi ni mi ze the need for wi ndows or el i mi nate the need for wi ndows al to- Figure 12. Anatomy of an FFT analyzer. Figure 13. a) Hammer tip not sufficient to excite all modes. b) Hammer tip adequate to excite all modes. Figure 14. Exponential window to minimize leakage effects. 40 60 0 800 Coherence Input Power Spectrum FRF Frequency, Hz M a g n i t u d e ,
d B a 40 60 0 200 Coherence Input Power Spectrum FRF Frequency, Hz M a g n i t u d e ,
d B b Antialiasing Filters Output Output Output Input Input Input Linear Output Spectrum Autorange Analyzer ADC Digitizes Signals Compute FFT Linear Spectra Averaging of Samples Apply Windows Computation of Averaged Input/Output/Cross Power Spectra Linear Input Spectrum Input Power Spectrum Cross Power Spectrum Output Power Spectrum Computation of FRF and Coherence Frequency Response Function Coherence Function Analog Signals What are the biggest things to think about when impact testing? What are the biggest things to think about when shaker testing? Actual Time Signal Sampled Signal Window Weighting Windowed Time Signal 7 SOUND AND VIBRATION/JANUARY 2001 gether. There are many other consi derati ons when performi ng shaker testi ng but a detai l ed expl anati on of al l of these i s far beyond the scope of thi s presentati on. One of the more common exci tati on techni ques sti l l used today i s random exci tati on due to i ts ease of i mpl ementati on. However, due to the nature of thi s exci tati on si gnal , l eakage i s a cri ti cal concern and the use of a Hanni ng wi ndow i s com- monly employed. This leakage effect is serious and causes dis- tortion of the measured frequency response function even when wi ndows are used. A typi cal random exci tati on si gnal wi th a Hanni ng wi ndow i s shown i n Fi gure 15. As seen i n the fi gure, the Hanni ng wi ndow wei ghti ng functi on hel ps to make the sampl ed si gnal appear to better sati sfy the peri odi ci ty requi re- ment of the FFT process, thereby mi ni mi zi ng the potenti al di storti on effects of l eakage. Whi l e thi s i mproves the di stor- ti on of the FRF due to l eakage, the wi ndow wi l l never total l y remove these effects; the measurements will still contain some di storti on effects due to l eakage. Two very common exci tati on si gnal s wi del y used today are burst random and si ne chi rp. Both of these exci tati ons have a special characteristic that do not require the need for windows to be appl i ed to the data si nce the si gnal s are i nherentl y l eak- age free i n al most al l testi ng si tuati ons. These exci tati ons are rel ati vel y si mpl e to empl oy and are commonl y found on most si gnal anal yzers avai l abl e today. These two si gnal s are shown schemati cal l y i n Fi gures 16 and 17. In the case of burst random, the peri odi ci ty requi rement of the FFT process i s sati sfi ed due to the fact that the enti re tran- si ent exci tati on and response are captured i n one sampl e i n- terval of the FFT. For the si ne chi rp exci tati on, the repeti ti on of the si gnal over the sampl e i nterval sati sfi es the peri odi ci ty requirement of the FFT process. While other excitation signals al so exi st, these are the most common exci tati on si gnal s used i n modal testi ng today. So now we have a better i dea how to make some measure- ments. Wi ndows are, i n many measurement si tuati ons, a necessary evi l . Whi l e I woul d rather not have to use any wi ndows at al l , the alternative of leakage is definitely not acceptable either. As di scussed above, there i s a vari ety of exci tati on methods that can be successful l y empl oyed whi ch wi l l provi de l eakage free measurements and do not require the use of any window. How- ever, there are many ti mes, especi al l y when performi ng fi el d testing and collecting operating data, where the use of windows i s necessary. So what are the most common wi ndows typi cal l y used. Basi cal l y, i n a nutshel l , the most common wi dows used to- day are the Uni form, Hanni ng, Fl at Top and Force/Exponen- tial windows. Rather than detail all the windows, lets just sim- pl y state when each i s used for experi mental modal testi ng. The Uni form Wi ndow (whi ch i s al so referred to as the Rect- angul ar Wi ndow, Boxcar or No Wi ndow) i s basi cal l y a uni ty gain weighting function that is applied to all the digitized data points in one sample or record of data. This window is applied to data where the enti re si gnal i s captured i n one sampl e or record of data or when the data are guaranteed to sati sfy the peri odi ci ty requi rement of the FFT process. Thi s wi ndow can be used for i mpact testi ng where the i nput and response si g- nal s are total l y observed i n one sampl e of col l ected data. Thi s wi ndow i s al so used when performi ng shaker exci tati on tests wi th si gnal s such as burst random, si ne chi rp, pseudo-random and di gi tal stepped si ne; al l of these si gnal s general l y sati sfy the peri odi ci ty requi rement of the FFT process. The Hanni ng wi ndow i s basi cal l y a cosi ne shaped wei ght- i ng functi on (bel l shaped) that forces the begi nni ng and end of the sampl e i nterval to be heavi l y wei ghted to zero. Thi s i s useful for si gnal s that general l y do not sati sfy the peri odi ci ty requi rement of the FFT process. Random exci tati ons and gen- Figure 15. Shaker testing excitation considerations using random excitation with Hanning window. Figure 16. Burst random excitation without a window. Figure 17. Sine chirp excitation without a window. Figure 18. Input-output measurement locations. 1 2 3 4 Autoranging Averaging 1 2 3 4 5 6 Input Output h 61 h out,in Tell me more about windows. They seem pretty important! eral fi el d si gnal s usual l y fal l i nto thi s category and requi re the use of a wi ndow such as the Hanni ng wi ndow. The Fl at Top wi ndow i s most useful for si nusoi dal si gnal s that do not sati sfy the peri odi ci ty requi rement of the FFT pro- cess. Most often thi s wi ndow i s used for cal i brati on purposes more than anythi ng el se i n experi mental modal anal ysi s. The force and exponential windows are typically used when performing impact excitation for acquiring FRFs. Basically, the force wi ndow i s a uni ty gai n wi ndow that acts over a porti on of the sampl e i nterval where the i mpul si ve exci tati on occurs. The exponential window is used when the response signal does not di e out wi thi n the sampl e i nterval . The exponenti al wi n- dow i s appl i ed to force the response to better sati sfy the peri - odi ci ty requi rement of the FFT process. 1 2 3 4 Averaging with Window Autoranging 1 2 3 4 Autoranging Averaging 8 SOUND AND VIBRATION/JANUARY 2001 Each of the wi ndows has an effect on the frequency repre- sentati on of the data. In general , the wi ndows wi l l cause a degradati on i n the accuracy of the peak ampl i tude of the func- ti on and wi l l appear to have more dampi ng than what real l y exi sts i n the actual measurement. Whi l e these errors are not total l y desi rabl e, they are far more acceptabl e than the si gni fi - cant di storti on that can resul t from l eakage. So now that we have di scussed vari ous aspects of acqui ri ng measurements, l ets go back to the pl ate structure previ ousl y di scussed and take several measurements on the structure. Lets consider 6 measurement locations on the plate. Now there are 6 possi bl e pl aces where forces can be appl i ed to the pl ate and 6 possi bl e pl aces where we can measure the response on the pl ate. Thi s means that there are a total of 36 possi bl e i nput output measurements that coul d be made. The frequency re- sponse functi on descri bes how the force appl i ed to the pl ate causes the plate to respond. If we applied a force to point 1 and measured the response at poi nt 6, then the transfer rel ati on between 1 and 6 descri bes how the system wi l l behave (Fi gure 18). Whi l e the techni que shown i n Fi gures 19 and 20 i s adequate for very si mpl e structures, mathemati cal al gori thms are typi - cally used to estimate the modal characteristics from measured data. The modal parameter esti mati on phase, whi ch i s often referred to as curvefi tti ng, i s i mpl emented usi ng computer software to si mpl i fy the extracti on process. The basi c param- eters that are extracted from the measurements are the fre- Figure 19. Plate mode shapes for mode 1 peak pick of FRF. Figure 20. Plate mode shapes for mode 2 peak pick of FRF. 1 2 3 4 5 6 Mode 1 So how do I get mode shapes from the plate FRFs? Mode 2 1 2 3 4 5 6 Figure 21. Breakdown of a frequency response function. a ij1 a ij2 a ij3
1
2
3
1
2
3 Figure 22. Curvefitting different bands using different methods. Figure 23. Curvefitting a typical FRF. Figure 24. Schematic overviewing the input-output structural response problem. SDOF MDOF Residual Effects Residual Effects How Many Points??? How Many Modes??? f(t) FFT y(t) IFT Input Time Force Input Spectrum Frequency Response Function Output Time Response f(j) h(j) y(j) Output Response quency, dampi ng and mode shapes the dynami c characteri s- ti cs. The measured FRF i s basi cal l y broken down i nto many si ngl e DOF systems as shown i n Fi gure 21. 9 SOUND AND VIBRATION/JANUARY 2001 Figure 26. Excitation close to mode 1. Figure 25. Measured operating displacements. These curvefitting techniques use a variety of different meth- ods to extract data. Some techniques employ time domain data whi l e others use frequency domai n data. The most common methods employ multiple mode analytical models but at times very simple single mode methods will produce reasonably good results for most engineering analyses (Figure 22). Basically, all of the esti mati on al gori thms attempt to break down measure- ments i nto the pri nci pal components that make up the mea- sured data namely the frequency, damping and mode shapes. The key i nputs that the anal yst must speci fy are the band over whi ch data are extracted, the number of modes contai ned in these data and the inclusion of residual compensation terms for the esti mati on al gori thm. Thi s i s schemati cal l y shown i n Figure 23. Much more could be said concerning the estimation of modal parameters from measured data, the tool s avai l abl e for deci pheri ng data and the val i dati on of the extracted model but a detai l ed expl anati on i s far beyond the scope of thi s ar- ti cl e. All structures respond to externally applied forces. But many ti mes the forces are not known or cannot be measured easi l y. We could still measure the response of a structural system even though the forces may not be measured. So the next questi on that i s often asked concerns operati ng data. We fi rst need to recogni ze that the system responds to the forces that are appl i ed to the system (whether or not I can measure them). So for expl anati on purposes, l ets assume for now that we know what the forces are. Whi l e the forces are actual l y appl i ed i n the ti me domai n, there are some i mportant mathematical advantages to describing the forces and response i n the frequency domai n. For a structure whi ch i s exposed to an arbi trary i nput exci tati on, the response can be computed usi ng the frequency response functi on mul ti pl i ed by the i nput forci ng functi on. Thi s i s very si mpl y shown i n Fi gure 24. The random exci tati on shown exci tes al l frequenci es. The most i mportant thi ng to note i s that the frequency response function acts as a filter on the input force which results in some output response. The exci tati on shown causes al l the modes to be acti vated and therefore, the response i s, i n general , the l i near superposi ti on of al l the modes that are acti vated by the input excitation. Now what would happen if the excitation did not contain all frequencies but rather only excited one particu- lar frequency (which is normally what we are concerned about when eval uati ng most operati ng condi ti ons). To i l l ustrate thi s, l ets use the si mpl e pl ate that we just di s- cussed. Lets assume that there is some operating condition that exi sts for the system; a fi xed frequency operati ng unbal ance wi l l be consi dered to be the exci tati on. It seems reasonabl e to use the same set of accel erometers that were on the pl ate to measure the response of the system. If we acquire data, we may see somethi ng that l ooks l i ke the deformati on pattern shown i n Fi gure 25. Looki ng at thi s deformati on, i t i s not very cl ear why the structure responds thi s way or what to do to change the response. Why does the plate behave in such a complicated fashi on anyway??? Thi s doesnt appear to be anythi ng l i ke any of the mode shapes that we measured before. In order to understand thi s, l ets take that pl ate and appl y a si mpl e si nusoi dal i nput at one corner of the pl ate. For the ex- What are operating data? Mode 1 Contribution Mode 2 Contribution ampl e here, we are onl y goi ng to consi der the response of the pl ate assumi ng that there are onl y 2 modes that are acti vated by the i nput exci tati on. (Of course there are more modes, but l ets keep i t si mpl e to start.) Now we real i ze that the key to determi ni ng the response i s the frequency response functi on between the i nput and output l ocati ons. Al so, we need to re- member that when we col l ect operati ng data, we dont mea- sure the i nput force on the system and we dont measure the system frequency response functi on we onl y measure the response of the system. Fi rst l ets exci te the system wi th a si nusoi d that i s ri ght at the fi rst natural frequency of the pl ate structure. The response of the system for one frequency response functi on i s shown i n Fi gure 26. So even though we exci te the system at onl y one frequency, we know that the frequency response function is the fi l ter that determi nes how the structure wi l l respond. We can see that the frequency response functi on i s made up of a con- tri buti on of both mode 1 and mode 2. We can al so see that the majority of the response, whether it be in the time or frequency domai n, i s domi nated by mode 1. Now i f we were to measure the response onl y at that one frequency and measure the re- sponse at many poi nts on the structure, then the operati ng defl ecti on pattern woul d l ook very much l i ke mode 1 but there i s a smal l contri buti on due to mode 2. Remember that wi th operati ng data, we never measure the i nput force or the frequency response functi on we onl y measure the output response, so that the deformati ons that are measured are the actual response of the structure due to the i nput exci tati on whatever i t may be. When we measure frequency response functi ons and esti - mate modal parameters, we actual l y determi ne the contri bu- ti on to the total frequency response functi on sol el y due to the effects of mode 1 acti ng al one, as shown i n bl ue, and mode 2 acting alone, as shown in red, and so on for all the other modes of the system. Noti ce that wi th operati ng data, we onl y l ook at the response of the structure at one parti cul ar frequency whi ch i s the l i near combi nati on of al l the modes that contri b- ute to the total response of the system. So we can now see that the operati ng defl ecti on pattern wi l l l ook very much l i ke the fi rst mode shape i f the exci tati on pri mari l y exci tes mode 1. Now l ets exci te the system ri ght at the second natural fre- quency. Figure 27 shows the same information as just discussed for mode 1. But now we see that we pri mari l y exci te the sec- ond mode of the system. Agai n, we must real i ze that the re- sponse l ooks l i ke mode 2 but there i s a smal l contri buti on due to mode 1. But what happens when we exci te the system away from a resonant frequency? Lets excite the system at a frequency mid- way between mode 1 and mode 2. Now here i s where we see the real di fference between modal data and operati ng data. Figure 28 shows the deformation shape of the structure. At first 10 SOUND AND VIBRATION/JANUARY 2001 Figure 27. Excitation close to mode 2. Figure 28. Excitation somewhere between modes 1 and 2. Figure 29. Broadband plate excitation. Figure 30. Schematic of the SDM process. Figure 31. Overall dynamic modeling process. Figure 32. Modal model characteristics. Rib Stiffener Dashpot Spring Mass Experimental Modal Testing Develop Modal Model Structural Changes Required Use SDM to Evaluate Structural Changes Repeat Until Desired Characteristics Are Obtained No Yes Done Perform Eigen Solution Modal Parameter Estimation Finite Element Modeling Spring Dashpot Rib Stiffener Mass Corrections Frequency Response Measurements Parameter Estimation Modal Parameters Model Validation Finite Element Model Eigenvalue Solver Modal Parameters Synthesis of a Dynamic Modal Model Mass, Damping Stiffness Changes Structural Dynamics Modification Modified Modal Data Forced Response Simulation Real World Forces Structural Response Output Spectrum Frequency Response Function Input Spectrum y(j) f(j) gl ance, i t appears that the deformati on doesnt l ook l i ke any- thi ng that we recogni ze. But i f we l ook at the deformati on pat- tern l ong enough, we can actual l y see a l i ttl e bi t of fi rst bend- i ng and a l i ttl e bi t of fi rst torsi on i n the deformati on. So the operati ng data are pri mari l y some combi nati on of the fi rst and second mode shapes. (Yes, there wi l l actual l y be other modes but pri mari l y mode 1 and 2 wi l l be the major parti ci pants i n the response of the system.) Now, we have di scussed al l of thi s by understandi ng the fre- quency response functi on contri buti on on a mode by mode basi s. When we actual l y col l ect operati ng data, we dont col - l ect frequency response functi ons but rather we col l ect output spectrums. If we l ooked at those, i t woul d not have been very cl ear as to why the operati ng data l ooked l i ke mode shapes. Fi gure 29 shows a resul ti ng output spectrum that woul d be measured at one l ocati on on the pl ate structure. Now the i n- put appl i ed to the structure i s much broader i n frequency and many modes are exci ted. But, by understandi ng how each of the modes contri butes to the operati ng data, i t i s much easi er to see how the modes al l contri bute to the total response of the system. So actual l y, there i s a bi g di fference between operati ng de- 11 SOUND AND VIBRATION/JANUARY 2001 how the structure behaves i n servi ce. Thi s i s extremel y useful i nformati on. However, many ti mes the operati ng shapes are confusi ng and do not necessari l y provi de cl ear gui dance as to how to sol ve or correct an operati ng probl em (and modi fi ca- ti on and response tool s cannot be uti l i zed on operati ng data) as shown i n Fi gure 33. The best situation exists when both operating data and modal data are used i n conjuncti on to sol ve structural dynami c prob- l ems. Some si mpl e expl anati ons were used to descri be structural vi brati on and the use of some of the avai l abl e tool s for the so- l uti on of structural dynami c probl ems. Thi s was al l achi eved wi thout the use of any detai l ed mathemati cal rel ati onshi ps. In order to better understand more of the detai l s of the data pre- sented here, a theoreti cal treatment of thi s materi al i s neces- sary. 1. Lecture Notes, Peter Avi tabi l e, Modal Anal ysi s I & II, Uni versi ty of Massachusetts Lowel l . 2. Semi nar Presentati on Notes, Peter Avi tabi l e. 3. The Modal Handbook, A Mul ti medi a Computer Based Trai ni ng and Reference Guide, Dynamic Decisions, Merrimack, NH. For additional i nformati on, contact: i nfo@dynami cdeci si ons.com. The author can be contacted at: peter_avi tabi l e@uml .edu. Figure 33. Operating data characteristics. So what good are modal data? So should I collect modal data or operating data? Summary Related Material fl ecti ons and mode shapes we can now see that the mode shapes are summed together in some linear fashion to form the operating deflection patterns. But typi cal l y we are i nterested i n the total deformati on or response of the system. Why do I even want to bother to col l ect modal data? It seems l i ke a l ot of work to acqui re measurements and extract data. Modal data are extremel y useful i nformati on that can assi st i n the desi gn of al most any structure. The understandi ng and vi sual i zati on of mode shapes i s i nval uabl e i n the desi gn pro- cess. It helps to identify areas of weakness in the design or areas where i mprovement i s needed. The devel opment of a modal model is useful for simulation and design studies. One of these studi es i s structural dynami c modi fi cati on. Thi s i s a mathemati cal process whi ch uses modal data (fre- quency, dampi ng and mode shapes) to determi ne the effects of changes in the system characteristics due to physical structural changes. These calculations can be performed without actually havi ng to physi cal l y modi fy the actual structure unti l a sui t- abl e set of desi gn changes i s achi eved. A schemati c of thi s i s shown i n Fi gure 30. There i s much more that coul d be di s- cussed concerni ng structural dynami c modi fi cati on but space l i mi tati ons restri ct thi s. In addition to structural dynamic modification studies, other si mul ati ons can be performed such as force response si mul a- ti on to predi ct system response due to appl i ed forces. And another very i mportant aspect of modal testi ng i s the correl a- ti on and correcti on of an anal yti cal model such as a fi ni te el e- ment model . These are a few of the more i mportant aspects rel ated to the use of a modal model whi ch are schemati cal l y shown i n Fi gure 31. And one of the fi nal questi ons that i s often asked i s whi ch test i s best to perform. Of course wi th ti ght schedul es and budgets, do I real l y need to col l ect both modal data and operati ng data? Thi s i s al ways di ffi cul t to answer but i t i s al ways better to have both when- ever possi bl e. If onl y one of the two i s avai l abl e, then many ti mes some engi neeri ng deci si ons may be made wi thout ful l knowl edge of the system characteri sti cs. To summari ze, l ets poi nt out the di fferences between each of the data sets. Modal data requi res that the force i s measured i n order to determine the frequency response function and resulting modal parameters. Onl y modal data wi l l gi ve the true pri nci pal char- acteristics of the system. In addition, structural dynamic modi- fi cati ons and forced response can onl y be studi ed usi ng modal data (operati ng data cannot be used for these types of studi es). Al so correl ati on wi th a fi ni te el ement model i s best performed usi ng modal data. But of course i t needs to be cl earl y stated that modal data al one does not i denti fy whether a structure i s adequate for an i ntended servi ce or appl i cati on si nce modal data are i ndependent of the forces on the system as shown i n Fi gure 32. Operati ng data on the other hand i s an actual depi cti on of