IRNATIONAL CENTRE FOR MECHANICAL SCIENCES COURSES AND LECTURES - No us JOHN D, ROBSON COLIN J. DODDS DONALD B, MACVEAN VINCENT R, PALING ‘AIVERSIFY oF oLAscoW i 4 a “RANDOM VIBRATIONS 4 COURSE HELD AT THE DEPARMENT. ‘OF GENERAL MECHANICS OCTOBER 1971 UDINE 1971 SPRINGER ~ VERLAG WIEN ~ NEW YORK,CONTENTS Preface chapter I: Introduction .. 1,1 The Nature of the Problem 1,2 The Problem of Description « 1.3 The Nature of the Subject 1.4 The Nature of the Course ... Bibliography: for Chapter I . % Chaptsr TT Description Tt Probability . 2.1 Probability ..... 6 2.2 Continuous Probability Distribution .... 2.3 Expectation 2.5 Gaussian Density Funct 2,6 Tabulation ....++ 2.7 Probability in Relation to Random Pro- cesses 60a 2,8 Stationary Random Process . 2.9 Ergodic Random Process 4 2.40 Time Averages for Ergodic Process ... 2,11 Autocorrelation Function Bibliography for Chapter IT . a2 Chaptior IIT Description II: spectral Density .. gel Frequency Analysis ..eseeeeeeee 3.2 Fourier Series . 3.3 Fourier Integral . 3.4 Mean Values in Terms of Fourier Trans - form and Spectral Density 3.5 Spectral Density .217 7.1 Introduction ., ease a ee Contents Contents Bib Page ~ j bate ‘ography for Chapter IIT ,.,, a 7.2 Response of Linear System to Random Input 68 | Chapter IV Vibration the. : 7.3 Mean and Variance teeee 70 1X on Theory T: Receptance .....,, 35 7.4 Random Input to First Order Mechanical 4,1 Forced Vibration of a Danped System Hay— System + 7 ing One Degree of Freedon , 35 7.5 Random Input to Second Order Mechanical 4.2 Receptance . 7 . System ........, ae 74 4.3 Quality Factor ter VIIT Vibration Theory IIT: Normal Modes .. 7 44 Roveptance in Comosiss Sassy » 2/Chapter VIII Vibration Theory IIT: No z X teristics of Random Processes . 65 11,1 Introduction 4.4... ca ed 6. App. 2 Unit Impulse Function 6 11,2 Response to Profile Excitation .......... 103 6 App. 3 table of Fourier Tronefecg 11,3 Simple Model, Single Input ......s.s00.5. 105 (Double Infinite tntegral) ... 6 11.4 Two Input Problem ...cceeeceeseeseseeeeees 107 | : 11.5 Caloulation of Response - Practical Exan— |XX ehePter VET Randon vibration Response T+ Gonersa Rela oe oe + 10 tionships .... eee 68 Bibliography for Chapter XI + 13219 a Contents contents Me ] Page © 46,3 standard Method or Blacksan-Tokey Weth- én for the tine-average ofx, and we can.write: Ele] = = tim 4 / xctyat . oe T | et If the probability distribution over the realisations is Gauss— ian, these two quantities define the distribution, As noted be- fore, we usually arrange for to be 0. The mean value of the product of two values of24 probability an interval ¢ becomes x, separated by Re) = Efxpte] = = fim A] xttett + vat. i tr 2.11. Autocorrelation funetion ‘his function R(t) plays an important part, and ye can note at this stage some of its properties. (1) From the sntegral, R(t) is an even function of ©» that is Rt) = R@). (2) Ife = 0,R(0) becomes which is equal tog?if = 0+ It is sometimes convenient to express R(t) in a non-dimen. sional form by dividing by RO). ‘ast increases, the Link between (3) Im any Tandon process, Aininishes; ultimately jagnitude of x(t +7) and x(t) the m ‘the two values are practically unrelated and the average value of the product tends to zero, since products wil) positive and negative. Hence R(t) tends and in fact R(t) is always less af be equally often to 0 as 7 becomes large, ‘than R(O); (this applies to strictly random process there is a periodic term present in x it will show uP in R). typical standard forms for R(C) 2.4. These ca which occur in theoretical work be represented by relations of are shown in Figs Autocorrelation function 25 the types: RG) = acl] and ae ltleoser, where a,b, and c are constants to be adjusted to suit the cir- On cumstances. If we wish to consider the function x(t) as the sum of two functions, say u(t)and v(t), repres— enting perhaps a random Fig 24. and a determina’ a determinate function, the auto-correlation function be- comes: I z im 2 tin + if fultdutt +2) + v(t +2) + ultv(t + c+ vue +a}dt Tre Rt) 4 Rue) + Ry(e) + Ruglt) + Rol) 5 where the first two tems are auto-correlation functions f ion functions for w and v, and the renainis and thi ng terms are.'cross-correlation functions! measuring the Link between u and v, These will be discussed later; meanvhile we note that they will be zero only if u and v are completely independent (as they will be if one is random and the other determinate).Libliography for Chapter 2 ALL the books Listed in the Chapter 1 contain some account of probability theory, and those by Bendat and Lin have quite a lot, To present the theory on a logical mathe natical basis, vhich will be valid for discrete and for contin uous distributions, requires a fairly sophisticated treatment, and this is adopted in most modem books on the subject. of those given below, Parzen's treatment is at a medium level, Gramér is more advanced. . 2.1, Parzen E. Modern probability theory and its applications. Wiley 1960. Stochastic Processes. Holden-Day 1962, (both these are available as paperbacks). 2.2, Parzen E. 2.3. Cranér H.: Mathematical methods of statistics. Princeton University Press 1958, Chapter 3. DESCRIPTION Il : SPECTRAL, DENSITY 3.1. Frequency analysis The probability functions which we have consider ed give an account of the part played in a random process by values of x of different magnitudes, but they do not reveal the course of the variation of x with time except through the auto- correlation function R(t)as a function of time interval t . We are accustomed in vibration work to think of tine variation as measured rather by frequency, and we now consider how we can apply this concept to a random function, through the method of Fourier analysis. We shall find that this leads us back to the auto-correlation function by a different path. 3.2, Fourier series We shall state briefly some results which we wish to use, If we have a function x(t) of t, periodic with a period T, we can express x as the sum of an infinite series of sines and cosines of integral multiples of the basic frequency w = en/T. Thus x(t) = ay + acoswt + apcos2ut +... + bysinut + bysin2ut +. in which the a's and b's are constant coefficients given by28 Spectral density the formulae: 2 i 2 co: = 2 (27 Q)sinnwt an 2 F fF iaWeesnutat ant by 8 fa innutdt x(t}dtn represents any positive integer. ‘hese integrals may extend over any complete period of x. the restrictions on the fon of the function x, and the sense in which the series can replace the function, are discussed in books on Fourier series, but they need not detain us here. Its sufficient to note that x(t) need not be sooth or continuous. The tine average of x can be expressed in tems of the coefficients in the series: hus = a a +pe +b). 2 wi i = 7 [20 dt = 3 For our purpose it is more convenient to express these results in terms of complex exponentials. Thus a,cosnut +b,sinnwt = ap will be cp, with ¢, = $,-ib,) and c., = £(a,+ib,), conjugate mumbers, We can write: x(t) = Sree! , with n taking all integral values, positive and negative, and zero. i i pourker integral ““ccqhe coefficients expressed as integrals are now: 29 ‘ 4a abn, c= PfPxtermar . abr We now have: = = Ele. eo lc, is the modulus of the complex number c,, so |c,) = = tae sbe = dafses. 3.3, Fourier tegral The results quoted are valid for periodic func ‘tions, and they can be applied to a finite stretch of a non-pe~ riodic function by treating this stretch as one cycle of a hypo- thetical periodic function, To apply then to the complete range of a non-periodic function, we require to extend the period to infinity and to reduce the basic frequency to zero, This leads to the concept of a Fourier integral, replacing the sum of terms in the series, Write the series in the form: 2TatterHetat tr xt) = -Yeyetot ma og =f where we have used Wp for the fixed basic frequency. We wish to express x(t) in terms of a continuous frequency & in a form:30, Frequency Analysis xtt) = al Aw)etdw, with ACiw) a complex function ofw. (The factor 2x will enter the calculation and it Niwas on to ¢, in the pervious series, so that the dimensions of A are hk Ayu, Aifferent frome. We therefore put ¢ = “EU ;then t " eR is convenient to place it here). The product corresponds x(t) = ATA and Ay = [xWemurat u,ekugso thatg = Wy4-8, = 805 AL since wpT=2x, Now put is now Mw,) sthen 1 joat “b)eninnt x) = EE Ku,)e ana Aw, = [xe dt. If we take the Limiting fom of this for wp = bu and T+ @ —————E—>————ev SO xt) = aa | Mea and Miu) = [renee] the first of these relations is a Fourier integral, and Au) is the Fourier transfor of x(t) .ACju)is a complex function of as we can see by writing it in trigonometric terms: Aiw) = Jxvoosur-as - i fx) sinut-ct = ale) — iB) , vere « and are veal functions of w , & being an even function and fi odd, Evidently these integrals will not exist unless3(t)=0 for t-emand -o. In the nearly-symmetrical relations between x(t) 31 “jaNiu), the factor 2 must occur soneshere, and the notation “VA vary according to where this factor is placed, For our pur pose, it is convenient to cLiminate it by expressing the forme Jae in terms of the frequency f in place of the circular fre- queney (pulsatance) ¥ , withw =2nf, This gives: [xtrerttttas 20) = [Mineltethar and MiD = 34, Mean values in terms of Fourier transfrom and spectral density We cannot obtain directly mean values over an infinite range; to overcone the difficulty we introduce @ modi fied function zt) such thatx,t) = x() over the range -}T to +t and x,(t) O outside this range. Having defined the required quantity in tems of x(t), we take the Linit for Teo. Thus the ean value of z,(t) over the range T is 1 pit th x,(i) dt Lf, 1 “4 i z,(Wat = $A,0) with Ay) = [moe dt and = fin [LA,].32. Frequency Analysis For the mean square value, we have in our previous notation = ee a a 1 [isa - THe Li, 7 3 sptwat = | “ [etna | wore [AGua[’ = fated] + [ptwi]’ , a real even function of w 5 so we can take the integral over positive frequencies only fova = Lfincwfas - 2fisintar : : Now introducing x¢(t), mean value over range T is "20a 1 ae 2 hi : z[xithdt = £ /|ACt)|-aF if i 2 = [ {tim n.c! Jer = [soa a wich Sf) = tm | [Ziascinl] ant Adit) = fide at Thiz function S ef the frequency is the ‘spectral density", which proves to be the key to our treatment of random functions. physical meaning of this quantity, namely that S()df for any range of frequency from f to (f+ df) measures the contribution to the mean square of x(t) coming from this frequency range. Since in many applications the quantity measures the mean power, § is sometines called the 'power spectral density!, Exe Perinentally we should have to define df by a narrowband filter and measure the value of x? transmitted, From the equations it will be seen that Sis a real, even function of f. The notation used for § varies accord ing as the integral is taken one-sided (as above) or double-sie- ea, wich introduces a factor of 2, and also according as $ is considered as a function of f or ofw, which introduces a factor of 2x, The convention used here is that used in the books by Bendat and Robson in tke bibliography to Chapter 1,ibliography for Chapter 3 Fourier series and integrals are discussed in many books on mathematics The two listed below are, however, written with engineering applications in view. 3.1. Lawlen D.¥.: Mathematics of Engineering Systens: Methuen 1959 (also paperback). ‘This applies the theory to random processes. - Operational Methods for Linear Systema: Addison-Wesley 1962. More advanced in treatment, but does not touch’ random processes. 3.2. Kaplan We \ Chapter 4 VIBRATION THEORY 1: RECEPTANCE 4.1. Forced vibration of a damped system having one degree of freedom When the configuration of a mechanical system is completley described by one independent variable x, it is said to have one degree of freedom. This quantity usually has the dimension of length or angle. In a real systen there will be inertia forces depending on the second time derivative 2, vis- cous damping forces depending on the velocity i , and restoring forces depending on x, The equation governing the vibration can usually be brought into the fom f,x,x) = PH), a) where the periodic excitation P(t) is regarded as the input to the system and the resulting motion x(t) the response, It should be noted in passing that in requiring the forcing term Pt) to appear alone in (1) we have ruled out the case of parametric excitation, The system is said to be linear when equation (1) can be written in the standard fora 4 tout + ube = KD), @ ie, when no higher power of x,%,2 occurs. (the systen is still Linear when either of the tvo parameters in (2) varies with tine,36 Receptance However, we will exclude such cases here). Above we have had in mind a force as input and a displacement output, There are cases where the excitation is a prescribed displacement causing another displacenent as re~ sponse. There are also cases where the governing equation has to be formulated in onler to find out what the input to the sys tm really is. For instance, the casing shown below contains a plunger m constrained by spring k and dashpot ¢. hmm wo ‘he prescribed (or unknown) displacenent yt) excites the syste, Vowever, since the governing equation is m(x-y)"+ci+kx = 0, or (3) ve must take Pt) = jas the input, For this reason the system is known as an accelerometer. Comparison of (2) and (3) yields the damping fac tor Cacl2Vkm) and the undanped natural frequency wn = Vk/arad/sec. The corresponding cyclic frequency is fo = wol2x hertz (or cycles per second). There are several ways of solving equation (2). 37 “In‘vibration theory we study stable systens, so any transients “ysl be damped out with time. Thus we search for the steady- state solution only. When Plt)=Pym'cos2aft =Pomtcosut, it is not too surprising that the calculations can be simplified by using the identity etl! — cosantt + isinaatt, where i?=-1, The panalty is that one must deal with complex valued quantities even though the original equation was a real one in real variables. This technique consists in studying net the original equation (2) but the analogous equation E+ ugk + vbx = mtorr, @)" in which x +iy 4s the complex response. Then, provided Re{&) = (Rely) = x and Rei) = Rel)” = ¥, x will be the sol tion of (2) with P(t) = Pomfcos2aft 4.2, Receptance ce(if) ‘The receptance «(if) of the syste is defined by writing the complex response in the form x(t) = clif)Py el@tt (4) Much confusion can be caused by the common prac tice of not distinguishing between the actual response x and thea 38 Receptance ” complex response x. We will emphasize the fact that the recep tance is complex by regarding its argunent not as f but as if, The magnitude cf the receptance |a(if)| is the ratio of the re~ sponse anpLituée at each frequency. This is the gain g(f) or glu) of the systen, The phase shift © between the response x and the input is given byarge=-O(u), Thus, the performance of the single | freedom systen is completely described once the real and inagin | ary components of % are known at all frequencies. (The receptanc ‘ is also known as the complex frequency response, and the gain as the magnification factor, frequency response, or transfer fync~ tion). Substitution of (4) into (2)! yields wi) = er (hf + 2ibfiom (5) o 4 GE — wo? + 2kowdm : Pe up - ut - 2ifwu = gue {ug = ut! + 4thutuz}n Hence, the gain g(w)and phase shite 0(0) are © glu) = miu - otf + 4p°utud}?, ”) Qu) = arctan(2twusius - w*)) - 39 curve and the dashed curve, respectively, in Fig, 4.1, Since the receptance is complex, it is able to describe both the gain and phase characteristic of the Linear system. Fig. 4.1 4.3. Quality factor Q First we note that the phase shift increases con tinually with frequency from zero at zero frequency to its great, est value of 180°. The phase shift is exactly 90° at the undamp— ed natural fregency wy ; from (7) tan® = © vhen w= uy The Q factor is a measure of the resonant magni~ fication of the systen and can be defined as the ratio [oxi] Je0(0)] soo 0)” = Te is to be noted in passing thet the maximum gain actually occurs at the frequency w=ujl'-20)!, woverer,40. Receptance | 0 — when the viscous danping is small the difference is negligible, and when the danping is large the hunp in the gain curve is so flat that the difference is again negligible. (In stecl at engiy eering strain levels, $ = 0.003). Definition (8) is not convenient for experimental purposes because of the difficulties in measuring g(0)at zero | frequency. Nevertheless, with the values mglu,) = 2Bui), m0) = uz! given by (6) we obtain the important theoretical relation ' (9) a= ze : , Using the above value of the damping factor for steel we obtain Q=160 . At lower strain levels the Q factor can be much higher. Amore conmorly used relation defining Q is found as follows: Let wand wy be the frequencies at which the phase leads or lags that at resonance by 45°, respectively. Using (7) we have 2uguihw§ — wi) , +1, uow,lus - 0) = Hence yy Gary + Gee and ty = Cup ++ Chu. rus w= wy = 2Bu9, (10) Tt can be verified that laCitp] = ge) = /amV2Quiwo , [eG] = fwd W220 , and ve know that [aGfo] = glo) = Wemtus . once again assuning that Q is fairly high, it can be said that the gain at w, and w, is 1/V2 that at resonance. This atten uation in decibels is 20logigV2 = 3db. Thus, these points, marked Pj and Py on Fig. 4.1, are often called the three-decibel points, Since the power during steady state conditions in a lin car systen is proportional to the square of the anplitude, Py and P, are also called che half-power points. Experinensal data on the danping capacity of ne- tals is often expressed not in terms of E but rather in tems of the logarithmic decrenent or, more commonly, in terms of the specific damping capacityAWW, where AWis the energy lost per cycle and W io the maximum potential energy during the cycle. (unfortunately there is a host of symbols used for these quanti- ties). If A, , Anat are the amplitudes of two successive positivee e in composite system 8 peaks in the damped transient, then she calculation of & because only SW is given as a function of ress levels to caleulate W ve need to know Young's modulus AWIW = (AD = AL OIAR #20 ~ AgsDlAnat > (eed wate at and set Wo= o%(peakV/2E = oXrms/E . However, if we take E=2x10"N/n?for mild stecl at ¢ =100MN/m? ve obtain 6WW=005and hence F = 0.002 b= tA SAL © (A, AnaillAy « Moreover, the transient solution of (2) to be given in Lecture 5, shows that | 4.4, Receptanee in composite system Consider the case of a danper on an elastic foun 1 Alaa = explant - oy? et eee ae bl dation, The equation for the spring is + exp@xt). , hence we have the conversion formulae i 1 ow : 29 = 1h, - 2W ! The data given in Fig. 4.2 is insufficient for t 4 Sree Dengan Emrsy 8M . q rT " pafeyeto) y [uate i . T tat ayes Oopntene at Spee Dnopng 1 eae! nargy AW on Shress mpi |W tatigue Coe 10 Caatariat sit Elimination of the intemal force F and displacement x, yields fatigue Unit ‘ xs(-ms? +k + imBk) = SieaaeAnplitde OM?) a a an ae 100 meet 408 Fig. 4.244: Receptanee { Hovever, by definition of es4, we have Ess onP 4 mks - uP + twk/c) © (the suffix notation is the sane as for influence numbers. The oy = finst suffix gives the station where the response occurs and the second suffix gives the location of the excitation). By applying a force Psat station 3 and calculat- ing the conplex response x; of the mass, we find from the relation X= MPs ‘that ey = tse ’ This is an important reciprocal property of receptances that holds even for continuous systens, Chapter 5 VIBRATION THEORY Il: IMPULSIVE RECEPTANCE. 5.1. Introduction In the preceding lecture the excitation P(t) of 4 Linear systen was considered as a superposition of sinusoidal components. By virtue of the Linearity of the systen, each sinus oidal component can be treated separately and the corresponding responses added thus leading in a natural vay to the concept of a frequency-dependent function, the receptance a(if), describing the behaviour of the systea, In the present Lecture a tine-uise decomposition of the excitation into a series of equivalent impulses will be considered, This leads to a time-dependent function, the so-cal led impulsive receptance W(i)describing the behaviour of the sys ten, It will be shown that for a given system aif) and W(t) are in general a Fourier transform pair, i.e., each is the Fourier transform of the other, thus establishing the theoretical equi valence of the two descriptions. Bach of the above’ two recoptances will be used at tines in the following lectures, and it will becone evident that the choice of one or other receptance can depend on (1) analytical. convenience, (2) whether interpretations in the tine donain or frequency-domain are easier in the physical problen in hand, (3) computational efficiency in the case of large-scale46 Impulsive receptance data handling. 5.2, Response to deterministic excitation P(t). Sampled data Tt is becoming quite usual to record the values of a variable P(t) only at unifomity spaced times t =nA, where n are the integers and A is the sampling period. For the purpose of calculating responses it is then convenient to assume that P(t) = P, = P(nd) during the sanpling interval (i-t)Af. Here we wish to54. much 0} is ofte enfin the pass band. « To see what @ consider the va how the phase of «& varies wn overlooked when choosing & mpulsive receptance sin(-a+ 2008 9 Ca + ent) Cent +2) radical effect the phase char: two rather extrene cases in the pass band. 5.4041) a sineat +f act u [ooscas2ndtat ay (this fact filter). Suppose aif) = Let us determine the cor- responding W(t) using = fowtortar gaset, a-0 sin2atit wt) = 2—35¢ eristic has on W(t) | pxamples case 2, a=x/2h, WO) now shifted distance 1/4f, to right. (4) Spring-danper system aif) = —L k+iwe te The behaviour of =] 55 @ for case 2 in the preceding example is an approxinatioa to that of the spring-danper sys- tea, However, in one case Wit) is oscillatory vhereas in the other it is periodic, i.e, g00d approximation to « does not imply good pointwise approximation to W(t), and con versely.Chapter 6 DESCRIPTION 111 : AUTOCORRELATION FUNCTION 6.1, Autocorrelation and spectral density Me now show that the spectral density S(f)is the Fourier transform of QR): that is br sf) = f ARclew*Feaee, wien KG) = fin £ fa(bx(teaat, Ty Making use again of our fonction aot restricted range, which fs equal to x(t) in the range-JT vo 4, and 0 outside, Jrcoe*ae = f {i ay(t)aylt + eat} ett ae I xy(t)ag(t + ce entlton pizattat gig im, ioe { [=iser athe lsdereteds a e with S=te? = LayinasGe 2 (a) = FlaGol (o) Ye have also the inverse relation: Ra)=[ 4 s(pe!*'rar, ee ee eters cree ee eee cme te covered separately by Wiener and by Khinchin, and is often re~ ferred to by their nanes. This can be taken as the definition of Sif desired, and the other properties deduced fron it. We may repeat here that the three quantities, R(t) , and S(f), are all defined in terms of a single realisa~ tion, of infinite duration, of a random process; but if the process is stationary and ergodic, they are characteristic of the process as a whole, The autocorrelation function and the spectral density measure essentially the same property of @ random process, aiid we use whichever is most convenient for our purpose at the tine, Physically, the relation between these tx foras is, that the presence of high frequency conponents in the random process, as shown in the spectral density, destroys the correlation between successive values of x except at short tine intervals, as show by the autocorrelation function decreasing rapidly with ¢6.2. Examples of spectral density We illustrate the meaning of these quantities by applying our analysis first: to determinate functions, and then to idealised random functions, A summary of the relations which we have obtained so far is given in an appendix to this chapter, together with some examples of Fourier transforms witch we shall require, and a note on the impulse function &(t) . 6.3. Sinusoidal function Let x(t)» acos@nfyt +4) where fo is a constant frequency and is the phase angle tr = [ atcost@aft+ddt = Ar 4 RG) = fm 4 / ab cosxfgt + Acos(2nfy(t +z) + Sat “jr = Lat cosentyr =i ty sf) = 2 [ie cosentye ede = ¥ fleteron, pteoejae = Sa*[a(t—to) + OF Ff] 59 t pepresents Ssolog RS 42 gu peats, of area Lat, ‘the points f= fy and Aas @ check 5 - Fig. 6.1 ‘These functions are shovn in Fig. 6.1, We note that a periodic component in x(t) appears as a periodic component in ‘R(t)and as a pair'of isolated peaks in S(f), The phase angle # does not appear at all. GA, General Periodic funtion We can vrite down at once the corresponding re- sults for any function which is represented by a Fourier series, the sunmation of terms in R being justified because the cross- correlation functions are zero. x(t) = ag + Ya,cos2anfot + Vib, sin2mnfot = Yic,¢%2antot 7 t : aah, Gi+bh) = He i RG) = ab + h(a + bheos 2anfye = Vic, cos2enfye ‘ 460, Autocorrelation functig. 7 re jom pulse signal SO) = Zab a(t) + Zp + By)LBC -nfa) + OCF + nfl] Let x(t) have the form shown in Fig. 6.3 cefined = BlesP[ote-nty + r+ mt) « malt : al 6.5. Square wave {UU et [™. As an application of the preceding result, suppox Fig. 6.3 X is a square wave of amplitude A and period T . i 72> eee eel | gy the following property: x is either +a or » the changes Then x(t) = DPlsinzx ts Leino d + bsin orks} of sign occurring at randon times, with an average time interval Z T T T, + The events (changes of sign) occur according to a Poisson Re) = OR feosent ++ cos6xt +L coston® +..] probability distribution, We shall discuss this later, ani for ® tas TF acest the moment we quote the result that the probability of the occur 2 wich represents a triangular wave of anplitute A and period T. rence of k events in a specified tine 7 is given by: Pk) = oo where oo which represents a series of pairs of isolated peaks of diminish The auto-correlation function is the average value ing area, shown in Fig. 6.2. of x()x(b+ 0) . tis product will have the value -a® vien t is chosen so that there are an odd munher 1, 3, 5, ete. of chan 7 4 math 7}V VO Fig. 6.2 = a°(P(0) + P(2)+...J-a®[ PU) +P) +...) . Ses of sign between t and $+, and the product will be +a for an even number 0, 2, 4, ete. changes. So = o?Pr[even number 22Pr[odd munber ]Autocorrelation function vaiform spectral density 63 @ Using the formula above this will be | i ¢ wae QD go> _allade*t + i afer OP ]-afoo rete is an even ction. ve therefore write Radaate™M" , since R ven func 2 ‘the mean square value is of course 2° iene ge Pit ent spectral density Sif) = 2 [tem = dat [o-P* cos2aftae a ae 20th ee ae jon and spectral density are this pair of autocorrelation function and =P! of frequent occurrence in theoretical work, { 6.1. Specified autocorselation function Tt is sometimes convenient to start from @ specif- il yeads to A Jed form of RZ) + We have just shown that R@) = ve ‘the form: z si) - 48K. z 4xth +k 4 and use she Fourier transfor we find Por coseitfot af we take RO) = of this function given in the table, sy = —Aatklantt ete] (ante? + #3) ai] o4ntt? eG If k is not too large, this has a maximum near f =f), as shown in Fig. 6.4, represent~ ing a predominating group of frequencies in x(t) * in the neighbourhood of ‘this value. 6.8. Uniform spectral ‘density onan Tf ye wish to start from a specified spectral den sity, dt is natural to think of a unifom value: S(f) =a®, inde- pendent of f. This would give Rt) = $0, representing a peak of area ie at ¢=0, and 0 elsewhere, In fact, however, this bears no resemblance to our usual image of a random function since x(t) would be dominated by indefinitely high frequencies; evidently the mean square has an infinite value, Nevertheless it is often found useful to use this form as a hypothetical “input to a sys~ ten when ve are really interested in the output. Brery systen ultimately rejects the highest frequencies to vhich it cannot respond; and, provided we have a plausible specification for the inpue over the effective frequency range, it does not much matterAutocorrelation function 64, i input is eall- hat input ve Gropose beyond this points Such an ATP as ed twhite noise’. ourse specify S(f) as constant over a We can of © ghis may be tern, and zero outside it. certain frequency TANEes — «tho following for RC). | ed bandlinited vhite noise, This give ays) = @ for \ =H Re) = Pegineatee ry x*(O> i, ; @s = # for fe = ay-f iach Bfy be ——r—"—_-TEF band of width fo ween fy and fy Z fp = fo-4bfo poe f= thee amd fe 8 inxlbt yt Ree) = Zzeoseafor sinx(of)t = FH - APPENDIX FOR CHAPTER 6 L. Summary of relation between characteristics of random processes Mean square value of x: Autocorrelation function: R(x) spectral, density: S(F) Fourier transform: ACif) AGt) ig-defined as Jrsvesenatyre xt) = Jane Kn is defined as tim £ tf =(Hdt tr + Rx) As defined as tin + “] x(t)a(t + a)dt chr S(P) as dotined as fim 2/0, wien AGH) = fayer**¥a sy) =X ic fT it { = RO) = ~ [sow | x= efrcoete ae | Raed famereer re,written 6(t), is represented by a BF the unit impulse function (Dirac delta function) , ‘aph vhich is zero everywhere xcept in the neighbourhood of the point F< 0 where it has a peak having an area of unity, The duration, amplitude and form of the peak are not prescribed, xcept that the duration is in- g(t) definitely small. Tt obviously corresponds to the concept of an impulse in dynamics, The notation A(t -t,) represents a similar 1 of the point t= to B(t) 48 not but within the a) as, and peak in the neighbourhood Se cused matinee ae inits of the definition &t can be frechy used in oauatie ext ivative and an integral. A formal jus- eo even provided with a de 3.2) who tification of the procedure is given in Kaplan (ref. tgeneralized functions! i F(t).8(t-tokdt = Flte) calls these things er tlcosenfot with k>0 We have the general result that at ett for any function F(t). vrom this we ‘gan deduce the Fourier transforms [ometerrat = 4 (as a function of f) and the Snverse | frotrtat = ot), it et with the corresponding pair cosentet fonersetat = (as a function of +), and ~ sin 2nfgt freeetar = of). impulse function 3, Table of Fourier tansforms (double infinite integral . (double infinite i wren = ve ehat at gf funean otf) ar) 2k att ae ak[ane? + #3) +k] Tate? ae cneye_ anteee [ane el ante 1S ersten ura) 1 $0 ~ 1) 4 $80 +10) $46 + 1) doer).Chapter 7 RANDOM VIBRATION RESPONSE I + GENERAL RELATIONSHIPS | | 7.1. Introduction We can now bring together the res m input to a syS~ wults of the pre~ ‘the effect of a randor coding chapters to study tem, of which the output is related to the input through a Lin- car differential equation with constant coefficients: We deal, igh the situation in which there 4s a single random in- first Jingle point in a’ constant for exanple a force applied at @ 5: puts ed in’more than one output. although we may be interest direction, superposition of random Pro- pecanse of the complications of the we defer the case of multiple in- cesses, referred to in § 2-11, puts till later (chapter 10). ye shall find that the auto-correlation functions .4 by the impulsive receptance of of input and output are Linke Linked by the receP- ‘the system, and the spectral densities are tance. stem to random input 7.2. Response of linear sy$ Let the input P(t) correlation function be a stationary, ergodic ran- don function, with an auto~ RAs) = + Response of Linear system 69 Assuming that the output is a similar random function x! . 1 (t) have RG) = . ‘The property o! i a perty of the system is defined by the impulsive recept: oe receptance sd in a preceding chapter Then the output x for any input P is x(t) = [WORE - var. a RS yee f ‘Wee)PCt - ade f WCEP = te + Udz,> « a mons distinction between them, — Setene cesta ene spect to ty s t, so if we change the order of integration we get Rc) = Fwcaol Faceperte-e; [ Ke [Wen .

= find [PODAL, wich o inte 7 Mean and variance | = Wa ti acor—ourfat eat Te _ = [ure -dode 3 = 40> fuoae since is a number, independent of f and 7. put we haver aif) = fwceder'*Fa a so emae [Wendie = «i0) _sed> =

, (0) is'a veal number, Aso ee = 2K () co the mean square of the output is finite, although that of the jnput would be infinite, This arises from the attemiation of the higher frequencies, visich 4s showm more clearly by considering che spectral. density ron the above, klif'= <1, 50 S(f)= oe keaete Keath e 4 For white noise input Sp(f) = a and S,(f) = ert Keanttd hese vill be recognised as the associated pair of R and S previously discussed (§ 6.6): For white noise input, = a! efits a which gives z&— as above. ack Integrals of this type are of frequent occurrence and are te- Gious to work out for more complex systems; a table of some stan- dard types is given in Crandall and Mark, page 72 (Ref, 1-1). Lf we suppose the input to be of the type eater tl 42eb R= ate, S()= wee ) applied to same systen, fo -alen we obtain R,(@) =74 and Sf) = these give <=> [note that the integrals for Ry since Rp is represented b itive and for negative valu 7.5. Random input to seco at all frequencies in of second order has will be reflectes ‘The situation Fig. 71s we General relationships 4a’b att + ante?) 2 2 ik+bo) * yy different algebraic es of the argument] . wnd order mechanical system In the previous example, s a natural frequency of its 4 in the output unless the damp: as it affects the spectral densit Fig. 7-1 passing-through the system; have to be done in sections forms for pos~ the input is attenuated but a system own, and this ving is too Large. y is shown in second order mechanical system 15 ‘Taking a conventional system consisting of a mass controlled by a spring and viscous damper, the equation of mo- tion will be mésck+kx = P(t). ‘this gives the impulsive receptance ectetlot ein 2atat wt y amt if the damping is small (& <1) s0 that the response to an in pulse is oscillatory. Here the natural frequency is fy = we Vi danping factor § = ZS Rig frequency of damped vibration fy = fy o For a white noise input wi white noise input with Rp(t) = ZaP6(E) , this gives deck Vi - ¢ R= with cos This gives = R,(0) = a°/4ck. The dan damped frequency fy of the system shows up in the output auto-correlation function. ‘The receptance for this system is:General relationships : so that if Sp(f) = a®, 2 SFr re 2 2felitdt = at/Ack as ab { We can verify that Fig. 7.2 hese results are shown a peak at frequencies plitude of the peak depending on the value of § «The of the value of S(fq) to $(0) 48 48" a* (see § 4.3). system with small danping the peak will be sharp, and # put will be a random motion with the frequencies in thi ourhood strongly predominating. in Fig, 7.2. The spectral density has | in the neighbourhood of f = fy » the am) For a the out- s neigh- Chapter 8 VIBRATION THEORY 1 : NORMAL MODES 8.1. Introduction In this chapter ve will be concerned with the modal analysis of the response of Lightly danped Linear’ systens ‘To begin with ve wish to consider systems of such complexity that they cannot be regarded as a collection of concentrated masses, beans, shells, ete, whose partial differential equations can be formulated on the basis of geometric dimensions and ma terial properties, Frequéntly it is found that the configuration of a machine = even as complicated as an aircraft - can be de- scribed adequately in terms of a certain set of n displacements or rotations qi(j=1,2,--.n) at judiciously chosen points P; - It is to be noted that in assuming that the q; describe the configuration independent of tine ve are to a certain extent ex- cluding systens with time varying parameters. For the sake of sinplicity it will be assuned that no forced displacements are imposed on the system. Thus, in the parlance of higher dynamics, ve will be considering scleronomic as opposed to rheonomic sys~ tens, After deciding on the n significant generalized coordt- ates qj» the next step would be to conduct vibration tests on ‘the given machine by applying sinusoidal forces at the points Pj. Afeer considerable effort one will be in a position toNormal modes 2B systen of second-order differential. equations governing : re been increased or (the necessary size of n may have : sual. test procedures decreased during the tests). sten from the theory of modal, response bt {ficial Lines of argument Since the u: ased on the governing equations, ve will follow the usual art: by starting with these governing equations naeagegg - 9 i.e. (1) 7” ”. ” 27 pedi Dudu + CGM = See ‘the Lagrangian method of setting up equations A and a stiffness matrix § that are : | (4) yields amass matrix i ‘is clear that A) must.be positive | symmetric, In addition it definite in-all real cases, since the Kinetic energy of a nov- | ro of negative. ing physical system can never be 76 BF Here as in Chapter 4 we will use the complex no- ent sinusoidally varying quantities. ve write 0, = (f,+49))° the response will be xf the sation to repres forces in Q are all at frequeney & io in view of the linearity, sexp(iwt) . : (ujrivPexpGod 5 4 oo ge (yt tpesptiut. subset s ation of the fac- of these expressions into (1) and cancel | tion tor, exp(w!) yields (Griug-wpPyrigd = frig. i Q@) ial introduction a which is equivalent to the two real equations Guy - wBy + Be ol a Here it 4s to be noted in passing that, being the ratio of the complex response to the complex: excitation, the receptance of the present system is given by g(ww)=(G+iu8—utAYs which is similar in form to the corresponding expression 4.25) for the one freedom case, If all the forces are in phase we can set g =0 with ff desoribing the distribution of amplitudes of che" txoi- tation, the system is said to be vibrating in a nomal node when the frequency wand {are such that all the displace ments have a common phase differing from the excitation by 60°, (Alternatively, the forces and velocities are in phase), Equa- tions (3) then reduce to “By = f> (a) G-o'py = g- 6) In order that (5) have non-zero roots y, #0 it is necessary and sufficient that det(G-ufA)=0, which is the characteristic equation for the associated conservative systen Ag+ Gq = 0. The vector yy is called the k-th modal shape and is obviously Andependent of the structure of the damping matrix B . The factNormal modes orthogonality 81 80 gative and symnetric guarantees that The zero roots correspond chat A) and Q are non-ne ail the roots wf are non-negntive, igid-body modes and are usually treated separately, Since Fg avo sancti an one at Last is sestine defines 27 proved eee incarly Sndepentent wodsh shapes Bi ) even if there are Tess than nm distinct modal e than one mod- fo one modal frequency, The probability Being Linearly indepen- are defined by (5) frequencies Wy » i.e, it is possible to have mo" a1 shape corresponding t fairly low. of such degenerate cases is the modal shapes is that any dent, the important thing abou n-dimensional vector v can be f the n modal shapes i.e. represented as a Linear combi- nation of gon auger aeyet et Rake * © 8.2. Onthogonality ‘Two veotars ft > Geqyie21) and Be (xasyerte) | —rr———— “=E™™”— zee | + yset H1%2 vanishes. calar product can be written If the above vectors are treated a= colt wectors, the 8 aegine opine! here superscript T indicates transposition to 4 TOW veetor. A | weighted scalar product can be defined by axtetbyiyet CHEE | which in matrix notation is written a 0.0][xe [enyoz]]/0 b Ollye] = riDee. 0 0 elle. Having gone so far, the last step in generalizing the idea of orthogonality is to aduit a symmetric non-diagonal matrix D for caloulating thé scalar product , riDr, = rJDr, ridge = pidge = duce t delve + may te. - : Thus in two dimensions the "Length" of a vector 2, y) becomes dyx*+2diozy + dee. This is simply equiv. alent to using ellipses instead of circles to measure the di tance from a point, Because D de- - fines how we measure distances, it is called the metric, After these prelin- inary remarks, let us show that the normal modes are orthogonal in this generalized sense, i ized ,, the mass and stiffness matrices playing the roles of metrics, For two distinct modal. freque quencies wy and Wp 8.1(5) reads 2 gu-uidys = 0, Gye-widy = 0-7) Premulesplyin ying the first equation by yf» the second by ylNormal modes ff 83 and subtracting the results we obtain otylam-eigidze * 9° due to the symmetry of and since wt-w) 40, for the Hence, y relations nodal vectors we have the orthogonal st @) pidge = O- ss of A means that worebver, the positive definitene (9) ylAy = m2 0 and gaye = mz >O- Now from (7) we have the additional orthogonality relations vhen tb 0 1 zy? co) vidas = Ore ARE | even bd wt i cto explain the weility of the tv0 oF a thogonality relations (8), (10) con on an elastic \ | sider two unequal masses cable, With’ a suitable choice of units 4 0 at wene hb, g42 3]. . b I & i 4 she modal veotors By 2 Be can b° found from the orthogonality relations (8) and (10) above: yi = 4% -3) orthogonality yy = (1,-vs-3 ae cree ) the orthogonality relations yield the modal shapes without re- quiring the calculation of the corresponding modal freq. al. frequencies. Zn the case of proportional damping 8 is a line ear combination of A and/or & , so in this case we Rave vim -{ 0 when jf # etapa spk. OD Let us now see how cae o see how tho response qin system (1) e expressed in terms of the moral modes. For the sake of deriving a model expression for the receptance ve et the +420) ication te Q=pexpliut). subseisucion of g= (ys +. explwt) yields aj =v i 2% AG vA +inBhy, =p - (a2) In ‘view of the relations (8), (9), (10) and (11) premltiplica— tion of both sides of (8.12) by yf yields a> yt 4 mAak- 8 + 2:8 0,0} ro (13) 4 . —BeBgpo_up HOON.84 Normal modes From this expression it is clear that any node yj can be SuP- pressed by making the force vector p orthogonal to ity dae. K C., Alternatively the amplitude of node wip = quency differences (w} wut). ‘the can be ine wR creased by reducing the fre receptance Gp, af station r due to & simisoidal excitation .d by inserting a foree vector p with at station s is obtaines vent and zeros elsewhere. If Vy; de~ a 4 in the s—th compon notes the *—th component of vector yj s it follows from (13) hat (4) ee = Mes GE oe + aie wyu} ia np{of 09 + 28,0.) A similac expression for the receptance of a continuous system ‘can be derived, For instance, in the case of a uniform bean of visco-elastic material we have $ Ww, . PA i) Mj {o* = w®) + 2iejoa) fiore the wodsl finetions W;(s) satisfy the equations \ EIW.%) - mut We) = 0 team, The mass per unit Length of and the end conditions on the the beam is denoted by m . The gencralized mass is defined as w}(s) dx. The modal damping factor is defined as gj Voy 2e where V ‘and E are the viscous and clastic mod~ Normal. coordinates 85, duli in the stress-strain relation 6 = EE + Vé ‘All modes above a certain : tain frequency will be over-critical danped, since the dé factor _ ees factor increases with W; 5 this is character istic of all systems in which the damping matrix is 8.3. Normal coordinates The n norm vaca 1 normal. coordinates of system (1) are re~ ated to the initial more or less arbitrarily chosen coordinat coordinates qty the matrix transformation go ME = St. + 8p, (15) where Yj is composed of the n colum vectors wv, 1. Bir be. Mi M2 Ms Y= Uwoperoga) = | + (6) », rat Substitution of (15) into (1) yields AVE+gye+cye - g, which after preluntiplying by V' and taking account of the or ing by V ing a thogonality relations (8), (9), (10), (11) becomes y (8), (9), (40), (41)B+ bet an : where the normal forces aseion (11) 0 equations (17) would ever, since the curr , are base’ masses ™; cous damping is more of @ jon of the usual. mechanics script: Little point in con cept as a last resol te is seen’ from the set rns coordinates $j G+ 20,048, + Habs mn the damping matrix were 1 28d + wits = 2Botwebe + ibe 7 components of YQ + Tf cone ot fulfilled the normal are the ‘be coupted through the velocity cerns, Hove ene methods of determining the generalized 41 on an wndanped model and the use of vis convenient artifice than a true de- al situation, there seems to be sidering other than proportional. damping - cond way of writing (as) are a measure of what proportion that the no! of each normal mods .e response q+ fe appears in the response 9 Chapter 9. RANDOM VIBRATION RESPONSE I: SIMPLE EXAMPLES 9.1. System with two degrees of freedom To illustrate the situation where a system is to be regarded as one with a few degrees of freedom, with lumped parameters, we consider the conventional system of Fig. 9.1 We suppose that the input consists of random movenent of the base, re~ presented by the coordinate x, and that we are interested in the result ing movements of the two masses re- presented by the coordinates x and y , We require the re~ ceptances i, and ty, » giving the outputs at x and y respective ly for a unit sinusoidal input at % . The receptances here are ra Fig. 9.1 tos of displacements instead of ratios of displacements to force, bat it is convenient to use the term in a wider sense to cover inputs and outputs of various natures. ‘The differential equations for this edse are tot ede t(kyt kx -cay-key = ck thyzSimple examples Normal: modes 7 88 = ma + ce +key ~ ce Kae = 0 iat wet the desired E yor z= Zeit, x— seh y= yore? , we get the tions of i ratios xo/% and yo/20 a8 funct! re = wXlingky + x62) — bw macy * D D and the denominator D is 'm, im, A 4 make + O62) + gots iwXmyce + myc, + Myce) ~ UPL + make + Make + O462 gat a iulkyee + kee) + Kike » ¢ there is no damping, the denominator reduces to a quadratic the d re a qf tl 5 natural frequen~ the roots of which give the two natural freq ‘ 2 ners the guantities| cies of the system, For 2 Jecgal” and [ety,[° wild bot modérate anounts of damping, hh have peaks in the neighbourhood of | ‘these frequencies, the shape of the peaks depending on the parane f ss e peaks depending on the P ies, sat is white noise at z ; eters of the system, If the random input is whi roporticnal to these sities at x and y will be Pp | the spectral densit tence sueputs can be quantities. The mean square of the output <> ; = 2 f Jeli de, as in 7.4 3 A detasled study sing the table of integrals referred to there. of this system, with reference to the effect of differént values of the paraneters, is given in Crandall and Mark (Ref. 1.1). 9.2, Use of normal modes of vibration The most general method of analysis for a system with many degrees of freedom is that based on the concept of nor- mal modes, as described in the last lecture, This method is best illustrated in tems of a system with distributed parameters, which we think of as having an infinite number of degrees of freedom. We shall again use the transverse vibration of a beam aS a convenient illustration, since for this case the idea of the nodes of vibration is a familiar one, aid the modes themselves can be easily visualized, We suppose then that we have a random force P applied to the beans at a point determined by a coordinate x, , measured along the length, and that we wish to find the movement at a point with coordinate x, It has been established that the movenent ww(tp,t) at this point for a sinusoidal input Pye ig given by F__ willed sat (x98 0) A= 28,050" = eafndPoe™t, where ciay(iu) 4s the receptance Linking output displacenent at 2 with input force at x4, In this formila, w,(x) is the mode associated