7 Response of SDOF Systems to Periodic Excitation: Frequency-Domain Analysis
Transcription
7 Response of SDOF Systems to Periodic Excitation: Frequency-Domain Analysis
7 Response of SDOF Systems to Periodic Excitation: Frequency-Domain Analysis In Chapter 4 you studied the response of SDOF systems to harmonic excitation and became familiar with important concepts such as resonance. You also learned how to simplify the analysis of viscous-damped systems by using complex frequency response. These concepts from Chapter 4 are now extended to determine the response of SDOF systems to periodic excitation. Upon completion of this chapter you should be able to: • • • • • • Determine the Fourier series representation of a periodic function using the real form of the Fourier series. Determine the Fourier series representation of a periodic function using the complex form of the Fourier series. Determine the steady-state response of an SDOF system to periodic excitation using either the real form or the complex form. Use the basic definition to determine the Fourier transform of a transient function p(t). Show that the impulse-response function for a linear system and the frequencyresponse function for the system form a Fourier transform pair. Apply the FFT algorithm to compute the Fourier transform of a periodic function p(t), and plot the magnitude and phase of the Fourier transform computed. 7.1 RESPONSE TO PERIODIC EXCITATION: REAL FOURIER SERIES Forces acting on structures are frequently periodic, or can be approximated closely by periodic forces. For example, the forces exerted on an automobile traveling at constant speed over certain roadway surfaces can be considered to be periodic. Figure 7.1 shows a periodic function with period T1 , that is, p(t + T1 ) = p(t) (7.1) A periodic function can be separated into its harmonic components by means of a Fourier series expansion. In this section we consider real Fourier series. In Section 7.2 184 7.1 Response to Periodic Excitation: Real Fourier Series 185 p(t ) t T1 Figure 7.1 Periodic function with period T1 . complex Fourier series are introduced. The complex form is very useful when combined with the complex frequency-response function of Chapter 4 to study the steady-state response of damped systems. 7.1.1 Real Fourier Series The periodic function p(t) may be separated into its harmonic components by means of a Fourier series expansion. The real Fourier series expansion of p(t) may be defined as p(t) = a0 + ∞ ! n=1 an cos n.1 t + ∞ ! bn sin n.1 t (7.2) n=1 where .1 = 2π T1 (7.3) is the fundamental frequency (in rad/s), and an and bn are the coefficients of the nth harmonic. The coefficients a0 , an , and bn are related to p(t) by the equations a0 = 1 T1 2 an = T1 bn = 2 T1 % τ +T1 τ % τ +T1 p(t) dt = average value of p(t) p(t) cos n.1 t dt, n = 1, 2, . . . p(t) sin n.1 t dt, n = 1, 2, . . . τ % τ τ +T1 (7.4) where τ is an arbitrary time. Although theoretically, a Fourier series representation of p(t) may require an infinite number of terms, in actual practice p(t) can generally be approximated with 186 Response of SDOF Systems to Periodic Excitation: Frequency-Domain Analysis sufficient accuracy by a relatively small number of terms. Example 7.1 illustrates the Fourier series representation of a square wave. Example 7.1 (a) Determine expressions for the coefficients of a real Fourier series representation of the square wave shown in Fig. 1. Write the Fourier series representation of p(t). (b) Plot truncated series employing, respectively one, two, and three terms of the Fourier series. p(t ) p0 t p0 T1 Figure 1 Square wave. SOLUTION (a) The integrals in Eqs. 7.4 can be evaluated over the period −T1 /2 < t < T1 /2 and thus written in the form % 1 T1 /2 p(t) dt (1) a0 = T1 −T1 /2 % 2 T1 /2 an = p(t) cos n.1 t dt (2) T1 −T1 /2 % 2 T1 /2 bn = p(t) sin n.1 t dt (3) T1 −T1 /2 where p(t) = < −p0 p0 − T1 /2 ≤ t < 0 0 ≤ t < T1 /2 (4) Substituting Eqs. 4 into Eqs. 1 and 2, we get a0 = an = 0 Ans. (a) (5) This results from the fact that p(t) is an odd function of t [i.e., p(t) = −p(−t)], whereas a0 and an are coefficients of even terms in the Fourier series. The coefficient for the odd terms is % 4 p0 T1 /2 sin n.1 t dt (6) bn = T1 0 so 9T1 /2 9 4p0 −1 bn = cos n.1 t 99 (7) T1 n .1 0 7.1 Response to Periodic Excitation: Real Fourier Series But .1 T1 = 2π, so bn = − 2p0 (cos nπ − 1) nπ 187 (8) or 4p0 , n = 1, 3, 5, . . . nπ The Fourier series representation of the square wave is thus 4p0 ! 1 sin n.1 t p(t) = π n bn = Ans. (a) (9) Ans. (a) (10) n=1,3,... (b) The plots in Fig. 2 show the contributions over one period of the first three nonzero terms of the Fourier series representation of the square wave in Fig. 1. p (t ) p0 t (a) T1 p0 (b) (c) (d) t t t Figure 2 Fourier-series representations of a square wave: (a) square wave; (b) sin(.1 t) term; (c) two-term representation; (d ) three-term representation. 7.1.2 Steady-State Response to Periodic Excitation Having determined the response of SDOF systems to harmonic excitation in Chapter 4, and having determined how to represent a periodic function in terms of its harmonic components, we can now determine the response of SDOF systems to periodic excitation. In Example 7.2, an undamped SDOF system is subjected to the square-wave excitation of Example 7.1. 188 Response of SDOF Systems to Periodic Excitation: Frequency-Domain Analysis Example 7.2 The undamped SDOF system in Fig. 1 is subjected to a square-wave excitation p(t) like the one in Example 7.1. Determine a Fourier series expression for the steady-state response of the system if ωn = 6.1 . u (t ) k m p (t ) Figure 1 Spring–mass oscillator. SOLUTION From Eq. 4.9, the steady-state response of an undamped SDOF system to cosine excitation p0 cos .t is p0 /k cos .t 1 − r2 u= (1) √ where r = ./ωn = . m/k. From Example 7.1 we can write the square-wave excitation p(t) in the form p(t) = where ∞ ! Pn sin n.1 t (2) n=1 4p0 nπ Pn = 0 n = odd (3) n = even Hence, the steady-state response can be written in the form u(t) = ∞ ! Un sin n.1 t (4) n=1 From Eq. 1, the coefficients Un of the steady-state-response series have the form Un = Pn k (1 − rn2 ) (5) n.1 ωn (6) where rn = Finally, combining Eqs. 3, 4, and 5, we get the following Fourier series expression for the steady-state response: u(t) = 4p0 kπ ! n=1,3,... 1 sin n.1 t n [1 − (n/6)2 ] since, for the present problem, ωn = 6.1 . Ans. (7) 7.2 Response to Periodic Excitation: Complex Fourier Series 189 pPn 4p0 1 wn = 6Ω1 Ωn Ω1 3Ω1 5Ω1 7Ω1 9Ω1 7Ω1 9Ω1 pkUn 4p0 1 Ω1 3Ω1 Ωn 5Ω1 Figure 7.2 Excitation and response spectra based on Example 7.2. It is very convenient to visualize periodic functions in terms of their spectra, that is, plots of the amplitude of each harmonic component versus frequency. The spectra of the periodic excitation p(t) and resulting steady-state response u(t) from Examples 7.1 and 7.2 are plotted in Fig. 7.2. From Eqs. 3 and 5 of Example 7.2, the following nondimensional forms are obtained: 4 n = odd Pn = nπ (7.5a) p0 n = even 0 4 n = odd Un nπ[1 − (n/6)2 ] = (7.5b) p0 /k n = even 0 and .n = n.1 (7.5c) Note that for the particular frequency ratio ωn = 6.1 , some of the Fourier components of the excitation are at frequencies below resonance, whereas others are above resonance. The response would have a very large Fourier component if n.1 were to fall close to ωn for some value of n. 7.2 RESPONSE TO PERIODIC EXCITATION: COMPLEX FOURIER SERIES In Section 4.3, the complex frequency-response function H (.) was introduced as a convenient means of representing the response of a viscous-damped system to harmonic 190 Response of SDOF Systems to Periodic Excitation: Frequency-Domain Analysis excitation. A complex Fourier series representation of periodic excitation and response functions will also prove to be very useful. 7.2.1 Complex Fourier Series Let the periodic function p(t) be separated into its harmonic components by means of a complex Fourier-series expansion, given by1 p(t) = ∞ ! P n (.)ei(n.1 t) (7.6) n=−∞ where the fundamental frequency .1 (rad/s) is related to the period of the function by .1 T1 = 2π. The bar over Pn symbolizes the fact that the coefficients of the series may be complex, even though the series as a whole represents a real function of time. To evaluate these components, note that : % τ +T1 0 n &= m i(n .1 t) −i(m .1 t) e e dt = (7.7) T1 n=m τ So, multiplying Eq. 7.6 by e−i(m .1 t) and integrating over one period, we get Pn = 1 T1 % τ +T1 p(t)e−i(n.1 t) dt, τ n = 0, ±1, . . . (7.8) Note that ∗ P −n = P n = complex conjugate of P n (7.9) and that 1 P0 = T1 % τ +T1 τ p(t) dt = average value of p(t) (7.10) [Actually, P 0 is real-valued since p(t) is real-valued.] Example 7.3 (a) Show that if p(t) is a real-valued function, the right-hand side of Eq. 7.6 will turn out to be real-valued, as it should be. (b) Show that if p(t) is an odd function, P n (.) is purely imaginary and the coefficients P n and P −n are related by the equation P −n = −P n SOLUTION (a) From Euler’s formula (Eq. 3.14), e±iθ = cos θ ± i sin θ 1 By and e0 = 1 (1) including negative n terms in this series, the Fourier series can represent a real function of time, p(t). 7.2 Response to Periodic Excitation: Complex Fourier Series 191 Therefore, Eq. 7.6 can be expanded into the following form: ∞ ! p(t) = P 0 + + n=1 ∞ ! P n (cos n.1 t + i sin n.1 t) (2) P −n (cos n.1 t − i sin n.1 t) n=1 Since P n is assumed to be complex, it can be expressed in terms of its real and imaginary components as P n = -(P n ) + i .(P n ) (3) Then, from Eq. 7.9, ∗ P −n = P n = -(P n ) − i .(P n ) (4) Combining Eqs. 2 through 4, we get p(t) = P 0 + 2 ∞ ! n=1 [-(P n ) cos n.1 t − .(P n ) sin n.1 t] which is real. Q.E.D. (b) Making use of Euler’s formula, Eq. 1, we can write Eq. 7.8 as % τ +T1 1 Pn = p(t) (cos n.1 t − i sin n.1 t) dt T1 τ Since p(t) is said to be an odd function of t, the cosine term drops out and % −i τ +T1 P +n = p(t) sin n.1 t dt T1 τ (5) (6) (7) Then P −n −i = T1 i = T1 % % τ +T1 p(t) sin(−n.1 t) dt τ τ +T1 (8) p(t) sin n.1 t dt τ Hence, from Eq. 7, P n is purely imaginary, and from Eqs. 7 and 8, P −n = −P n . Q.E.D. By comparing Eq. 5 of Example 7.3 with Eq. 7.2, you will notice that the coefficients of the real Fourier series and the coefficients of the complex Fourier series are related by the following equations: ao = P 0 , an = 2-(P n ) bn = −2.(P n ) (7.11) 192 Response of SDOF Systems to Periodic Excitation: Frequency-Domain Analysis Example 7.4 (a) Determine an expression for the Fourier coefficients P n of the complex Fourier series representation for the square wave of Example 7.1. (b) Sketch spectra of -(P n ), .(P n ), and |P n |. SOLUTION (a) Equation 7.8 can be evaluated over the period 0 < t < T1 , giving % % 1 T1 /2 1 T1 Pn = (p0 )e−i(n.1 t) dt + (−p0 )e−i(n.1 t) dt (1) T1 0 T1 T1 /2 9 ' −p0 & −i(n.1 t) 99T1 /2 −i(n.1 t) 9T1 e (2) Pn = − e 0 T1 /2 in.1 T1 But .1 T1 = 2π, so < +1 n = even e−i(n.1 T1 /2) = e−i(nπ) = −1 n = odd (3) e−i(n.1 T1 ) = e−i(2nπ) = 1 Therefore, the complex Fourier coefficients are given by ip0 [2e−i(nπ) − 1 − e−i(2nπ) ] Pn = 2πn or n = even 0 ip0 −i(nπ) (2e Pn = − 2) = −2ip0 2nπ n = odd nπ (b) See Fig. 1. (4) Ans. (1) (5) (Pn) n −5 2p0 5p −5 −3 −1 1 2p0 p (Pn) 3 5 3 5 2p0 3p 1 −3 −1 −2p0 3p n −2p0 5p −2p0 p |Pn| 2p0 p −5 Figure 1 −3 −1 1 2p0 3p 2p0 5p 3 5 n Various spectra for the square wave of Example 7.1. 7.2 Response to Periodic Excitation: Complex Fourier Series 7.2.2 193 Complex Frequency Response The topic of complex frequency response was introduced in Section 4.3. Here we extend that discussion to cover the complex representation for steady-state response of an SDOF system to periodic excitation. From Eqs. 4.30 and 4.33, the steady-state response can be written in complex form as u(t) = U (.)ei .t = H (.)p0 ei.t (7.12) where the frequency-response function H is given by2 H (.) ≡ H u/p (.) = 1/k [1 − (./ωn ) ] + i [2ζ(./ωn )] 2 (7.13) where . is the forcing frequency and ωn is the undamped natural frequency of the SDOF system. When the excitation is periodic, we can use the complex Fourier series representation of Eq. 7.6, repeated here: p(t) = ∞ ! P n ei(n.1 t) (7.6) n=−∞ Then the steady-state response can be written as u(t) = ∞ ! U n ei(n.1 t) (7.14) n=−∞ Noting from Eq. 7.12 that for harmonic excitation U = H p0 , we see that the corresponding expression for periodic response is U n = H n P n = |H n ||P n | ei(αHn +αPn ) (7.15) where for a viscous-damped SDOF system,3 H n ≡ H (n.1 ) = 1/k [1 − (n .1 /ωn ) ] + i [2ζ(n .1 /ωn )] 2 (7.16) Example 7.5 illustrates the use of complex Fourier series in determining the steadystate response of a SDOF system subjected to periodic excitation. The method would be even more beneficial if the system were a damped system. Example 7.5 (a) Repeat Example 7.2 by determining an expression for the Fourier coefficients U n for the undamped SDOF system subjected to square-wave excitation with ωn = 6.1 . (b) Sketch magnitude and phase spectra, |U n | and αUn . 2 Here we use the dimensional form of H rather than the nondimensional form given in Eq. 4.33. The subscript n of the undamped natural frequency ωn should not be confused with the Fourier series index n. 3 194 Response of SDOF Systems to Periodic Excitation: Frequency-Domain Analysis SOLUTION (a) From Eq. 7.15, U n = H n P n = |H n ||P n | ei(αHn +αPn ) (1) and from Eq. 7.16, with ζ = 0, Hn = 1/k 1/k = 1 − (n.1 /ωn )2 1 − (n/6)2 (2) From Eq. 5 of Example 7.4, the Fourier coefficients of the square wave are 0 n = even P n = −2ip0 n = odd nπ Hence, the nth term of the frequency response is 0 U n = H nP n = −i(2p0 ) nkπ[1 − (n/6)2 ] n = even (3) Ans. (a) (4) n = odd (b) From Eq. 4 we can evaluate expressions for the magnitude and phase angle. |U n | = 2p0 /kπ |n [1 − (n/6)2 ]| (5) Because the nonzero terms in Eq. 4 are pure imaginary, the phase angles will all be ±π/2. Hence, : −π/2 n = +1, +3, +5, −7, −9, . . . αUn = (6) π/2 n = −1, −3, −5, +7, +9, . . . Note that sign changes occur as the resonance frequency is passed at ωn = 6.1 , that is, at n = 6. mn = pk |Un| zp0 1 n −9 −7 −5 −3 −1 1 3 5 7 9 an p/2 n −wn = −6Ω1 −p/2 wn = 6Ω1 Figure 1 Response spectrum (magnitude and phase). 7.3 Response to Nonperiodic Excitation: Fourier Integral 195 For sketching purposes, evaluate the nondimensionalized magnitude µn ≡ (π/2) |U n | |1/n | = , p0 /k |[1 − (n/6)2 ]| n = ±1, ±3, . . . (7) The values to be plotted are 36 = 1.029, 35 36 = = 0.396, 91 4 = 0.444, 9 4 = = 0.089 45 µ1 = µ−1 = µ3 = µ−3 = µ7 = µ−7 µ9 = µ−9 µ5 = µ−5 = 36 = 0.655 55 The results are shown in Fig. 1. Compare the sketch of µn in Example 7.5 with Fig. 7.2b and note that the complex coefficients µn in Example 7.5 have an amplitude that is half that of the corresponding real coefficients, which are plotted in Fig. 7.2b. The contribution of the −n terms in the complex Fourier series accounts for this difference. It is convenient to think of problems such as Example 7.5 in terms of transformations from the time domain to the frequency domain (spectrum), and from the frequency domain to the time domain. Figure 7.3 illustrates this. 7.3 RESPONSE TO NONPERIODIC EXCITATION: FOURIER INTEGRAL In previous sections you have seen that a periodic function can be represented by a Fourier series, as in Eq. 7.2 or 7.6. When the function to be represented is not periodic, it can be represented by a Fourier integral. In developing expressions for the Fourier integral transform pair, it will be convenient to employ the complex Fourier series, letting the period T1 approach infinity. 7.3.1 Fourier Integral Transforms Equation 7.6, which defines the complex Fourier series, is repeated here: p(t) = ∞ ! P n ei(n.1 t) Time domain p (t ) (7.6) n=−∞ Frequency domain Eq. 7.8 Pn (Ω) Eq. 7.15 u (t ) Figure 7.3 Eq. 7.14 Un (Ω) Solution of a periodic response problem by transformation to the frequency domain. 196 Response of SDOF Systems to Periodic Excitation: Frequency-Domain Analysis where, as given by Eq. 7.8, the Fourier coefficients P n are related to p(t) by % τ +T1 1 Pn = p(t)e−i(n .1 t) dt, n = 0, ±1, . . . (7.8) T1 τ provided that the integral exists. In letting T1 → ∞, it will be convenient to introduce the following notation4 : .1 = /., .n = n .1 P (.n ) = T1 P n = 2π Pn /. (7.17) (7.18) Then Eq. 7.6 can be written in the form p(t) = ∞ 1 ! P (.n )ei(.n t) /. 2π n=−∞ (7.19) where, from Eqs. 7.8 and Eq. 7.18, P (.n ) = % T1 /2 p(t)e−i(.n t) dt (7.20) −T1 /2 The limits of integration on Eq. 7.20 have been taken as shown so that when T1 → ∞, the entire time history of p(t) will be included regardless of the specific form of p(t). As T1 → ∞, .n becomes the continuous variable ., and /. becomes the differential d.. Then Eqs. 7.20 and 7.19, respectively, can be written as % ∞ P (.) = p(t)e−i.t dt (7.21) −∞ 1 p(t) = 2π % ∞ P (.)ei.t d. (7.22) −∞ Equations 7.21 and 7.22 are called a Fourier transform pair. P (.) is known as the Fourier transform of p(t); and p(t) is called the inverse Fourier transform of P (.). The representation of p(t) by its Fourier transform requires the existence of the integral in Eq. 7.21. Conditions that must be satisfied for the Fourier transform integral to exist are discussed in texts on integral transforms (e.g., Ref. [7.1]). These conditions are met by most physically realizable functions representing forces, displacements, and so on. Finally, the Fourier transform pair can be written in a more symmetric form if written in terms of the frequency f = ./2π. Then P (f ) ≡ F [p(t)] = −1 % p(t) ≡ F [P (f )] = 4 Note that Eq. 7.18 introduces a scaling factor, T1 . ∞ p(t)e−i(2πf t) dt (7.23) −∞ % ∞ −∞ P (f )ei(2πf t) df (7.24) 7.3 Response to Nonperiodic Excitation: Fourier Integral 197 Example 7.6 illustrates the use of straightforward time-domain integration to determine the Fourier transform of a rectangular pulse that is symmetric about t = 0. Note that the resulting Fourier transform is a real function of frequency. Example 7.6 Let p(t) be the rectangular pulse defined by t < −T 0 −T ≤ t ≤ T p(t) = p0 t >T 0 (a) Determine the Fourier transform of this rectangular pulse. Express the transform as a function of the frequency variable .. (b) Plot the Fourier transform. SOLUTION (a) From Eq. 7.21, % ∞ % −i.t P (.) = p(t)e dt = −∞ Therefore, P (.) = T p0 e−i.t dt (1) −T p0 −i.T − ei.T ) (e −i . (2) which can also be written in terms of the sinc function (sin θ )/θ . P (.) = 2 p0 T sin .T .T Ans. (a) (3) (b) The Fourier transform P (.) is therefore a real function. It can be plotted versus the frequency variable . (Fig. 1) and compared with the discrete Fourier series of Example 7.4. P (Ω) 2p0T Ω 0 −4p/T −2p/T 2p/T 4p/T Figure 1 Fourier transform of a symmetric rectangular pulse. Reference [7.1] contains a table of a number of Fourier transform pairs. 198 7.3.2 Response of SDOF Systems to Periodic Excitation: Frequency-Domain Analysis Frequency-Response Functions In Eq. 7.14 we found that the response of an SDOF system to periodic excitation can be expressed in the form u(t) = ∞ ! U n ei(n.1 t) (7.14) n=−∞ where, from Eq. 7.15, U n = H nP n (7.15) Following the procedure of Eqs. 7.17 through 7.24, we obtain the following Fourier transform pair for the response of an SDOF system: U (f ) = u(t) = where % ∞ u(t)e−i(2πf t) dt (7.25) U (f )ei(2πf t) df (7.26) −∞ % ∞ −∞ U (f ) = H (f )P (f ) (7.27) which is the product of the system frequency-response function, H (f ), and the Fourier transform of the excitation, P (f ). Therefore, the response can be expressed by the following inverse Fourier transform: % ∞ H (f )P (f )ei(2πf t) df (7.28) u(t) = −∞ In some cases a table of Fourier transform pairs can be used to evaluate this inverse transform.[7.1,7.2] However, evaluation of this definite integral generally involves contour integration in the complex plane, which is beyond the scope of this book. 7.3.3 Parameter Identification Of great importance (as demonstrated in Chapter 18) is the fact that Eq. 7.27 can be written symbolically in the form H (f ) = U (f ) P (f ) (7.29) The system frequency-response function, H (f ), can thus be obtained from Fourier transforms of the measured time histories of excitation p(t) and response u(t). Important system parameters (e.g., the undamped natural frequency ωn and damping factor ζ of the system) can then be extracted from this system frequency-response function. For example, for a viscous-damped SDOF system, H (f ) is given by (Eq. 7.13) 1/k H (f ) = (7.30) [1 − (f/fn )2 ] + i(2ζf/fn ) Such parameter identification procedures are discussed in Chapter 18 and in modal analysis references such as Refs. [7.4] and [7.5]. 7.4 Relationship Between Complex Frequency Response and Unit Impulse Response 199 In Section 7.5 we describe how the Fourier integral can be approximated by the discrete Fourier transform (DFT), which, in turn, can be evaluated numerically by use of the fast Fourier transform (FFT) algorithm. 7.4 RELATIONSHIP BETWEEN COMPLEX FREQUENCY RESPONSE AND UNIT IMPULSE RESPONSE The complex frequency response, H (.) or H (f ), of a linear system describes its response characteristics in the frequency domain, and the unit impulse response h(t) describes the system’s response in the time domain. We now show that the unit impulse function and the system frequency-response function form a Fourier transform pair. From Eq. 7.23, the Fourier transform of the unit impulse excitation is % ∞ P (f ) = p(t)e−i(2πf t) dt = 1 (7.31) −∞ Therefore, the response to this unit impulse is, from Eq. 7.28, h(t) = % ∞ H (f )ei(2πf t) df (7.32) −∞ But from Eq. 7.24, the equation that defines the inverse Fourier transform, this unit impulse response function is just the expression for the inverse Fourier transform of the system frequency-response function H (f ). Conversely, it follows that the system frequency-response function H (f ) is the Fourier transform of the unit impulse response function h(t), that is, H (f ) ≡ F [h(t)] = % ∞ h(t)e−i(2πf t) dt (7.33) −∞ This Fourier transform pair relationship between a system’s impulse-response function in the time domain and its frequency-response function in the frequency domain is illustrated in Fig. 7.4. For example, for a viscous-damped SDOF system, H (f ) is given by Eq. 7.30 and h(t) by Eq. 5.30, both repeated here. H (f ) = h(t) = Time Domain h(t) Impulse Response < 1/k [1 − (f/fn )2 ] + i(2ζf/fn ) (7.30) 1 −ζ ωn t e sin ωd t, m ωd (5.30) F −→ ←− F −1 0 t >0 Frequency Domain H (f ) Frequency-Response Function Figure 7.4 Relationship between impulse response and frequency response. 200 Response of SDOF Systems to Periodic Excitation: Frequency-Domain Analysis It is left as an exercise for the reader to show that these SDOF system functions satisfy Eq. 7.33. In Chapter 18 it will be noted that some parameter-estimation algorithms work directly with the frequency-response function H (f ); these are called frequency-domain algorithms. Other parameter-estimation algorithms make use of the impulse-response function h(t); these are called time-domain algorithms. 7.5 DISCRETE FOURIER TRANSFORM AND FAST FOURIER TRANSFORM Although the Fourier integral techniques discussed in Section 7.3 provide a means for determining the transient response of a system, numerical implementation of the Fourier integral became a practical reality only with the publication of the Cooley–Tukey algorithm for the fast Fourier transform in 1965.[7.3] Since that date, the FFT has virtually led to a revolution in many areas of technology, including the area of vibration testing (e.g., see Chapter 18 and Refs. [7.4] and [7.5]). Two steps are involved in the numerical evaluation of Fourier transforms.[7.1] First, discrete Fourier transforms (DFTs), which correspond to Eqs. 7.23 and 7.24, are derived. Then an efficient numerical algorithm, the fast Fourier transform (FFT), is used to the compute the DFTs. 7.5.1 Discrete Fourier Transform Pair For numerical treatment of the Fourier transform, it is necessary first to define a discrete Fourier transform pair corresponding to the continuous Fourier transform pair of Eqs. 7.23 and 7.24. To be transformed, a continuous function must first be sampled at discrete time intervals /t. Second, due to computer memory and execution-time limitations, only a finite number N of these sampled values can be utilized. Figure 7.5a illustrates a sampled waveform with N = 16 samples taken at /t = 0.25-sec intervals. 1.5 5 Mag(P ) p(t ) 1 0.5 0 0 −0.5 −1 0 1 2 t (sec) (a) 3 4 5 −5 0 4 8 n 12 16 (b) Figure 7.5 Sampled waveform (e−t with /t = 0.25 sec, N = 16): (a) discrete-time representation; (b) discretefrequency representation. 7.5 Discrete Fourier Transform and Fast Fourier Transform 201 The effects of this sampling and truncation are to approximate the continuous signal by a periodic signal of period T1 = N /t, sampled at times tm = m /t, m = 0, 1, . . . , (N − 1). The total sample time is T1 , so the fundamental-frequency sinusoid that fits within this sample time has a period T1 . Therefore, the frequency interval of the discrete Fourier transform is /f = 1 1 = T1 N /t (7.34) The DFT consists of N samples extending up to a maximum frequency of N /f . Specific features of sampling and truncation that are illustrated in Fig. 7.5 are discussed following Example 7.7. Sampling and truncation are discussed in greater detail in Section 18.4. Since a period T1 consists of N samples, the integral of Eq. 7.23 is replaced by the following finite sum: P (fn ) = N−1 ! p(tm )e−i2π(m/t)(n/f ) /t (7.35) m=0 Finally, the discrete Fourier transform (DFT) can be written P (fn ) = /t N−1 ! p(tm )e−i(2πm(n/N)) , m=0 n = 0, 1, . . . , N − 1 (7.36) The inverse DFT can be obtained from Eq. 7.24 in a similar manner. Thus, p(tm ) = N−1 ! P (fn )ei2π(m/t)(n/f ) /f (7.37) n=0 This expression for the inverse Fourier transform (IFT) can be written as p(tm ) = N −1 1 ! P (fn )ei(2πm(n/N)) , N /t n=0 m = 0, 1, . . . , N − 1 (7.38) Equations 7.36 and 7.38 define a discrete Fourier transform pair that is consistent with the continuous Fourier transform. The discrete Fourier transform approximates the continuous Fourier transform at discrete frequencies fn . The accuracy of a DFT representation depends on the sampling interval /t and the number of samples, or block size, N . In Section 18.4 we discuss these effects in much greater detail. Reference [7.1] presents a graphical derivation and a theoretical derivation of the DFT pair. The resulting discrete Fourier transform (DFT), written in present notation, is P (fn ) = N−1 ! m=0 p(tm )e−i(2πm(n/N)) , n = 0, 1, . . . , N − 1 (7.39) 202 Response of SDOF Systems to Periodic Excitation: Frequency-Domain Analysis where fn = n /f . The corresponding inverse DFT is p(tm ) = N−1 1 ! P (fn )ei(2πm(n/N)) , N n=0 m = 0, 1, . . . , N − 1 (7.40) This form is the one that is utilized in computing FFTs.[7.6] However, a scale factor /t is required to produce an equivalence between this DFT form and the continuous Fourier transform. 7.5.2 Fast Fourier Transform Algorithm The fast Fourier transform (FFT) is not a new type of transform but rather, an efficient numerical algorithm for evaluating the DFT. Its importance lies in the fact that by eliminating most of the repetition in the calculation of a DFT, it permits much more rapid computation of the DFT. Either Eq. 7.39 or 7.40 can be cast in the form Am = where N−1 ! Bn WNmn , n=0 m = 0, 1, . . . , N − 1 WN = e−i(2π/N ) (7.41) (7.42) A measure of the amount of computation involved in Eq. 7.41 is the number of complex products implied by the form of the equation and the range of m. It is clear that there are N sums, each of which requires N complex products, or there are N 2 products required for computing all of the Am ’s. By taking advantage of the cyclical nature of powers of WN , the total computational effort can be drastically reduced. Figure 7.6 shows the repetition cycle for W8mn . The number of complex products for the FFT algorithm is given by (N/2) log2 N . For example, if N = 512, the number of FFT operations is less than 1% of the corresponding number of DFT operations. Signal-processing software products invariably provide for computation of the FFT.[7.6] W 68 = W 814 = … W 78 = W 815 = … W 58 = W 813 = … W 08 = W 88 = … W 48 = W 812 = … W 18 = W 89 = … W 38 = W 811 = … W 28 = W 810 = … Figure 7.6 Cyclical nature of WNmn for N = 8. 7.5 203 FFT Computations In Example 7.7 the FFT command in the Matlab computer program is used to compute the Fourier transform of a square wave, similar to the one treated analytically in Example 7.4 in Section 7.2. Computation of FFTs of nonperiodic signals is discussed in Section 18.4. Example 7.7 Let p(t) be the square wave defined over one period by the function : 2 0.0 ≤ t < 0.5 sec p(t) = 0 0.5 sec ≤ t < 1.0 sec represented by 16 samples over the finite interval 0 sec ≤ t < 1 sec. Use the FFT command in Matlab to determine the Fourier transform of this square wave, and plot the real and imaginary parts of the resulting Fourier transform. SOLUTION Relative to an offset average value of 1.0, the square wave is antisymmetric in the time window 0.5 sec ≤ t ≤ 0.5 sec. The 16 discrete-time sample points are shown in Fig. 1a. These 16 data points constitute the input to the FFT algorithm. (See the Comments following this example.) 1.5 (P ) 1 0.5 0 2 −0.5 0 2 4 6 8 6 8 f (Hz) 1 0 1 0 0.5 t (sec) (a ) 0.5 1 (P ) p (t ) 7.5.3 Discrete Fourier Transform and Fast Fourier Transform 0 −0.5 −1 0 2 4 f (Hz) (b) Figure 1 (a) Time- and (b) frequency-domain representations of a square wave. 204 Response of SDOF Systems to Periodic Excitation: Frequency-Domain Analysis In Fig. 1b the results of a 16-point FFT are plotted versus the discrete frequency fn for 0 ≤ fn ≤ fN/2 . The real part (top right) and the imaginary part(lower right) can be compared with the plots in Fig. 1 of Example 7.4, noting that the square wave in Example 7.4 is an antisymmetric function with an average value of zero, whereas the square wave in this example is an antisymmetric function relative to an average value of 1.0. Other differences are explained in the Comments that follow. Comments Regarding Matlab FFT Input/Output 1. The FFT algorithm is executed with the command y = fft (x, N ) where N is the block size and x is the input sequence, that is, the vector of discrete-time samples of the input function. For structural dynamics applications, x will be a vector of real numbers. It is desirable to let N be a power of 2. 2. The FFT treats the first N numbers in the input sequence as one period of a periodic function. For example, the rectangular pulse in Fig. 1a of Example 7.7 will correspond to a square wave that is represented by a Fourier series whose fundamental frequency is 1.0 sec, with the sampled value at t = 1 sec repeating the sampled value at t = 0 sec. This topic is treated in greater detail in Section 18.4. 3. Where there is a step discontinuity in the input function, the sampled value is taken as the average of the value prior to the jump and the value after the jump, as shown in Fig. 1a of Example 7.7 and in Fig. 7.5a. 4. Because P (fn ) and P (f−n ) are complex conjugates of each other, the second half of the Fourier coefficients are usually not plotted (e.g., as in Fig. 1b of Example 7.7). Thus, the complex values in the output vector are sequenced by Matlab as follows: y(1) = P (f0 ) = average value of input (real) y(2) = P (f+1 ) .. . y(N/2 + 1) = P (f+N/2 ) y(N/2 + 2) = P (f−N/2+1 ) .. . y(N ) = P (f−1 ) 5. To treat the FFT output values as approximations to the complex Fourier coefficients given by Eq. 7.8, it is necessary to divide the output values y(n) by the block size N .5 5 This scaling is necessitated by the scaling factor T1 introduced in Eq. 7.18, together with the additional scaling factor of /t between Eqs. 7.36 and 7.39. Problems 205 6. To convert the input sequence number m to input sample time tm in seconds, it is necessary to use tm = m/t. Similarly, to convert the frequency scale of the output from coefficient number n to frequency fn in hertz, one must use fn = n/f = n/N /t. REFERENCES [7.1] E. O. Brigham, The Fast Fourier Transform, Prentice-Hall, Englewood Cliffs, NJ, 1974. [7.2] R. A. Gabel and R. A. Roberts, Signals and Systems, 3rd ed., Wiley, New York, 1987. [7.3] J. W. Cooley and J. W. Tukey, “An Algorithm for Machine Calculation of Complex Fourier Series,” Math Computation, Vol. 19, 1965, pp. 297–301. [7.4] D. J. Ewins, Modal Testing: Theory, Practice and Application, 2nd ed., Research Studies Press, Baldock, Hertfordshire, England, 2000. [7.5] N. M. M. Maia and J. M. M. Silva, ed., Theoretical and Experimental Modal Analysis, Wiley, New York, 1997. [7.6] Using MATLAB, Version 6, The MathWorks, Natick, MA, 2002, pp. 12-41 to 12-48. PROBLEMS p (t ) For problems whose number is preceded by a C, you are to write a computer program and use it to produce the plot(s) requested. Note: Matlab .m-files for many of the plots in this book may be found on the book’s website. Half-sine waves p0 t T1/2 T1 Figure P7.2 Problem Set 7.1 7.1 Determine the real Fourier series for the periodic excitation function in Fig. P7.1. p (t ) p0 t −T1/4 3T1/4 −p0 C 7.3 (a) Determine the real Fourier series for the square wave shown in Fig. P7.3. (b) By comparing your result from part (a) with the answer in Eq. 10 of Example 7.1, what do you observe to be the effect(s) of the phase shift of the square wave? (c) Run the Matlab .m-file sd2hw7 3.m, and print out the plots produced. Modify this .m-file to produce the input signal of Example 7.1, and print out the plots produced. p (t ) p0 T1/4 5T1/4 3T1/4 Figure P7.1 −p0 7.2 Determine the real Fourier series for the periodic excitation function in Fig. P7.2. Figure P7.3 t 206 Response of SDOF Systems to Periodic Excitation: Frequency-Domain Analysis 7.4 (a) Determine the real Fourier series for the periodic excitation function in Fig. P7.4. (b) By comparing your result from part (a) with the answer in Eq. 10 of Example 7.1, what do you observe to be the effect(s) of the upward shift of the square wave? 7.8 (a) Determine the coefficients of the complex Fourier series that corresponds to the excitation p(t) in Fig. P7.3. (b) Sketch the real and imaginary parts and the magnitude of the complex coefficients as in Example 7.4. 7.9 (1) Determine the coefficients of the complex Fourier series that corresponds to the excitation p(t) in Fig. P7.4. (2) Sketch the real and imaginary parts and the magnitude of the complex coefficients as in Example 7.4. p(t) 2p0 t T0/2 T0/2 7.10 The undamped SDOF system of Fig. P7.10a is subjected to the sawtooth base motion z(t) illustrated in Fig. 7.10b. Let z(t) = Figure P7.4 7.5 The undamped SDOF system shown in Fig. P7.5 is subjected to the excitation p(t) given in Problem 7.1. (a) Determine a Fourier series expression for the steadystate response of the system if ωn = 4.1 . (b) Sketch the spectra for Pn and for Un , similar to those shown in Fig. 7.2. ∞ ! Z n ei(n.1 t) , n=−∞ m ∞ ! W n ei(n.1 t) n=−∞ where W n = H n Z n . (a) Determine H n . (b) Determine Z n . (These will have the same form as P n of Problem 7.1.) (c) Determine W n , and sketch |W n |/Z versus frequency order, as in Example 7.4. z (t ) u (t ) k k w(t) = w=u–z m p (t ) ( a) Figure P7.5 z (t ) 7.6 The undamped SDOF system shown in Fig. P7.5 is subjected to the excitation p(t) given in Fig. P7.4. (a) Determine a Fourier series expression for the steadystate response of the system if ωn = 6.1 . (b) Sketch the spectra for Pn and for Un similar to those shown in Fig. 7.2. Z t −Z T1 (b) Figure P7.10 Problem Set 7.2 7.7 (a) Determine the coefficients of the complex Fourier series that corresponds to the excitation p(t) given in Fig. P7.1. (b) Sketch the real and imaginary parts and the magnitude of the complex coefficients as in Example 7.4. Problem Set 7.3 7.11 Determine the continuous Fourier transform for the rectangular pulse in Fig. P7.11. (Note that this involves a time shifting of the pulse in Example 7.6.) Problems p (t ) p0 t 2T Figure P7.11 7.12 Determine the real and imaginary parts of the continuous Fourier transform of each of the following functions: Note: Absolute value of time t. (a) p(t) = p0 e−α|t| 0, t <0 1 0≤t ≤ (b) p(t) = p0 sin(2πf0 t), 2f 0 1 0, t> 2f0 207 compute a 16-point Fourier transform to verify the (real and imaginary) results shown in Figs. 1b. (Note: You can find the Matlab .m-file for Example 7.7 as sd2ex7 7.m on the book’s website.) (b) What is the resulting value of the coefficient (P 1 )16pt ? Compare this answer with the value given in Eq. 5 of Example 7.4. (c) Repeat part (a) to calculate a 32-point transform and a 64point transform. Are the values of (P 1 )16pt , (P 1 )32pt , and (P 1 )64pt converging to the value given in Eq. 5 of Example 7.4? C 7.15 Let p(t) be the periodic function defined by the 16-point discrete-time sawtooth wave shown in Fig. P7.15. (a) Use the FFT command in Matlab to determine the 16-point discrete Fourier transform (DFT) of this sawtooth wave. (b) Plot the real and imaginary parts of this DFT transform. 1 0.5 7.13 Using the unit impulse response function h(t) of Eq. 5.30 and the definition in Eq. 7.33 of the continuous Fourier transform H (f ), show that the Fourier transform of the unit impulse response function is the frequency response function H (f ) given by Eq. 7.30. That is, using Eq. 7.33, show that h(t) and H (f ) form a Fourier transform pair, as illustrated in Fig. 7.4. p (t ) Problem Set 7.4 0 −0.5 −1 0 0.25 0.5 Time Problem Set 7.5 C 7.14 (a) Using the input sequence given in Fig. 1a of Example 7.7, use the Matlab FFT command to Figure P7.15 0.75 1