THERMALLY INDUCED VIBRATIONS OF THE HUBBLE SPACE TELESCOPE'S SOLAR ARRAY 3 IN A TEST SIMULATED SPACE ENVIRONMENT Derrick A. Early(n),William B. Haile°), Mark T. Turczyn (2) C°Swales Aerospace, 5050 Powder Mill Road, Beltsville, Maryland20705 USA, Email: [email protected] (2)NASA GSFC, Mailstop 442.O,Greenbelt, Maryland 2077I USA, Email: mturczyn@hst, nasa.gov ABSTRACT/RESUME To qualify for flight on HST, a test was designed to verify that the SA3 would not disturb the HST after NASA Goddard Space Flight Center and the European installation. NASA called this test the Disturbance Space Agency (ESA) conducted a disturbance Verification Test (DVT) of Solar Array 3. The test verification test on a flight Solar Array 3 (SA3) for the objective was to measure thermally induced vibrations Hubble Space Telescope using the ESA Large Space of SA3 in a test simulated space environment. The Simulator (LSS) in Noordwijk, the Netherlands. The events of interest were transient vibrational responses. LSS cyclically illuminated the SA3 to simulate orbital It was mandatory to measure these vibrations if they temperature changes in a vacuum environment. Data had been large enough to cause Loss of Lock (LOL) on acquisition systems measured signals from force the HST guide stars. [1] transducers and accelerometers resulting from thermally induced vibrations of the SA3. The LSS with To conduct this test, NASA Goddard Space Flight its seismic mass boundary provided an excellent Center (GSFC) used the European Space Agency background environment for this test. This paper (ESA) Large Space Simulator (LSS) inNoordwijk, The discusses the analysis performed on the measured Netherlands. This chamber was the best in the world transient SA3 responses and provides asummary of the for providing thermal, vacuum and solar simulation results. with a dynamically quiet seismic mass boundary. As shown in Fig. 1, SA3 consists of two wings. NASA placed one of the flight wings in the LSS on a stiff 1 INTRODUCTION pedestal and rot_atedthe wing 49 degrees to the solar beam. This rotation and the 2.9 m tall pedestal was This year during the Hubble Space Telescope (HST) necessary to position the wing for full illumination by servicing mission 3B (SM3B), NASA astronauts will the solar simulator with its 6m beam. Fig. 2shows the install Solar Array 3 (SA3) on HST. This will replace SA3 flight wing mounted on the pedestal. The 656 lb the nearly 7 year old Solar Array 2. The SA3 will SA3 spans a width of 7.6 m and a height of 4.1 m provide more power with less surface area by using above the top of the pedestal. Gallium Arsenide Solar Cells mounted on honeycomb panels. NASA expects the new SA3 to provide power The testing was conducted over four days using two to HST for the remainder of its mission. Fig. 1 shifts to cover 24 hours per day. After contamination illustrates how HST will appear after SM3B. bakeout and thermal equilibrium, the LSS simulated 20 orbital cycles. During the orbit simulations, data acquisition systems measured 40 hours of signals from instrumentation. This paper describes the methodology used to screen the data to find any thermally induced vibrations of the SA3 wing. 2 INSTRUMENTATION For this test, the instrumentation consisted of accelerometers, force transducers, thermocouples, a vacuum gage and solar sensors. Fig. 2 shows the location of the accelerometers with labels al through a7. All of the locations had three axes instrumented with the exception of a5. Only the Xs, and Zs, axes were instrumented at a5. Fig. 1 HST with Solar Array 3Installed a2 a3 / \ j SA3 Interface B°it"""_i__ 7 A';ill;; I Force 7_-- _*'*_L lSteel 2, Transducer i_====__ ]_ i Pedestal // 12in // " _ _ Force Transducers Fig. 4 Force Transducer Installation 3 SIGNAL PROCESSING All of the primary and secondary bending and torsion modes of the SA3 exist between I and 10 Hz. Fig. 2 Flight SoIar Array 3 on Test Fixture Therefore, the data was band pass filtered in the following manner. An 8tn order digital filter removed Fig. 3 shows the locations of the four force transducers background noise above 10 Hz from the measured located at the top of the pedestal. They are labelled fl 1 data with parameters set to emulate an analog elliptic through f14. Each force transducer produced three low pass filter. Then, a 4thorder digital filter removed orthogonal force signals oriented in the pedestal transducer noise below I Hz with parameters set to coordinate system. Two of the pedestal axes are emulate a Butterworth high pass filter. Application of labelled Xpedand Zped with Xp_dpointing away from the these filters is described in [3]. sun simulator. The 131° angle between the SA3 and pedestal coordinate systems is shown. Eight of the The energy of the filtered signals Yi may be described force transducer signals were collinear; therefore, the in relation to amean square value, W_z, on p. 14 of [2]. signals were summed before the charge amplifier. This reduced twelve force signals down to a total of eight. This is simply the average of all the squared values in the time history as shown in Eq. 1. Table 1 contains filtered signal results of Eq. 1 for the two hour long 13thorbit simulation. Table 1 uses units ofmilli-pounds (mlb) and micro-g's (p.g). iif,, tFl2= lim 1 Iory'2(t)dt (1) X ped In order to find short duration non-stationary events, Eq. 2 defines a short period average over time T to f13_ estimate the mean square values as a function of time. ( | et+TI2 2 W,2(t) = -_ J,_ri2y ,(t)dt (2) Fig. 3 Force Transducers Using the square roots of Eqs. 1 and 2, Eq. 3 normalizes the root mean square time histories of the Fig. 4 shows the installation of the force transducers. filtered signals. They were sandwiched between an aluminum plate and the top of the stainless steel pedestal. A bolt through w--(-t,)=q,,(t)Iw, (3) the center of the transducer was torqued while monitoring the force signal from the transducer. This technique allowed each bolt to be preloaded to 2000 lb. ;, J Table 1 Orbit 13 Signal Root Mean Square (rms) moment in time. This is reasonable, since any primary and secondary mode wing vibrations would result in multiple responses. Finally, the triple screen used the tp i Label Axis (mlb) Label Axis (pg) boundary accelerations measured at the top of the fll+fl4 Xp_t 14 al Xsa 73 pedestal to analytically predict the force transducer f12+f13 X_d 8 al Ys= 95 signals resulting from chamber vibrations. Whenever fl I+ fl2 Zp,,d 1! al Z_ 85 the predicted force signals exceeded a threshold during an event found in the second screen, the event was f13 + f14 Zpod 13 a5 Xsa 62 removed from consideration. fl I Yp_ 76 a5 Zsa 107 fl2 Yp_ 29 a6 Xs= 17 4.1 First Screen fl3 Y_,,a 78 a6 Y_a 14 fl 4 Yp_ 26 a6 Z_a 16 After computing q'¢, these results were screened to find events that might cause HST to have a loss of lock 4 SIGNAL SCREENING event. For this screen, only the force transducer signals were examined. The screen flagged each time With 40 hours of data to process, a screening method when any one of the signals had an upward crossing of for finding significant events of short duration was threshold -. This screen can be described with Eqs. 4-7. needed. The background noise level was fairly Eq. 4 generates a threshold flag time history of ones stationary with little variation from orbit to orbit when tFj exceeds z and zeros otherwise. Eq. 5 sums all simulation. This is evident in Fig. 5 showing the the force transducer threshold flag time histories to minimum, maximum and mean of all the orbit produce a time history that provides a count of all the simulation power spectral densities (PSD) for the Y force signals that exceeded z. Eq. 6 generates a screen axis signal of force transducer f14. logic time history of ones for the times when any one of the force signals _i exceeds_-. Finally, Eq. 7 10-1 produces an up tick time history of only the rising lff_ transitions of p]. - if W,(t) >_z (4) P_' = if _, (t) < z 10_ p/= _ p._,(t) (5) 1100"-3s_ i_f 10_ P] = {_ iiff pp//((tt))>_l<1 (6) 10"0;' 5 lb 1_ 20 f(Hz) p_,={1 if(p,(t + At)-p,(t)) > Oother(w7i)se Fig. 5 PSD of Force Transducer f14 Ysa Axis Signal On p. 234 of [2], the reference suggests the use of short The up tick peJ time histories were summed for each of period mean square estimates to test for stationarity. In the orbit simulations. Table 2 lists a summary of the this analysis, time histories of normalized short period results of this screen. For az of 4, a total of 25 events root mean square values were used in screening the were found. These events were identified during the data to uncover short duration events of non- test using another methodology and were analyzed by stationarity. HST pointing and control engineers. Their pointing and control system simulation results predicted no In the following sections, three different screens were LOL for HST. This was very good news, and as a developed. The first simply looked for any force result the LOL disturbance recluirement for SA3 was transducer signal exceeding an arbitrary threshold over verified. In addition, the test objective was met. the noise floor. Next, the double screen added the requirement that two force signals and two In Table 2, orbit 13 was observed to be more active accelerometers must all exceed a threshold at the same than the other orbits.. Th!s_was 91so noticed during the test. At the time, the engineers hypothesized that during an earlier orbit the activation of a heater on the _ "°,o aluminum interface plate, shown on Fig. 4, was the cause. Based on this hypothesis the test director disabled the heater. O0 1000 2000 3000 4000 5000 6000 7000 Table 2 Maximum Force rms Events 10 .... Xp_t(z)with T=Is orbit 2.0 2.5 3.0 3.5 4.0 1 93 9 2 1 1 2 147 25 3 1 1 ,. _,,I_ ,_ud_, _,,_,u_J0al _ 'La 34 18407 144 1 1 1 -0 t000 2000 3000 4000 5000 6000 7000 5 134 20 7 3 2 4 , • 7 181 18 4 8 163 11 2 I 1 9 178 17 2 10 89 9 2 1 1 II 174 8 1 o; i0'002o'00_ _ _ _ 7000 12 157 8 2 1 1 13 168 30 13 10 8 Time (s) 14 139 14 4 3 3 15 150 I1 3 3 2 Fig. 6 Orbit 13Normalized Signal rms 16 155 19 6 1 I 1"7 151 14 2 18 164 15 3 3 3 P2 = {_ if p.(t) >o2thearnwdisep:(t) > 2 (9) Z 2470 246 57 29 25 Table 3 summarizes the results of the double screen. Fig. 6 contains time history plots of the maximum The number of events with a z of 4 was reduced from normalized rms signals grouped by SA3 accelerometers 25 to only 2 events. All of these events occurred in the top plot, force transducers in the middle and during the 5thorbit simulation. boundary accelerometers in the bottom plot. The eight events counted in Table 2 for orbit 13with az of 4 Can Table 3 Double Screen Events be seen clearly in the middle plot of Fig. 6. Notice that the major peaks on the top plot do not line up with the- Epe2(z) with T=ls major peaks of the middle plot. Since the orbit 2.0 2.5 3.0 3.5 4.0 accelerometer and force transducer peaks did not 1 42 3 correlate, they cannot be the result of SA3 vibrations. 2 64 7 3 41 3 4.2 Double Sc.reen 4 68 2 5 70 13 4 3 2 Although the first screen identified no events that 7 87 10 3 would cause LOL, any thermally induced vibrations 8 82 6 were still interesting for their jitter effects. In a further 9 94 10 4 attempt to find SA3 vibrations, this double screen was 10 45 7 ! devised to find events with at !_east.two force .... 11 74 4 transducer signals and at least two SA3 accelerometer 12 62 5 signals greater than .-. Eqs. 4-9 define the screen. Eq. 8 .......---1_3 92 6 sums all the SA3 accelerometer signals from ai to a5 14 77 4 that have a _o greater than z. Eq. 9 defines the screen 15 79 6 logic time history. 16 74 4 82 6 86 5 p. = ___p_,(t) (8) 1219 101 12 3 2 #=G r Fig. 7 plots the maximum _, time histories for 5thorbit For the final triple screen, the predicted force transducer responses were used in Eq. I1to sum the simulation. Again, the top plot is the maximum of all the SA3 accelerometers. The middle is the force number of _g greater than zp. Eq. 12 defines the transducers, and the bottom plot is the seismic mass. logic time history. It is the same as the double screen The two events with a : of 4 are clearly evident on all with the added requirement on the predicted forces j,f. three plots at 3500 and 4000 seconds. Since the seismic mass was moving, the SA3 responses were the result of base motion. ppx = __,p,,(t) (11) J=pf 6 ..... (12) PJ = 01 if P2(t) >ot1hearnwdispepz(t)> 1 Table 4 lists the results from the triple screen. The _0 100020oo300o4oo05ooo6o00 number of events with a .- of 2.5 was significantly reduced from 101 to a total of 4. On closer inspection 6 ..... " of the time histories for these 4events, they still appear to be caused by boundary motion. Therefore, no 4 thermally induced vibrations of the SA3 were detected I__ 2 _,_a_L_ at zof 2.5. f_L Table 4 Triple Screen Events VO 1000 2000 3000 4000 5000 6000 Zp,_(z)withr=*sand ze=2 6 orbit 2.0 2.5 3.0 3.5 4.O 1 26! 2 42 3 30 1 4 45, 0 1000 2000 3000 4000 5000 6000 5 41 Time (s) 7 52 1 Fig. 7 Orbit 5Normalized Signal rms 8 59 9 56 This assumption was verified by simulating impulse 10 40 responses of the SA3. In the simulations, responses 59 were observed on the accelerometers and the force 45 transducers, but they were not observed on the top of 13 66 the pedestal. 14 61 1 15 57 ! 4.3 Triple Screen 16 47 17 63 In order to continue the search for thermally induced 18 69 vibrations of the SA3, this third and final screen 858 4 eliminates boundary excitation events by using a test data derived linear model to predict SA3 responses. A 5 SYSTEM IDENTIFICATION system identification of SA3 using the test data was performed and resulted in the A, B, C and D matrices The triple screen above used predicted SA3 responses contained inEq. 10for computing linear predictions of resulting from base excitation. In order to perform the SA3 responses )5. The methodology for accurate predictions, a system identification was performing the system identification is detailed in performed to estimate the state space matrices in section 5. Eq. 10. The elements of the matrices were cast into a modal parameter grey model form to be used in a ic= Ax + Bu parameter estimation algorithm detailed in [4]. ) = Cx + Du (10) Using the accelerometer signals from location a6 as an input source and all the other signals as output, the modal parameters were extracted from the test data. Table 5 lists natural frequency and percent critical Since no interior forcesfa are applied the second row damping coefficient results from the system of Eq. 16 may be written as Eq. 17. If a thermally identification for both ambient and vacuum conditions. induced vibration event occurred, this assumption The right two columns contain the analytical frequency would be false momentarily; however, the averaging values from the finite element model of the test described in the next section will still make this a configured SA3 and critical damping coefficients from reasonable assumption. SA3 modal survey data for comparison• Table 5 Modal Parameters = -2_co._-co#¢ --m¢b£ h (17) Ambient Vacuum Predicted For the SA3, the boundary may be considered kinematic. Therefore, k-bbis zero, and the first row of f°l Mode (Hfzn) I(%_) if]z) (%) (Hfz) (%) Eq. 16 may be written as Eq. 18. Eq. 19 is formed by ! 1.2 3.1 1.3 4.7 I.i 4.7 taking the second derivative of the second row of 2 1.5 4.4 1.7 4.6 !.4 3.9 Eq. 15. 3 - 1.8 3.0 1.8 2.3 5.3 0.2 5.5 0.4 5.2 1.7 _ 6.2 0.9 6.4 1.4 5.9 1.4 f b = mb¢_ + m---bhXb (18) 6 6.3 0.9 6.6 0.6 6.5 1.7 7 7.4 1.0 7.8 0.7 7.5 1.6 = + 9) 8 12.9 0.7 13.7 1.9 12.9 2.4 9 13.9 2.7 14.6 0.8 13.4 2.9 Finally, Eqs. 20 and 21 define the grey model state space matrices from Eqs. 17-19. 5.1 Modal Parameter Grey Model Eq. !3 represents the physical motion for the SA3. The degrees of freedom can be partitioned as shown in Eq. 14, where the xb is the boundary and Xa are the =:1-'-<oo;1=o,:Io-1o interior degrees of freedom. M'2/+ C_ + Kx = F (13) Cba X b C=-I.F mb_2(cO" mbcc02"] (21) F[_mmt,',bbmbma-o]o(j_Jb_]£,, ]_+ [LCchobbcaa](2a} +LFkkbb k kJb_[,xlfXoblJ:{_ } (14) _ mbb -- mb_m4t_ Substitute Eq. 15 into Eq. 14 and postmultiply by the transpose of the transformation. This Craig-Bampton transformation is described in [5]. 5.2 Spectral Analysis To estimate the modal parameters in Eq. 21, a spectral analysis of the data was performed. The objective was (15) to compute frequency response functions using the boundary accelerations as areference. Assuming diagonal damping and no boundary damping The power spectral density functions and cross spectral coupling, Eq. 16 shows the results of the transformation. density functions in Eqs. 22 and 23 may be calculated from the cross-correlation function in Eq. 24, as on p. 435 in [6]. I meb I Jl.4J Lo 2(co.JlCJ (16) S_ (co) = _ R,, (r)e-'_'_dr (22) o,J 4" 1 Sv (co) = .[-LR'j(r)e-t_dr (23) 10 15 20 In practice, the cross-correlation function in Eq. 24 is 180 .... ,.... not calculated, and time cannot extend from-oo to+oo. An approximation of the spectral Eqs. 22 and 23 is made byaveraging methods detailed in [3] and [7]. Now the frequency response functions may be 18%" k 1'o'" 15 20 computed with Eq. 25, as in p. 162 of [2]. For this analysis, the input degrees of freedom, denoted by the subscript u, were the signals from accelerometer a6 t-- test shown in Fig. 2. Since only the translational .._t5 I ....it.It,. 'tZ'::::fit 1/I accelerations were measured, 5',, is a three by three matrix at each frequency, 09. /-/)_, (co) = S,-] (CO)S_ (09) (25) fl4y /a6z, Frequency (Hz) Fig. 8 Orbit 13Force Signal FRF and Coherence Eq. 26 computed the multi-input coherence Cy of the frequency response functions. The operator on the third term of the numerator in this equation takes transpose of the complex conjugate of Hy,. Eq. 27 computed the phase of the frequency response functions (FRF) by 8(00r)b'(or) 00 (30) 0 ,Y(r) taking the imaginary part of its natural logarithm and converting to degrees. Fig. 8 shows a FRF plot of the force transducer f14 Ysa axis signal referenced to the Fig. 9contains a plot of the impulse response functions Z_ axis ofaccelerometer a6. computed for the three FRF's of the Y_a axis force transducer signal from location f14 referenced to the ¢, three accelerometer signals from location a6. Cy (co) - Hy, (co)S,, (co)Hy, (co) Sy(co) (26) 0.1 0.05 Op = 81____0Ir0a( In Hy. ) (27) Y 7/" -0.0 5.3 Modal Parameter Estimation -0.1 5 10 15 20 25 30 t(s) This section will describe how the parameters were Fig. 9Impulse Response Functions for fl4 Y,. computed to form the fit shown in Fig. 8. Using the FRF's computed by Eq. 25, impulse response functions The state matrices in Eq. 21 are a function of the in Eq. 28 were calculated by using the real portion of parameters contained in the parameter vector defined in their inverse Fourier transformation. Then, the impulse Eq. 31. Here the elements in matrices _, _bb, rnb_and response functions were truncated and tra!ned together in Eq. 29. For the input signals, three impulses are Tb, are reordered into single dimensioned vectors. Initial values for (, co, and _ are required to start the defined inEq. 30 where 5is the Dirac delta function. parameter estimation process. h,, (t) = Re( _._ H,, (co)e'_' dco) (28) 0=[( co, (b m---bbrnb, Tb,] (31) _, = [h,1(r) h,2(r) h,3(r)] with r _ [0 T] (29) To obtain starting values of _"and o-.b, a subspace method estimated a black box of the A, B, C and D ! matrices with Kand _, as objective inputs and outputs, Finally, the triple screen used the boundary accelerations to predict the force transducer signals. as in [4]. Then the complex eigenvalues, 2, of A were Whenever the predicted signals exceeded a threshold, computed. The natural frequencies, co,, were computed the events were removed. The results of this final from the magnitude of 2, and the critical damping screen essentially eliminated all the events that were coefficients, _', were computed by taking the negative above the background levels. In other words, no cosine of the imaginary portion of the logarithm of 2 in thermally induced vibrations of the SA3 were detected Eq. 32. above the background noise. co. = ]21 ( = - cos(Im(In 2)) (32) At this point in the analysis of the test data, no significant Solar Array 3 induced disturbances have From these values, only reasonable results were been found in the test data that would impact the retained. Then, Eq. 33 computed an initial value of g_ performance of the Hubble Space Telescope. by using the imaginary part of the peaks where j was selected to give the largest response at the natural 7 ACKNOWLEDGEMENTS frequency of cont. The remaining parameters in 0 may be started at zero. This paper and analysis were supported under contract NAS5-01090 with NASA Goddard Space Flight q_ik= Im H_ (co.k) (33) Center. The authors are thankful to Thomas Griffin, the HST Observatory Development Manager, for The state space matrices inEq. 21 were estimated with encouraging the publication of this work. The test team performed wonderfully preparing and conducting the a parameter estimation routine that minimizes the error test, and they deserve appreciation for their many long between the predicted time history _ and the impulse hours of hard work. Also, the authors cannot express response functions ._, detailed in [4]. Results of the enough gratitude to our very accommodating hosts at parameter estimation are shown in Fig. 8 and in ESA's ESTEC for the use of their excellent LSS Table 5, and they are considered to be very good. facility. 6 CONCLUSIONS 8 REFERENCES The disturbance verification test objective was to 1. Halle W. and Early D., Report of the Disturbance measure thermally induced vibrations of SA3 in a test Verification Test of Solar Array 3, SAI-RPT-367, simulated space environment. The events of interest Swales Aerospace, 2001. were transient vibrational responses. It was mandatory to measure these vibrations if they had been large 2. Bendat J. and Piersol A., Random Data: Analysis enough to cause Loss of Lock (LOL) on the HST guide and Measurement Procedures, Wiley-lnterscience, stars. [1] New York, 1971. The objective of the test was met. Events were 3. Signal Processing Toolbox User's Guide, The identified inthe data. However, based on HST pointing MathWorks, Inc., Nantick, Mass., 1988. control system simulations, no measured disturbances were large enough to cause the HST to experience 4. Ljung L., System Identification Toolbox User's LOL. Furthermore, many of the events identified by Guide, The MathWorks, Inc., Nantlck, Mass., 1988. the first screen were not attributed tothermally induced vibrations of the SA3, since the accelerometer and 5. Craig R.R. Jr., and Bampton M.C.C., Coupling of force transducer signals did not correlate. Substructures for Dynamic Analysis, AIAA Journal, Vol. 6,No. 7, 1313-1319, July 1968. The double screen added the requirement that two force signals and two accelerometers must all exceed a 6. Meirovitch L., Elements of Vibration Analysis, threshold. This requirement was an attempt to McGraw-Hill, New York, 1975. eliminate noise on a single transducer signal from triggering and event. This screen found many events 7. Welch, P.D, The Use of Fast Fourier Transform for that were easily identified as boundary motion, since the Estimation of Power Spectra: A Method Based on boundary motion will generate SA3 responses on both Time Averaging Over Short, Modified Periodograms, the force transducers and accelerometers. IEEE Trans. Audio Electroacoustics, Voi. AU- 15June, 70-73, 1967.