Home

identification of model aircraft dynamic using flight testing

Contents

1. 74 8 IDENTIFICATION FROM SIMULATED DATA cccsssssssssssssssssssssssssessssssssssessssesesssessssesssesessnsesessesesssessesesssssessesess 76 LONGITUDINAL IDENTIFICATION tere hee ertet ge dna dee Eee teer eee ena een nea e eese e ee ER eve eh ee 76 EAXTBRAE IDENTIBICATION ERR PROIN eR eO RU ente 80 THE EFFECT OF MEASUREMENT NOISE sssssssssessessesesceseecescesescescesessescnsenceecasescsecasecnesaesecasssenecassecnessesecaeseeseceeeeaeeeeseeaeens 85 THE EFFECT DIFFERENT INPUT FORMS epit enti teet tr Rr RE Aene n 87 9 FLIGHT TEST RESULT ssssssssssssssssssssssssssssssssesscsssesssssssssssssssssnsesevsssesssssssnsessssesesesevssesssesessesessvsssessessassessssssssesess 89 BRIGHT DAWA 89 PATA PRE PROGESS ING 3 25 s Noses EAT Gas NREL GRE ERE 90 STABILITY AND CONTROL DERIVATIVE ESTIMATION c csccsscsscsscssessessessesccescesessesseseecescescesesaesaecaecaecaessessesesesensersensenss 91 Longitudinal stability and control derivatives estimation 91 Lateral stability and control derivative eroian a i RAA E E ri a Ai 97 10 DISCUSSION E 104 ESTIMATED AIRCRAFT DYNAMICS decet aa ERR RE Ee YR EE ERES STER EVA 104 FLIGHT DATA PROGESSING 5 Hemden eiie 106 PEIGHT TEST MANOEUVRE oie ERBEN E TEE NERO EROR HE ERR RETINET 108
2. seen eee 39 4 2 DATA COMPATIBILITY ANALYSIS FLIGHT DATA RECONSTRUCTION cccsssssssssessessssssessessesseseeseesesneaseanssneaneeess 41 5 MODEL DESCRIPTION amp 8 40 9 44 3 4 MODEL DESCRIPTION ettet ee atate eade gato ce ctt de mide nee 44 S2 ENGINE TESTING ette rendi EURIBOR 45 CENTRE OF GRAVITY AND MOMENT OF INERTIA eene enn eene 49 5 4 THEORETICAL STABILITY AND CONTROL DERIVATIVE ESTIMATION seeeeenete ee eene eene eene 53 6 DATA ACQUISITION AND INSTRUMENTATION SYSTEMS sccssscsssssssscsensssnsensessssnscnsesssenseascansenseassensensesssenses 55 6 1 DESCRIPTION AND SPECIFICATION 6 1 1 The data acquisition 6 1 2 Instrumentation systems 62 CALIBRATION 7 PRACTICE AND IMPLEMENTATION PROBLEMS 69 7 1 PROBLEMS ENCOUNTERED DURING THE DEVELOPMENT OF THE INSTRUMENTATION 5 69 7 2 TRANSMITTER AND DAS DATA ACQUISITION SYSTEM INTERFERENCE esee eene rennes 70 7 3 PROBLEMS ON THE AIRCRAFT MODEL BE TESTED cccccssessescesessescssesceecssesenecacsecaesecuecassecneseesecaeseeseceeeeaeeeeaeeaeens 72 7 4 PROBLEMS IN FLYING THE
3. Cy W nz Dates rudder 0 I pV2Sb pon ur 58 x 6 saca FG sO ais 1 f V B r gcos 8 b Or in the state space form as 22V 2 Poy PSH 1 0 0 IB 2m 4 1 Sb V Sb V Sb 0 1 2204 P p 0 p 0 Cl p 0 CI 1 21 41 4 41 r 2 5 0 1 PV Sb Cn PV Sb Ch pV Sb Cn 1 2r 41 pV 5 pSb psb C cos0 2 y 8 aileron 2 8 udder Lm Yo V 2 Sb V 2 Sb Sb aileron P F Ch i P CT P Ch rudder 21 aileron 2 21 2 2 pV Sb Sb Cr sae pV Sb Cn 2 21 21 Cn Bis 3 26 33 For preliminary analysis the model can be further simplified into three simple single plane models equations 3 28 to 3 30 These mathematical models are valid if we assume that small perturbations are made about one axis only and that the motion is confined to that plane only 3 q Peso Cz a Pitch only model lt 3 28 pV Sc C a y 0 2 P Pa T A Roll only model Sr UOTA ee EM 3 29 B P os p 6 r 3s 5 Hapa s bon 2D Yaw roll model A V A 3 30 OnB Cn B A To model any non
4. boom error due to the present of boom support at the vicinity of the vanes The vane boom effect is normally obtained from wind tunnel calibration of the installed boom vanes system Approximation using potential flow theory is given as Oyane 1 location Byane 1 Kg Where the correction Ka and Kg are vane location 2 1 vane max oe B _ J dr max vane min For the T240 vanes r boom 0 7 Y vane min 4cm 9 The correction values and Kgcalculated to be 0 0136 1 36 of boom error D Static vane alignment error M Laban 1994 commented that there seems to be no need to accurately measure the vane misalignments This error can be combined with the induced flow distortion error and simply stated as wane Cody axis AQ alignment Prodi axis ABalignment E Alignment error due to fuselage and boom bending This error can be calculates as 5 2 cep EC HOD For the T240 the alignment error due to fuselage and boom bending is calculated below D Fuselage dL Length 1 meter Diameter r meter Structural efficiency n Stiffness E Density p Offset from Cg Ax meter PY ezS 3x105 4 16x105 dq dt e 9 28x10 4Ax dq dt 3 13x10 4 for max f 25 m s 0 006 due to the fuselage and dq dt 5 rad s e 0 157 0 151
5. for the T240 7 4 cm q max 2 rad s V 30 m s produces kinematics error of AP 3 Pascal 4 35x10 psi C Probe error Probe error in total pressure can be neglected Wuest 1980 as long as the flow angle is less than 10 degree However probe error in static pressure is significant and this includes error due to shear and error due to flow interference Flow friction along the probe ahead of static pressure transducer reduces the static pressure AP and is given as AP hear 1 5 4 1 0 0576 mes uus Pimpact The value of fn depends on the design of the orifice and the wall characteristics A severe value of fn 2 5 is taken for the T240 At Reynold number 3 45x10 at V 30 m s and viscosity of 1 456 10 the shear error is calculated as 0 016 Pimpact or 1 6 impact pressure This error is caused by the interference between flow in and out of the orifice with the external flow For calculations in a flow interference error refer to ESDU 85011 D Pressure tubes error A2 4 This error includes change in pressure inside the tube which is caused by acoustic effect air friction and rotational speed of the aircraft Present theoretical methods for determining this error is not reliable M Laban 1994 A typical value of 3 msec delay per one meter tube length is normally assumed E Pressure traducers error For electric transducer this error is normally small and can therefore b
6. Flight data Error criterion The assumed mathematical model dynamic Graphical plots Mean errors Standard deviations 4 Sliders representing derivative values Figure 3 1 The Interactive Curve Matching algorithm 30 0 06067 0 007606 Figure 3 2 Longitudinal Curve Matching Menu programmed Matlab and it s corresponding error layout 3 3 Model dynamics Selection of an adequate model in the analysis of flight test data is critical to the success of the identification process The criteria for the adequate model are however difficult to quantify The model is said to be adequate if it is simple and yet has a physical meaningful interpretation A priori information such as that from wind tunnel testing is normally used to assist in the selection of the right model For a rigid aircraft its dynamics can be represented by a six degree of freedom non linear mathematical model This model consists of 6 equations which couple the longitudinal and lateral motion of the aircraft Due to the complexity of the equations this model is not normally used in the extraction of stability and control derivatives from flight data Instead reduced linear uncoupled equations of motions are frequently used 3l Similarly in this project the linearised uncoupled longitudinal and lateral equations of motion selected for the analysis of the flight data These equations have been used extensively and succe
7. Howel 1994 2 2 Flight data analysis 2 2 1 Model dynamics The linear mathematical model is adequate for small perturbation analysis of a conventional configuration UAV The linear model has also been extensively used for the extraction of stability and control derivatives from flight test data of general aviation aircraft Iliff 1976 Colemann 1981 and Budd 1993 The model should be selected so as to give the simplest meaningful model of the vehicle s dynamic for a particular manoeuvre Validation of the assumed model is then carried out by utilising a statistical analysis e g residual analysis A non linear model becomes important in critical flight regimes where consideration of non attached flow is assumed Examples of such conditions are post stall regimes high angle of attack flights unconventional configuration and rapid manoeuvres Eulrich 1974 and Raisinghani 1993 discuss such non linear modelling However the non linear analysis was not considered necessary for the Telemaster T 240 project 2 2 2 Parameter estimation methods Many papers have been written which discuss the parameter estimation methods such as Klein 1973 Ross 1979 Maine 1986 and Iliff 1989 Klein 1973 and Ross 1979 discussed in particular the estimation of stability and control derivatives from flight data In principle the parameter estimation method is divided into 2 main approaches in respect to model structures equation
8. Converting raw data into engineering units using the sensor calibration in appendix 4 Checking for outliers and missing data and correcting them as appropriate The correction of data was carried out manually using a text editor for ASCII Whereas the plotting of all data were carried out using Matlab Cropping the flight data according to the length of data to be analysed Filtering the flight data by a low pass filter Note that all data records should be filtered with the same filter to avoid any time shifts in data records which would degrade the parameter identification process A program called filtcoba m has been prepared to perform this 90 operation However in analysing flight record 3 and 4 no software filtering were necessary since the MLM estimation produced a good convergence even without filtering e Smoothing any selected flight data record use a program called smooth m Correcting incidence angles 0 and B for rates effect see appendix 2 Removing the non zero steady state values from each record 9 3 Stability and control derivative estimation The linear regression analysis has failed to give satisfactory results since no acceleration measurements were available Estimating these measurements by differentiating angle of attack and pitch rate did not help The noise in the data was actually attenuated by the differentiation process Hence we proceed with the maximum likelihood method MLM for analysing of al
9. Figure 8 8 Sideslip response using ICM analysis estimated 0 2 4 6 8 secon d Figure 8 9 Roll rate response using ICM analysis estimated 0 2 4 6 8 secon d Figure 8 10 Yaw rate response using ICM analysis estimated 84 8 3 The effect of measurement noise Measurement noises were added to all the flight records Then the MLM was used to estimate the longitudinal and lateral derivatives from the noisy records We assume a white Gaussian noise with zero mean and standard deviations as shown in Table 8 6 below Note that these values were taken as the scatter of the sensor calibrations Record Standard deviation Maximum signal to noise ratio Angle of attack 0 7 deg Sideslip angle 0 7 deg 1 2 deg s Rollrate 145 69 Table 8 6 Measurement noise level used in the simulation The results of the MLM algorithm are shown below No noise With noise Derivatives True Airport Estimated Cramer Estimated Cramer values parameter Rao parameter Rao Bound Bound 4399 6 4371 01373 4126 0 858 Ca 581 0 I775 2080 9401 2511 Cass 054 0 1146 005 1399 0909 onec ___ 685 66 9 Es losa p 0 n bV ue p Table 8 7 Estimated Longitudinal Derivatives Using MLM algorithm for cases with and without measurement noise 85 No noise With noise Derivatives True Airport Estimated Cramer Rao Estimated Cramer Rao values parameter
10. J Airspeed sensor Directional flow vanes Inertia pack Aoa amp sideslip 3 accelerometers amp 3 rate gyros Engine rpm Figure 6 1 Sensor location on the T240 model Figure 6 1 shows the location of each sensor on the aircraft The individual sensors used in this project are described as follows Acceleration measurement Linear accelerometers are used to measure longitudinal lateral and vertical accelerations of the vehicle The accelerometers are of the SETRA systems type model 104 with 0 7 critical damping and 350 Hz natural frequency They have an excellent static and dynamic response with unlimited resolution limited only by output noise level low transverse sensitivity 0 005 g g compact and light weight In 6g nominal range they have 1 linearity and produce a flat response from static up to 22 Hz The accelerometers are normally used in vibrations shock and impact measurements 60 Angular rate measurement Pitch roll and yaw rate measurements are obtained using 3 piezo rate gyros type NE 1000 These gyros use flexural vibration of a piezo electric triangular bar see Howell amp William 1994 These rate gyros have a Figure 6 2 Inertia unit consists of 3 linear accelerometers silver and 3 rate gyros black linearity range up to 720 sec Since the accelerometers and rate gyros are not located exactly on the center of gravity of the aircraft then corrections should
11. cscsessssesesesesseseecseeecenseseseeecaesesenseeesees 84 FIGURE 8 17 YAW RATE RESPONSE USING ICM ANALYSIS 201 224 000 020004000 84 FIGURE 8 18 THE EFFECT OF NOISE ON THE ANGLE OF ATTACK RESPONSE SOLID LINE TRUE RESPONSE NONOISE WITA NOISE J o Roe t RARE ER IER TENERENT gere ERE 86 FIGURE 8 19 THE EFFECT OF NOISE ON PITCH RATE RESPONSE SOLID LINE TRUE RESPONSE NO NOISE 2 WITH NOISE eter Neo en eene ei Dee er De Nee Se eade ERE DERE EER 86 FIGURE 8 20 THE THREE DIFFERENT INPUT FORMS USED IN THE SIMULATION eene nennen 87 FIGURE 9 1 ESTIMATED LONGITUDINAL RESPONSES AND THEIR RESIDUALS FROM MANOEUVRE 1 RECORDS 94 FIGURE 9 2 ESTIMATED LONGITUDINAL RESPONSES AND THEIR RESIDUALS FROM MANOEUVRE 2 RECORDS 96 FIGURE 9 3 ESTIMATED LATERAL RESPONSES AND THEIR RESIDUALS FROM MANOEUVRE 3 RECORDS 101 FIGURE 9 4 ESTIMATED LATERAL RESPONSES AND THEIR RESIDUALS FROM MANOEUVRE 4 RECORDS 103 LIST OF TABLES TABLE 2 1 SENSORS FREQUENTLY USED IN THE EXTRACTION OF STABILITY amp CONTROL DERIVATIVES 12 TABLE 3 2 STABILITY AND CONTROL PARAMETERS USED IN THE LINEAR DYNAMIC MODEL 2 TABLE 5 1 WEIGHT BREAKDOWN OF THE T240 AIRCRAFT MODBL esee enne TABLE 5 2 RESULTS OF MOMENT INERTIA EXPERIMENTS s
12. 46 The Wackett Centre 1995 Airborne CO Analyser Development for MAFV Jabiru Internal Memorandum The Wackett Centre for Aerospace Design amp Technology Tel 86 3090 15 December 1995 47 Thompson L A Abanteriba S and Bil C 1993 A Multi purpose Autonomous Flight Vehicle System Proceedings of the 5th Australian Aeronautical Conference Melbourne 13 15 September 1993 48 Valentinis F Bil C Riseborough P 1996 Development and Trials of an Autonomous Flight Control System for UAVs Paper to be presented at the ICAS 1996 in Italy 49 Ward Donald T 1993 Introduction to Flight Test Engineering Elsevier Science Publisher ISBN 0 444 881476 50 Wingrove R C 1973 Quasi Lineriazation Technique For Estimating Aircraft States From Flight Data Journal of Aircraft vol 10 no 5 p 303 307 51 Wolowicz C H Yancey R B 1974 Experimental determination of Airplane Mass and Inertial Characteristics NASA TR R 433 October 1974 52 Wong K C Newman D M 1989 Exploratory Study Into The Use Of A Remotely Piloted Vehicle RPV For Aerodynamic Research Proceeding of The Australian Aeronautical Conference 1989 Melbourne 9 11 Oct 1989 p 27 30 53 Yip L P Ross H M Robelen D B 1992 Model Flight Test Of A Spin Resistant Trainer Configuration Journal of Aircraft vol 29 no 5 Sept Oct 54 Young Shin en Kneen John 1995 Data Acquisition Systems of The Multi Purpose Autonomous Flight Vehicle Project Re
13. Ca Elevator _ pU Sb 1 21 _ 2 LE I Cl pU Sb E eus Oo oAileron pU Sb Dus E CL uns Cz 9 dz U T 9 9 4 cy 20 ES 24 _ pUSc M Cm 2I s 4 2 M yy om AT pU Sc Elevator 2 1 CM gtevator 2 EM SP ns _ pUSP N 2 202 pU Sb Naso 22 eo pU Sb N Rudder 21 Maudie Gas U Jud U 4 qc 4 Oklevator he a Cie 0n Pu 7 20 _ ORudder Cy dAileron CYarudder Cl sr 2U Cl cron d rudder 22 2 ee 20 Chas an CNapudder ch 1 Introduction Dynamic characteristics of an aircraft are normally described in terms of its stability and control derivative values These values are determined either theoretically empirical or semi empirical computational fluid dynamics or experimentally wind tunnel or flight test This project involves estimating stability and control derivatives of a model aircraft from flight data using parameter identification PI techniques The unknown stability and control derivatives are inferred from the modelled vehicle dynamic equat
14. Apriory values and Marquardt Filtered unfiltered model as their standard Constrained Cramer Rao Bounds a phi gam c d q x0 dt r deviation optional Newton Sensitivities owing b filename p Indices of parameters GDOP Geometric to be identified with Dillution of Precision Quadratic Lavenberg Innovation covariance marquardt or matrix Constrained Newton Kalman filter gain Perturbation size Correlation coefficient Tnitial estimate of matrix innovation covariance matrix 220 OPT Max iteration and convergence criteria for the minimization algorithm Note 1N wersumisdefinedas x 2 RR 2 and it converges to 1 at the minimum The logarithmic likelihood wersum log gel Figure 4 1 Summary on the use of MMLE3 toolbox in MATLAB 39 The parameters being identified are given in variable name p pid To ensure that the algorithm has reached the global minimum point and therefore the p pid is the maximum likelihood parameters then the following properties are worth investigating The value of LLF has reached the minimum value The value of wersum 1 i e residual Gaussian as the cost function minimum e gradient approaches zero i e variable MaxGrad 0 No further change in parameter i e max dP 0 e The plot of output data and response estimate yest are matched plot of innovation inovt should show an uncorrelated Gaussian noise e The RRnsum su
15. Bound parameter Bound Oa 0 004 _ 004 0049 0688 0451 Cy 002 0002 _ 0003 00 0 _____ 0006 a 0 006 0008 0008 00 00039 rad s 138 138 1417 Qa 075 o5 J joe 20 Tae 052 os 904 4 Table 8 8 Estimated Lateral Derivatives Using MLM algorithm for cases with and without measurement noise Figure 8 1 The effect of noise on the angle of attack response solid line true response no noise with noise Figure 8 2 The effect of noise on pitch rate response solid line true response no noise with noise 86 The presence of measurement noise increases the uncertainty in the estimated parameters This increased uncertainty is reflected in the increase of CRB values for both longitudinal and lateral derivatives However no significant change was noticed in the predicted responses as shown in Figure 8 1 and Figure 8 2 The most affected parameters in the presence of noise are pitch rate derivatives Cz and in longitudinal mode and sideforce derivatives and in lateral mode 8 4 The effect of different input forms The effect of different input forms Figure 8 1 to the estimated longitudinal dynamics was studied The LS and MLM algorithm were then used to extract the derivatives Table 8 9 and Table 8 10 show the results from the LS and MLM esti
16. CRB ua ur betapr betadot pdot rdot ua ur beta pr beta estp estr est P est CRB preprocess m Vax Vay Vaz Vp Vq Vr axcg aycg azcg pcg qcg rcg dc da dR rps Vdal Vdar VdR thrust vcg alphacg betacg Vrps Vdyn Vstat Valpha Vbeta 5 1 APPENDIX 6 TESTINGS 1 Engine test Test specification Engine type RC 80 approx 1 7 Hp Propeller diameter 14 inches Pitch 6 inches Engine speed range 0 9000 rpm Air speed range 0 25 m s Apparatus e thrust balance with the thrust calibration Thrust Newton 1 9833 x balance reading 0 454 The accuracy of the balance is approximately 0 2 Newton e A pitot static tube and an inclined manometer with SG 0 785 and inclination of 36 degrees The wind tunnel speed is calculated as v 2x9 81xsin 36 xSGxAH where is the manometer reading in mm e A digital tachometer with engine speed reading in rpm revolution per minute where rpm 100x displayed value on the tachometer e remote control system Note To avoid overheating inside the wind tunnel the engine exhaust is channelled out of the wind tunnel through a flexible hose Measurements Manometer Tachometer Balance Airspeed Engine Advance Thrust readings readings readings m s speed ratio coefficient mm rpm J v nD 1 2 3 4 5 6 7 8 9 Results Thrust model Ct 0 065 0 089 J or Thrust 1 84x10 n 6 46x10 V n At sea level where p 1 225 wh
17. INSTRUMENTATION AND DATA ACQUISITION SYSTEMG ccsscssssssssssscssssssessssesscesceecseeseaecsecsecseeseeseeeeeeeseeseeseassesaeeess 109 TE CONGTIUSTON BN 111 REFERENCE 112 APPENDIX 1 SENSOR CHARACTERISTICS USED IN THE TELEMASTER T240 FLIGHT TEST PROGRAM A1 1 APPENDIX 2 SENSOR ERROR 515 2 1 APPENDIX 3 CHARACTERISTICS OF THE TELEMASTER T240 MODBEL eee ene eaten A3 1 APPENDIX 4 FLIGHT TEST SENSOR 4 1 APPENDIX 5 FLIGHT TEST SOFTWARE DESCRIPTION e en en ento to tonta inen nane AS 1 APPENDIX 6 5 68 nins i 6 1 APPENDIX 7 FLIGHT TEST PROCEDURES AND RECORDS rere LIST OF FIGURES FIGURE 2 1 OF CONTROL INPUTS FOR DYNAMIC FLIGHT TESTINQG sese eene ertet nennen tenentes 19 FIGURE 3 1 OUTPUT ERROR ALGORITHM c ccssssssssssseseesssesesssessnsesescseceesesessssessseseesesescssesssesesceusesssaacsesesensessuaeecasseseeaeessees FIGURE 3 2 FLIGHT DYNAMIC TEST ACTIVITI
18. Murata 10 volt 6 volt 5 K Ohm LPO6M3RIHA torque 5 gr cm Angle of sideslip Flow vane potensio type Murata 10 volt 6 volt 5 max 6 rotational torque 5 gr cm dynamic range 0 720 s dynamic range 0 720 s max rotational MED od potensio type RS 173 574 Z 9 potensio type RS 173 574 potensio type RS 173 574 13 Airspeed Differential pressure sensor 5KOhm SENSYM SCCOS5DN Engine rotational speed Hall effect IC Switch RC 307 446 25V 4 5 to 24V 6mA NENNEN 12 Left aileron deflection Control position transducer potensio type RS 173 574 gt 1 5 E APPENDIX 2 SENSOR ERROR ANALYSIS 1 Linear acceleration measurement A Transducer error Transducer error for the accelerometer is modelled as bias error and scale error These errors are determined from the calibration B Kinematics error Any cg offsets and misalignments produce errors in acceleration measurements These errors are calculates as follows Laban 1994 Cg Offset error 7 Oey phe Zar Pr 4 Xeg Xmeasured a a ts TVo p G z Xqr p x4 Xa 2 2 7E LN Zeg Zaz ES Xog 4 E Yeg x Yaz X Tp p Yax and Za are longitudinal accelerometer positions and Zay are lateral accelerometer positions Xa Yaz and Za are vertical accelerometer positions Misalignm
19. be carried out to their readings These corrections are dealt with in appendix 2 The accelerometer readings are required to perform this correction Airflow direction measurement The airflow directions angle of attack ot and sideslip B are obtained using noseboom mounted flow vanes A low rotational friction potentiometers are used to measure the vane angular deflections The Murata MPO06NGRIHA potentiometers have a very low minimum torque of 5 gr cm which is an Figure 6 3 The angle of attack flow vane mounted essential feature in measuring the airflow on a low friction potentiometer direction The potentiometers are also shielded 61 against any electromagnetic interference From calculation of the vane dynamics the vane has a natural frequency of 108 rad s and damping of 0 2 The vane s natural frequency is well above the vehicles dynamic and hence should not pose any problem Figure 6 8 Flow vanes boom mounted on the wing of the aircraft Air speed measurement A pressure transducer and a pitot static tube were used in the airspeed measurement The pressure transducer senses the different between total and static pressures from the pitot static tube and converts this into an equivalent airspeed The pressure transducer Sensym SCXO1DNC operates at 0 1 psi differential pressure range with a static sensitivity of 18 mV psi An amplifying circuit has been built to provide a 2 5Volt output for a 0 10 inche
20. engine idle Similarly a poor quality of flow directional readings B were found in the presence of air turbulence Hying the aircraft very early in the morning has a better chance of having no air turbulence during the manoeuvres Despite all the problems in conducting the required manoeuvres summarised in chapter 7 4 the response with the pulse and doublet input produced a reasonably good matching This indicated that the required manoeuvre for the identification of stability and control derivatives estimation was not very strict Practically any input that adequately excites the mode of interest is acceptable 108 In short it is recommended that the dynamic manoeuvre for estimating the stability and control derivatives of a model aircraft should be performed at engine idle at calm air preferable in the morning in the form of pulse or doublet inputs Alternatively if the thrust model of the engine is available at a very good accuracy then the test can be conducted at any engine setting Muhammad 1995 10 4 Instrumentation and data acquisition systems An important objective of the project has been to develop and demonstrate the instrumentation systems needed for the dynamic testing of a model aircraft The obtained flight result has shown that the whole system can be used to obtain a reasonably good quality of flight data All the measurements were recorded on board the aircraft Hence no significant signal no
21. error approach and output error approach Equation error techniques such as linear regression solve simultaneous linear algebraic equations The equation of the form is solved to find the unknown matrix A Here x is the state matrix and y is the output matrix This technique is quite simple However it requires a large number of measurements namely the system s states as well as the input output All those measurements ought to be measured with a relatively high accuracy instrumentation system The performance of this technique degrades drastically in the presence of bias errors in the instrumentation Examples of this technique can be found in Laban 1994 and Mulder 1994 The Delft University of Technology has also developed Two Step Method which is a combination of Flight Path Reconstruction and Data Compatibility Check with regression analysis The output error approach is more popular in the field of parameter estimation than the equation error The output error approach requires fewer numbers of sensors Generalised least square or weighted least square Maximum Likelihood method and Bayes method are based on this output error approach The difference among the three methods described above lies in the selection of the cost function The Generalised Least Square allows only a near zero level of noise or known noise level of the various instrumentation used The Maximum Likelihood Method MLM assumes a White Ga
22. invaluable in the processing of the flight data Pre processing include converting filtering smoothing cropping removing outliers etc was a lengthy process Yet it was crucial in the success of the whole identification process Some outliers were present in the recorded data No dropouts of data were apparent Filtering the angular rate measurements with software has no considerable effect on the estimated derivatives Hence we allowed all the recorded measurements unfiltered when performing the MLM algorithm The MLM was the main algorithm used in estimating the stability and control derivatives of the T240 The LS has failed to give a good match since no acceleration measurements were available For the longitudinal LS we need angle of attack rate and pitch acceleration measurements For the lateral LS we need sideslip rate roll and yaw acceleration measurements In cases when the MLM could not identify some weak derivatives such as CZ and Cz the ICM method was used as a fine tuning to estimate these weak derivatives The GUI graphical user interface facility in Matlab has helped to speed up the MLM estimation process For example the values and the parameter to estimate can be easily changed through the click of the mouse 106 Some typical problems encountered during the MLM estimation was that the algorithm sometimes did not converge satisfactorily A minimum logarithmic value could not be achieved There
23. knife mass kg number of degree of freedom mass of the model Kg engine rotational speed rev s number of time points angular rate about X body axis rad s roll pitch and yaw rates rad s angular rate about Z body axis rad s distance of the strings from the centre of gravity radius of gyrations wing area m time s period of oscillation seconds Vo airspeed m s weight Xax Xay Xaz Xap distances of instruments forward of the centre of gravity m Zeg centre of gravity locations m 4 measurement vector Rye oe b L b L 2 non dimensional moment of inertias Subscripts a aileron al left aileron am apparent mass ar nght aileron e elevator i time index m measured bias or initial condition p q t 0t 06 B 88 8 derivatives with respect to indicated quantity r rudder Superscript T matrix transpose Dimensional Stability and Control Derivatives Definitions X P Cx 2 X Px pUSc X TS Cx OElevator mo pU S 2m pUSb 1 4m C y pUSb C 4 pU S T MC YaAileron pU S ORudder am Cue where Cr X U i Gi qc du TUM al t Cy ___9 _ UA pus Z 2 Cz m 2 7 8 Cz _ _ pu s Elevator 7 2
24. linear effect Eulrich and Rynasky 1974 and Raisinghani 1993 discuss some of the non linear modelling However this non linear modelling is outside the scope of this project Table 3 2 Stability and control parameters used in the linear dynamic model Longitudinal 6 paranee Lateral 15 parameters p neue Ciis S es fe Css Tertiary Cy Cy aiteron peg 34 3 4 Flight Test Manoeuvres The following were taken into account when choosing the type of control inputs and manoeuvres to be performed by the pilot Most dynamic derivatives can be extracted successfully from manoeuvre with only a doublet input with the input frequency near the vehicle s natural frequency which is approximately 5 rad s for the T240 model This form of input is the most practical Maine 1986 e Alternatively the 3211 form input should be performed since this input has a wider frequency content and thus produces a better estimate of parameters The wider the frequency spectrum the more likely the aircraft is to be excited However this type of input is rather difficult to realise in practice than the pulse or doublet forms e Minimise any cross coupling between the longitudinal and lateral motions e The manoeuvre should be performed in the linearity range i e and B excursions should not exceed 5 degrees and of constant speed so that the validity of the linear equation
25. main reasons contributed for this unsatisfactory convergence wrong a priory wrong parameter to estimate or wrong mathematical model Since the accelerometers were not working no linear acceleration readings were available Had these readings were available we would have been able to perform some corrections to the angle of attack and sideslip data compatibility checking 107 10 3 Flight test manoeuvre Most manoeuvres conducted in this project were of pulse or doublet type inputs These inputs were reasonably easy to perform Yet the recorded responses contained a sufficient information to enable the MLM algorithm to extract some dominant derivatives One major point to consider is a need to compromise between a large magnitude of input and a small magnitude of responses On one hand we need a large input to excite the response On the other hand the resulted responses should remain within a linear region So that the validity of the uncoupled linear model can be preserved This proved to be not an easy task for the pilot Both manoeuvre 3 and 4 produced quite large sideslip responses Hence the validity of the linear model used was under question Another significant problem was the present of engine vibration noise in the angular rate readings When the manoeuvre was conducted at a throttle setting the rate readings were buried in noise Significant improvements in rate readings were achieved by conducting the manoeuvre with
26. the accelerometer drove the amplifiers into saturation Sensor and DAS adjustments such as reading range and resolution were an elaborate process The process had to be carried out in two different places The sensor calibration was conducted at the Aerospace Engineering Department and the adjustment of the sensor sensitivity in the DAS was carried out at The Computer System Engineering Department 69 7 2 Transmitter and DAS Data Acquisition System interference We experienced an interference problem between the transmitter signal and the DAS During preliminary flight tests the transmitter signal has momentanly lost twice The existent of the interference was also noticed during the ground range test The transmitter signal terminated immediately when the DAS was switched on Two immediate actions were taken before continuing the flight test First the whole DAS was placed in an enclosed metal box and grounded to the battery Second the receiver and antenna were moved to the bottom of the fuselage so that their positions are as far away from the DAS as possible However no significant improvement was noticed from these two actions The interference problem was solved after many trials and errors There was substantial assistance and suggestions from John Kneen Mal Wilson and Mitchell Lennard The steps taken to reduce the interference are described below They are listed chronologically Enclosing the whole DAS in
27. 0 0434 standard deviation r 04275 0 7946 Table 8 4 Mean and standard deviation of the fitted error response for the various identification algorithms standard deviation p 1 6374 1 0633 From the simulation the sensitivity of each derivative to the flight responses can be studied Table 8 5 presents the result from the sensitivity study This table is very useful in assisting which parameters to be held fixed during the MLM estimation 81 degis Degree of sensitivity Moderate Ga NE SRM RR GN Lo __ ___ 2 LE Cy amp udder needs high pe pu Gua __ 0 l Table 8 5 Sensitivity of each derivative to the flight responses 20 0 2 4 6 8 Figure 8 3 Roll rate response using regression analysis estimated 82 degrees a 8 asa nc d sa 8 a 0 2 4 secon ds Figure 8 4 Yaw rate response using regression analysis estimated 15 a 5 5 6 0 2 4 secon d Figure 8 5 Sideslip response using MLM analysis estimated 100 80 60 40 20 0 20 6 2 4 secon d Figure 8 6 Roll rate response using MLM analysis estimated 20 15 10 5 0 5 10 15 20 25 30 0 2 4 secon d Figure 8 7 Yaw rate response using MLM analysis estimated 83 degis 0 2 4 6 8 secon d
28. 0 160 170 130 140 150 160 170 180 190 error plot error plot 2 150 160 170 180 190 130 150 160 170 180 190 counts counts fitted curve fitted curve 0 120 130 140 150 1 9 1 92 error plot error plot 120 130 140 150 counts Figure A4 3 Roll Rate Gyro Chn 4 Calibration Figure A4 4 Airspeed Sensor Calibration 4 2 fitted curve fitted curve 100 120 140 160 error plot 150 200 error plot 120 140 counts 150 counts Figure A4 5 Yaw Vane Calibration fitted curve fitted curve 150 100 150 200 250 error plot error plot 150 100 150 200 250 counts counts Figure A4 7 Left Aileron Calibration Figure A4 8 Right Aileron Calibration 4 3 fitted curve 150 200 error plot 150 200 error plot counts Figure A4 11 Rudder Calibration fitted curve 40 60 80 100 error plot 80 100 120 140 160 counts Figure A4 10 Right Elevator Calibration A4 4 APPENDIX 5 FLIGHT TEST SOFTWARE DESCRIPTION Input and output variables in the subprogram m files M file Input variables Output variables NNI u alpha q alphadot qdot u alpha q alphadot qdot ua ur beta p r betadot pdot rdot u alpha q u alpha q alphadot qdot alpha_est q_est Cz Cm Cz_est Cm_est P_est Islatgui m ua ur beta pr betadot pdot rdot ua ur beta p r betadot pdot rdot Cy Cl Cn Cy_est Cl_est Cn_est P_est beta_cal p_cal r cal STD u alpha alphadot qdot u alpha q alpha estq est P est
29. 108 rad s and damping of 0 19 The approximate time delay tis 19 msec Flow vane geometry Flow Vane sources of errors A Aerodynamic position error due to flow perturbation in the presence of nose or body Hence the local angle will not represent the free stream flow directions The flow vanes in this project were located far from the nose twice the fuselage diameter hence this error is assumed to be negligible B Kinematics error due to offset vane locations from the centre of gravity This offset location produces angular velocities which affect the flow angle measurements x 5 Oyane are tan x V y arc Where V V and V represents velocities relative to the air p vane X Then both the aerodynamic position error and kinematics error can be formulated as Vz AV aicinduced d Xyane Wane location 476 AV AVxz a cinduced amp vane cg B Co location AQ 5 induced 4 V Vy AVY a c induced vane Xcg P Zvane cg Pose location 8 81 Vx AVxz cg 7 cg Zvane B Bes location AB co induced V TE V In a typical doublet manoeuvre the T240 may experience a maximum pitch rate of 2 0 rad s from simulation If the vanes are located at 1 meter forward of the c g this introduces kinematics error of q Xvane Xce V which corresponds to a 4 error in angle of attack A2 2 C
30. 15 2 5549 10 0 4075 deg 2 6412 x 10 0 40 deg 0 17 deg 6 34 5 9816 10 1 02 X 41 9118x10 X 5 4749 17 18 channel 18 45 87 Left aileron chn 13 Note is the corresponding channel reading Flap Channel 14 68 T Practice and implementation problems 7 1 Problems encountered during the development of the instrumentation systems We originally planned to use a Remtron RTS 1 Telemetry System for collecting the flight data The system was developed by the Computer System Engineering Department RMIT Howell and Wiliams 1994 However since we were anticipating more problems in trying to make the system works e g signal interference with the receiver then we decided to develop an onboard DAS instead The onboard DAS would also produce a better flight data reading than the telemetry system Due to memory devices problems in the DAS we could not have the 256 Kbytes corresponds to a 10 minutes of data acquisition onboard memory originally planned Instead we have a 16 second of flight data recording The rate gyros consume a lot of current In the calibration all the three gyros gave an inconsistent result A large drift was noticed especially on roll gyro Eventually an extra power supply was added using a 7 2 NiCad battery We had problems in getting the accelerometers working There was no provision made in the DAS for the accelerometer offset voltage and hence
31. AL 5 5 5 0 0 0 0 42 2 103 TABLE 10 1 ESTIMATED LONGITUDINAL DERIVATIVES OF THE TELEMASTER T240Q sse 104 TABLE 10 2 ESTIMATED LATERAL DERIVATIVES OF THE TELEMASTER T240Q sese 104 iii Abstract The project involves estimating stability and control derivatives of a remote control aircraft model from flight test data using parameter identification techniques The stability and control derivatives are inferred based on the modelled vehicles dynamic equations and the measured inputs and aircraft responses during a predetermined manoeuvre Computer programs necessary to perform the identification processes have been developed using Matlab a matrix manipulation software The identification from simulated data has been carried out to assess the effectiveness of the identification algorithms In addition instrumentation and data acquisition systems for conducting the flight test program have also been developed in collaboration with the Computer System Engineering Department RMIT Implementation challenges encountered during the development of the whole flight test systems are presented The capability of the whole system was then demonstrated by conducting a dynamic flight test program on the Telemaster T240 aircraft model Six longitudinal and fifteen lateral derivatives have been extracted from several recorded flight test data The estimated derivatives will then be used in the design of
32. ES ccscsssssesessescesescescesessescssescsecassecnecaesecaesacnecassecnessesecaeseesecaeneaeeneaeeasens FIGURE 3 3 TRANSIENT PEAK RATIO METHOD DIAGRAM Bs FIGURE 3 4 TRANSIENT PEAK RATIO MEASUREMENTS cccccsssssssssscsesessesesesesececsesescssescuesecaesesesaacsesecensesenaecaeseseeaeecseas FIGURE 3 5 RELATIONSHIP AMONG THE DIFFERENT TECHNIQUES USED IN THIS PROJECT 24 FIGURE 3 6 INPUT OUTPUT FOR THE THREE DIFFERENT IDENTIFICATION METHODS eene 26 FIGURE 3 7 THE INTERACTIVE CURVE MATCHING ALGORITHM seen ener entente nenne tentent tentent nennen 30 FIGURE 3 8 LONGITUDINAL CURVE MATCHING MENU PROGRAMMED IN MATLAB AND IT S CORRESPONDING ERROR EA YOU tenter rre Re tsb sen 31 FIGURE 4 1 THE STRUCTURE OF THE FLIGHT TEST COMPUTER PROGRAM DEVELOPED FOR THE PROJECT 37 FIGURE 4 2 INTERCONNECTION BETWEEN M AND MAT FILES IN THE PROGRAM cccscssssssesescseeeceesesesteecsesesenseecsees 38 FIGURE 4 3 SUMMARY ON THE USE MMLE3 TOOLBOX IN MATLAB esses 39 FIGURE 4 4 COMPATIBILITY CHECKING ALGORITHM USED IN THIS PROJECT 1 12 4 0 0000000000000000000003 41 FIGURE 5 1 THE TELEMASTER T240 AIRCRAFT MODEL TO BE FLIGHT TESTED eese nennen 1 FIGURE 5 2 EXPERIMENT SET UP FOR THE ENGINE TEST 46 FIGURE 5 3 THRUST MEASUREMENT IN THE 50X50CM AER
33. ESTIMATED LONGITUDINAL PARAMETERS USING LINEAR REGRESSION ALGORITHM 5 erre reete eee reto neret ee A exuere rne aenea ee reser E e 88 TABLE 8 10 THE EFFECT OF DIFFERENT INPUT FORMS TO THE ESTIMATED LONGITUDINAL PARAMETERS USING MAXIMUM LIKELIHOOD ALGORITHM TABLE 9 FLIGHT DESCRIPTION t ERE REPRE UH EEUU RU E REDE Eheu TABLE 9 2 MANOEUVRE DESCRIPTION cic i tet teen RH RUE Re Rei Pe ee ve E HERE ane ete peo RE cag TABLE 9 3 FLIGHT TEST CONDITIONS FOR EVERY MANOEUVRE eret 90 TABLE 9 4 ESTIMATED LONGITUDINAL PARAMETER FROM RECORDED DATA MANOEUVRE 1 WITH TWO DIFFERENT we entere hte ey eee Senses e ore ome EE eee eee 91 TABLE 9 5 ESTIMATED LONGITUDINAL PARAMETER FROM RECORDED DATA MANOEUVRE 2 WITH TWO DIFFERENT SETS OF A PRIORI VALUES tn etn nee exeo eye ue conden eerte e eere cute TABLE 9 6 RESIDUAL CHARACTERISTICS OF THE ESTIMATED LONGITUDINAL RESPONSES TABLE 9 7 ESTIMATED LATERAL PARAMETER FROM RECORDED MANOEUVRE 3 WITH TWO DIFFERENT SETS OF APR ORT VA U S eterne trente eR Pr PRETEREA Ye te EHE RUE C RUE PERIERE 97 TABLE 9 8 ESTIMATED LATERAL PARAMETER FROM RECORDED MANOEUVRE 4 WITH TWO DIFFERENT SETS OF APR ORT VA UD Yero E EE SEN Tee even age 97 TABLE 9 9 RESIDUAL CHARACTERISTICS OF THE ESTIMATED LATER
34. IDENTIFICATION OF MODEL AIRCRAFT DYNAMIC USING FLIGHT TESTING by Edi Sofyan Thesis submitted in accordance with the regulation for the degree of master engineering Supervised by Robert Danaher Aerospace Engineering Department Royal Melbourne Institute of Technology Victoria Australia September 1996 Declaration I Edi sofyan declare that this thesis is my own work except where dully acknowledged to others and has not been submitted previously in whole or in part in respect to any other award All work has been carried out since the official date of commencement of this research program Trade of manufacture s names are used where essentials to this applied research Endorsement of those names is not intended E Sofyan September 1996 Acknowledgements The author wishes to express sincere appreciation to the following people for their inspiration and help during the work of this thesis Robert Danaher Dr Cees Bil Associate Professor John Kneen Mal Wilson Lachlan Thompson Professor Vladislav Klein and the MAFV personnel and for their support tolerance and understanding my wife Andhika Purnamasari and my daughter Amanda Haruminori Sofyan TABLE OF CONTENTS ASBTRAGCT x cce t E ne EN Ri A HE SOR 1 NOMENGEATURE soeben home 2 TINT ROD UC TION eder re 6 2 10 2 INSTRUMENT ACTION E A p Oe
35. M algorithm on the other hand estimated most of the longitudinal derivatives satisfactorily except for the Cz The large Cramer Rao value for the Cz indicates that this derivative is weakly identified The amp Figure 8 4 and pitch rate Figure 8 5 show a good fit between the actual and estimated responses The SPO characteristics were also well identified The ICM algorithm estimated Cm Cmbtevator and SPO characteristics quite well Those parameters which do not change the and q responses significantly such as and Cmq are poorly estimated Figure 8 6 and Figure 8 7 show the result of the fit Figure 8 2 Angle of attack response using regression analysis estimated 78 degrees deg s degrees a i a a o o o o n o a e 2 a a a o Figure 8 6 Angle of attack response using ICM analysis estimated 79 degis Figure 8 7 Pitch rate response using ICM analysis estimated 8 2 Lateral identification The aircraft was excited by a rudder doublet Figure 8 1 followed immediately by an aileron pulse Figure 8 2 The responses of the model lasted about 8 seconds seo Figure 8 1 Rudder deflection The results of the identification using various identification techniques are summarised in Table 8 3 Also shown in the table are the characteristics o
36. MLM is widely used in the extraction of stability and control derivatives of either small or large UAV and other types of aircraft routine to perform algorithm is available in either MATLAB toolbox Milne 1992 or Xmath Matrix X identification module Both Matlab and Xmath software are accessible at the Aerospace Engineering Department RMIT Some problems commonly encountered in using the maximum likelihood analysis occur if e There is a linear dependency between the unknown parameters e There is aeroelastic coupling between flight mechanics and structural modes e g structural vibration Drifts in the states e g caused by variation in flight conditions Improper specification of instrumentation and inaccurate modelling 2 3 Input forms The most widely used inputs for dynamic flight testing are single pulse and doublet 1976 Colemann 1981 Howard 1991 and Yip 1992 Both inputs are relatively easy to execute while at the same time producing responses with a relatively rich information about the dynamics of the vehicle Other commonly used inputs are PRBS sine sweep and 3211 type see Figure 2 1 Several papers have also been written in formulating a mathematically optimal input Chen 1975 However this type of input is rather complex and difficult to execute during flights Other constraints that dictate the input form selection are safety envelope coverage hardware constraints an
37. OSPACE ENGINEERING WIND TUNNEL RMIT 47 FIGURE 5 4 THRUST COEFFICIENT TO ADVANCE RATIO RELATIONSHIP FOR THE PROPELLER 48 FIGURE 5 5 COMPARISON OF THE THRUST CHART FROM THE EXPERIMENT AND THE DERIVED THRUST MODEL 48 FIGURE 5 6 EXPERIMENTAL TECHNIQUE FOR DETERMINING WEIGHT AND CG POSITIONS eee FIGURE 5 7 RESULTS FROM THE CG EXPERIMENT cccssssscsssscsssescscescessescssesessuesecsesescsseacueseseesesesaacsesecesesenaecaesesenaeecseas FIGURE 5 8 PITCHING MOMENT OF INERTIA DETERMINATION USING A KNIFE EDGE METHOD FIGURE 5 9 YAW AND ROLL MOMENT OF INERTIAS DETERMINATION USING BIFILAR SUSPENSION METHOD 53 FIGURE 6 1 ON BOARD DATA ACQUISITION SYSTEMS ccccsssssssssssssescscsessssesessececsesescusescsesecaesesesaeecsesecensesenanecaeseesaeesseas 56 FIGURE 6 2 ON GROUND DATA SYSTEM essere ennt nnne tente 56 FIGURE 6 3 THE ON BOARD DATA ACQUISITION BLOCK DIAGRAM FOR THE T240 FLIGHT TEST PROGRAM 57 FIGURE 6 4 THE DAS CARD USED IN THE FLIGHT TEST nre tenente tnit teniente 57 FIGURE 6 5 SENSOR LOCATION ON THE T240 MODEL essere teens tenter 60 FIGURE 6 6 INERTIA UNIT CONSISTS OF 3 LINEAR ACCELEROMETERS SILVER AND 3 RATE GYROS BLACK 61 FIGURE 6 7 THE ANGLE OF ATTACK FLOW VANE MOUNTED ON A LO
38. PO was even estimated quite accurately In the analysis the results from the LR are used as a priori values for the MLM and ICM Among the three 76 techniques the MLM produces the best estimate of the derivatives Table 8 2 shows that the MLM produces the smallest error criterion Algorithms Derivatives True Estimated Standard Estimated Cramer Estimated 68 6n em Ee Table 8 1 Results from various estimation algorithms LR MLM ICM standard deviation 0 2924 0 0154 0 0179 0 0148 0 1226 0 3013 standard deviation q 1 8822 0 7197 1 1549 Table 8 2 Mean and standard deviation of the fitted error response for the various identification algorithms Since the acceleration measurements were not available when performing LS algorithm the and records were differentiated to produce Cz and Cm respectively This explains the reason why significant errors are observed from the LS result The differentiation of and q have introduced significant noise The LR estimated Cz Cm and Cm quite well However Cz CZetevator and Cm were poorly estimated The resulting fits to and q are shown in Figure 8 2 and Figure 8 3 respectively The estimated responses show a significant error after the elevator input was removed after 3 seconds The damping was underestimated 50 down but the frequency was closely estimated 1 5 down 77 The ML
39. W FRICTION 61 FIGURE 6 8 ENGINE RPM SENSOR AND THE ROTATING DISC entente tn tentent tentent nnns 63 FIGURE 6 9 PROPELLER ROTATIONAL SPEED MEASUREMENT USING HALL EFFECT IC SWITCH DEVICE 63 FIGURE 6 10 RUDDER DEFLECTION SENSOR sees entere nnne intrent nen 64 FIGURE 6 11 RATE GYRO CALIBRATION USING RATE TABLE eerte entente 66 FIGURE 6 12 RATE GYRO CALIBRATION TRACE nns 66 FIGURE 6 13 RESULTS OF THE SENSOR CALIBRATIONS eese nennen teet 68 FIGURE 7 1 THE HALE SCALE MA EY enisi rrr Hr RERO HR dte Er EE RO d PU ERR eterne 72 FIGURE 7 2 THE TELEMASTER PRECEDENT T240 essent entente trennen tenent 72 FIGURE 7 3 ROLL RATE READING BURIED IN ENGINE NOISE DURING A FLIGHT MANOEUMVRE eere 74 FIGURE 7 4 ROLL RATE READING WITH ENGINE IDLE FIGURE 7 5 ANGLE OF ATTACK READING BURIED IN TURBULENCE DURING AN ELEVATOR DOUBLET UE NOISE 75 FIGURE 7 6 ANGLE OF ATTACK RESPONSE IN A REASONABLE CALM AIR eerte tnnt nennen 75 FIGURE 8 1 ELEVATOR DEFLECTION e e ee
40. ___ 0165 fixed 123 0886 C _____ 0 0745 0 0930 1 086 0 1989 0 177 1 694 2 555 8123 1745 ed ed Cy amp aileron fixed ed fixed fixed 4 196 fixed fixed Chu 0 fixed 0256 920 ba o 0M Eo E sed s 0 033 0 fed 00395 0380 fied BRENNEN gt O o k Table 9 8 Estimated lateral parameter from recorded manoeuvre 4 with two different sets of a priori values piral sec 0 05 97 A similar procedure as that for the longitudinal estimation was used Since more parameters were to be estimated in lateral case the estimation process was slightly more difficult It involved trying to fix any weakly derivatives and to find a good starting value for the dominant derivatives Results from analysis 3 2 was better than the other 3 lateral analysis Analysis 3 2 produced a better estimated parameters with smaller CRB and a better matching of flight data Analysis 3 2 estimated all the 15 lateral derivatives with a reasonable degree of confidence The only exceptions are for the sideforce derivatives and Claas The simulation result had predicted these derivatives would be hard to estimate No sideforce information can be accurately extracted from a low frequency excitation Coleman 1981 A high frequency input with lateral acceleration readings are required to estimate these
41. al linkages connecting the two sides of the control surfaces Due to this error it becomes impossible to have a perfectly symmetrical movement of the left and right control surfaces Ideally deflection sensors should be placed on all control surfaces However due to limited number of channels available only left and right ailerons are measured separately In this project since the linear accelerometers were not working then no angular rate correction can be performed Kinematics errors for angle of attack and sideslip were corrected From the error analysis above other low vane errors can be neglected A2 6 APPENDIX 3 CHARACTERISTICS OF THE TELEMASTER T240 MODEL Wing Value w e Ec NEN 288 ae Aileron Value 5 1 Dihedral angle deg Inboard station half span 0 15 Swept angle deg B Outboard station half span 100 Tailplane Value ea 22 Lco NN Elevator Value mee LLL ____ Vertical fin Value Area including rudder 820 Span cm 39 0 1 1 2 D lift curve slope per degree 2 D drag curve slope per degree Rudder Value Outboard station half span 4 3 3 90 9 13 100 A3 1 Flap Value ENGINE AND PROPELLER Value DISTANCES Value CHARACTERISTICS Area of each flap cm 348 Fuselage length cm Engine type Irvi
42. an aluminium box to prevent any radiation from the DAS 2 Moving the receiver and antenna to the bottom of the aircraft s fuselage as far away as possible from the DAS 3 Collecting all the sensor ports into a single port and hence reducing the complexity of the sensor wiring going into the DAS 4 Replacing all the cables parallel to the antenna those of rudder elevator and engine rpm by shielded computer data cables Then all these sensors had to be recalibrated Associate Professor John Kneen is a senior lecturer at the Computer System Engineering Department RMIT He has built the DAS for this project and currently supervising 2 Phd s in flight control systems Mal Wilson is a technical staff at the Aerospace Engineering RMIT formerly electrical technician with RAAF He has flown model aircrafts for more than 15 years and has a lot of experience in electrical and communication Mitchell Lennard is an avionic design consultant with Mikley system integration 70 5 Moving all the power supplies into the aluminium box together with the DAS This was done since the power supplies might radiate signals which interfered with that of the transmitter 6 Moving all the switches that for data retrieval power supplies and rate gyro into the aluminium box By this time all the cables were contained inside the box except that from the sensors located around the aircraft 7 Installing a digital low pass filter D
43. ar system the model can be represented in a polynomial form as y t O X 40 x 4 49 x 3 5 Or as a regression equation 3 6 where X x x 2 and 9 8 9 9 11 3 7 X regressor matrix N x n N number of parameter Y measured Y matrix N x 1 N number of data points equation error estimated parameters 26 The parameter estimate 0 is obtained by minimising the error cost function J given as J y s Which produces the parameter estimate 6 as Mz 40 42 95 42 3 8 The spread of parameter estimate covariance is calculated as covariance 26 X7 X ee 3 9 T Where 3 10 N n The quantity of information in the data that can be explained by the model is given in the coefficient of determination R where f _ SUMOF 590 _ peepee sum of square 3 11 and y estimate of The correlation between the regressor is given as T X X Where R T Www and X is the centred data X X 12 w diagonal elements of X X matrix 3 12 The adequacy of the model can be assessed by looking at the R and PRESS values Oo x y Ny VEE ee 3 13 n 1 s GO y 0r PRESS 000 3 14 2 variance y i pe se 3 15 1 1 R A better model is indicated by high values of the above var
44. board memory records flight data for intervals of 16 seconds The recording is initiated by means of a microswitch operated from the radio transmitter and terminated automatically after 16 seconds of data acquisition At the end of every flight the data is 57 downloaded into a personal computer RS232 for further processing using Telemate communication software The system is equipped with 2 control input buttons see Figure 6 3 sample green button and dump blue button When the sample button 15 activated the DAS will record one set of sample for 16 seconds When the dump button is activated the DAS will transmit the contents of its memory over the RS232 channel This dump button has a secondary function i e for a calibration mode If the button is pressed during reset the microprocessor is reset by removing and applying power the DAS will then go into calibration mode In this mode the input channels are continuously monitored and the results are transmitted via the RS232 line to a monitor The Telemate communication software is used to display and save the results for further analysis The DAS collects 3 different type of input data potentiometer inputs voltage inputs and timer The potentiometer inputs can deviate positive or negative For maximum sensitivity the potentiometers should be mounted so that to give reading close to 000 at minimum negative potentiometer deflection and close to 255 at maximum positive deflecti
45. connector to all the cables carrying currents The attenuation of the filter was approximately 20dB at 40 MHz 8 Elimination of all intermittent ground loops 9 Replacing the on off relay switch by a microswitch to trigger the DAS The microswitch was operated by a servomotor through the gear channel on the radio transmitter The idea was to eliminate any direct cable connection between the receiver and the DAS Also by using a separate motor we would have an option to use a separate transmitter to trigger the DAS By this time the ground range was considerable improved to approximately 150 meter However this was still not yet considered adequate for the aircraft to fly safely 10 Changing the radio transmitter frequency from a PCM 36 MHz to TF FM 29 725 MHz This was done since we suspected that the DAS clock at 3 6864 MHz somehow interferes with the transmitter signal at 36 Mho 1 10 harmonic Another option was to change the internal clock of the DAS However this would create problems in retrieving the data from the DAS since the 3 6864 divides down to give the standard serial baud rates of 9600 By this point significant increase in the transmitter range satisfied us to resume our flight testing 71 7 3 Problems on the aircraft model to be tested e first aircraft model to be flight tested was the half scale Figure 7 1 A pitot static and flow vane s boom was mounted on the nose of the mod
46. csssssscssssssesesescsececsesescsseseseseceesesesascsesecenseseneecaeseesaeesseas TABLE 8 1 RESULTS FROM VARIOUS ESTIMATION 5 TABLE 8 2 MEAN AND STANDARD DEVIATION OF THE FITTED ERROR RESPONSE FOR THE VARIOUS IDENTIFICATION ALGORITHM S Rere terea eerte re EN FOROR IP Peer KE NER sen DX eee PUPA TI TABLE 8 3 RESULTS USING VARIOUS ESTIMATION ALGORITHMG cscsssssssssesesesesesceseseuesecaesesesaeecsesecensesesaecaesesesaeesseas 81 TABLE 8 4 MEAN AND STANDARD DEVIATION OF THE FITTED ERROR RESPONSE FOR THE VARIOUS IDENTIFICATION ALGORITHMS trae oie sees tettcesssustecensescnesussuset e euusetceceduusecdeesebcucceecuascobersectescstenceess 81 TABLE 8 5 SENSITIVITY OF EACH DERIVATIVE TO THE FLIGHT 5 5 85 1 2 40 2044 6 00 82 TABLE 8 6 MEASUREMENT NOISE LEVEL USED IN THE SIMULATION ccccssssssssssssssseseseesceesesescescsesecensesesaeeceeseeesaeesseas 85 TABLE 8 7 ESTIMATED LONGITUDINAL DERIVATIVES USING MLM ALGORITHM FOR CASES WITH AND WITHOUT MEASUREMEN T NOISE eerte eerte ee eese desee eee STEE TE SEESE PSR 85 TABLE 8 8 ESTIMATED LATERAL DERIVATIVES USING MLM ALGORITHM FOR CASES WITH AND WITHOUT MEASUREMENT NOISE decer a e evertere iei ve EE C TE eter 86 TABLE 8 9 THE EFFECT OF DIFFERENT INPUT FORMS TO THE
47. d control systems influence 40 doublet Sine sweep Figure 2 1 Type of control inputs for dynamic flight testing 19 3 Overview of the method Selection of a particular method in flight testing a model aircraft depends on the objective of the test number of measurements taken and their type of accuracy and means of computational available In this project the stability and control derivatives of the Telemaster T240 model are estimated from flight test data using an output error method The output error method is used in extracting the stability and control derivative of the aircraft Figure 3 1 The method minimises a defined error cost function J to produce the best fit between the flight data and its simulated responses of the assumed mathematical model Since the assumed mathematical model consists of several unknown parameters that have to be identified the method 15 also commonly known as the parameter identification input responses error Figure 3 1 Output error algorithm 20 The whole activity in flight testing the UAV model is depicted in Figure 3 2 Test planning mass characteristics determination and calibration of instrumentation are categorised as pre flight activities whereas data processing amp analysis data compatibility check and parameter identification as post flight activities A priori information about the derivatives is used to either complement or assist in the process
48. derivatives Two different manoeuvres were conducted for the lateral identification Manoeuvre 3 had a combined aileron and rudder inputs Whereas manoeuvre 4 had only rudder input Estimated parameters from the rudder input only had a significant larger CRB This was because that the rudder only produced a less rich information content The data analysed in manoeuvre 4 was also shorter only 1 84 s The values of residual characteristics in Table 9 9 also support this argument The mean and standard deviation in manoeuvre 3 was generally less that those in manoeuvre 4 Another point to notice was that the roll derivatives could not be extracted from a rudder only manoeuvre as seen from Table 9 8 This suggests that little roll information was contained in the data Data from an aileron only manoeuvre would certainly be used to extract the roll roll derivatives Hight 1 was designed to extract the roll derivatives But since the data was covered by engine vibration noise the data could not be used 98 reasonably good matching for manoeuvre 3 and 4 were obtained and shown in Figure 9 1 and Figure 9 2 Matching flight data flight data estimated manoeuvre 3 1 estimated manoeuvre 3 2 25 20 15 10 5 Ej 0 5 10 15 20 25 3 4 6 8 9 10 seconds Rudder deflection 0 5 0 0 5 1 3 2 2 5 3 3 5 4 WS g 4 5 6 7 8 9 10 seconds Right aileron deflection 20 15 10 8 5 E 0 5 10 15 AO 5 6 8 9 s
49. due to the boom A2 3 Assessment of the flow for the T240 The table below compares several existing flow vanes Velocity Natural freq Damping m s rad s Sydney University RPV 10 J e _____ Swearingen Metro II 1240 model Source of errors for the T240 flow vanes Error source Magnitude Comment Flow perturbation assumed negligible Verify with the press distribution at the nose Determinate error Static alignment obtain from wind tunnel calibration The vane design is acceptable since its damping is relatively high and its natural frequency is well above aircraft s mode is typically lt 18 rad s The expected error from the vane systems is small and remains inside the required resolution of the sensor which is 19 Note that the kinematics error is quite significant and should be accounted for during the analysis of flight test data 4 Air pressure measurement A Aerodynamic position error due to the presence of the nose or body This error normally dominates the static pressure errors However since the static pressure is located far away from the nose in the T240 configuration then this error is assumed to be negligible B Kinematics position error due to the offset position from the cg The kinematics error for the total pressure measurement is given as 1 2 P P 5 PWeg Ta Veg AVpitot Y P 5 E Pr T Yeg r Z pitot Zeg 4 8
50. dynamic of a model aircraft can be estimated with a reasonable confidence using flight testing 111 References 1 Budd G D Gilamn R L 1993 Operational And Research Aspects Of A Radio Controlled Model Flight Test Program paper 93 0625 January 1993 or NASA TM 104266 2 Butter R W Langhan T F 1976 Aircraft Motion Sensitivity To Variations In Dynamic Stability Parameters Agard Stability amp Control Conference Florence Italy October 3 Chen R T N 1975 Input Design For Aircraft Parameter Identification Using Time Optimal Control Formulation Methods For Aircraft State And Parameter Identification AGARD CP 172 paper 13 May Ist 4 Chow 1996 Stability and control derivative of the T240 model aircraft AV408 undergraduate project Aerospace Engineering RMIT 5 Coleman R Robins Frary D J Stephenson R 1981 Mini RPV Research The Aeronautical Journal Feb 1981 p39 47 6 DARoorporation 1996 Advanced Aircraft Analysis User s Manual Version 1 7 February 1996 7 De Jong Mulder 1987 Accurate Estimation of Aircraft Inertia Characteristics From a Single Suspension Experiment Joumal of Aircraft vol 24 no 6 p 362 370 June 1987 8 Draper N R Smith H 1981 Applied Regression Analysis 2nd edition John Wiley amp Sons Inc NY 9 Eshelby M E 1991 Short Course in Experimental Mechanics of Flight Cranfield College of Aeronautics 10 Eulrich Rynas
51. e calm air Coupled longitudinal and lateral motions during the test e Very short dynamic response of the model due to a high inherent stability of the model 75 8 Identification from simulated data A simulated data has been generated using equations 3 24 and 3 27 to study the effectiveness of the various parameter identification techniques We divided the work into separate longitudinal and lateral derivatives identification In order to resemble the actual flight manoeuvre the control inputs used for the simulation are taken from the real flight test data The resulted responses were then analysed using several parameter identification techniques The work was also extended to study the effect of measurement noise and different input forms to the estimated parameters 8 1 Longitudinal identification The aircraft was excited by an elevator doublet as shown in Figure 8 1 The response of the model lasted about 6 seconds 1 2 3 4 5 6 secon ds Figure 8 1 Elevator deflection The results of the longitudinal identification using various identification techniques are summarised in Table 8 1 Also shown in the table are the characteristics of the Short Period Oscillation mode Note that the ICM does not give a measure of uncertainty for each estimated parameter The LR MLM and ICM have successfully identified the six longitudinal derivatives and the SPO characteristics of the model The frequency of the S
52. e ehe Ie E eR e e 76 FIGURE 8 2 ANGLE OF ATTACK RESPONSE USING REGRESSION ANALYSIS 78 FIGURE 8 3 PITCH RATE RESPONSE USING REGRESSION ANALYSIS ESTIMATED esee 79 FIGURE 8 4 ANGLE OF ATTACK RESPONSE USING MLM ANALYSIS eres 79 FIGURE 8 5 PITCH RATE RESPONSE USING MLM ANALYSIS ESTIMATED eere nennen 79 FIGURE 8 6 ANGLE OF ATTACK RESPONSE USING ICM ANALYSIS ESTIMATED eene 79 FIGURE 8 7 PITCH RATE RESPONSE USING ICM ANALYSIS 5 2 2 1 0 2 2 20 000000080 000000000001 80 FIGURE 8 8 RUDDER DEFLECTION eren FIGURE 8 9 AILERON DEFLECTION FIGURE 8 10 ROLL RATE RESPONSE USING REGRESSION ANALYSIS ESTIMATED 82 FIGURE 8 11 YAW RATE RESPONSE USING REGRESSION ANALYSIS ESTIMATED eere 83 FIGURE 8 12 SIDESLIP RESPONSE USING MLM ANALYSIS ESTIMATED essere eene nennen enne 83 FIGURE 8 13 ROLL RATE RESPONSE USING MLM ANALYSIS 2 02 2 0001000680 83 FIGURE 8 14 YAW RATE RESPONSE USING MLM ANALYSIS ESTIMATED eere nnns 83 FIGURE 8 15 SIDESLIP RESPONSE USING ICM ANALYSIS ESTIMATED esee nnns 84 FIGURE 8 16 ROLL RATE RESPONSE USING ICM ANALYSIS ESTIMATED
53. e neglected Airspeed measurement The airspeed value V is obtained indirectly from the measurement of static pressure Ps total pressure Pt air temperature Ts and is related as follows Yi 1 Y ps 1 2yRT where y 1 4 vary very little with humidity yis also insensitive to V value RER ay air 287 05 J Kg K The value of R depends on the dew point temperature Assuming a constant value of produces a speed uncertainty of AV 0 5 m s at temperature of 20 C Laban 1994 Airspeed sensitivity due to changes in air parameters are given as 1 AR AV V 2 T in Kelvin 1 AT AV 2 V 2 5 Suppose an error of 2 degrees in temperature measurement at an airfield say T 18 C to measure aircraft s velocity of 30 m s This temperature measurement error is equivalent to uncertainty in speed measurement of AV 0 1 m s In addition A 10 degree variation in due point which corresponds to AR 5 J Kg K produce uncertainty in speed measurement of AV 0 26 m s For low speed flight The actual calibration of the airspeed sensor is carried out in the wind tunnel A2 5 5 Control surface deflection measurement 1 Transducer error This error is obtained from the calibration Control surfaces errors deg full scale Right elevator Rudder _____ 034 06 Right aileron 2 Mechanical linkage error This error is caused by the elasticity lag and imperfection of the mechanic
54. econds Rudder deflection 99 deg s degrees degrees deg s deg s 100 150 seconds Sideslip matching o 3 4 5 6 i 8 9 10 seconds Residual in sideslip matching 150 100 50 50 3 4 6 8 9 10 seconds Roll rate matching 60 40 20 20 40 60 50 3 4 6 8 9 1 seconds Residual in roll rate matching seconds Yaw rate matching 100 deg s 30 20 10 0 40 20 30 4 5 6 7 8 9 10 seconds Residual in yaw rate matching Figure 9 1 Estimated lateral responses and their residuals from manoeuvre 3 records flight data estimated manoeuvre 4 1 estimated manoeuvre 4 2 5 0 5 10 15 20 25 10 5 11 11 5 12 12 5 13 seconds Rudder deflection degrees degrees seconds Right aileron deflection degrees seconds Rudder deflection 101 degrees deg s deg s deg s degrees 30 25 20 15 10 5 0 5 10 75 105 Ti 115 12 12 5 13 seconds Sideslip matching 8 6 4 2 0 2 4 40 5 Ti 115 12 12 5 18 seconds Residual in sideslip matching 120 100 seconds Roll rate matching seconds Residual in roll rate matching 40 20 10 5 1 TT 12 gt 1 seconds Yaw rate matching 102 15 10 deg s 10 15 20 25 30105 11 11 5 12 12 5 13 seconds Residual in yaw rate matching Figure 9 2 Estimated lateral responses and their residuals from manoeuv
55. ed response that shows little oscillatory behaviour References such as Ward 1993 and Eshelby 1991 deal with the practical application of these conventional method of dynamic flight testing The following two conventional methods are selected for this project since they are simple practical and easy to program in Matlab script language TPR Transient Peak Ratio method The process involved in the TPR method is depicted in Figure 3 1 below Read chart Flight Transient Peak Ratio TPR TPR vs damping E trace Damped period T Figure 3 1 Transient Peak Ratio method diagram 22 Where TPR LL 3 1 X X And 3 2 complete detail theory be found Ward 1993 from page 211 to 225 The method has been automated by the author using Matlab To execute the program simply type TPR at the Matlab prompt Figure 3 2 Transient Peak Ratio M t rx easurements 2 Curve Fitting This method is based on fitting a first or second order curve to the flight response Newton minimisation algorithm is used to minimise the error between the fitted curve and the flight response The first order system is given as y K Ke eni 3 3 The second order system is given as Ke cos amp J 3 4 The Matlab programs needed to perform these methods are contained in files ordel1 m orde12 m orde21 m and orde22 m Type 12 to perform a first order curve fitting or orde22 t
56. el The model was tested in the wind tunnel The lift drag and pitching moment were measured using strain gages However the result was not satisfactory Except for the lift significant scatters on the drag and pitching moment were apparent Unfortunately the half scale model crashed on the first flight trial The propeller hit the ground on take off After a few seconds of flight the pilot felt that the canard control was too sensitive and decided to cut off the engine However the aircraft became nose heavy and crashed to the ground To proceed with the project we purchased and assembled a Telemaster Precedent T240 aircraft model Figure 72 In approximately 2 Months the aircraft was ready to undergo its first flight testing Figure 7 1 The half scale MAFV Figure 7 2 The Telemaster Precedent T240 We also had a problem with engine to power the Telemaster T240 We originally used the RC 80 engine However getting this engine to work was a difficult task Eventually we decided to 72 purchase another engine Irvine 150 22cc This was a very good engine and proved easy to Start The Telemaster T240 crashed at the 13 flight due to an undetected flat battery The model suffered a major damage to its engine mounting amp cowling its right fuselage low directional vanes engine rpm rudder and elevator sensors We spent 3 weeks to rebuilt the model and recalibrate most of the sensors 73 7 4 P
57. ent error 1 Yo A og V 1 Q A measured 9 P 1 Where Wo Oo and are yaw pitch and roll misalignments Since the kinematics acceleration error is a systematic error it can therefore be minimised by locating the three accelerometers as close as possible to the centre gravity 2 Angular rate measurement A Transducer error The averages of 5 deg s transducer error for the rate gyros were obtained from the calibration on the rate table B Kinematics error The kinematics error due to misalignment is given as 1 Wo WV 1 Po measured 0 1 3 Airflow direction measurement From Laban 1994 page 216 the vane dynamics is given as shaft 6 5 1 G pV 5 CI 1 0 0 1 v shaft The lift curve slope can be approximated as A CI E From the above 2nd order approximation the damping and natural frequencies of the vane are 2 1 A simpler low frequency approximation to the vane dynamic can sometime be useful and is given in a lag time form as follows snap 1 t tT vane 0 5 SEL 5 Pimpact The and Bvanes for the half scale model has the following characteristics Aspectratio A 2 66 Area S 9 68 cm Arm length 1 5 2 Mass 3 84 grams Inertia I 15 3 gr cm ___ y Calculation at 15 m s gives a vane gom 4 84mm gt natural frequency of
58. ere J v and Thrust p n D Ct D Propeller diameter 0 39 meter n rotational speed in rev s The data is fitted with the standard deviation of 1 Newton 2 Centre of gravity determination The distance between the two support points d 1250mm Thickness mm equivalent theta Rm kg Rn Kg tan theta Rnd wcos as Result xcg 15 74cm from the datum 47 5 cm behind the nose zcg 25 79cm from the datum ie 14 cm above ref point 3 Moment of inertia determination om ly 0 14 m x Ly 0 44 10 3 kg 2 26 r 0 662 R_y 0 2475 m L 1 55 10 5 10 Average Period 904167 Average period 119444 period 1 945679 22 21 21 10 17 17 17 2 21 21 10 10 17 7 19 19 T Results Rolling inertia I 1 15 Ptiching inertia I 1 30 Yawing inertia Lj 1 28 A6 2 Appendix 7 Flight test procedures and the collected records Sample record of flight data Test name Longitudinal dynamic Input Manoeuvre Elevator doublet J gt Test model Telemaster T240 Flap setting deg Date 13 6 96 Approx Speed m s T O time 13 00 Landing time Manoeuvre flight no Filename test2 1 T O fuel Kg T O weight Kg 11 Cg x y z in cm 47 5 0 14 Inertia Kgm2 Ixx 1 15 Izz 1 28 Ixz 0 Ground temp deg C 16 Ground pressure mmHg 76 Apparatus checklist 1 Inclinometer 2 Scal
59. erivatives Espo A priori Estimated Cramer Rao A priori Estimated Cramer Rao parameter Bound parameter Bound 5608 Table 9 5 Estimated longitudinal parameter from recorded data manoeuvre 2 with two Matching flight data degrees degrees degrees different sets of a priori values flight data estimated manoeuvre 2 1 estimated manoeuvre 2 2 1 2 seconds Elevator deflection seconds Rudder deflection seconds Right aileron deflection 93 0 degrees seconds Angle of attack matching degrees seconds Residual in angle of attack matching deg sec seconds Pitch rate matching de 9 40 20 20 40 60 seconds Residual in pitch rate matching Figure 9 1 Estimated longitudinal responses and their residuals from manoeuvre 1 records flight data estimated manoeuvre 2 1 estimated manoeuvre 2 2 94 degrees degrees degrees degrees degrees seconds Elevator deflection 10 8 6 4 2 0 2 4 6 8 70 6 1 3 4 6 seconds Rudder deflection 20 15 10 5 0 5 40 15 20 4 1 2 3 5 6 7 seconds Right aileron deflection seconds Angle of attack matching 4 6 8 seconds Residual in angle of attack matching 95 4 seconds Pitch rate matching deg sec 40 30 20 10 0 40 20 30 0 2 4 6 8 seconds Residual in pitch rate matching Figure 9 2 Es
60. es 3 Spare vanes 4 Metering tape 5 Stopwatch 6 Temperature and pressure measuring devices 7 Laptop A7 1 Flight test procedure to perform dynamic manoeuvre Step no Description Click the gear control on the transmitter forward and then backward to check that the DAS records the data The DAS indicator should stop blinking 10 Close the file by pressing Alt L 11 Close the file by pressing Alt L 2 Repeat step 2 to 5 3 Taxi the aircraft into take off position on runway 14 Take off 5 Climb to altitude then perform a turn Prepare for doublet manoeuvre Maintain heading and wing level Click the gear control on the transmitter to start recording the flight data Throttle idle perform doublet manoeuvre throttle maximum 9 Climb out 0 Perform a turn and prepare for landing Landing then engine off 22 Repeat step 7 to 11 Lael 10 ____ uoo 13 ____ 14 e 215 16 17 ____ 17 18 jai N degrees degs degs degrees degrees degrees Records of flight data Also contained in the accompanying disc 0 5 70 g 7 seconds Elevator deflection b a 0 5 10 E 7 second Aileron right top and aileron left bottom deflection 8 6 4 2 o 2 4 5 EI 40 second Pitch rate Roll rate Yaw ate 4 of attack EP 7 3 degre
61. es degs degs degs degrees degrees degrees Figure A7 1 Flight_1 records Elevator HM Aileron EN top and left bottom Pitch rate Roll Yaw RM Ew A7 4 deg sec degrees degrees degrees deg sec Figure 7 1 Flight 2 records 5 7 17 g 7 Elevator deflection F E 17 T 7 secon Right aileron deflection B E T g 7 secon ds Rudder deflection 7 5 degrees degrees degrees degrees degrees 3 17 B secon Angle of attack 3 ri 0 E T Sideslip angle Figure 7 2 Flight 3 records 7 17 T 7 secon ds Elevator deflection 7 5 rj 17 E Right aileron deflection 3 5 ri 17 E T secon ds Rudder deflection 3 5 rj 17 E 14 secon gt Pitch rate 7 6 degisec degrees 888 3 ri 17 g ds Yaw rate 7 7 g E secon Angle of attack 7 7 g E seconds Sideslip angle Figure A7 3 Flight 4 records 7 7
62. es essential guidance in establishing these requirements 6 1 1 The data acquisition system The data acquisition system DAS in this project is divided into that on board Figure 6 1 and on ground Figure 6 2 The system was developed separately by the Department of Computer Systems Engineering at the RMIT Kneen 1994 35 Sensors Signal conditioner ENDO Data collection is triggered by Multiplexer he transmitter printer or flight data processing software Down loading the flight data is triggered by the transmitter Z channel 36 MHz Figure 6 2 On ground data system Figure 6 3 shows the block diagram for the on board DAS The system is based on an Intel 8031 microprocessor operating at 3 6864 MHz There are two 8 channel analog to digital converters plus one timer input channel 17 amp 18 The DAS samples every 4 mill seconds 25 Hz with 8 bit data resolution i e 256 counts for a full range data calibration A total of 256 Kbytes onboard memory allows up to 10 minutes of data acquisition for each flight 56 Data download button Inertia unit power Rate 9 vro e Sample button green e Dump button blue Microswitch Accelerometers Servo motor gear Control surfaces channel Figure 6 3 The on board data acquisition block diagram for the T240 flight test program Receiver Figure 6 4 The DAS card used in the flight test During flight the on
63. f attack and yaw vanes were conducted at the 3x2m Mechanical Engineering RMIT The whole model was mounted on a sting with an adjustable pivot for changing the angle of attack For one particular angle of attack setting the model was yawed from 25 deg to 25 deg by rotating the table on which the model was supported The procedure was repeated for several angle of attack settings 66 The speed correction due to blockage effect in the wind tunnel was carried out using a formula taken from Pope 1947 page 220 as follows Ib where Vol model volume length of side of tunnel parallel to wingspan height of tunnel The calculated blocking correction for this experiment turned out to be 0 0054 Control surfaces The control surface calibrations were conducted by deflecting the appropriate control surfaces while noting the output from the corresponding channels The control deflections were measured by a digital inclinometer which has an accuracy of 0 2 deg Pressure sensor Airspeed indicator Calibration of the airspeed indicator pitot static boom was conducted in the 50x50cm Aerospace Engineering wind tunnel RMIT The pitot static boom was removed from the aircraft and placed inside the wind tunnel for calibration An inclined manometer was used to measure the tunnel speeds Voltage outputs from the airspeed sensor were noted for several tunnel speeds and plotted to obtain the sensor calibration To
64. f the Dutch and spiral modes The LR MLM and ICM have successfully identified the fifteen lateral derivatives Dutch and Spiral mode characteristics of the model In the analysis the results from the LR are used as a priori values for the MLM and ICM Among the three techniques the MLM produces the best estimate of the 80 derivatives Table 8 4 shows that the MLM produces the smallest error criterion Ideally the estimated parameters from LR should give exactly similar values as the true parameters However the differentiation process of angular rate in the simulation has introduced errors in the estimation Algorithms Derivatives True Estimate Standard Estimate Cramer Estimate parameter d deviation d Rao d paramete paramete Bound paramete 0 349 0 0080 0 126 00529 Cy 0153 042 00059 027 00902 Cy 009 0172 00015 0343 00013 0041 0 0049 0 043 0 038 0 0009 0 045 0 0012 0 038 0 751 0 0017 0291 0 0131 0 006 0 0007 0330 0 0027 0 003 0 0001 0 065 0 0019 0 095 0 0014 0 044 0 0002 Cm 0 006 00013 0 008 0 0008 __ 5 138 1422 __ 13 052 060 Table 8 3 Results using various estimation algorithms LR 0 0040 00005 02445 standard deviation 3 0 6698 0 0088 0 4815 0 0612 MLM ICM 00134 __ 00195
65. flight control system for the aircraft The project has shown that the dynamic of a model aircraft can be estimated with a reasonable confidence using flight testing procedure Nomenclature EP g DE w Ch Ch Clauader Cmggtevator Cng Cn Cngaiterons Cy Cyaudder CZiglevator L L Fx puce edge experiment ZBB 2 3 3 angle of attack rate rad s initial pitch rate rad s roll and pitch accelerations rad s sideslip rate roll rate and pitch rate rad s air density kg m angle of attack rad control surface deflection initial yaw angle rad pitch angle rad roll angle rad sideslip angle rad roll pitch and yaw acceleration rad s acceleration m s system matrices wingspan m wing chord m non dimensional roll derivatives non dimensional pitch derivatives non dimensional yaw derivatives thrust coefficient thrust p n D non dimensional side force derivatives non dimensional lift derivatives propeller diameter m gravity constant 9 81 kg m moment of inertia advance ratio nV D aircraft length length of the string meter vertical distance between cg and pivot point in bifilar suspension experiment vertical distance between cg and pivot point in
66. for the engine test 46 Figure 5 2 Thrust measurement in the 50x50cm Aerospace Engineering wind tunnel RMIT Figure 5 2 shows the testing of the RC 80 engine in the wind tunnel However since we encountered many problems with the RC 80 in time of T2405 first flight testing we just had to change the engine to Irvine 150 Though the engine test result could still be used since we utilised the same type of propeller The result of the engine test is given in Figure 5 3 and Figure 5 4 Figure 5 3 shows that the thrust coefficient is linearly related to the advance ratio J The graph covers most of the advance ratio operating range for the actual flight From this graph a good linear model can be extracted Figure 5 4 shows a good agreement between the experimental results and those predicted by the model 47 Thrust coefficient Ct mode 0 0 5 Advance ratio J Figure 5 3 Thrust coefficient to advance ratio relationship for the propeller model Thrust chart Experiment fitted line Speed m s Figure 5 4 Comparison of the thrust chart from the experiment and the derived thrust model Comments on the result An adequate thrust model has been derived from the experiment i e Ct 0 065 0 089J with the standard deviation of the fitted line to the experiment data of 1 Newton error of 396 48 Some of the possible sources of errors during the experiment were e The unsteady thrust
67. iables The following assumptions are used when using the linear regression method 1 X is deterministic no noise 1 is uncorrelated with X iii is identically distributed and uncorrelated with zero mean and variance O i e white noise 27 3 2 2 Maximum likelihood method A linear dynamic model of an aircraft can be given in a state space form as x t Ax t Bu t ud 3 16 z t Cx t Du t Gn t Where x t State at time t Dynamic matrix z t Measurement at time t Control distribution matrix u t Input at time t State measurement matrix Transmission matrix T Square root of the state noise spectral density FF Square root of measurement noise covariance matrix n t Yt 2 Gaussian noise gt The maximum likelihood estimator maximises the conditional probability density function of the output given the set of parameter i e maximising P y P y 9 is normally given in logarithmic form and known as the logarithmic likelihood function LLF 0 N Nm RR Z log RR log gt 3 17 Where RR E Z 27 To minimise the likelihood function above a Marquart Constrained Newton or other minimisation technique can then be used to predict the successive estimate of the unknown parameters The detail computational technique used in this project is described in chapter 4 Suppose the parameter
68. ions and the measured inputs and system responses during flight manoeuvres In contrast to conventional estimation the PI technique provides for reduced test time more flexibility in manoeuvre requirements and more parameters including those unobtainable using conventional techniques are obtained from a single manoeuvre The significance aspects of the project are First the obtained derivatives will be used in the design of an autonomous flight control system The design of the control system 1s currently carried out by another post graduate student Valentinis 1996 Second the project will assess the capability of the flight test instrumentation systems designed in collaboration with the Computer System Department at RMIT Kneen 1994 Third this project will provide a statistical stability and control derivative data base extracted from flight test measurements which will extend the confidence in existing stability and control derivative estimation techniques when applied to UAV s Unmanned Air Vehicles and other small flight vehicles There are many potential benefits in using UAV as aerial platforms for either commercial or research applications UAVs have a low operating cost as compared to manned aircraft operations UAVs can perform hazardous tasks such as close monitoring of fires hurricane tracking observation of radiation contaminated areas and volcano eruptions UAVs are suited to long endurance tasks that 6 are generally
69. is used in the calculation The equation to calculate the moment of inertia is given as wolowicz 1974 _ T MgR For the yaw mode Aj 5 9 51 and for the roll mode 1 47 Figure 5 9 Experimental set up to determine and roll inertias 52 Figure 5 10 Yaw and Roll moment of inertias determination using bifilar suspension method Table 5 2 shows the results of the inertia experiments Mode Period sec Calculated Radius of Non inertia Kgm2 gyration R m dimensional R Roll 1 Table 5 2 Results of moment inertia experiments 5 4 Theoretical Stability and Control Derivative Estimation Theoretical stability and control derivatives estimation was conducted for the following purposes To construct simulated flight test data and analyse the effectiveness of the various parameter identification methods prepared in this project To provide a priori information for the Maximum Likelihood and Interactive Curve Matching methods To compare with the derivatives estimated from flight test data Two different theoretical methods were used to estimate the stability and control derivatives of the T240 model 53 1 The AAA Advanced Aircraft Analysis version 1 7 software program DARcorporation 1996 This program is based on the theory given in the book written by Roskam 1985 The software provides a user friendly iterative calculations of stability and control derivat
70. ises were present This would certainly not the case had we used the telemetry system Coleman 1981 found a significant noise in his flight data obtained from the telemetry system The only significant noises contaminating the recorded flight data were from the engine vibration and air turbulence A soft damper wrapped around the IMU unit would certainly reduce the vibration noise The turbulence noise can only be reduced by flying the aircraft in a calm air Other major problem with the system was that of transmitter signal interference as described in chapter 7 We spent months trying to reduce this interference In the end changing the transmitter frequency from 36 MHz to 29 725 MHz solved the problem The sampling rate 25 HZ resolutions and accuracy of the sensors were adequate for dynamic flight testing However for a better result a resolution of 12 bit could be used in which case the resolution would be increased by 16 times 109 A sixteen seconds of data acquisition has proved to be sufficient for recording two different manoeuvres A memory devices with 256 Kbytes correspond to 30x16 seconds of data acquisition would record 60 different manoeuvres in one flight This would certainly make the dynamic flight testing process quicker and less expensive There is a huge potential in using the already developed system for other research in model flight testings For example by adding three axes linear accelerometers to
71. ivatives of their STABILEYE RPV Coleman 1981 In Australia Sydney University has developed a series of small RPV for aerodynamic research Wong 1989 and Newman 1995 Present and future research in this field concentrates on 3 different key areas First the development in the instrumentation systems Hamory 1994 second the development of system modelling and various estimation techniques for the extraction of the derivatives Iliff 1989 A recent research topic in the estimation technique is in the application of computational neural networks to identify several aerodynamic derivatives Linse 1993 and third the search for an optimal input design and a more practical flight test manoeuvres Plaetschke 1979 The following sections survey the above three key areas namely instrumentation flight data analysis and input forms 2 1 Instrumentation The flight test instrumentation includes sensors data acquisition system DAS and Telemetry systems With the present technology it is possible to have flight test instrumentation system that is small and light Most components are commercially available for model aircraft hobbiers to construct their models These components have been used by the University of Sydney Wong 1989 and NASA Hamory 1994 2 1 1 Sensors Parameters to be measured in flight can be categorised in to two groups inertial or dynamic data and air data Typical sensors needed to extract stabilit
72. ive of any aircraft In addition a data base approach of the software allows the user to use common sets of aircraft parameters when the parameters of the calculated aircraft are not yet available The estimation of the T240 stability and control derivatives was partly undertaken by an undergraduate student working on his final year project Chow 1996 2 A computer program written by the author based on the theory given in Smetana 1984 The theory has been proved successful in estimating the derivatives of several conventional subsonic light aircraft 54 6 Data Acquisition and Instrumentation Systems 6 1 Description And Specification The data acquisition and instrumentation systems required to collect flight data depend on several factors such as test objectives method of analysis and hardware limitations The system for extracting aircraft stability and control parameters may have different requirements than those for performance testing In the former testing for example thrust and longitudinal acceleration measurements can be of secondary important when a Maximum Likelihood method is used In contrast the thrust and longitudinal acceleration measurements are critical in performance testing In this project the data acquisition and instrumentation requirement is established by looking at other similar research in model flight testing Coleman 1981 Wong 1989 Hamory 1994 and Budd 1993 Also NASA 1168 Maine 1986 provid
73. ki E G 1974 Identification Of Non Linear Aerodynamic Stability And Control Parameters At High Angle Of Attack Agard Conference Proceeding AGARD FMP Specialist meeting NASA Langley Nov 1974 paper 1 2 11 Gracey W 1948 The Experimental Determination of the Moments of Inertia of Airplanes by a Simplified Compound Pendulum Method NACA TN 1629 12 Hamory P J Murray J E 1994 Flight Experience With Light Weight Low Power Miniaturised Instrumentation Systems Journal of Aircraft Vol 31 No 5 Sept Oct 1994 p 1016 1021 13 Hodge W F Bryant W H 1975 Monte Carlo Analysis Of Inaccuracies In Estimated Aircraft Parameters Caused By Unmodelled Flight Instrumentation Errors NASA TN D 7712 Feb 1975 14 Holcomb M L Tumlinson R R 1977 Evaluation Of A Radio Control Model For Spin Simulation Society Of Automotive Engineers Business Aircraft Meeting Century II Wichita March 29 April 1 15 Howard R M TrainerJ C Lyons D F 1991 Flight Test Of A Half Scale Unmanned Air Vehicle Journal Of Aircraft Vol 28 no 12 December 16 Howell S and Williams B 1994 Stealmouth Telemetry Systems RMIT Engineering Conference Journal ENGenius 04 p 43 46 17 Iliff 1989 Parameter Estimation For Flight Vehicles Joumal of Guidance Control and Dynamics vol 12 no 5 Sept Oct 1989 18 K W Maine 1979 Practical Aspects Of Using A Maximum Likelihood Estimation Methods To Extract Stability And Contr
74. l the flight test data For the MLM analysis the a priori values for each parameter are obtained either from theoretical method calculated from AAA software or earlier flight data analysis 9 3 1 Longitudinal stability and control derivatives estimation The estimated longitudinal stability and control derivatives are given in Table 9 4 and Table 9 5 and the matchings of flight data are presented in Figure 9 1 and Figure 9 2 Derivatives A priori Estimated Cramer Rao A priori Estimated Cramer Rao parameter Bound parameter Bound p mu 5 ee ee comp 8 6 1 BEL NAE NEN Table 9 4 Estimated longitudinal parameter from recorded data manoeuvre 1 with two different sets of a priori values 91 Table 9 4 shows the estimation results from manoeuvre 1 data using two different sets of values In the first set analysis 1 1 we used the a priori based on the theoretical work AAA software However the first attempt to estimate all the six longitudinal derivatives simultaneously has failed The maximum Likelihood Method MLM did not converge into solutions From the simulation study chapter 8 we found that the Cz was weakly identified and hence should be kept fixed during the identification process In the second attempt we fix both Cz and Cz to these a priori values The value of Cz was also available with quite a reliable accuracy The MLM then converged to solutions in 20 iterati
75. lage length More detailed characteristics of the model are given in appendix 3 Figure 5 1 The Telemaster T240 aircraft model to be flight tested The model weighs about 10Kg of which 60 constitutes the structural weight Table 5 1 shows the complete weight breakdown of the model Table 5 1 Weight breakdown of the T240 aircraft model Components Mass gram of total weight Total The whole vehicle is constructed from commercial home built components The structure of the T240 is balsa wood covered with composite skin The main wing structure consists of a single plywood spar strengthened by several balsa wood ribs along the wing span The vehicle is powered by a small 22cc aeromodelling glow plug engine Irvine 150 A two bladed fixed pitch propeller 16 diameter and 8 pitch is used With this engine and propeller combination approximately 15N thrust can be produced for cruise at engine speed of 7500rpm This was measured in the wind tunnel as can be seen in Figure 5 4 Three main control surfaces elevator rudder and aileron are used to control the aircraft A flap 15 also added as to generate more lift if needed The control surfaces are driven by electrically servo actuators All these servos are controlled by Futaba RC Max 7 system which uses PCM encoding at frequency of 36 MHz However at a later stage in flight testing the PCM transmitter was replaced by a TF FM at 29 725 MHz due to interference The control
76. light number Description Length of data Filename 16 00 seconds Elevator and aileron rudder manoeuvre 12 64 seconds Elevator and aileron rudder manoeuvre 12 76 seconds Table 9 1 Flight description 16 00 seconds From the above four flights four sections of manoeuvre data were analysed successfully The manoeuvres are described in Table 9 2 All the filenames for these manoeuvres are saved in the accompanying disc in subdirectory c data Manoeuvre no Taken from Description Length of data Filename flight no analysed manoeuvre manoeuvre aileron manoeuvre manoeuvre Table 9 2 Manoeuvre description 89 Data from flight number 1 and 2 could not be analysed This was due to the presence of a significant vibration noise in the angular rate measurements and turbulence noise in the angle of attack and sideslip See records of flight 1 and 2 in appendix 7 The relevant flight test conditions and flight configurations are summarised in the Table 9 3 and all the collected flight data are presented in appendix 7 Airspeed m s 15 Approximate reference altitude m 30 5 Manoeuvre number Flight parameters 1 Dor T 30 Hlap setting deg CANNE 5 leading edge Table 9 3 Flight test conditions for every manoeuvre 9 2 Data pre processing Before proceeding with the estimation of stability amp control derivatives the following data pre processing was carried out
77. m of residual covariance a fixed value as the cost function minimum Small value of Cramer Rao Bounds 2fcramer and insensitivities Large values of these variables indicate a poor information content in the data to identify a particular parameter 1 insensitive to parameter These parameters should then be fixed or supplied with information from wind tunnel or previous flight test data e scatter of parameter estimates from repeated experiments 15 approximately 1 2 times the filtered CUR filtered Cramer Rao Bound where CR 40 4 2 Data Compatibility Analysis Flight Data Reconstruction Data compatibility analysis to the measured outputs is becoming an important procedure prior to processing flight dynamic test data The analysis gives estimates to any unmeasured variables acts as a state estimator and also estimates any biased errors in the measured response data Papers written by Wingrove 1973 and Klein 1977 present several methods in conducting the compatibility analysis The proposed compatibility checking in this project is described in the Figure 4 1 below Minimization techniques Estimated states uvw eov XYZ uvwO wyxyz Figure 4 1 Compatibility checking algorithm used in this project The complete kinematics equation is given as a qw rv gsinO v a ru 4 1 z gcos
78. mation respectively practical Figure 8 1 The three different input forms used in the simulation The results from the simulation suggested that there seems to be no significant different in the estimated parameters under this flight condition However the CRB values obtained from practical input form were generally higher and hence more uncertainty in the results 87 Input forms 3211 True Estimated Standard Estimated Standard Estimated Standard parameter parameter deviation parameter deviation parameter deviation Derivatives Guo ad 6 677 65 6 eec E 40366 E Table 8 9 The effect of different input forms to the estimated longitudinal parameters using linear regression algorithm Input forms 3211 Derivatives True Estimated Cramer Estimated Cramer Estimated Cramer parameter parameter Rao Bound parameter Rao Bound parameter Rao Bound Gygo rad s 6 687 685 685 ___ SS aaa ee Table 8 10 The effect of different input forms to the estimated longitudinal parameters using maximum likelihood algorithm 88 9 Flight Test Results 9 1 Flight data The flight test was conducted at the Weribee flying field Melbourne Four flight sets of data were gathered The complete recorded flight data are given in appendix 7 Table 9 1 lists the description of the flights F
79. minimize pressure errors caused by the boom installation on the wing the boom length was designed to be at least four times the wing thickness Gracey 1981 Hence no pressure error was considered in this project except the kinematics position error due to offsets from the aircraft s center of gravity see appendix 2 67 From the calibration the obtained sensor characteristics are summarized in Figure 6 3 below Figure 6 3 Results of the sensor calibrations 216 1 0005 deg s 170 to 170 1 4 deg s 0 27 deg s Rate gyro 2 chn 3 1 1977 deg s 170 to 170 1 5 deg s 0 30 deg s Rate gyro 1 chn 2 1 4283X 216 88 Accelerometers ae Rate gyro 3 chn 4 3 9375 deg s 170 to 170 1 3 deg s 1 14 deg s Angle of attack vane 0 6745 deg 0 4 deg chn 12 0 66 Yaw vane chn 11 0 41327X 57 99 0 7515 deg 0 4 deg 0 74 Elevator chn 15 1 0155x10 0 2096 deg 0 7905x10 30 to 11 deg 0 25 deg 5X3 3 1717x10 0 5 0 3346 deg 1 5258 x 10 30 to 30 deg 0 25 deg 0 6 2 2612 106 9 3349x10 X Pressure inc water 2 40867x10 Speed m s Channels 8 and 9 are spares Rudder chn 16 Right aileron chn 14 3X745 9756x107X 33 208 6 1266x10 6 342 4028 103 X249 3398x10 X 0 5094 deg 5 9164 x 10 20 to 21 deg 0 25 deg 1 2 7 5499x10 X 21 917 2 767 10 0 1747 deg 1 790 x 10 13 to 17 deg 0 25 deg 3741 3356x10 X7 0 58 4 477x10 X 18 1
80. n With a typical payload of 20 Kg the aim is to achieve mission endurance ranging from 3 hours at 60 000ft to 5 days at 7 500ft The design of the will be in close co operation with CSIRO to accommodate their mission requirements One of their specific missions 1s to measure the atmospheric abundance of CO and its stable isotopes The Wackett Centre 1995 In its development stage a Telemaster T240 model aircraft has been purchased and assembled for use as an electronic test bed for the full scale MAFV The model will perform several flight trials for dynamic flight testing and autonomous flight testing This project deals with the dynamic flight testing of the T240 model aircraft to obtain the stability and control derivatives of the vehicle The specific objectives of this project are To provide stability and control derivative values for the aircraft model To determine the necessary measurements and flight manoeuvre required in estimating the stability derivatives To prepare the instrumentation and data acquisition system e To determine inertial characteristics mass centre of gravity and inertia To select an appropriate model structure and parameter identification algorithms To develop a computer program to extract stability and control derivatives from recorded flight test data To determine the accuracy or confidence of the parameters obtained The project has several limitati
81. ne 150 Span cm 43 5 Fuselage width cm Taper ratio C g to wing a c chordwise in cm 46 Xcg Inboard station half span LEN C g to wing a c vertical in cm 14 7 Hub diameter mm Outboard station half span 44 7 C g to thrust axis cm THE Nose to wing quarter chord cm Nose to tail quarter chord cm Wing to tail quarter chord cm Vertical distance from wing to tail cm ie A3 2 APPENDIX 4 FLIGHT TEST SENSOR CALIBRATIONS Results of the flight test sensor calibrations are presented in figures A4 1 to A4 11 Each figure contains 2 different graphs the top graph shows the experimental result and its fitted curve the bottom graph shows the corresponding calibration error Channel allocations and calibration results are shown below Chann Sensors Calibration Standard deviation full scale ares ene 1 5342 234 38 Accelerometer Accelerometer 7 14 18903X 354455 1720 120 6 92667 10 1 i 1890 3X 3544 6 95 J 1 57 1 1 Left aileron 2 767 10 3 1 3356 10 2 14 Right aileron 2 2612x10 5x2 9 3349x 10 X 7 5499x10 X 21 917 l4 Flap 2 5549x10 3 5 9816 10 0 4075 1 02 2 6412 10 4 9 Elevator 1 0155 105 3 3 1717 10 Be X 45 9756x10 X 33 208 6 1266x10 6 3 2 4028 103 1 5258 x 10 X249 3398x10 X 45 87 speed 0 1 2 3 Yaw vane 0 41327X 57 99 07515074 0962x10 A4 1 fitted curve fitted curve 50 15
82. nel Testing John Wiley amp Sons New York 38 Raisinghani Singh Jatinder S C 1993 Aileron And Sideslip Induced Unsteady Aerodynamic Modelling For Lateral Parameter Identification Journal of aircraft vol 30 no 4 July Aug 39 Reed R Dale 1974 RPRV S The First And Future Flights Astronaut amp Aeronaut Vol 12 No 4 April 1974 113 40 Roskam J 1985 Component Weight And Estimation Airplane Design Part V Roskam Aviation and Engineering Coorp 41 Roskam J 1987 Aiplane Design Part VI Preliminary Calculation Aerodynamics Thrust and Power Characteristics Roskam Aviation and Engineering Ottawa Kansas 42 Ross J A 1979 Application Of Parameter Identification Techniques To Analysis Of Flight Data Progress in Aerospace Science Vol 18 p 325 350 43 Smetana F O 1984 Computer Assisted Analysis of Aircraft Performance Stability and Control McGeaw Hill Book Court N Y ISBN 0 07 058441 9 44 Sofyan E Bil C Danaher R 1996 Aircraft Model Flight Test for Parameter Identification Proceeding of the 2nd ISASTI Intemational Symposium on Aeronautical Science and Technology in Indonesia 27 July 1996 Jakarta Indonesia vol 1 page 118 129 45 Sofyan E Danaher R Thompson L Bil C 1995 A Half Scale MAFV Multi Purpose Autonomous Flight Vehicle Flight Test Program Proceeding of the 7th IASE The Indonesian Aerospace Students in Europe 12 14 July 1995 Manchester England
83. o perform a second order curve fitting 23 3 2 Parameter identification techniques Three different parameter identification methods The Linear Regression LR Maximum Likelihood ML and Interactive Curve Matching ICM are selected for identifying the aerodynamic stability and control parameters from flight data The ML method is the main algorithm in this project whereas the and ICM are complements The parameters obtained from the LR and ICM analyses are used as initial estimates for the Maximum Likelihood ICM Maximum Likelihood Derivatives Hand calculation Previous flight tests Figure 3 1 Relationship among the different techniques used in this project Linear regression analysis treats the aircraft equation of motion separately see equation 3 19 and 3 20 The parameter estimates are obtained by minimising the error cost function for that particular equation However when the regressors independent variables are contaminated with measurement noise the method produces a biased estimate of parameters In contrast to LR the ML method minimises a combined cost function of several equations The method produces an asymptotically unbiased efficient and consistent estimate of parameters The method is more complex than the regression Also a good initial estimate of parameters is required when extracting parameters from poorly excited responses in the flight data Iliff 1989 24 In contra
84. of extracting derivatives from flight data This a priori information may be derived from several sources such as hand calculation pure theory or semi empirical wind tunnel testing computational fluid dynamic or other independent flight tests In this project only the hand calculation performed using Advance Aircraft Analysis AAA V 1 7 software program and results from previous flight tests are used as a priori information for the subsequent analysis Nm Figure 3 2 Flight dynamic test activities As for comparison to the parameter identification techniques several existing conventional techniques have also been automated Chapter 3 1 describes briefly the theory behind these selected conventional techniques 21 3 1 Conventional Methods There are several existing conventional methods to analyse dynamic flight data such as TPR Transient Peak Ratio MTPR Modified Transient Peak Ratio TR Transient ratio MS Maximum Slopes and SRR Separated Real Roots of these methods are based on extracting dynamic characteristics such as damping ratio and natural frequency from the recorded system responses For example one can extract the natural frequency and damping ratio of a short period mode from a recorded pitch rate Similarly the Spiral and Dutch characteristics can be estimated from the recorded yaw rate One main difficulty when using these methods is that it is sometimes difficult to analyse data from a well damped record
85. of motions are preserved e The manoeuvre should be performed on smooth air i e no turbulence present Turbulence can introduce modelling errors since no turbulence model is incorporated in the flight data processing software e The manoeuvres are best performed at engine idle thus minimising any effect of the engine loads and vibrations increase the statistical confidence of the parameter estimates every manoeuvre should be repeated at least twice 35 4 Flight Test Software Development To process and analyse data from the flight tests a computer program has been developed specifically for this project The program must perform the following tasks Dynamic simulation of the model aircraft Signal processing of the flight data Graphical representation of set of data Identification of stability and control derivatives State estimation of unmeasured variables Flight reconstruction The MATLAB software has been selected since it has several beneficial features such as A powerful computing capability A good graphic capabilities Graphical User Interface GUI capabilities Many built in functions Relatively easy to program in the form of script M files A Personal Computer version is available Simulation program is supported SIMULINK It has a special toolbox for Maximum Likelihood Algorithm The Fortran version of this program MMLE3 is normally used in aircraft industry
86. ol Derivatives From Flight Data NASA TN D8209 April 1979 112 19 Iliff K W Maine R E Montgomery T D 1979 Important Factors The Maximum Likelihood Analysis Of Flight Test Manoeuvres NASA TP 1459 April 1979 20 Iliff K W Maine R E Shafer M 1976 Subsonic Stability And Control Derivatives For An Unpowered Remotely Piloted 3 8 Scale F 15 Airlane Model Obtained From Flight Test NASA TN D 8136 21 Klein V and Williams D A 1973 On Some Problems Related To The Identification Of Aircraft Parameters Identification amp System estimation Proceeding 3rd IFAC Symposium 22 Klein V 1975 On The Adequate Model For Aircraft Parameter Estimation Cranfield Report Aero No 28 Cranfield Institute of Technology March 1975 23 Klein V Schiess J R 1977 Compatibility Check Of Measured Aircraft Responses Using Kinematic Equations And Extended Kalman Filter NASA TN D 8514 24 Kneen J 1994 Avionics Projects at RMIT AOPA Aircraft Owners and Pilots Association of Australia Vol 47 No 8 August 1994 p 45 48 25 Laban M 1994 On Line Aircraft Aerodynamic Model Identification ISBN 90 6275 987 4 26 Linse D J Stengel 1993 Identification of Aerodynamic Coefficients Using Computational neural Networks Journal of Guidance Control and Dynamics v 16 0 6 Nov Dec 1993 27 Maine R E 1981 Programmer s Manual for MMLE3 A General Fortran Program for Maximum Likelihood Parameter Estima
87. on If the required range can not be achieved the resistors in the DAS circuit can be changed to alter the sensitivity One channel is organized to handle voltage input This channel will indicate 000 with no input applied and 255 with the maximum As for the potentiometer input the sensitivity of this input can also be altered by changing the resistor in the DAS circuit The timer input is used to measure the time between input pulses In practice these pulses will be obtained from a hall effect switch which measures the rotational speed of the aircraft s propeller The timer provides two sets of outputs The full result is obtained by combining 256 x first reading second reading These readings will indicate the propeller rotational speed and is obtained from the 58 calibration In contrast to potentiometer and voltage inputs the sensitivity of the timer can not be easily changed 6 1 2 Instrumentation systems There are 14 sensors used to measure inertia and air data during flight maneuvers The characteristics of these sensors are listed in appendix 1 Most of the sensors are sufficiently accurate and commercially available at a relatively low cost The rate piezo gyro for example is the hobby type normally used in helicopter models and has an acceptable linearity range up to 720 deg s 59 Rudder deflectidBi Left amp aileron deflection Elevator deflection Right flap amp aileron deflection m
88. ons Further iteration did not change the values of the estimated derivatives The maximum gradient of 0 0 was achieved with the minimum logarithmic value of 398 33 In the second set analysis 1 2 we used values which were obtained from estimating the derivatives one at a time We first estimate Cm by fixing all other derivatives constant Then the estimated Cmy was used as a priori for the next estimation and tries to estimate Cm while fixing the other derivatives constant The process was repeated until all the derivatives were estimated It should be noted here that this approach is very much dependant on the accuracy of those parameters held fixed It is however one alternative way to get the MLM converge into a solution Looking at the two sets of result in Table 9 4 the analysis 1 1 produced a smaller CRB Cramer Rao Bound for each parameter than those in analysis 1 2 Hence we can place more confidence in the analysis 1 1 results than those of analysis 1 2 However the two sets produced almost similar and q responses as shown in Figure 9 1 Table 9 5 shows the estimation results from the manoeuvre 2 The same process as in manoeuvre 1 was performed to arrive to the shown results Figure 9 2 shows the estimated responses The fit was reasonable good except for the pitch rate matching The poor pitch rate matching might be caused by an unintentional aileron input during this manoeuvre as shown in Figure 9 2 92 D
89. ons that include e Only dominant linear stability and control derivatives are to be estimated No coupling between longitudinal and lateral modes are considered e Limited accuracy and number of sensors are available e Limited time and budget for conducting the experiments Parts of the thesis have been presented at the 95 Sofyan 1995 ISASTT96 Sofyan 1996 seminars The content of the thesis is divided into 3 major sections The first section provides an introduction to the project chapter 1 literature review chapter 2 and method of flight testing chapter 3 The next section addresses the works undertaken prior to the actual flight test chapter 4 to 6 and some hardware problems encountered during the course of this project chapter 7 The last section presents the simulation and flight test results chapter 8 and 9 followed by discussion and conclusion All the raw data from the pre flight flight and post flight are collected in the appendices and computer files A computer disc that is included with the thesis contains a number of Matlab script programs necessary to process the flight data 2 Literature review In the past the role of model aircraft in dynamic flight testing was not so popular The instrumentation was either too heavy or too large to be housed in the RPV Reed 1974 Also the technology in the off the shelf aircraft modelling was not as advanced as today Now however an inexpen
90. port at the Department of Aerospace Engineering and Computer System Engineering RMIT November 1995 114 APPENDIX 1 SENSOR CHARACTERISTICS USED IN THE TELEMASTER T240 FLIGHT TEST PROGRAM No Quantity measured Transducer Static sensitivity Resolution Rms measurement error of full range 1 Longitudinal acceleration Accelerometer Setra systems model 0 05g pore uam DENEN KG RM 2 Lateral acceleration Accelerometer Setra systems model 0 05g ee ee eee a 141 Pitching velocity Rate piezo gyro NE J 1000 250 s Yawing velocity Rate piezo gyro NE J 1000 250 s Rolling velocity Rate piezo gyro NE J 1000 2508 __ 250 08 5 j Angle of attack Flow vane potensio type Murata 50 to 60 441 0 4 0 7 LPO6MG3RIHA LPO6MG3RIHA pw p om type RS 173 574 calibration P Wa ovo type RS 173 574 calibration type RS 173 574 calibration 173 574 calibration MEUM SCCOSDN 10 inc H20 Engine rotational speed Hall effect IC Switch RC 307 446 02500 pm 1 J No Quantity measured Transducer Max applied Normally applied Resistance Zero offset Others voltage or voltage or current current 141 350 Hz 141 350 Hz 141 350 Hz 4 8 6V 80 mAh dynamic range 0 720 s Yawing velocity Rate piezo gyro NE J 1000 4 8 6V 80 mAh m Rolling velocity Rate piezo gyro NE J 1000 4 8 6V 80 mAh EMEN J Angle of attack Flow vane potensio type
91. qa eR ERR as 11 12 2 1 2 Data dcquisitiOn System diode m CRIME e e EEE Ee RO drehen e 14 21 3 Telemetry System erae RD DIOS OO IE eR RT e REA OR 15 2 2 BBIGHT DATA ANALYSIS inscr rr rte RR EU HER BI 15 2 21 MO TIAE 15 2 2 2 Parameter estimation amp o rennen treten entente 16 2 3 INPUT FORMS NO ARR EE AE NE NTEGER EE este 18 S OVERVIEW OE THE METHOD iens repete esesadesbsasesnessuce tosusescensesaduaysoonsasooapesscnonsecees E SESE EES 20 3 1 CONVENTIONAL METHODS eis eae inea aetas pestes ta sin 3 2 PARAMETER IDENTIFICATION TECHNIQUES ccccccssssssesecsecsecssesscsecceseescssscessssesseseecescescssesaesaecaecaecaesaesaeseseesensersenseass 3 2 1 Linear regression 3 2 2 Maximum likelihood method 3 2 3 Interactive Curve Matching 3 3 MODEL 6 3 4 FLIGHT TEST MANOEUVRES ccccsssessescssescsscssescsscssescssesesecsesscnsssssecaeseseessesesaessessscassssaesassesassessesacsessessesesacseesesscaseaeseeeseaes 4 FLIGHT TEST SOFTWARE DEVELOPMENT 4 1 THE MMLE3 STATE SPACE IDENTIFICATION TOOL BOX MATLAB
92. r in the time tagging degrades the estimation process This error should be less than 10 msec Hodge 1975 in his paper pointed out that the worst inaccuracy in the estimated parameters is found when there is a time shift in the control surface measurements 2 Aliasing and prefiltering The antialising and prefiltering should be performed before sampling for example by using a 40 Nyquist frequency filter 3 Sample rate Normally the data are sampled at 100 200Hz Then the data 15 filtered out and thinned to 25 50Hz for post flight data analysis However in a radio controlled model flight test a sampling rate of 25 60 Hz is commonly used Coleman 1981 Wong 1989 and Yip 1992 4 Resolution Butter 1976 pointed out that the dominant factor effecting the errors in the estimated derivatives is the control surface deflection errors Hence the resolution of the control surfaces should be as good as possible typically 1 100 1 200 of the full scale 2 1 3 Telemetry system There are a number of telemetry systems available such as FM AM PCW PCM etc However the pulse coded modulation PCM is the most frequently used in the flight test program Iliff 1976 Colemann 1981 and Wong 1989 used PCM telemetry system Remtron RTS 1 system is one of the commercially available PCM typed telemetry systems This system is the one that the Computer System Department at the Royal Melbourne Institute of Technology is developing
93. re 4 records Mean Standard deviation Roll rate deg s Yaw rate deg s Table 9 9 Residual characteristics of the estimated lateral responses 103 10 Discussion 10 1 Estimated aircraft dynamics The project has estimated 6 longitudinal and 15 lateral derivatives from 4 flight manoeuvres data Only records of control inputs and vehicles responses were used in the analysis The results are summarised in Table 10 1 and Table 10 2 Cny Cimgelevators Cng and Cigrudder are strongly identified whereas Cz Cys Cyp and Cy are weakly identified The rest are moderately identified AAA Analysis 1 1 Analysis 1 2 Analysis 2 1 Analysis 2 2 _ 439 __ e 5851 0 _ _ _ _ 1103 7 742 1 894 9 6820 168 14 18 4 66 Table 10 1 Estimated longitudinal derivatives of the Telemaster T240 AAA Analysis 3 1 Analysis 3 2 Analysis 4 1 Analysis 4 2 Ch O 19810 1478 0 120 0 0123 0945 fixed 0 945 fixed Vg Yp yr B l ng 5 0 21 3012fixed 0 037 0 0288 0 043 0 121 fixed 0 012 0 0463 0 034 0 176 0 221 fixed 0 733 0 125 0 0025 2 895 0 4088 0 165 0 165 fixed 0 096 0 114 fixed 2 788 0 9158 0 038 fixed 0 033 0 033 0 221 0 072 fixed 0 024 0 020 0 886 0 0745 1 088 0 0930 0 045 0 121 fixed 0 16600 0177 0 002 0 107 fi
94. reading due to the engine vibration e presence of the wall in the working section blockage effect This error was calculated using a formula taken from Pope 1947 page 256 as follows A y AI 2 1 where thrust p AV V 241 27 P propeller disc area and tunnel cross sectional area A typical blockage effect of 4 was obtained from the calculation at the thrust value of 8 3N and tunnel speed of 10 4m s This value was small enough to be neglected in the analysis Extraneous drag produced by the engine support and the exhaust hose Thrust misalignment between the engine body and the airflow 2 4 misalignment results in approximately 0 8N error in thrust measurement Limited accuracy of the instrumentation The accuracy of the tachometer and the manometer are equivalent to 0 2N and 0 3N error in the thrust measurement respectively Centre of Gravity CG and Moment Of Inertia Determination The centre of gravity CG locations and the moment of inertias were determined experimentally see appendix 6 for the results Horizontal and vertical CG locations were determined by placing the T240 model on weighing scales at two different points and measuring the reaction forces at these points The model was then tilted and the scale readings were noted The experiment was repeated for a number of tilt angles 49 The equation for determining the CG is given in Wolowic
95. red separately since they are driven by a separate servomotor On the other hand the elevator deflection is obtained by measuring only one side of the control surface deflection since the left and right elevators are mechanically connected Figure 6 6 Rudder deflection sensor The analysis of the sensor errors deterministic and random is described in appendix 2 6 2 Calibration Follow this procedure to carry out sensor calibrations e Connect the sensor to the allocated channel number on the DAS refer to Figure 6 3 e Run the Telemate Communication software and connect the DAS output port to the RS232 on the computer e Apply power to the DAS and at the same time press the dump button blue e You should then be able to monitor all the sensor readings continuously on the monitor The results of the sensor calibration are given in appendix 4 including the fitted and the associated error curves Rate gyros rate table was used to calibrate the pitch yaw and roll rate gyros Figure 6 1 Since there was no rotational speed measurement available on the rate table a switch potentiometer was used to measure the time taken for every revolution Figure 6 2 shows a typical sample of rate gyro calibration result 65 Computer count 300 20 30 seconds Figure 6 2 Rate gyro calibration trace Figure 6 1 Rate gyro calibration using a rate table Airflow direction indicator The calibrations for angle o
96. roblems in flying the aircraft The following lists some of the problems encountered in the actual flying and conducting the required manoeuvres Weather dependent The model should be flown in a calm air free of turbulence preferable early in the morning However since we conducted most of the flying in the winter we would be fortunate to have one perfect day to fly out of one week e Limited visual range and lack of information on the model s flying condition Difficulty in getting an exact trim condition Inability of the model to perform a required manoeuvre to produce a rich information response e Structural vibration due to engine rpm degrades the angular rate readings Figure 7 1 shows a contaminated roll rate reading during a flight manoeuvre with engine on Figure 7 2 shows the roll rate response with engine idle deg s 0 5 10 15 20 seconds Figure 7 1 Roll rate reading buried in engine noise during a flight manoeuvre 74 deg sec 0 2 4 6 8 10 12 14 seconds Figure 7 2 Roll rate reading with engine idle presence of air turbulence during the test manoeuvre Figure 7 3 and Figure 7 4 show the difference in the recorded angle of attack in a turbulence and calm air degrees 0 5 10 15 20 seconds Figure 7 3 Angle of attack reading buried in turbulence during an elevator doublet manoeuvre degrees seconds Figure 7 4 Angle of attack response in a reasonabl
97. s of water pressure measurement equivalent to 0 65 m s of airspeed Figure 6 9 Differential pressure sensor used as speed indicator 62 Engine rotational speed measurement A hall effect IC switch is used as a sensor to measure the engine rotational speed The IC produces a bounce free switching when influenced by a magnetic field Hence by mounting a magnet on a disc which rotates with the engine the IC will produce a pulse train which corresponds to the rotational speed of the engine The hall effect IC switch was selected since it was reliable small in size inexpensive robust to environmental contamination such as heat and light and can operate up to a high repetition rate 100 KHertz engine rpm engine shaft Balancing mass Figure 6 5 Propeller rotational speed measurement using a hall effect IC switch device Figure 6 4 Engine rpm sensor and the rotating disc Holcomb and Tumlison from NASA 1977 used a hall effect device to measure their engine rotational speed successfully A light sensor device can also be used however a direct light from the sun may introduce an error as experienced by Sydney University RPV Wong 1989 63 Control surface deflection measurement Servo potentiometers RS173 574 are used to measure the angular deflections of the control surfaces The deflections to be measured are those of elevator rudder left and right ailerons The left and right ailerons are measu
98. set to be estimated is 5 then the estimate of at iteration L 1 is given as ae 3 18 28 For a fixed RR the first and second gradient are given as N LLF amp GRAD amp RRT Bo 3 19 i l N N VELLE VZ RR viz Viz Z 1 i l N HES VEZ RR 7 i l N Where HES V Z Vg m 3 21 i l The accuracy of the parameter estimates can be assessed by determining their Cramer Rao Bounds CR bouna Which gives an estimate of the standard deviation of each parameter The CR bound is calculated via the information matrix H as follows 2 1 _ 2 LLF minimum ES N 1 AES goo 3 22 A more detail explanation of the method can be found in and Maine 1979 and 1989 29 3 2 3 Interactive Curve Matching The basic idea of this method is to interactively change the value of stability and control derivatives of the assumed mathematical model to obtain a good fit between the calculated responses and those of flight data The algorithm for this method is given in Figure 3 7 The application of this technique is possible due to a facility known as GUI Graphical User Interface offered in MATLAB software Figure 32 shows a longitudinal ICM with 8 different sliders representing 8 different derivative values Also shown is the corresponding error between the flight data and the fitted curve
99. sinO 4 p qsin tan rcos tanO ae 4 2 W 0 gsing cos 41 h 510 wcosO cos X ucosO cosy sin cosy cosdsiny sinO cosy sindsiny 4 3 y ucosOsiny 51 sinO siny cos pcos w cos 1 0 siny sin cosy And the output equation as AG v o 4 4 X tar en u If p q r ax ay and a are measured without error i e deterministic systems then the unknown bias b ba bg and scalar errors Ay Ag can be obtained using a linear regression to the above equations To simplify the analysis the complete non linear kinematics equations above are reduces to uncoupled longitudinal and lateral equations as Assuming constant velocity V then gt a E aja NN 4 5 ihe Assuming a ay q and are measured without error then 42 1 A 06 b measurement noise A B 5 measurement noise The scale and bias errors are then estimated by minimising o Pog and gt p 43 5 Model Description 4 Testing 5 1 Model Description The aircraft model to be flight tested is the Telemaster T240 Figure 5 1 It is a conventional wing tail configuration with elevator flap aileron and rudder as the aerodynamic control surfaces The model has 2 26m wingspan and 1 55m fuse
100. sive and a reliable small RPV can be easily built in which necessary flight test instrumentation can be incorporated Hamony 1994 reported on a state of the art light weight low power miniaturised instrumentation system which is used to gather information during flight test Beside the progress in the instrumentation systems several common problems in using a radio controlled model aircraft to conduct dynamic flight testing still remain Budd 1993 These problems include e Inability of the model to perform a required manoeuvre Coleman 1981 e Limited visual range and lack of flying conditions Wong 1989 e Signal interferences Hamory 1994 Errors in the obtained sensor data Coleman 1981 Typical errors in the sensor mostly originate from engine and other structural vibration cg offsets and misalignments transducer errors coupled longitudinal and lateral motions and the presence of air turbulence Budd 1993 Despite the above problems some have reported successful flight test programs in determining several dominant stability and control derivatives NASA has been using RPV s extensively to study the dynamic behaviour of their research aircraft such as the X 29 drop model Klein 1975 F 15 model Iliff 1976 and HIMAT Mathew 1981 The US NAVY was also researching with their RPV Howard 1991 In England British Aerospace conducted a similar flight test program to extract stability and control der
101. ssfully to analyse flight data Iliff Maine and Montgomery 1979 Coleman 1981 Budd 1993 Since the aircraft is of a conventional configuration and the manoeuvres conducted are of small perturbation these reduced equations should prove to be adequate Theoretically from these equations 6 longitudinal and 15 lateral derivatives can be extracted However in practice it is not always possible to get all the 21 derivatives from a single manoeuvre A low information content of the flight data is a typical cause of the problem a The longitudinal motion expressing perturbation from a horizontal steady flight is written as Klein 1994 V S Q q p Cz a elevator Cz V 2m XV 3 23 pV Sc qc 0 Cm a Cm C 5 C 4 21 2V M5 elevator elevator Or in a state space form as VS S VS VS pr 5 Cz amp m lo 2m elevator 2m 9 B 2 gt 4 elevator 3 24 d pV Sc Cite pV Sc la Sc C pV Sc C 1 21 si Dp ee Ap Note that in the moment equation the Cm derivatives are the compound effect of several variables as follows Cm 9 9 canara 32 Gn c Oni PS end c 4m 5 ie 1 PS es NT 3 25 4m pSc C E C m delevator t C m CZ iso elevator 4m b The lateral motion expressing perturbation from a steady flight condition pb rb aps ater
102. st to the previous two numerical approaches LR and the ICM is entirely a graphical technique The Interactive Curve Matching as the name suggests is a method of trying to fit the measured aircraft flight test responses with computed responses by interactively adjusting the values of the derivatives This method is very simple and allows a graphical observation during the identification process It is the basic principle of all the output error methods The only different is that the criterion of fit is decided by the operator by observing the goodness of fit on the computer monitor and hence is a subjective matter Figure 3 2 illustrates the inputs and outputs of the three different methods described above The MLM and ICM methods require fewer measurements in both longitudinal and lateral variables than the LR 25 INPUT a Selevator aq INPUT REGRESSION ANALYSIS MAXIMUM LIKELIHOOD INTERACTICE CURVE MATCHING Cz CZSetevator Cma Cm Cmajeator Cy Cng Cn CY aileron CY udder Claieron Cla udaer Cn amp uaaer Figure 3 2 Input Output for the three different identification methods 3 2 1 Linear regression This section describes briefly the linear regression technique the solution and its statistical accuracy Draper and Smith 1981 give a more detailed explanation of the technique For a line
103. systems including the on board receivers are powered by a 7 2 Ni Cad battery with capacity of 1 2 ampere hour 5 2 Engine testing The following section describes the thrust measurement in the wind tunnel This is needed to make correction when the flight test is conducted with engine on However when the test 15 conducted at engine idle the correction will not be necessary Various thrust measurement techniques are available such as direct force measurement propeller slipstream measurement propeller models and combined propeller and engine models Laban 1990 page 57 The propeller model technique was selected in this project due to its simplicity Also the technique requires only measurement of propeller operating conditions 45 The propeller model technique relates wind tunnel measurements of airspeed v and propeller rotational speed n to the thrust generated by the propeller For a fixed pitch propeller the blade element theory shows that the thrust produced is directly proportional to the advance ratio J where J v nD see Laban 1990 page 71 The set up of the experiment is given in Figure 5 1 The engine was supported on the thrust balance This balance measured the change in thrust as the engine rpm and wind tunnel speeds were varied Thrust Inclined Wind Tunnel balance manometer PCM TEE reciever 7 2 NiCd 400 mA hr pnm mm 5 Volt JR remote regulator 7 channel 36 MHz Figure 5 1 Experiment set up
104. that need to be installed When regression analysis is used each term involved in the regression equation has to be measured This means that for example to extract longitudinal derivatives 5 variables X q 4 and Selevator need to be measured or derived However a fewer number of sensors are needed when an output error technique is used such as maximum likelihood method Maine 1986 The only requirement is the availability of input and output variable measurements If redundant measurements are available then a data compatibility analysis can be performed to the obtained flight data The analysis can reveal any bias scale factor and other errors thus enabling correction to the flight data prior to estimation of the control and stability derivatives This is known as flight reconstruction Klein 1977 and Wingrove 1973 Generally among all the inertia sensors accelerometers produce the noisiest signals The structural and the engine vibration noises are the two major contributors to the accelerometer signal noise Maine 1986 Therefore a low pass filter should filter the signal before analysis The accelerometers should also be mounted on a rigid attachment to reduce noise from any structural vibrations 2 1 2 Data acquisition system The most common problems with the data acquisition systems are Maine 1986 1 Time tags Time tagging ensures that all the measurements are taken at the same time reference Erro
105. the IMU a performance testing can then be conducted Consequently the range and drag polar of the aircraft can then be determined 110 10 Conclusion The potential benefit of using UAV Unmanned Air Vehicles has prompted The Sir Lawrence Wackett Center for Aerospace Design and Technology to initiate a project referred to MAFV The objective of the project is to develop an unmanned autonomous flight vehicle This thesis is a part of the MAFV project with the objective of estimating a dynamic characteristic of a model aircraft from flight data using parameter identification techniques A Telemaster T240 model has been assembled and equipped with necessary flight test instrumentation The on board data acquisition system based on Intel 8030 has been developed in collaboration with The Computer System Engineering Department RMIT In addition the flight data processing software has been written using Matlab The whole system has been demonstrated by conducting a dynamic flight test program on the Telemaster T240 During the project the model has performed 17 number of flights through the whole development of the flight test system Four sets of maneuver data a total of 26 seconds of data have been successfully analyzed to estimate the T240 s dynamics A reasonably good flight data matchings have been achieved and 21 stability and control derivatives 5 longitudinal and 16 lateral have been estimated The project has shown that the
106. the program and the users is made as friendly as possible Thanks to the facility known as GUI in MATLAB that makes it possible The user can change any values and click any buttons to perform any required functions The complete window menus available in the program are given in appendix 5 38 4 1 The MMLES State Space Identification Tool box on Matlab The tool box contains functions for the parameter estimation of continuous linear time invariant multi input multi output state space models from observed input output data using either the maximum likelihood or output error method The tool box is an enhanced Matlab implementation of the widely used parameter identification program in processing flight data Maine 1981 It runs on a personal computer under the environment of Matlab software The tool box is very user friendly accessible and easy to modify or incorporated with other data processing functions in Matlab The steps needed for the estimation is given in Figure 4 1 Computation of Calculation of Minimisation of gradiert and wersum amp wersum amp NEUE hessian of cost likellihood likellihood function function function Parameter values Predicted output Gradient and Hessian Dynamical model by Input output data Choices of algorithm of the cost function creating file to Initial estimate of Quadratic Filtered innovation convert parameter parameters Lavenberg sequence vector into state space
107. timated longitudinal responses and their residuals from manoeuvre 2 records Mean Standard deviation Angle of attack deg 0 0265 1 888 0 6210 2 549 Pitch rate deg s 6 3934 11 80 6 9690 14 64 Table 9 6 Residual characteristics of the estimated longitudinal responses 9 3 2 Lateral stability and control derivative The estimated lateral stability and control derivatives are given in Table 9 7 and Table 9 8 and the matchings of flight data are presented in Figure 9 1 and Figure 9 2 Analysis 3 1 Analysis 3 2 Derivatives A priori Estimated Cramer Rao A priori Estimated Cramer Rao parameter Bound parameter Bound fixed fixed Cng Cn Cn Cy fixed fixed 5473 1 1 0 135 0231 0 126 Cop oni 015 4 933 Cn 002 fixed 0 109 _______ 0108 fixed 0 103 fixed 0 023 fixed 0 337 0 272 0 062 fixed 4 832 fixed 0 1323 0 090 0 065 4 _____ 448 _ 44 0 ia 0I Ty sec ____ 019 p r aL _____ Table 9 7 Estimated lateral parameter from recorded manoeuvre 3 with two different sets of a priori values Analysis 4 1 Analysis 4 2 Derivatives A priori Estimated Cramer Rao A priori Estimated Cramer Rao parameter Bound parameter Bound 9 79 2 75 43 42 8 15 2 289 11 52 fixed 0355 fixed ___ fixed 0945 fied Cl 0 176 021 Cng Cn Cn 0 034 fix fix CO ch ____
108. tion NASA TP 1690 28 Maine R E 1981 The Theory And Practice Of Estimating The Accuracy Of Dynamic Flight Determined Coefficient NASA RP 1077 29 Maine R E Iliff K W 1986 Application Of Parameter Estimation To Aircraft Stability And Control The Output Error Approach NASA Reference Publication 1168 30 Malvestuto F F Gale L J Jr 1947 Formulas for Additional Mass Corrections to the Moment of Inertias of Airplanes NACA TN 1187 31 Mathew N W Pangeas G N 1981 HIMAT Aerodynamic Design And Flight Test Experience AIAA Paper 81 2433 Nov 1981 32 Milne Garth 1992 State Space Identification Toolbox For Use With Matlab The Mathworks 33 Mohammad M 1995 Identification of Turboprop Thrust From Flight Test Data Phd Dissertation Delft University Press ISBN 90 9009058 4 34 Mulder Sridar J K Breeman J H 1994 Identification of Dynamic Systems Application to aircraft Part 2 Nonlinear Analysis and Manoeuvre Design AGARD AG 300 Vol 3 Part 2 35 Newman D M Wong K C 1995 An Atmospheric Disturbance Model for Small Remotely Piloted Vehicle Simulation and Analysis Proceeding of the 2nd pacific International Conference on Aerospace Science and Technology PICAST2 Melbourne March 1995 Vol 2 Page 579 583 36 Plaetschke E Schulz G 1979 Practical Input Signal Design Parameter Identification AGARD LS 104 paper 3 29 Oct 2 Nov 37 Pope Alan 1947 Wind Tun
109. tiring and strenuous on aircraft crew The present state of technology allows the development of relatively small lightweight and accurate remote sensing equipment that will provide a wide range of different payload packages suitable for incorporation into UAVs With the advent of a reliable and low cost GPS Global Positioning System an autonomous unmanned air vehicle becomes technically and economically feasible for survey or surveillance missions With self contained navigation and control systems these vehicles have the potential to carry out their mission according to a pre programmed set of instructions Future developments on built in intelligence open the way to true autonomous missions whereby the on board equipment senses anomalies and can take independent action The potential benefit of UAV technology has prompted The Sir Lawrence Wackett Centre for Aerospace Design Technology to initiate a project with the objective to develop an unmanned autonomous flight vehicle referred to as Multi Purpose Autonomous Flight Vehicle MAFV The vehicle will be designed to suit a wide range of missions such as aerial photography coastal surveillance geological and agricultural survey atmospheric research and weather soundings Thompson Abanteriba and Bill 1993 The Division of Atmospheric Research of the CSIRO in Australia has expressed particular interest in the MAFV as a potential platform for their equipment for monitoring of atmosphenc pollutio
110. to perform their parameter identification process 36 The structure of the program is given in Figure 4 1 It has 5 main categories dynamic simulation flight data stability amp control derivative estimation data compatibility analysis and a priori Each category consists of several functions which perform the necessary calculations for that particular category Wind tunnel e calculation Dynamic Simulation Longitudinal Stability amp Control Derivative Estimation Prior Flight test e Curve Matching Regression Stability amp Control Derivative Data Base Flight Data e Maximum Likelihood Conditioning Analysis Figure 4 1 The structure of the flight test computer program developed for the project The whole program contains several sub programs in the form of M script files These M files perform just as subroutines in programming languages such as C or Fortran The result of the calculation from each M file is saved in a binary form with extension mat therefore named as mat files The interconnection between M files and mat files in the program is described in Figure 4 2 37 parameter identification files Dynamic simulation files londata mat 2 Lslatgui m ST Filtsig m m Orde21 m m d Ordell m Matchlat m Data prepocessing files Figure 4 2 Interconnection between M and mat files in the program Interaction between
111. ussian noise in the instrumentation However if a process or input noise is present the method fails to converge into a solution A Kalman filter should then be incorporated to the MLM to enable the estimation of the system s states Milne 1992 The MLM is the most widely used method on the extraction of stability and control derivatives from flight test data NASA has developed a computer code MMLE2 to perform this algorithm Maine 1981 The Bayes method is not widely used in the estimation of stability and control derivatives The reason for this is that the method assumes a known a priori statistical noise In practice this a priori statistical noise is not always available In this project the Maximum Likelihood Method is selected as the main algorithm to extract stability and control derivatives of a small UAV from the flight test data This method has several beneficial features such as the following t gives asymptotically unbiased and consistent estimates Only input and output data is required and hence less number of sensors needed e Good performance even in the presence of output noise If input or process noise is present then a MLM Kalman filter is used A Cramer Rao bound which is by product of the algorithm can be used as a measure of accuracy of the individual estimated parameters Maine and Iliff 1981 e information can be incorporated e g from wind tunnel results e
112. vre This might be due to the difference in the pulse width of the rudder input As the Dutch mode is a combination of yawing and rolling oscillations then the combined rudder and aileron inputs should excite the Dutch mode better thus resulting in better estimation than the rudder input alone The Dutch damping on the other hand was quite consistent throughout different estimation process The scatter in the obtained longitudinal derivatives was quite low and hence a reasonable confidence in the results The scatter on the lateral derivatives on the other hand varied significantly Those derivatives with high CRB values show high scatter in the results Dominant derivatives such as Clgaiteron and have low CRB values and hence better estimated The matching between the prediction and the flight data was generally good Even in the presence of significant engine vibration noise in the angular rate measurements the MLM algorithm predicted the 105 response quite well When a good match could not be achieved then one of the following reasons might be causing the problem modelling errors uncorrected bias errors or a small excitation in the mode of interest 10 2 Flight data processing Four recorded manoeuvres a total of 26 seconds of data have been processed and analysed to obtain the stability and control derivatives of the Telemaster T240 The software written in Matlab has undoubtedly eased and proved
113. xed 0 305 0 0958 4 196 1 694 8 123 1 745 0 084 0 272 0 1247 0 120 018 0 0 090 0 0024 0 099 0 0049 0 256 0 0395 0 380 Table 10 2 Estimated lateral derivatives of the Telemaster T240 0 001 0 108 fixed 0 103 0 0432 1 086 0 1989 1 894 0 177 The estimated values were not always in a good agreement with those predicted by AAA The AAA software is normally used for estimating derivatives of a conventional aircraft with minimum mass of 50 Kg not for a small RPV type aircraft Here no direct comparison could actually be made However most of the flight test results were in the same order of the AAA In addition the AAA predictions have assisted in starting the MLM algorithm One interesting point to comment is on the values of CZgelevator and CMselevator The AAA predicted the wrong sign of derivatives since it assumed a conventional horizontal tail In fact the T2408 tailplane is a flat top aerofoil which generates lift when the elevator is deflected upward hence a positive values of Cz amp ievaior ANd pitch up manoeuvre is achieved by a positive downward elevator deflection The SPO Short Period Oscillation mode characteristics were estimated reasonably well However the lateral modes showed a little inconsistency The rudder manoeuvre estimated higher Dutch mode frequency 10096 higher than the combined rudder and aileron manoeu
114. y and control derivatives are given in Table 2 1 This table is summarised from Maine 1986 Wong 1989 and Yip 1992 Table 2 1 Sensors frequently used in the extraction of stability amp control derivatives 2 Yawing velocity Rate gyro 6 Rolling velocity Rategyro Es Angle of attack Flow direction velocity sensor Angle of sideslip Flow direction velocity sensor Control deflections Control position transducer transducer transducer Digital clock Importance _______________ Secondary 0 to 0 to 60 knt 0 25 15 psi em em mem Tem The type of sensor needed depends upon the purpose of the flight test and the capability of the instrumentation systems Coleman 1981 with his STABILEYE RPV conducted flight tests with only body rates and control deflection measurements With this limited number of sensors he failed to get several dominant lateral derivatives He then proposed to add a lateral accelerometer to the aircraft Due to the limitation in the number of sensors in their first flight test Howard 1991 at the US NAVY measured only engine rpm and angle of attack onboard the vehicle The airspeed was 12 measured by observation on the ground Only lift and drag plots were obtained from this experiment and a significant scatter in the drag measurement was apparent The method of flight data analysis also dictates the type of sensors
115. z 1974 as N x rtanO whee 5 7 W z 15 7cm d n Ss Datum gt 2 d _ B C B s hl 7 Ec _ d Figure 5 1 Experimental technique for determining weight and CG positions y 0 2579x 0 1574 0 9992 5 0 2 0 15 o 0 1 9 0 05 c oc 0 0 0 05 0 1 0 15 0 2 tan theta Figure 5 2 Results from the cg experiment The result of the CG test indicated that the centre of gravity was located at 25 8 cm aft of the datum 15 7 cm above the datum point see Figure 5 1 The pitching moment of inertia was determined by using a knife edge method The model was supported on two knife edges along the y axis and allowed to oscillate Figure 5 8 The time taken 50 for several oscillations were noted and averaged The pitching moment of inertia was then calculated as De Jong 1987 T Mgl Ar Where ly is the vertical distance between the cg and pivot point in metre Figure 5 3 Pitching moment of inertia determination using a knife edge method The yaw and roll inertias were determined experimentally using bifilar suspension method In this method the model was suspended by two thin strings equidistant from the centre of gravity and allowed to oscillate freely about the vertical axis passing through the centre of gravity During the experiment several samples were taken and an average reading

Download Pdf Manuals

Related Search

Related Contents

OV-1422 - Tristar  Hanoukah mode d`emploi  NS フラップデスク組立説明書 - Garage  4. For Cytoscape App Developers  IAN 97734  Graco PD243033A User's Manual  Philips HD8824  AL-LHQ-112-3  EPP-200 Consola de Corte “Precision Plasmarc”  Pages from [MOPDE user manual]  

Copyright © All rights reserved. Failed to retrieve file