art and science of institute of flight research system ... a300-600 3 1982 ... a/c landing gears ......

Inst

itu

te o

f Fl

igh

t R

esea

rch

Art and Science of System Identification

OutputInput

System

Art and Science of System Identification 2000

Institute of Flight Research 1

hrough innovative ideas in flight maneuver tech-niques, parameter estima-

tion methods, and aerody-namic modeling, the Institute has established itself interna-tionally as a major contributor to the field of flight vehicle sys-tem identification. The inverse-principle based system identifi-cation methodology provides a scientific approach to modeling and estimation. Over the last thirty years, aircraft parameter estimation techniques have been successfully applied to various tasks such as:

• Data correlation for increas-ing the confidence in flight mechanical prediction tech-niques,

• Generation of high fidelity aerodynamic databases for training and in-flight simula-tors,

• Flight envelope expansion through isolation and identi-fication of non-anticipated aerodynamic effects,

• Handling qualities assess-ment, and

• Rotorcraft modeling with high bandwidth require-ments.

Table 1 provides a concise summary of the various appli-cations covering a broad spec-trum from simple aircraft to high performance, highly augmented modern flight ve-hicles and other miscellaneous dynamic systems.

System identification is an in-terdisciplinary task requiring a coordinated approach encom-passing the important aspects of:

• Flight test instrumentation, • Flight test techniques, and • Analysis of flight data.

The art and science of these strongly interdependent topics

practiced in the Institute have led to a well established Quad-M (Maneuvers, Measure-ments, Methods, and Models) approach, which has been the key to the highly successful application of system identifi-cation methodology to numer-ous dynamic systems at DLR.

Flight Measurements and Data Pre-Processing

The accuracy of the parameter estimates is directly dependent on the quality of the flight measured data, and, hence, high accuracy measurements of the control inputs and of the motion variables are a

Flight Vehicle Period Type Reports 1970 - 1973 Do 27 5

1975 - 1988 HFB 320-S1 FLISI 10

1976 CASA C-212 1

1978 - 1983 DHC-2 Beaver 4

1981 A300-600 3

1982 - 1985 Do 28 TNT OLGA 8

1984 A310 1

1986 - 1996 VFW 614 ATTAS 23

1990 DA Falcon E 2

1995 Grob G 850 Strato 2C 1

1997 - 1998 Dornier 328-110 7

1997 A300-600ST Beluga 1

Civil A/C

1997 - 1998 A330-200 5

1976 - 1983 MRCA - Tornado 13

1977 CCV F 104 - G 1

1981 Do Alpha-Jet TST 2

1984 Do Alpha-Jet-DSFC 1

1989 - 1997 C-160 Transall 14

1993 - 1998 Dasa/Rockwell X-31A 31

Military A/C and Technology Dem-onstrators

1997 ==> EF 2000 Eurofighter 2

1975 ==> Bo 105 S123 20

1986 - 1989 Bell XV-15 tilt rotor 3

1989 - 1990 AH 64 Apache 2

1989 - 1990 SA 330 Puma 2

1989 - 1995 Bo 105 S3-ATTHeS 15

1998 SA 365 Dauphin 1

Helicopter and Tilt Rotor A/C

1999 ==> EC 135 HESTOR -

1978 - 1980 ATA free flight model 5

1980 Do 28 TNT wind tunnel model 1

A/C Models

1982 - 1983 Do 28 TNT free flight model 3

1989 Falke 2 Reentry Models

1998 USERS 1

1995 NASA Spacewedge Parafoil 1

1997 D 160 1

Parafoil Payload

1998 ==> ALEX 1

1986 EPHAG 2 Rockets and Projectiles 1987 - 1988 EPHRAM 4

1992 Rolls Royce Tyne R.Ty.20 in C-160 Transall

1

1997 Pratt and Whitney PW 119A in Dornier 328-110

1

1997 Rolls Royce M45H MK501 in VFW 614 ATTAS

1

A/C Propulsion Systems

1998 ==> F404-GE-400 in X-31A 1

1996 - 1997 C-160 Transall 2

1997 Do 328-100 1

1997 VFW 614 ATTAS 1

A/C Landing Gears

1998 ==> Dasa/Rockwell X-31A -

1985 Submarine, Class 206 �U 13� 2

1997 ==> Aircraft pilot coupling 1

Miscellaneous

1998 ==> Combustion Plant RWE - VVA 1

Table 1: DLR system identification experience



prerequisite for successful ap-plication of the modern meth-ods of system identification. Considerable effort has been spent on this aspect at the In-stitute. Commercially available sensors, standard signal proc-essors, and in-house developed data processing systems are judiciously combined to opti-mize this laborious and time-consuming process. Capability to design, fabricate, and install a full-fledged flight test instru-mentation, including its flight certification, was successfully demonstrated for a high-

fidelity flight simulator data gathering with the C-160 Transall aircraft. A total of 91 physical parameters were measured and recorded, Figs. 1 and 2. A few signals were picked off from the basic air-craft instrumentation, but the bulk of them required addi-tionally installed sensors. The test aircraft was one from the operational fleet of the Ger-man Air Force. Due to safety reasons, it was not permitted to drill holes into the aircraft, and, hence, installation of ad-ditional sensors and test

equipment was ingeniously done using clamps, brackets, etc.

Optimal Inputs for Dynamic Motion

The scope of identification is limited by the flight tests, since the physical phenomenon pos-tulated as a cause-effect model cannot be identified, if it is not in the data. Hence, a proper experiment design is impor-tant. Moreover, accuracy and reliability of parameter esti-mates depend heavily on the amount of information avail-able in the vehicle response. In general, an optimal input is that which best excites the fre-quency range of interest. Purely from this view point, the direct choice may appear to be a frequency sweep. However, sweep testing is mostly re-stricted to single axis excita-tion. Moreover, it leads to rela-tively long maneuver times and thus the vehicle has a tendency to depart from the nominal trim. Based on these practical considerations, several signals, for example, multistep 3211, Mehra, Schulz, and DUT, were designed in the seventies. The doublet input is also often used due to its simplicity.

Fig. 3 shows Koehler�s 3211 input developed in the Institute and its power spectrum in comparison to step and dou-blet inputs. The 3211, Mehra, DUT, and the recently designed Langley inputs are efficient; amongst them the multistep 3211 signal is the easiest to re-alize and relatively easy to fly manually by pilots. In addition, the frequency contents can be readily adapted to match changing flight conditions. It is for these reasons that the 3211 signal is widely used in-ternationally for system iden-

Fig. 1: 5-hole probe and on board data acquisition for C-160 data gathering

� � � � ��

� � � � � �

� � � � ��

� � � �

� � � � � �

� � � � � � � �

� � � � � � � � � � � � � � � �

� � � �

� � � � � �

� � � ��

� � � � � � � � � � �

� �

� � �

� �

� � �

� �

� � ��

� � �

��

Fig. 3: Frequency spectra of typical inputs

Fig. 2: Instrumented C-160 Transall



tification, see Table 2 for the application statistics of the 3211 signal [1].

The more recent emphasis on expanding the design tech-niques based on additional practical considerations has re-sulted in new inputs having improved properties. These in-puts are, however, difficult to fly manually and can be best realized through onboard com-puter implementation. Thus, although it may appear that the current trend is towards more complex computerized inputs, the simpler inputs like doublet or 3211 will continue to be as accepted in the future as in the past.

Parameter Estimation Methods

Recognizing early the need for efficient analytical tools for analysis of flight data from flight vehicles of increasing complexity, as a long term per-spective the research and de-velopmental work on aircraft

parameter estimation methods was initiated at the Institute in the early seventies. This work has gone through various phases of innovations, up-dates, and consolidations. This process continues even today with the current goal being to meet new demands and pro-ject specific requirements. Fig. 4 shows at a glance the evolution of the software tools for parameter estimation in the time domain.

Although the recent focus has been on the time domain methods, parameter estimation methods in the frequency do-main have also been developed and used extensively in the In-stitute, particularly for ro-torcraft identification. Similarly, other techniques of frequency analysis based on spectral, co-herence, and frequency re-sponse functions catering to multi-input/multi-output sys-tems have also been devel-oped.

Fig. 4: Evolution of estimation program packages

Period Flight Vehicle Institution Period Flight Vehicle Institution

Fixed wing 1993 Convair CV 580 NRC

1974 � 1983 HFB 320 FLISI DFVLR 1994 Challenger CL601/3 NRC

1974 � 1976 CASA C-212 INTA/DLR 1994 Gulfstream G IV NRC

1976 � 1982 DHC-2 Beaver DUT/NLR 1994 A300-600ST Beluga SATIC

1978 Tu-154M FRI 1994 F/A-18A/D US Navy

1978 � 1980 Tornado P11 E-61 1994 � 1995 EF 2000, Eurofighter WTD 61, BAe

1980 � 1981 Alpha Jet TST E-61 1995 Strato2C Grob/DLR

1980 � 1981 CCV F104 G MBB 1996 S-3B Viking US Navy

1980 � 1982 Do 28 TNT Model, OLGA DFVLR 1997 F-18 HARV NASA

1980 � 1982 Do 28 TNT Model IMFL 1998 � 1999 F/A-18E/F US Navy

1981 � 1983 Do TNT-Experimental Dornier Rotorcraft

1983 FBW Jaguar BAe 1977 Bo 105 S3 MBB

1983 � 1984 Alpha Jet DSFC E-61 1978 � 1987 Bo 105 S123 DFVLR

1984 F-8 DFBW NASA 1980 � 1987 Bo 105 ATTHeS DFVLR

1984 A300-600 AI 1982 UH 60 A US Army

1984 � 1986 Do 28 TU-BS 1983 CH-53 A US Army

1985 Tornado BAe 1983 RSRA NASA

1985 � 1990 VFW 614 ATTAS DLR 1985 Bell 206 A NRC

1986 � 1991 EAP BAe 1986 Bell 205 A�1 NRC

1987 A310-300 AI 1986 � 1987 XV-15 US Army

1987 Cessna Citation 500 NLR 1988 � 1990 AH 64 Apache MDHC

1990 Dash 8 Series 100 NRC 1988 � 1990 SA 330 PUMA RAE

1990 � 1995 X-31A RI/Dasa/DLR 1992 Bell 412 HP NRC

1991 Dash 8 Series 300 NRC 1997 � 1998 Dauphin 6075 CEV/ONERA

1992 � 1993 C-160 Transall DLR 1998 ==> EC 135 ECD

Table 2: DLR 3211 input signal statistics

� � � � � ��

� � � � ��

� � � � ��

� � � ��

� � � � ��

� � � � ��

� � � � � ��

� � � � � � � �

� � � � � � �

� � � � � �

� � � � � � � � � � � � �

� � � � � � � � � � � � � �

� � � � ��

� � � � ��

� � � � � � � � � � � � �

� � � � � � � � � � � � � � � � �

� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �

� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �

� � � � � � � � � � � � � � � � � � � � � � � � � �

� � � � ! � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �

� �

� � � � � � � � � � � � � � � �

" # � � � � � � � � � � � � � � � � � � � � � � � � � � � � �

� � " # � � � � � � � � � � � � � � � � $ � � � � � � � � � � � � � � � � � � � � � � � � � � �

� � % " & � � � � ' � � � � ( � � � � ) � $ � � � � � � � � � � � � � � � � � � � � � � � � � � �

* # % " & � � � � % + � � � � � � � " � � � � $ � � � � � � � � � � � � � ) � � � � � � � �

� � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � � �



The two milestones in the de-velopment of aircraft parame-ter estimation methods in the Institute are: i) extension of the widely used output error method to general nonlinear retarded systems and ii) development of filter error algorithms for such systems accounting for atmospheric turbulence and measurement noise. This systematic work on algorithmic development was supported by the Special Re-search Project SFB-212 of the Technical University of Braun-schweig. Today�s comprehen-sive software package, called ESTIMA, is a modular system of algorithms for aircraft parame-ter estimation in the time do-main and is a unique tool of the Institute, see Table 3.

Aerodynamic Model Identi-fication � Last Decade�s Highlights

Since it would be a mammoth task to provide a complete de-scription of the various applica-tions listed in Table 1, only a few highlights of the last dec-ade are presented here. It does

not in anyway imply that the other cases were less impor-tant or simple.

High Fidelity Databases for Training and In-Flight Simu-lators

With the evolution of modern high performance aircraft, spi-raling developmental and ex-perimental costs, and increas-ing flight safety issues, the im-portance of ground-based flight and in-flight simulators has increased significantly in

the recent past. Simulators are increasingly used not only for pilot training, but also for other applications such as flight planning, envelope ex-pansion, design and analysis of control laws, handling qualities investigations, and pilot-in-the-loop studies. The majority of these cases demand high-fi-delity aerodynamic databases of the flight vehicle and the importance of flight validation of such a database is well rec-ognized. Applying system iden-

Fig. 5: C-160 Transall

Fig. 6: Dornier 328 commuter aircraft

Scope of application: Main features:

ESTIMA is optimized to meet the heavy demands of: - processing large amount of flight data (up to 80000

data points from 80 flight maneuvers) - estimation of large number of parameters (up to 1200) - complex dynamical systems with large number of state,

observation, and control input variables (up to 30, 60, and 60 respectively)

It caters for and is applicable to: - linear and general nonlinear retarded systems - unstable aircraft - systems with measurement noise and atmospheric

turbulence (process noise) - multi run evaluation concatenating several maneuvers - estimation of time delays and hysteresis - different optimization and integration methods - hybridization of integration methods - special option for proof-of-match computations - classical simulation mode providing possibilities to trim

the model and carry out simulation

Modes: - Output Output error method - Filter Filter error method - EKF Estimation by Extended Kalman Filter - TRIM_FV Trim the model to specified flight measured SIM_1P or arbitrarily specified flight conditions - UMRECH Data pre-processing

Optimization methods: - Gauss-Newton (Modified Newton-Raphson) method - Levenberg- Marquardt method - Quasi-Newton and Conjugate gradient methods - Direct search methods of Powell and Jacob - Simplex method of Nelder and Mead - Subspace searching Simplex method

Integration methods: - Euler, Runge-Kutta 2nd, 3rd, and 4th order - Runge-Kutta-Fehlberg 4th/5th order with step size control - Backward differentiation formula for stiff systems

Table 3: Scope of application and main features of estimation package ESTIMA



tification methods, aerody-namic databases were gener-ated by the Institute for the military transport aircraft C-160 Transall, Fig. 5, and the commuter aircraft Dornier 328, Fig. 6; in both cases meeting the Level D quality criteria, the highest fidelity class specified by FAA [2, 3].

The In-Flight Simulator ATTAS (Advanced Technologies Test-ing Aircraft System) is a fly-by-wire aircraft, Fig. 7, and has been the testbed for numerous investigations, such as model-ing of direct-lift control, sepa-rate wing and tail identifi-cation, separation of pitch damping derivatives, control surface interference effects, and modeling of actuation sys-tems of primary controls as well as propulsion systems. Ex-tensive tests and identification efforts led to a high fidelity global aerodynamic database [4, 5].

Unsteady Aerodynamic Modeling

Although unsteady aerody-namics has been a subject of extensive investigations using computational fluid dynamic methods, wind tunnel tests, and a number of semi-empiri-cal models, flight validation of the postulated models has been difficult in the past. The recent advances in aerody-namic modeling have led to analytical models for complex processes such as ground ef-fects and separated flow in-cluding stall hysteresis, making identification and validation from flight data amenable. As a typical example, Fig. 8 shows a comparison of the total lift and pitching moment coeffi-cients derived from flight measurements with those es-timated using system identifi-

cation. Applying Kirchhoff�s theory of flow separation, a parametric model with only three unknown parameters adequately characterizes the quasi-steady stall behavior. This innovative approach was suc-cessfully applied in the de-velopment of aerodynamic da-tabases for the aforemen-tioned Level D training simula-tors C-160 Transall and Dornier 328 as well as to the In-Flight Simulator ATTAS [6].

X-31A System Identification

The USA/German experimental aircraft X-31A is a highly aug-mented technology demon-strator with enhanced maneu-verability, incorporating ad-vanced technologies such as high angle-of-attack aerody-namics and an integrated flight control system including thrust vectoring, Fig. 9. The X-31A system identification, for which the Institute was primarily re-sponsible, posed special chal-lenges mainly because: i) the basic aircraft is inherently un-

Fig. 7: In-Flight Simulator VFW 614 ATTAS

�

� � � �

� � �

� � �

� � � � � �

� � �

� � �

� �

� � �

�

� !" � � � � # " � � � $ % & � � �

� � � � � � � � � � � � ' � � ( � � � � ) � � � � * � � � � � � � � � # � � ��

� � � ! "

� � � � � �� !

� �

" � � � � # " � � � $ % & � � �

Fig. 8: Modeling of Dornier 328 stall hysterisis

Fig. 9: X-31A in post stall flight



stable leading to error propa-gation and divergence in the numerical integration during open loop identification and ii) the flight control laws sup-press oscillatory and transient motions as well as lead to cor-related input and motion vari-ables resulting in identifiability problems. The solution to the first problem of numerical di-vergence is provided by dili-gent choice of appropriate es-timation algorithms from the wide variety available in the modular system of algorithms ESTIMA. The solution to the second problem of identifiabil-ity was provided through sepa-rate surface excitation (SSE). In this approach the computer generated standard inputs such as doublet or 3211 are fed directly to the control sur-faces. The electronics for single surface excitation were de-signed and provided by the In-stitute.

As a typical example, Fig. 10 shows estimates of the canard control effectiveness obtained from X-31A flight data for two cases, namely the pilot input and separate surface excitation maneuvers. As evident from Fig. 10, the pilot input maneu-vers yield estimates with large standard deviations shown by vertical bars and, moreover, the scatter is also large. On the other hand, the separate sur-face excitation maneuvers yield well identifiable estimates.

The results of X-31A system identification have been used to validate and in several cases update the original database. Fig. 11 shows two typical ex-amples of database update. The flight estimates did not confirm the wind-tunnel pre-dicted large negative value of the dihedral effect between 30°-45° of angle of attack.

Similarly, considerable discrep-ancies were observed in the di-rectional stability at higher Mach numbers. System identi-fication provided improved models of aerodynamic charac-teristics including post stall re-gime and thrust vector control for flight test planning, flight envelope expansion, and a validated database for simula-tion and control law modifica-tions and verification [7, 8].

EF 2000 Model Validation

During the development of Eu-rofighter, a joint venture be-tween England, Germany, Italy, and Spain, validation of wind tunnel predictions played an important role, particularly to

confirm the nonlinear rudder effectiveness and power rever-sal, Fig. 12. Flight investiga-tions and the subsequent sys-tem identification using the specially tailored methods of DLR confirmed these predic-tions. Besides the specialized task, the Institute also provided project support to the Dasa flight test center.

Rotorcraft High Bandwidth Modeling

The philosophy of model-following control based on feedforward regulation pro-vides safer and more accurate mode control; the perform-ance, however, depends

, - .. , / , 0 , 1 ,

# � ) � � � � $ � # � � � 2 � � � � )

� � � � � � � � 3 � � � � � � � ( � �, - 0

, - /

, - .

,

4 � � �

. , / , 0 , 1 ,# � ) � � � � $ � # � � � 2 � � � � )

� � 3 � � � � � � � $ � � � � + � � � � � � �

� � � � � � � � � �

� � � � � � # � � �

+ � � � � � � � � � � ) � ��

� � � � � � # � � �

+ � � � � � � � � � � ) � �

Fig. 10: X-31A canard control effectiveness

, - /

,

, - .

. , / , 0 , 1 ,# � ) � � � � $ � # � � � 2 � � � � )

+ � � � � � � � � � � ) � �� # � � � & � � ,� � � � ) � � � � � � � �

�4 �

5 � � � � � � � � � � � � 6 � � � � �

, - .

, - 7

, - !

, - ,, - 1 ! - , ! - .

4 ��

5 � � � � � � � � $ $ � � �

� � �

+ � � � � � � � � � � ) � �� # � � �� ) � � � � � � � �

Fig. 11: Examples of X-31A database update

Fig. 12: EF 2000 Eurofighter



strongly on the quality of the mathematical model of the host vehicle. For such an appli-cation, the rotorcraft flight control has to meet high bandwidth requirements, which demand augmentation of the lower to medium fre-quency range rigid-body model through higher-order rotor dy-namics. In the specific case of the DLR In-Flight Helicopter Simulator ATTHeS, Fig. 13, the application of the flight vali-dated higher order models for the model-following control was a major prerequisite for the achieved excellent fidelity. The frequency response of roll rate to lateral stick input shown in Fig. 14 clearly brings out that the high frequency helicopter rotor characteristics cannot be adequately de-scribed by rigid-body model alone, but that a 9 degree-of-freedom model combining the rigid-body and rotor dynamics is necessary [9].

XV-15 Tilt-Rotor Model Validation

The tilt-rotor aircraft configura-tions ideally combine the aero-dynamic characteristics and the advantages of both helicopter rotors and aircraft perform-ance. Identification of tilt-rotor flight vehicle in hovering flight is, however, particularly diffi-cult because of the highly nonlinear and coupled dynam-ics, and also because of the open-loop instabilities at such flight conditions. As a part of a USA/FRG Memorandum of Understanding on helicopter flight control, an extensive joint study was conducted to analyze XV-15 flight data, Fig. 15. Parameter estimation techniques in the time and fre-quency domain were applied to develop a unified approach for rotorcraft identification.

The results were used for da-tabase validation and for im-proving model fidelity [10].

Dissemination of Expertise

Promulgation of the expertise acquired in the Institute has been consistently one of the focal points over the last three decades. Besides the numerous technical papers in the reputed international journals, key con-tributions through AGARD ac-tivities are noteworthy, for ex-ample, Conference Proceed-ings CP-172 in 1975, CP-235 in 1978, Lecture Series LS-104 in 1979, LS-114 in 1981, LS-178 in 1991, and RTO MP-11

in 1998. The activities of the AGARD Advisory Report AR-280 in 1991 were also coordi-nated by the Institute. The in-vited survey paper in the AIAA Journal of Aircraft on �Evolu-tion of Flight Vehicle System Identification� [11] giving a historical perspective and a consolidated account of vari-ous contributions and the overview paper on �Advances in Rotorcraft System Identifica-tion� [12] have become stan-dard reference material on the subject and are examples of outstanding contributions to the field.

Fig. 13: In-Flight Simulator Bo 105 ATTHeS

. ,� � 8 �& � � 9 � � � � �

# � 3 � � � � � �

� � � �

� � )

! ! ,

,

! 1 ,

7 ,

: ,

� ;

� � � � � � � � � ) � � � 6 � � � � � � � � � � � � � � 5 < &

Fig. 14: Bo 105 frequency response comparison

Fig. 15: XV-15 tilt-rotor in hover flight



Software Systems and Dedicated Codes

Modeling, analysis, and simula-tion are the main thrust areas of the Institute. A considerable expertise has been developed in the application of system identification methodology to model flight vehicles, be it a conventional civil or military aircraft, a highly augmented aircraft flying at high angles of attack, a rotorcraft or a missile. Similarly, data analysis and simulations incorporating wind tunnel predicted, flight identi-fied or generic databases are integral parts of the research work. The powerful and vali-dated software tools devel-oped in the Institute for each of these tasks form the back-bone of the various applica-tions.

ESTIMA � A Modular System of Algorithms for Parameter Estimation

System identification involves analyzing complex multiple in-put/multiple output (MIMO) dynamic systems and a large amount of flight data. The ad-vanced methods of parameter estimation, like filter error method and extended Kalman filter, are necessary besides the classical output error and Least Squares methods. Efficient im-plementation of computational aspects such as state estima-tion, optimization, and gradi-ents is important, see Fig. 16. Flexibility to handle general nonlinear retarded systems and a wide variety of algorithms is a pre-requisite to cover the broad spectrum of applica-tions, see chapter 5.

The dedicated software tool ESTIMA, a modular system of algorithms for parameter esti-mation, has evolved over the time and caters for:

• Linear and general nonlinear retarded systems,

• Optimized processing of large amount of flight data (up to 80000 data points from 80 flight maneuvers),

• Complex dynamic systems with large number of state, control input, and output variables,

• Wide variety of gradient and non gradient optimization methods (Gauss-Newton, Levenberg-Marquardt, Sim-plex, Subplex, Extrem, Quasi-Newton, Conjugate gradient etc.),

• Various integration methods to suit different types of non stiff and stiff dynamic sys-tems, and

• Hybridization of integration and optimization methods.

The modular software provides a common environment having a uniform structure and a sin-gle set of peripheral support, like reading of flight data, generating outputs from esti-mation program, plot routines, statistical error analysis etc. for the various modes of opera-tion. ESTIMA has been devel-oped using standard FORTRAN 77 language and is currently implemented under VMS and UNIX operating systems. It is contemplated to extend the scope of this software to opti-mization with simple bounds

and to cater for batch mode operation as well as having a graphical surface for user in-teraction and display of estima-tion results [13, 14, 15].

RAPID � Modeling and Analysis with Local Model Networks

Identification of nonlinear physical models presents many problems since both model structure and parameters must be determined. For several ap-plications, like calibration of air data sensing systems or control of combustion plants, it is too costly and time-consuming to deduce possible model struc-tures from engineering knowl-edge of the system. The soft-ware tool RAPID realizes the concept of local model net-works by using locally defined partial models to approximate an unknown non-linear func-tion. The structure of the model network is not deter-mined a priori, but results from successive subdivision of the input area into hyper-rectan-gles. A smooth transition be-tween the local models is achieved by the fuzzy ap-proach of Takagi and Sugeno. For easy user interactions, a OSF/Motif-based graphical user interface and sophisticated visualization modules are de-signed [16].

# � � � � $ �

� � � � � � � � � ��

� � � � � � ��

� � � � � � �� 3 � � � � �

% � � � � � � �� 3 � � � �

� � � � � � � �

& � � � � ��

� � 3 � � �

� � � � ��

� � � � � � � � 3 � � � � � � � 6 ��

� � � � � � � � � 3 � � �6 � � � � � ! � � � � � � $� � 2 � � � � � � � � $ � � � � � � �

Fig. 16: Schematic of parameter estimation, MIMO aircraft systems



PENSUM � Parameter Esti-mation of Nonlinear Systems Using MATLAB

Recently, more and more re-search work at the Institute is done using the MATLAB lan-guage. In addition, there is re-newed interest in frequency domain identification, mainly in the handling qualities re-search area. Therefore, a tool-box PENSUM was developed which implements Maximum Likelihood parameter estima-tion method and caters to general nonlinear and linear models written in MAT-LAB/Simulink. PENSUM works similar to the software package ESTIMA and currently imple-ments the

• Output error method and • Gauss-Newton optimization.

In addition to handling nonlin-ear systems in the time do-main, PENSUM also allows identification of linear systems in both the time and frequency domain. The linear systems are implemented as MATLAB state space linear time invariant sys-tems and handled using rou-tines from the Control System Toolbox.

DIVA � Dialog Program for Interactive Data Analysis

DIVA is a tool for flight data analysis in the time domain as well as in the frequency do-main. The graphical interface OSF/Motif allowed the creation of a common version - XDIVA - for different operating systems. For making it platform inde-pendent, a new data format, the so-called Institute�s Stan-dard IS3 was introduced, based on the Common Data Format (CDF) of NASA, and with direct access to the data. The dia-logue of XDIVA is written in C, the computing is done with

FORTRAN programs called from C, and the graphics sys-tem is GINO-F.

The main features of DIVA in the time and frequency do-main are:

• Quick look plots, • Time history plots • One and two dimensional

classification, • Auto and cross correlation

functions, • Linear regression, • Spectral analysis, • Frequency response and co-

herence functions, • Transfer function approxima-

tion, • Handling qualities analysis,

and • Multiple input/multiple out-

put (MIMO) analysis.

Besides these main features, DIVA contains Utilities for data pre-processing and generation, and a plot program with free layout. Outside DLR, XDIVA is used by DaimlerChrysler Aero-space and by the integrated team member WTD 61 at Manching [17].

eXpert � A Data Bank for Support of Flight Data Analysis and Parameter Es-timation

The program eXpert is a sup-port tool specially developed for the management of large amounts of maneuver related information and the results of system identification. Other additional features such as plotting are also included. It of-fers various modules to create, manage, and visualize data from the system identification process. The main modules are:

• Database for flight maneu-vers (trim conditions, type of

inputs, classification of ma-neuvers, flight data file de-tails, etc.),

• Database for collation of pa-rameter estimation results,

• Automatic generation of in-put data files compatible with ESTIMA,

• Generation, comparison, and plotting of aerodynamic da-tabase in tables from flight identified parameters,

• Convergence plots of pa-rameter estimates to visual-ize the iteration progress, and

• Online documentation.

eXpert frees the user from rou-tine work. It works under the operating system VMS on an Alpha-station. The data bank uses the DEC-Rdb server and is written in the programming languages C, FORTRAN, and SQL. The graphical user inter-face is based on the OSF/Motif widget set, plotting is done with GINO-F, and the online documentation is available in HTML.

Outlook

System identification will play an ever-increasing role in the sophisticated complex of mod-eling and simulation (M&S) during the flight vehicle design and evaluation phases. The in-tegrated utilization of system identification tools and exper-tise will provide ultimate an-swers to unmasking the modeling deficiencies, reducing developmental risks, and improving flight safety issues. It appears that the present challenge to flight vehicle system identification may be formulated as to determine a high-fidelity aerodynamic model of high performance, highly aug-mented vehicles valid over the entire operational envelope.



Such a global model is, in gen-eral, of unknown structure, highly nonlinear, and affected by elastic structure, unsteady aerodynamics, and erroneous

air data measurements. It is evident from this retrospective that the Institute has been in the forefront of the develop-ment in this field and is in a

position to face the current and future challenges of flight vehicle system identification.

List of References

[1] Koehler, R. and Wilhelm, K., �Auslegung von Eingangssignalen für die Kennwertermittlung�, DFVLR-IB 154-77/40, Dec. 1977.

[2] Jategaonkar, R. V., Mönnich, W., Fischenberg, D., and Krag, B., �Identification of C-160 Simulator Data Base from Flight Data�, Pro-ceedings of the 10th IFAC Sym-posium on �System Identification�, Copenhagen, Denmark, Elsevier Sciences Ltd., IFAC Publications, Oxford, UK, 1994, pp. 1031-1038.

[3] Jategaonkar, R. V. and Mönnich, W., �Identification of Do-328 Aerodynamic Database for a Level D Flight Simulator�, AIAA Model-ing and Simulation Technologies Conference, New Orleans, LA, USA, Aug. 11-13, 1997, Paper No. AIAA-97-3729.

[4] Jategaonkar, R. V., �Identification of the Aerodynamic Model of the DLR Research Aircraft ATTAS from Flight Data�, DLR-FB 90-40, July 1990.

[5] Jategaonkar, R. V., �Identification of Actuation System and Aerody-namic Effects of Direct-Lift-Control Flaps�, Journal of Aircraft, Vol. 30, No. 5, Sept.-Oct. 1993, pp. 636-643.

[6] Fischenberg, D. and Jategaonkar, R. V., �Identification of Aircraft Stall Behavior from Flight Data�, RTO-MP-11, 1999, Paper No. 17.

[7] Rohlf, D., �Direct Update of a Global Simulation Model with In-crements via System Identifica-tion�, RTO-MP-11, 1999, Paper No. 28.

[8] Weiss, S., Friehmelt, H., Plaet-schke, E., and Rohlf, D., �X-31A System Identification Using Single-Surface Excitation at High Angles of Attack�, Journal of Aircraft, Vol. 33, No. 3, May-June 1996, pp. 485-490.

[9] Kaletka, J. and Grünhagen, W. von, �System Identification of Mathematical Models for the De-sign of a Model Following Control System�, Vertica, Vol. 13, No. 2, Pergamon Press, Oxford, UK, 1989, pp. 213-228.

[10] Tischler, M. B. and Kaletka, J., �Modeling XV-15 Tilt Rotor Air-craft Dynamics by Frequency and Time Domain Identification Meth-ods�, AGARD CP-423, 1986, Pa-per No. 9.

[11] Hamel, P. G. and Jategaonkar, R. V., �Evolution of Flight Vehicle System Identification�, Journal of Aircraft, Vol. 33, No. 1, Jan.-Feb. 1996, pp. 9-28.

[12] Hamel, P. G. and Kaletka, J., �Ad-vances in Rotorcraft System Iden-tification�, Progress in Aerospace Sciences, Vol. 33, 1997, pp. 259-284.

[13] Jategaonkar, R. V. and Plaetschke, E., �Maximum Likelihood Parame-ter Estimation from Flight Data for General Nonlinear Systems�, DFVLR-FB 83-14, April 1983.

[14] Jategaonkar, R. V. and Plaetschke, E., �Algorithms for Aircraft Pa-rameter Estimation Accounting for Process and Measurement Noise�, Journal of Aircraft, Vol. 26, No. 4, April 1989, pp. 360-372.

[15] Jategaonkar, R. V., �User�s Man-ual for NLMLKL, a General FOR-TRAN Program for Parameter Es-timation in Time Domain�, DLR IB 111-95/03, March 1995.

[16] Giesemann, P. and Thielecke, F., �Prediction of Gas Concentrations and Temperatures for an Experi-mental Combustion Plant with Local Model Networks�, Proceed-ings of Fourth International Work-shop on �Neural Networks in Ap-plications � NN 99�, Otto-von-Guericke-University Magdeburg, Germany, March 1999, pp. 199-206.

[17] Doherr, K.-F., Wulff, G., and Zöll-ner, M., �Data Gathering, Pre-processing and Analysis�, DLR-Mitt. 93-14, Dec. 1993, pp. 263-288.

art and science of institute of flight research system ... a300-600 3 1982 ... a/c landing gears ......

Documents