One-Day Tutorial on

Flight Vehicle System Identification in Time Domain

Dr. Ravindra Jategaonkar

InstituteofFlightSystems
DLR- GermanAerospaceCenter
Lilienthalplatz 7
38108Braunschweig,Germany
Email: [email protected]
Phone: +49531295-2684
Fax:
+49531295-2647
AIAA Professional Development Tutorial, Keystone, CO
24 August 2006
Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/1

This page is left intentionally blank.

Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/2

Contents of Tutorial
Introduction
Overview: Definitions, what is and why SysId, Historical background and Quad-M Basics

Maneuvers
General Aspects, Flight Tests/Maneuvers, Various Types of Inputs

Measurements
General Aspects, Data Compatibility Check, Different Types of Problems

Methods
Model Formulation, Cost Function, Optimization Methods, Numerical Aspects, ...

Models
General Aspects, Linear and Nonlinear Models, Current Trends

Validation
General Aspects, Different Criteria, Procedures, Examples

Examples
C-160 Aerodynamic Data Base,
DO-328 Proof of Match
X-31A High Angle of Attack Modeling,
Phoenix Reusable Orbital Glider
Bo-105 and EC-135 Helicopter: High Bandwidth Models

Software Tools
Fortran and Matlab based
Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/3

This page is left intentionally blank.

Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/4

Objectives and Key Topics of the Tutorial

Overview of key methods of parameter estimation in time domain
Not highly mathematical, rather emphasis on practical utility
Large scale systems
Cover aspects of parameter estimation and model validation
Several examples from real flight data to bring out wide applicability
of time domain methods to highly nonlinear phenomenon
Review some available tools
Goals:
- Better understand the importance of coordinated Quad-M approach
- Get to know the intricacies in the application of time domain method
- Be in a better position to address individual problems
Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/5

This page is left intentionally blank.

Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/6

What is System Identification? (1)

Classification of Problems in System Theory
Inputs
u

State
Equations
.
x = f (x, u, )

Outputs
z/y

Classical problem (Simulation):

given u and f, find y

Control problem:
given y and f, find u

Identification problem:
given u and z, find f

Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/7

What is System Identification? (2)

Dynamic System

Mathematical Model
u

x& (t ) = f ( x (t ), u (t ), )
y (t ) = g ( x (t ), u (t ), )

AIM:
To determine unknown model parameters
such that the model response y matches well
with the measured system response z.
Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/8

What is System Identification? (3)

Technical Definition

Philosophical Definition

System Identification is the determination,

on the basis of observation of input and
output, of a system within a Specified class
of systems , to which the system under test
is equivalent.
Zadeh 1962

Given the answer, what are the questions,

i.e., look at the results and try to figure out
what situation caused those results.
Iliff 1994

SysID: an Inverse Problem

(1) System Identification

Concerned with the mathematical

Structure of a flight vehicle model

(2) Parameter Estimation

Quantifying of parameters for a
selected flight vehicle model

In the commonly used terminology PID appropriate?

Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/9

What is System Identification? (4)

Definitions: Simulation, Parameter Estimation, and System Identification
Dynamic System
Model structure
and parameters
known a-priori

Simulation
Concerned with the
computation of
system responses

Numerical integration
Model structure
and parameters
unknown

Mathematical Model ?
Model
structure
fixed
Parameter estimation
Concerned with the
quantification of
parameter values
Statistical estimation
of parameters

Dr. Ravindra Jategaonkar

System identification
Concerned with the
model structure
determination and
estimation of
parameters

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/10

What is System Identification? (5)

Interdisciplinary Flight Vehicle Modeling
Low Order Dynamics

Higher Order Dynamics

Transfer
Function
(Magnitude)
Model Parameter
Uncertainty

Envelope of
"True" Systems
Model
Structure
Uncertainty
Nominal Model
Possible
"True" System
Frequency

Flight Mechanics Modeling

Flight Control Modeling
Structural Dynamics Modeling
Aeroservoelastic Modeling
Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/11

What is System Identification? (6)

Block Schematic of System Model
u

Dynamic System

Process noise
(turbulence)

Measurement
noise

State Equations
Inputs

x& (t ) = f ( x(t ), u (t ), )

Aircraft mass
characteristics

Measurement Eq.
States

Aerodynamics
(unknown
parameters)

y (t ) = g ( x(t ), u (t ), )

Sensor
locations

Outputs

Sensor model
(calibration factors,
bias errors)

AIM: To determine unknown model parameters such that the model

response y matches well with the measured system response z.
Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/12

Some Fundamental Assumptions

True state of dynamic system is deterministic (unchanging):
- real valued system functions
- iterative experimentation and data analysis converges to the truth

It is possible to carry out specific experiments:

- different modes of dynamic motion (flight vehicle; economic systems?)
- design experiments

Measurements of system inputs and outputs are available:

- directly measured or derived quantities

Physical principles underlying the dynamic process can be modeled:

- model implies mathematical description of the process
- phenomenon purported to underlie the process
- black-box models (Neural networks)

Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/13

Model Characterization (1)

Model characterization: a critical aspect of paramount importance
1) Phenomenological models:

a) Parametric models:

- knowledge based,
- built from basic principles,
- involves physics of the process

2) Behavioral models
- approximate observed behavior,
- no physical meaning

- model structure and order assumed,

- state space models, transfer functions

b) Nonparametric models
- No model structure or order assumed,

Phenomenological
Parameters
Simulation
A priori info
Validity

Dr. Ravindra Jategaonkar

- impulse response, frequency response

Behavioral

physical meaning
no concrete meaning
complex and difficult quick and easy
included
not necessary
large
restricted

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/14

Model Characterization (2)

Three types of models
White Box models
- Derived from theoretical formulation of phenomenon purported
under lie the process under investigation
(Newtonian mechanics, parameters having physical interpretation)
- To reproduce system structure and match the system response

Black-Box models
- Input-Output subspace matching
(Neural networks)
- To Reproduce the system response

Grey-Box models
- Combination of above two models

Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/15

Model Characterization (3)

Parsimonious models
Principle of parsimony (Principle of Simplicity; Ockhams Razor)
The number of entities should not be increased beyond what is
necessary to explain anything.

Methodological principle
- minimizes redundancies and inconsistencies in the model
- helps to determine the best model
- model representation with minimum number of parameters,
yet having fidelity within specified tolerances

Ockham: English theologian and Philosopher, early 14th Century

Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/16

Why System Identification?

Need and quest to better understand the system
- Cause-effect relationship purported to underlie the physical phenomenon

Mathematical models required for:

- Investigation of system performance and characteristics
- Aerodynamic databases valid over operational envelope for flight simulators
- High-fidelity / high-bandwidth models for in-flight simulators
- Flight control law design
- Analysis of handling qualities compliance

Aerodynamic databases from flight data

- Analytical estimates: validity and inadequate theory !
- Wind-tunnel predictions: model scaling, Reynold's number,
dynamic derivatives, cross coupling,
aero-servo-elastic effects !!
Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/17

This page is left intentionally blank.

Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/18

Historical Background (1)

The Apple and Newtonian gravity:
Sir Isaac Newton (1642-1727):
Observed process => model => numerical values

C. F. Gauss (1777-1855):
problem during evaluation of astronomical measurements:
True values q1, q2, ...., qr of physical constants are unknown (trajectory
parameters of a planet). q1, ...., qr are however not measured.
Related parameters are observed, whose true values f1, ...., fr depend
on q1, ...., qr according to some rule: fi = fi(q1, ...., qr ).

Q: Which values of i define the observations at best?

Least Squares method (1795)

Daniel Bernoulli (1700-1782):

The most probable choice between several discrepant
observations and the formation of the most likely induction (1777)
Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/19

Historical Background (2)

Dynamic Response Flight Testing

Time-Vector Method
r

1919 -1923 (Glauert, Norton)

Step input:
- Sand Dropping from Wing Tips
1940's (Milliken)
Steady State Sinusoidal Response
- Circle Diagram: Effective Damping
and Spring Constant
- Analytical Method - Aero. Derivatives

Early 1950 (Seamans)

Pulse Transient Response
- Fourier Transformation
- Electro-mechanical Synthesizer
1950's (Doetsch, Breuhaus, Wolowicz)
Time Vector Method
1960's (Rampy)
Analog Matching
Dr. Ravindra Jategaonkar

-C l r
-

-I xz

r
r

1/2 V Sb r

-C l

p
I xx
2
1/2 V Sb r

-Cl p

p
r

Analysis of Dutch Roll Oscillation:

- Graphical method
- Time invariance of amplitude and phase
between the degrees of freedom
- C l , C lp, C lr : Only two derivatives

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/20

Historical Background (3)

Estimation of Damping-in-Roll Derivative:
NACA Report 167, by F. H. Norton, 1923
Flight Test Technique:
- Load sand boxes on each wing tip,
one pound each, distance to CG 14.7 ft
- Steady flight
- Excitation:
Suddenly release sand in one box,
box emptied in < 0.5 sec
Sand box

- Allow aircraft to roll up to 90 bank

with neutral controls
- Rudder was kicked over and then
other box emptied.
- Tests were carried out in smooth air
- Carefully executed repeat runs
(test to test scatter within 0.01 rad/s)

Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/21

Historical Background (4)

Estimation of Damping-in-Roll Derivative:
NACA Report 167, by F. H. Norton, 1923
Recordings:
- angular rate: electrically driven gyro
- Angular velocity recorder;
was calibrated frequently on a
revolving table, accuracy 0.01 rad/s

Aircraft mass characteristics:

- Sand was weighed out in every
case to within 1%

Methods and Models:

Perceptions of SysID 80 years back !

C-160 Example will be shown later

- Simple basic formula for estimation:

Lp = M / (mass*p) ==> M =150 x 14.7

Results
- Flight estimate 40% lower than the
WT prediction from small oscillations
Dr. Ravindra Jategaonkar

More complex methods and models

in the modern Era 1966-2006

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/22

Historical Background (5)

Simplified-Equations and Analog-Matching Methods
Simplified-Equations Method

Analog-Matching Method

Principle:
For selected types of responses the
effect of only a few coefficients
dominates.

Principle:
Solve equations of motion on
analog computer; manually tune
parameters to match the response
to flight data.

r&
Nr
r

Ruder pulse

p&
p
La
Lp
a
a

Aileron pulse

Lp La

a
p

Aileron step

d
r
L Lr
La a
d
d

Steady sideslip

N 2 + L
d

Dutch roll

Dr. Ravindra Jategaonkar

- limited to a few primary derivatives

- time consuming
- ingenuity of operator

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/23

This page is left intentionally blank.

Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/24

Transition Phase
Classical Approach
1919 - mid 1960's

Late 1960's

Classical
Methods

Modern Era
1966 - 2006
Advanced
Methods

- Deterministic
- Graphical
(Paper & Pencil)

- Statistical analysis

- Frequency domain
- Analog computation

Dinosauric
Digital
Computation

- Time domain
- Frequency domain
- Digital computation

Fortran
Laptops

Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Matlab

Introduction/25

Unified Approach to Flight Vehicle System Identification

Quad-M Basics
Maneuver
Input

Optimized
Input

Actual
Response

Flight Vehicle

Methods
A Priori Values,
lower/upper
bounds

Estimation
Algorithm /
Optimization

Models
Model
Structure

Measurements

Mathematical
Model /
Simulation

Data Collection
& Compatibility

Parameter Estimation
Identification
Criteria

Response
Error

Parameter
Adjustments
Model Response

Identification Phase
Model
Complementary
Flight Data
Validation
Dr. Ravindra Jategaonkar

Validation Phase

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/26

References (1)

Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/27

References (2)
Jategaonkar, R. V.,
Flight Vehicle System Identification: A Time Domain Methodology,
Volume 216, AIAA Progress in Astronautics and Aeronautics Series
Published by AIAA Reston, VA, Aug. 2006, ISBN: 1-56347-836-6
http://www.aiaa.org/content.cfm?pageid=360&id=1447

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.
Hamel, P. G. and Jategaonkar, R. V., The Role of System Identification for Flight Vehicle Applications Revisited, RTO-MP-11, March 1999, Paper No. 2.
Iliff K. W., Parameter Estimation for Flight Vehicles Journal of Guidance, Control, and Dynamics,
Vol. 12, No. 5, Sept.-Oct. 1989, pp. 609-622.
Klein, V., Estimation of Aircraft Aerodynamic Parameters from Flight Data, Progress in Aerospace
Sciences, Vol. 26, Pergamon, Oxford, UK, 1989, pp. 1-77.
Maine, R. E. and Iliff, K. W., Identification of Dynamic Systems, AGARD AG-300, Vol. 2, Jan. 1985.
Maine, R. E. and Iliff, K. W., Identification of Dynamic Systems - Applications to Aircraft. Part 1:
The Output Error Approach, AGARD AG-300, Vol. 3, Pt. 1, Dec. 1986.
Walter, . And Pronzato, L., Identification of Parametric Models, Springer, Berlin, 1997.
Jategaonkar, R. V., (Guest ed.), Special Section: Flight Vehicle System ID - Part 1,
Journal of Aircraft, Vol. 41, No. 4, 2004, pp. 681-764.
Jategaonkar, R. V., (Guest ed.), Special Section: Flight Vehicle System ID - Part 2,
Journal of Aircraft, Vol. 42, No. 1, 2005, pp. 11-92.

Dr. Ravindra Jategaonkar

AIAA Short Course: Flight Vehicle System Identification in Time Domain, Aug. 2006

Introduction/28