Вы находитесь на странице: 1из 20

Validation of Flight Vehicle System Identification

Maneuver
Input

Optimized
Input

Actual
Response

FlightVehicle

Measurements
DataCollection
&Compatibility

Methods
APrioriValues,
lower/upper
bounds

Estimation
Algorithm/
Optimization

Models
Mathematical
Model/
Simulation

Model
Structure

Identification
Criteria

Response
Error

Parameter
Adjustments
ModelResponse

Identification Phase
Complementary
Flight Data

Dr. Ravindra Jategaonkar

Model
Validation

Validation Phase

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

Validation/1

This page is left intentionally blank.

Dr. Ravindra Jategaonkar

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

Validation/2

Data Base Validation (1)


System identification:
Given the system inputs and responses, what is the model?
Model Validation:
How do you know that you got the right answer?
Definition:
Validation refers to the process of confirming that the conceptual
model is applicable or useful by demonstrating an adequate
correspondence between the computational results of the model and
the actual data (if it exists) or other theoretical data.
Broad classification:
1) Statistical properties of the estimates,
2) Residual analysis, and
3) Model predictive quality

Dr. Ravindra Jategaonkar

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

Validation/3

Data Base Validation (2)


Several criteria
1) Standard derivations
2) Correlation among the estimates
3) Goodness of fit
4) Plausibility of estimates (WT data base)
5) Statistical analysis of residuals (bias, variance, covariance, and PSD)
6) Model deficiencies in terms of residual control inputs (inverse simulation)
7) Model predictive capability
"ACID TEST
Simulation and comparison with flight data not used in identification

- Criteria to be used in conjunction with each other


- Basic philosophy remains same for simple as well as for global models
Dr. Ravindra Jategaonkar

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

Validation/4

Statistical Accuracy of Parameter Estimates


Measure of accuracy
Clues into the effectiveness, or lack thereof, of model parameters
Fischer information matrix provides a good approximation to the
parameter error covariance matrix P:
N y (t )T 1 y (t )
k
k
P
R


k =1

Standard deviations (Cramer-Rao bounds):


i =

pii

Correlation coefficients:
i j =

Dr. Ravindra Jategaonkar

pij

pi p j

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

Validation/5

Statistical Accuracy: Practical difficulties


Standard deviations:
- Estimated error bounds are too optimistic
-> use fudge factor of 5-10
(ad hoc approach to bridge the gap between theory and practice).

-> Scatter in the estimates larger than standard deviations


(non trivial issue; scatter can not be avoided)

- Variances of the estimated parameters and the covariances


(off diagonal elements of the error matrix) must be small.
- Correlation coefficients:
-> Linear dependency (> 0.9; >0.95; >0.98..)

Dr. Ravindra Jategaonkar

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

Validation/6

Residual Analysis: Goodness of fit


Cost function value:
- ML estimation: determinant of R (covariance matrix of residuals)
- Most direct way to evaluate model quality
- Cost function value should be small:
Value depends on the number of system outputs, noise level, and units
of the variables (radians or degrees; g or m/s2; m/s or kts)

- Absolute value alone of limited use


- No consistent workable criteria has been put forward
- Old adage: Goodness of Fit is no criteria.
It is a necessary but not a sufficient condition.
Peculiarity of cost function:
Very low value does not guarantee equally good matching of all variables;
In worst case a perfect match of one variable yields minimum.
Dr. Ravindra Jategaonkar

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

Validation/7

Residual Analysis: Overall Fit and


Decomposition of Fit Error
Theils inequality coefficient
TIC =

2
z( t ) y( t )
1 N

k
N k=1 k
2
z ( t )
1 N
+

N k=1 k

2
y( t )
1 N

N k=1 k

Ratio of the root mean square fit error and the root mean square
values of the measured and estimated signals summed together
TIC is normalized: between 0 and 1
0 --> perfect fit
1 --> worst case ==> two time series are very significantly different
Thumb rule: Acceptable TIC value depends on the application, in
general 0. 25 to 0.3 indicates a good agreement
Dr. Ravindra Jategaonkar

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

Validation/8

Residual Analysis: Decomposition of Fit Error


Bias proportion:

U iM

( z i yi ) 2

(1 / N )

measure of systematic

[ zi (t k ) yi (t k )]2

error in the identified model

( zi y ) 2
i

represents the models ability

[ zi (t k ) yi (t k )]2

to duplicate the variability in


the true system

k =1

Variance proportion:

U iS

(1 / N )

k =1

2 (1 i ) z i yi

Covariance proportion: U iC =

(1 / N )

[ zi (t k ) yi (t k )]2

k =1

measure of non-systematic
error

U iM + U iC + U iS = 1

zi and y i denote the mean vales;

and the standard deviations and correlation coefficients.

Bias and variance proportions should be very small (typically less than 0.1),
in an ideal case, zero;
Covariance proportion should be close to one.
Dr. Ravindra Jategaonkar

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

Validation/9

Inverse Simulation
Simulation:

given inputs u and system model f, find system output y

Inverse
simulation

Process of calculating desired controls, for the given


system model and response.
Input

Aircraft

u
u

Feedback
controller
Mathematical
model

Measured Response

z
+
_

y
Computed Response

Closely similar to the control problem (namely, given system model f and
output y, find control input u).

Subtle difference:
Classical control problem does not require measured system responses.
Inverse simulation explicitly needs measured responses, and leads to
controls based on residuals.
Dr. Ravindra Jategaonkar

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

Validation/10

Model Plausibility (1)


Example: Control surface effectiveness - ATTAS VFW-614
Elevator
-1.0

-.16

/rad

/rad

-1.2

-.20

Cm

Aileron

Rudder

-.14
/rad

Cn
-.16

Cl
-.24

-1.4

-.18
-1.6

-.28
0
10
20
Angle of attack, deg

0
10
20
Angle of attack, deg

-10
0
10
Angle of sideslip, deg

Wind tunnel / analytical prediction


Estimates from flight data
Dr. Ravindra Jategaonkar

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

Validation/11

Model Plausibility (2)


Damping Derivatives - ATTAS VFW-614
Short period motion

Rolling motion

Dutch roll motion

-4.0

-.80

-.10

/rad

/rad

/rad

Cm.
Cn.

-.90

-6.0

Clp

-.40

-1.0

-8.0

Cmq
-10.0

Cnr
0

0.2

0.4

0.6

-1.1

Mach number

0.2

0.4

Mach number

0.6

-.70
0

0.2

0.4

0.6

Mach number

Wind tunnel / analytical prediction


Estimates from flight data
Dr. Ravindra Jategaonkar

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

Validation/12

Model Plausibility (3)

Eigenvalues and eigenvectors completely characterize


the system behavior

Eigenvalue = ( -1.61196 , + - 0 )

Linearized system matrix

0.3

0.2

WK

THE
UK

VK

PSI

PHI

Eigenvalue = ( -0.965076 , + - 1.75372 )

0.1

Dutch roll

UK

WK

THE

VK
Q

PSI

PHI

Im [rad/s]

Eigenvalue = ( -0.142527 , + - 1.54814 )

Short period
1

Dr. Ravindra Jategaonkar

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

UK

-0.2

WK

-0.4

THE

Re [1/s]

-0.6

-0.8

PSI
VK

-1

PHI

-1.2

-1.4

-1.6

Validation/13

Model Predictive Capability (1)


Proof-of-Match (POM):
An important part of flight simulator certification and acceptance.
Compare the flight measured system responses with those predicted
by the model for the same (identical) control inputs.
In this POM process, the identified (aerodynamic) model is kept fixed.
Other important issues:
1) Proper choice of the data set to compare against,
2) Initial conditions on the state variables (trim), and
3) Criteria to check adequate correspondence between
model response and measured aircraft outputs.

Dr. Ravindra Jategaonkar

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

Validation/14

Model Predictive Capability (2)


Proof-of-Match (POM):
Data set: Complementary flight data, i.e., flight maneuvers not used
in the estimation the of aerodynamic (Acid test)
FAA has defined a set of roughly a little more than 100
different cases, covering different modes of aircraft motion,
and configurations
Initial conditions: Ideally, the simulation is to be started from the
same initial conditions as in the flight
Criterion: FAA tolerances (to avoid subjective evaluation)
Non trivial task: Measurements corrupted by noise
Presence of turbulence
Modeling errors
Allow small biases on the selected initial conditions and on the
measured control deflections
Dr. Ravindra Jategaonkar

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

Validation/15

Model Predictive Capability (3)


Stand-alone and integrated models
Measured
Aircraft
Motion
Variables

Reversible
Flight Control
Measured
Dynamics

Control
Surface
Deflection

Measured
Control
Surface
Deflect.

Rigid Body
Dynamics

Aircraft
Motion
Variables

Pilot Input
Forces

Flight controls stand-alone

Pilot
Input
Forces

Reversible
Flight Control
Dynamics

Rigid-body stand-alone

Control Surface
Deflection

Rigid Body
Dynamics

Aircraft
Motion
Variables

Integrated model
Dr. Ravindra Jategaonkar

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

Validation/16

Model Predictive Capability (4)


Validation test 2c11: Short period dynamics
10
deg

Time domain verification:


Elevator
deflection

Majority of VT in time domain


e.g., 1.5/s for rates,
0.1 g for accelerations

0
-5
15

deg/s

Pitch rate

Few are in terms of damping


ratio and frequency; e.g.,
0.02 damping ration,
10% period

-10
0.5
g

Recent effort:
Frequency domain criteria

Vertical
acceleration

-2.5
0

Dr. Ravindra Jategaonkar

2.5

time

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

7.5

sec

10

Validation/17

Model Predictive Capability (5)


Frequency domain criteria
Bode plots of measured to
model estimated response
e.g., qm/q (error function)

Validation test 2c11: Short period dynamics


10
dB

Magnitude 0

Ideally, for a perfect match:


0 dB magnitude and
0 deg phase angle
over the frequency range
Boundaries are based on the
LOES (Low Order Equivalent
System) mismatch criteria:
unnoticeable dynamics
Brings our more clearly the
range of model applicability
Important for high authority
FCS where aeroservoelastic
effect may be dominant
Dr. Ravindra Jategaonkar

-5
-10
40
deg

Phase
-40
-80
0.3

10

Frequency

rad/s 20

More restrictive tolerance band in the range of 1 to 5 rad/s


pilot cross-over frequency, allows more error outside of it
(LOES phase-lead leniency), and are asymmetric at low
frequencies (allow more phase lead error than phase lag
error) to have better fidelity with respect to lag error which
might cause pilot-induced oscillations.

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

Validation/18

Model Predictive Capability (6)

From Identified model:

P / AILERON and P / AILERON


dB

Linearized system matrices A, B, C, D

5
0

-10

Frequency response matrix

Magnitude

G( j ) = C ( jI A )1 B + D

-20
-30

-180
deg

From measured flight responses:


Approximation of frequency responses
through FFT techniques

-240

Phase
-300

-360

Noisy measured data: smoothing

Dr. Ravindra Jategaonkar

Frequency

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

10

rad/s

Validation/19

References
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.
Hodgkinson, J. and Mitchell, D., Handling Qualities, in Flight Control Systems, Pratt, R. W. (Ed.),
AIAA Progress in Astronautics and Aeronautics Series, Vol. 184, 2000, Chapter 4.
Jategaonkar, R.V., Determination of Aerodynamic Characteristics from ATTAS Flight Data Gathering for
Ground-Based Simulator, DLR-FB 91-15, 1991.
Murray-Smith, D. J., Methods for the External Validation of Continuous System Simulation Models:
A Review, Journal of Mathematical and Computer Modelling of Dynamical Systems, Vol. 4, No. 1,
1998, pp. 5-31.
N.N., Airplane Simulator Qualification, FAA Advisory Circular, AC 120-40C,Interim Version, Jan. 1995.
N.N., Joint Aviation Requirements - Aeroplane Flight Simulators, JAR-STD 1A, Westward Digital Ltd.,
Cheltenham, England, April 1997.
Tischler, M. B., System Identification Methods for Aircraft Flight Control Development and Validation,
NASA TM 110369, Oct. 1995.
see also the References from the Section Examples
Dr. Ravindra Jategaonkar

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

Validation/20

Вам также может понравиться