Você está na página 1de 37

MOSCOW AVIATION INSTITUTE

MOSCOW AVIATION INSTITUTE EXPERIENCE ON PILOT-IN-THE LOOP INVESTIGATIONS Prof. A.V. Efremov, Ph. D, Dean of Aeronautical school

Manching, EADS, Germany, 11-13 November, 2008

Content
1. MAI fundamental investigation on pilotvehicle system 2. MAI applied results in manual control area

MAI fundamental investigations on pilotvehicle system (PVS)


Specification PVS parameters

Display

Controlled element dynamics

Goals: 1. Development of reliable basis


database of knowledge ; technique; models for investigation of single loop, multiloop, multimodality , systems (stationary and unstationary).

2. Use of the basis for solution of applied manual control tasks:


flight control system design; display design; FQ prediction

Developed technique for experimental investigations:


Unified Fourier coefficient technique
(frequency, spectral, integral characteristics of pilot, closedloop, openloop system )

Technique for preliminary definition of CooperHarper scale metrics in groundbased investigation

Experience in pilotvehicle system investigations exposed the following problems:

limited potentialities of wellknown mathematical pilot models for description of experimental data received with real input spectrums and aircraft dynamics considerable influence of different factors and task variables on ground based evaluation of pilotrating and on PVS characteristics

Examples 1. Pilots adaptation in low frequency range


W ,
dB

40 20

S ii
0 -20 0,1

( 2 +

i2 )

i = 1.5 sec i = 0.5 sec


1,0 10

-1 -1

, sec-1

2. Pilots ability to generate complicated actions in crossover frequency range


W ,
dB

W ,
dB

40 20 0

40 20 0

-20 0,1

LAHOS 1.4 LAHOS 2.10 1,0 10

, sec-1

-20 0,1

= 0 sec = 0.3 sec 1,0 10

, sec-1

d lg Wp d lg
C

40 dB

sek
6

3. Disagreement between groundbased and inflight simulation


ground

PR

averaged results
10

2-B 2-1 2-5 2-7 2-8 3-D 3-1 3-3 3-6 3-8 3-12 3-13 4-1 4-2 5-1

PRg

disagreement between the results in I and III levels of pilot ratings, decrease of pilot rating interval PR = PRworst PRbest in groundbased simulation, decrease of sensitivity of flying qualities estimation in ground-based simulation to FQ change.
PR

PRf
1 2 3 4 5 6 7 8 9 10

5-9 5-10 5-11

4. Influence of motivations (requirement to the accuracy on PR)


d [sm] r [dB] PR 0.5 8.15 8.5 1.0 7.53 8.0 2.0 2.3 3.5 e

flight

5. Influence of additional channel(s): singleloop task PR = 2 in multiloop task PR = 5

Problems and tasks: determination of optimal aircraft dynamics; exposition of regularities in pilot evaluation of Flying qualities; understanding of complicated behavior and ways for its simplification; definition of rules for taking into account the different factors.

Solution of problems and tasks


Optimization of aircraft dynamics for each piloting task
task

Technique for definition of Wc opt


Wcopt (task , PL)

Pilots limitations (PL)

Application of optimal aircraft dynamic 1. 2. 3. Development of criteria for prediction of flying qualities. Agreement between groundbased and inflight investigations. Flight control system design.
9

Regularities in FQ evaluation
1. Agreement between CooperHarper pilot rating (PR) and WeberFechner
9 7 5 3 1 1 2 3 4 5
PR PR=1+5.36 ln( d )
variability PR

Data base: 1. Neal Smith 2. Have PIO 3. LAHOS

2. Pilot workload and pilot-vehicle system parameters correlated with PR.


r= r rW
Copt

PR = f (r , p )

normalized resonance peak of closed loop system

p = max p p WCopt WC

B p << p
w

10

3.Relationship between CHPR and PIOR

PIOR = 0,5PR + 0,25


PR = 3,5 PR = 6,5 PR = 9,5 PIOR = 2 PIOR = 3,5 PIOR = 5,0

11

4.Evaluation of FQ in multichannel task


PR

PR = max( PR , PR )
12

5. Distribution of PR
Pilot actions variability pilot rating variability
PR
10 9

PR random value
max PR = 3 5

Ex. No 1 PR = 6

Wc1

8 7 6 5 4 3 2

3D 4.1 5.11

3.13

Peculiarities of random value PR: PR whole number PR a number contained in the limited set of numbers

5.1

3.3

Ex. No 2 PR = 9

Wc1

Configuration

Conclusion: Random value PR has to be characterized by binomial law p(PR) = C9PR1pPR1(1 p)10 PR C9PR1 = 9! (PR 1) ! (10 PR) ! (PR 1) (10 PR) 9 p= PR 1 9

PR =

13

EXPERIMENTAL TEST ON POSSIBILITY TO USE BINOMIAL LAW FOR DESCRIPTION OF PILOT RATING p(PR) Configurations Number of experiments

PR

2.1 22 2.86

4.1 22 2.75

3.8 24 3.1

3.8 20 3.7

3.12 19 6.4

5.10 17 7.35

Total

124

1.8 1.6 1.4

PR

Binomial law Experiment

1.2

1 0.8

0.6

0.4 0.2

0 1 2 3 4 5 6 7 8 9 10

PR

14

Understanding of reasons of complicated pilot behavior


W c
W 1

W vis p

nm

Wc

W ad p

aW

W pe ( s ) = W pvis ( s )

Wnm ( s ) 1 + W pad ( s )aW f ( s )Wnm ( s )

Measurements of a set of characteristics

W pvis ( jk , j ); Wpad ( jk , j ); W pe ( jk , j )

Exposed regularities: 1. Pilot uses additional cues 2. He does it more actively when W f = W1 , where W1 WC = WC W1

15

Mathematical modeling
a. Modified structural approach

New features:
complicated form of FVIS the different procedure for the choice of parameters: c, FVIS, FPF dependence of neuromuscular system on PVS task variables taking into account pilot remnant
16

b. Modified optimal control model


Disturbance u(t) Vehicle dynamics x(t) Display y(t) = Cx (t) + Du (t)

U c(t) 1 TN s + 1 V u(t) L
*

x(t) Predictor

x(t - ) Kalman estimator

Y p(t) Time delay V y(t)

Human operator model

New features:
modified cost function recommendation for the choice of weighting coefficients modified model of remnant spectral density
17

Composite approach to pilot modeling based on Neural network


Stages for development of Composite model:
development of pilot neural network models (NNM) W p ( j ) Wc ( j)

} {

development of composite model based on pilot NNM allowed to predict PVS characteristics

18

Stages for development of neural network pilot model


1. Selection of technique and definition of parameters for pilot training: NNP /MATLAB, inverse distribution technique 2. Definition of training set 2400 points 3. Definition of model structure: architecture type of model: Time Delay Neural Network (TDNN) type, set of inputs for model: for linear Wc ei(t), ci(t), yi(t), for nonlinear system e(t), c(t), y(t), numbers of layers, numbers of neurons, types of neuron actuation functions: for linear controlled element dynamicsactuation function is linear (F = 1), for nonlinear controlled element dynamicsactuation function is nonlinear.
19

General structure of pilot neural network model


e e ( t-0.25) e ( t-0.40)-e ( t -0.45) e ( t-0.45)-e ( t -0.50) y 1 T ys +1 y* y *( t- y )- y *(t - y -0.10) y *( t- y -0.10 )-y *(t - y -0.20) y *( t- y -0.20 )-y *(t - y -0.30) w1 w2 w3 w4 w5 w6 -b
W

c ( t)

( wi , b) = f (W c )

Comparison of mathematical modeling with experiment


4 experiment model

20 15 10 5 0

10 0 -50

-1

10 (1/c)

10

+1

-2

-100 -150 -200 -250 -1 10


experiment NNPM optimal model structural model

-4

10

15

20

25 30 time ( c)

35

40

45

50

10 0 (1/c)

10 +1

20

Composite approach for prediction of pilot dynamics response


1.Selection of the configuration Wck(j) Wcm(j) {Wci(j)} close to Wc(j):
J=

2 [ci (k ) c (k )]2 [ Aci (k ) Ac (k )] + 180 k =1


Ak (i ) Am (i ) I Ak (i ); I Ak (i ) I Am (i )

2.Calculation of composite pilot model WP corresponding to WC


Ap (i ) = Ak (i )

I A x (i ) = Ac x (i ) Ac (i ) I F x (i ) = Fc x (i ) Fc (i )

p (i ) = k (i )

k (i ) m (i ) I F k (i ); I F k (i ) I F m (i )

3. Development of pilot neural network model ei (t ) A( ) Inverse ci (t ) ( ) Fourier yi (t )


transform
40 20 0 -20 -40 -1 10
(experiment) 10075 ( ) (experiment) 24075 ( )

Neural Network model

10 (1/c)

10

+1

50 0 -50 -100 -150 -200 -250 10- 1

experiment interpolation composite model

21
10
+1

10 (1/c)

MAI investigations on applied manual control tasks

22

I. DEVELOPMENT OF CRITERIA FOR PREDICTION OF FLYING QUALITIES (FQ) AND PIO TENDENCY 1. Criteria the requirements to pilot workload and pilot-vehicle system characteristics Potentialities: Prediction of FQ level 2. Criteria the requirements to FQ by calculation of PR Potentialities: the possibility to define a value of PR for the selection of FQ FIRST TYPE OF CRITERIA Criteria for prediction of FQ and PIO tendency in longitudinal angular motion Criteria for prediction of FQ in longitudinal path motion (refueling task)

-2

Pilot phase compensation

Definition of r and W: Experiment Mathematical modeling (optimal or structural approach to PVS modeling 23

Criteria as a requirements to HQSF


Visual Block Central Processing Time Delay Neuromuscular System Controlled Object

e -

ne

FVIS

e0s

FNM
Proprioceptive Feedback

FAC

Um

HQSF =

Um 1 ( j ) C KL

FPF

Modified levels of HQSF


6 5 4 HQSF 3 2 1 0 0 1 2 3 4 5 6 7 8 9 10 , rad/sec

Original levels of HQSF


15 deg/sec 150 deg/sec
6 5 4

Level 2

Level 2

Level 1

HQSF

3 2 1 0 0 1 2 3 4 5 6 7 8 9 10 , rad/sec

Level 1

24

SECOND TYPE OF CRITERIA

Criteria for FQ prediction in pitch tracking task Exposed regularities


11 PR 9
II

PR = max (PRa , PRb ) PRa = F (a); PRb = F (b)


I group of configurations: PR = f(e) II group of configurations: PR = f(pilot workload)
0
e)

P R= 11 *(1+ Ln( e))

5
I

1 -1.2 -1.1 -1 -0.9 -0.8 -0.7

.2.1

Structural pilot model Wp(j)


PR = 11 ( 1 + ln ( 0.4 + 1.68 e )
PR
5 4.5 4

Criteria PR = max (PR, PR) Prediction of PR by mathematical modeling


-0.6 -0.5 -0.4 -0.3 -0.2 -0.1 L n(

PR = 0.11 (14 + )

PR = 11 ( 1 + ln ( 0.052 + 1.126 e)
PR 4.5 4 3.5 3

Optimal pilot model

PR = 0.11 ( 0.952 p )

1b 3.5 3 2.5 2 1c 1d 2d 2c 2f 7c

3.5
2_1

2.7 2.5 2.3 2 1.5 2

3d 4_1 5_1

1.5

1
1 1 1.5 2 2.5 3 3.5 4 4.5 5

1.5

2.5

3.5

PR 4.5

25

PR

DEVELOPMENT OF CRITERIA FOR PREDICTION OF FQ IN LATERAL CHANNEL Problem: Disagreement of FQ requirements developed in ground and in-flight simulation

From: J.R.Wood AIAA-83-2105

Reason: Lateral acceleration caused by rotation

Suggestion: PR = f [PRacc, PR vis]

PR =max[PRacc, PR vis]

PRvis = f (ln ) PRvest = f (ln n y )

26

CRITERIA FOR PREDICTION OF FQ IN DUALCHANNEL CONTROL TASK


PR = max PR , PR PR , PR ratings from dualchannel system investigations

Experiment
opt ( ) , ( )

( PR, =1 +5. 36 ln opt )


( )

Modeling

from pilot optimal control model corresponding to dualchannel system


9 8 7 6 5 4 3 2 1 1 2 3 4 5 6 7 8 9 10

exp PR 10

mod PR 27

II. AGREEMENT BETWEEN GROUNDBASED AND INFLIGHT INVESTIGATIONS ON FLYING QUALITIES ESTIMATION

10

2-B 2-1 2-5 2-7 2-8

PR
8

PR=PR-PR PR = PRworst PRbest

Calspan

3-D 3-1
6

W L

3-3 3-6 3-8 3-12 3-13 4-1 4-2 5-1


2 5

NASA VMS
4

MAI

5-9 5-10
1

1 1 2 3 4 5 6 7 8 9 10

5-11

0 no motion motion LAMARS MS-1 TL-39 In-flight

Reasons of disagreement In the first level of PR ) a noise of estimation process due to inaccurate simulation of the different factors of flight, b) The wrong (absence) instructions about the Cooper-Harper metrics In the third level of PR inability to simulate the stress situation typical for 3 level 28

THE WAYS FOR ACHIEVEMENT OF AGREEMENT Definition of Cooper-Harper scale metrics on base of developed technique for calibration, Simultaneous estimation of PR in longitudinal and lateral channels, Increase of 3D objects on simulated visual scene (for the landing task)
9 7 5 3 1 1 ddes dad

PR

Workstation
PR=1+5.36 ln( d )
- variability PR

d ad

des dW

dopt

d des FromWFL d opt

53

d ad d ad d opt

d, sm

Without metrics
PR . . 10
2_1

With full set of metrics


PR . . 10 9 8 7 6 5 4 3 2 1 1 2 3 4 5 6 7 8 9

9 8 7 6 5 4 3 2 1 1 2 3 4 5 6 7 8 9

2-1 2-5 2-7 3-D 3-6 3-8 3-13 4-1 4-2 5-1 5-9 5-10

2_5 2_7 3d 3_6 3_8 3_13 4_1 4_2 5_1 5_9 5_10

10 PR

PR10

Result : increase of PR interval from PR=2.5 up to PR=8

29

THE AGREEMENT BETWEEN IN FLIGHT AND GROUND BASED INVESTIGATIONS FULFILLED ACCORDING THE DEVELOPED TECHNIQUE

Landing
Longitudinal error X Desired Adequate 75 m 150 m Lateral error Y 1.5 m 78 m Touchdown velocity VTD 1.5 m/s 2.5 m/s

Refueling
Longitudinal error X Desired Adequate Lateral error Y Contact velocity 0.9 1.4 m/s 0.5 1.8 m/s

Less then 40% radius of basket Less then 60% radius of basket

Aimtoaim tracking
Angular error Desired Adequate 5.0 mrad 1.5 mrad

30

III. Means for improvement of pilot actions and FCS system conjunction Goal: To suppress exposition of flight control system limited potentialities
WAYS FOR SOLUTION OF PROBLEM:

SYNCHRONIZED PREFILTER
LOGIC OF SYNCHRONYZED PREFILTER TO SYNCHRONIZE PILOT ACTION AND FLIGHT CONTROL WITH LIMITED POTENTIALITIES BY LINEARIZATION OF PILOTAIRCRAFT SYSTEM CHARACTERISTICS
1.
max &

law 1: quick change of K f 1/s law 2: restoration of initial gain coefficient K f


o

Kf 2.

max &
o

Kf

MANIPULATOR WITH VARIABLE STIFFNESS


1

Kf

1/p

W1 nonlinear standard prefilter Additional force regulation law

P T p + (1 a ) = PX X T + 1

31

Effect of synchronized prefilter


&
emax

= 30 deg/sec

PILOTAIRCRAFT SYSTEM CHARACTERISTICS

e2 error
[sm2]

r resonance peak

BEFORE FAILURE AFTER FAILURE (st. pref) AFTER FAILURE (sync. pref)

& without failure ( emax = 80 deg/sec )

& with failure ( emax = 30 deg/sec )

& with failure ( emax = 30 deg/sec )

PR normal = 34

PR st. pref = 9

PR sync. pref = 45

32

Effect of manipulator with proposed regulation of force


After failure

& (limiter output) Px=const

& (limiter output) variable stiffness

Control surface deflection


Px=const

P X = const
variable force

variable stiffness

33

IV. Some aspects of direct lift control use


Direct lift control surfaces flapperrons, canards

Direct lift control allows to: improve short period dynamics conserve flying qualities in case of FCS failure to suppress the speed instability

34

INVESTIGATION OF DLC EFFECTIVENESS IN REFUELING

Improvement of accuracy

Without DLC

Aircraft with DLC

35

Effectiveness of direct lift control in carrier landing


DLC allowes to suppres speed instability in carrier landing (because =const)
Without DLC With DLC

Results of experiments
PR 10 8 6 4 2 0 Without DLC With DLC

Touchdown
2 , 2 X
50 40 30 20 10 0

47 3.5

36

V. Additional information on path angle

Without

With

Integration of DLC and path angle indication gives an improvement of performances up to 20 30%

37

Você também pode gostar