Upload
others
View
1
Download
0
Embed Size (px)
Citation preview
D. Bieniek, R. Luckner – 2 June 2010WakeNet3 - Europe Specific Workshop: Models and Methods for WVE Simulations
Berlin Institute ofTechnology
Probabilistic Pilot Model Approach for Wake Vortex Encounter Simulations
Slide 2D. Bieniek, R. Luckner 2 June 2010
Content
Objective / Motivation
Pilot Model Requirements
Development Process
Model Structure
Types of Simulation Data
Exemplary Results
Conclusions
Slide 3D. Bieniek, R. Luckner 2 June 2010
40 45 50 55 60 65 70-1
0
1Pilot Side-Stick Roll Input [-]
Why is a probabilistic PM needed?Why is a probabilistic PM needed?
Variation observed for:
a) magnitude
b) phase of control inputs
Modelling these variations via a probabilistic pilot model expands worst case simulation capabilities
40 45 50 55 60 65 70-20
0
20Roll Angle [deg]
Scatter in pilot inputs results in scattered A/C response
Severity model approaches are based on parameters of A/C response (e.g. roll angle, roll rate)
Pilot model quality directly affects inputs to severity model
For identical vortex encounters a variation in pilot control inputs can be observed for one single pilot and in-between pilots.
Worst case Worst caseencounter scenario
Worst casepilot reaction
= +
b)a)
Objective / Motivation
Probabilistic pilot model contributes toseverity assessment
Slide 4D. Bieniek, R. Luckner 2 June 2010
WVE Risk AssessmentWVE Risk Assessment
Objective / Motivation
Pilot Model
A/C Simulation Model
PilotInputs
WV Simulation
Model-40-30-20-10010203040
-20
-15
-10
-5
0
5
10
15
20
Lamb-OseenBurnham & HallockProctorWinckelmans
Vortex Disturbance
Time t
0.00
1.000.660.33
crit
Time t
0.00
1.000.660.33
crit
Time t
0.00
1.000.660.33
Time t
0.00
1.000.660.33
crit
Severity CriteriaTime t
0.00
1.000.660.33
crit
Time t
0.00
1.000.660.33
crit
Time t
0.00
1.000.660.33
Time t
0.00
1.000.660.33
crit
Severity Criteria
A/CResponse
„probabilistic“
deterministicdeterministic
orprobabilistic
deterministic
RiskScenario
deterministic
A/C - WVDistance
Pilot Model
Deterministic / Probabilistic
A/C ResponseDeterministic / Probabilistic
Severity RatingDeterministic / Probabilistic
Probabilistic PM contributes to
risk assessment results
Slide 5D. Bieniek, R. Luckner 2 June 2010
The deterministicdeterministic pilot model shall result in pilot control inputs and aircraft reactions that are well-within (< 1σ) the natural, statistical scatter observed in piloted WVE simulations.
Qualitative Requirements
The probabilisticprobabilistic pilot model shall result in pilot control inputs and aircraft reactions with means and variations about the means comparable to the natural, statistical scatter observed in piloted WVE simulation.
time of peak t(x)
peak value x
range around mean valuedefined by standard deviation s[t(x)]
range around mean valuedefined by standard deviation s(x)
(x)t
x
t [sec]
Quantitative RequirementsMeans and standard deviations of first two significant roll axispeaks (magnitude and time of peak) for:
1) Pilot control input2) A/C response
Peaks contribute directly to severity model results
Pilot Model Requirements
25 30 35 40 45 50 55 60 65-2
0
2
SS
RO
[-]
25 30 35 40 45 50 55 60 65-50
0
50
Φ [
°]
25 30 35 40 45 50 55 60 65-50
0
50
p [°
/s]
25 30 35 40 45 50 55 60 65-2
0
2
4x 106
Δ L W
V [
°/s]
time [s]
Peaks contribute directly to severity model results
Slide 6D. Bieniek, R. Luckner 2 June 2010
PilotModel
Criterion
Optimization
PilotPM
OFF
LIN
E
ON
LIN
E
Pilot-in-the-loop Simulator tests
Test evaluation(Time histories &Questionnaires)
Pilot task
Wake vortex
Normal / special procedure
FCOM
Flight Crew Operational Manual
A/C
R
eact
ions
Pilo
tIn
puts
Piloted Model Development
Models of human pilot behaviour (e.g. Crossover Model)
RMSE optimization criteria
Quality assessment via AAM, R, DTW
PMIn
puts
Piloted WVESimulationData Base
e.g.S-WAKECREDOS
Test
/ Va
lidat
ion
(Offl
ine
sim
ulat
ion,
PM
-in-th
e-Lo
op)
Slide 7D. Bieniek, R. Luckner 2 June 2010
Developed by D. McRuer / E. Krendel in 1960ies
Derived from Crossover Law which describes adaptation capabilities of human pilot
Three elements:
Four parameters: KP, TLead, TLag, τe
WVE pilot modelling: generate side-stick roll input (SSRO) from roll angle Φ
Pilot Model Structure
Crossover Pilot ModelCrossover Pilot Model
- Pilot Gain- Lead / Lag Filter- Time Delay
Effects of KP and TLead on Amplitude [1]
Parameters are correlated with respect to their effects on dynamic characteristics of describing function
Parameter correlations may affect numerical optimizations
Correlation of KP and TLead as found in [1]
[1] Source: Airbus
NOTICE!
1jT1jT
Lag
Lead
+⋅+⋅
ωω
Φ SSROPK eje τω ⋅−
Slide 8D. Bieniek, R. Luckner 2 June 2010
-5 0 5 10 15 20-1
0
1Pilot Side-Stick Roll Input [-]
-5 0 5 10 15 20-10
0
10
20Roll Angle [deg]
-5 0 5 10 15 20-4
-2
0
2x 106 Vortex Induced Rolling Moment [Nm]
Free / Forced WVE Simulation Data
Simulation Characteristics
Free evolution of vortex induced disturbances in time depending on A/C flight path
Vortex disturbances comply exactly with simulated A/C motion
Pilot inputs before encounter affect vortex disturbance via flight path
Recorded data includes pilot reactions for a large number of different encounter scenariosbutEach encounter is “unique”, no exact reproducibility
Variation in pilot inputs may be a result of factors other than behaviour (e.g. WVE scenario)
Characteristics w.r.t. PM Development
Suited for deterministicdeterministic pilot model tuning
Complete simulation of WVE
Slide 9D. Bieniek, R. Luckner 2 June 2010
40 45 50 55 60 65 70-1
0
1Pilot Side-Stick Roll Input [-]
40 45 50 55 60 65 70-20
0
20Roll Angle [deg]
40 45 50 55 60 65 70-4
-2
0
2x 106 Vortex Induced Rolling Moment [Nm]
Fixed WVE Simulation Data
Simulation Characteristics
Characteristics w.r.t. PM Development
Vortex induced disturbances are realistic but fixed in time
Vortex disturbances do not comply exactly with simulated A/C motion
Pilot inputs before encounter do not affect vortex disturbance via flight path
Pilot reactions for small number of representative encounter scenarios,butMultiple recordings of pilot inputs for identically reproduced encounters
Pilot inputs only affected by probabilistic behaviour
Suited for probabilisticprobabilistic pilot model tuningSuited for pilot model testing / validation
Pre-recorded time series of vortex disturbances
Slide 10D. Bieniek, R. Luckner 2 June 2010
Parameter Optimizations
Deterministic PM“Multi-Case”: combine all encounters during optimization
Probabilistic PM“Single-Case”: optimization for each recorded encounter
PilotInputs
A/CResponse
PMInputs
Encounter 1 Encounter 2 Encounter n
One set of parameters that gives best overall result for all piloted encounters
Encounter 1 Encounter 2 Encounter n
Distributions of pilot model parameters
KP = 2.63 TLead = 1.73
TLag = 0.04τe = 0.66 sec
KP =TLead =TLag =
τe =
Example :
Pilot Model
Criterion
Optimization WVE Data Base
Slide 11D. Bieniek, R. Luckner 2 June 2010
Deterministic PM Results - Offline
-6 -4 -2 0 2 4 6 8 10 12
-1
-0.5
0
0.5
1
SS
RO
[-]
R
mean = 0.91 P
mean = 0 MRE
mean = 0.627 AAM
mean = 0.663 RMSE
mean = 0.21777 PRE
mean = 0.268
-6 -4 -2 0 2 4 6 8 10 12-30
-20
-10
0
10
20
30
Rig
ht O
uter
Aile
ron
[°]
time [sec] (synchronized, t=0 at large(st) induced rolling moment)
Piloted SimulationsCO Pilot Model
-150-100-50050100150
-40
-20
0
20
40
60
80
wd_
zacv
txm
[m]
wd_yacvtxm [m]
Γ = 565-965 m 2/sec ΔΨWV = 10 ° ΔγWV = -3 °
WVE Start: Square WVE End: Dot
-30 -20 -10 0 10 20 30-3
-2
-1
0
1
2
3x 106
Δ LW
V [N
m]
time [sec] (synchronized, t=0 at large(st) induced rolling moment)
-4 -2 0 2 4 6 8 10 12
-1
-0.5
0
0.5
1
SS
RO
[-]
R
mean = 0.893 P
mean = 0 MRE
mean = 0.671 AAM
mean = 0.62 RMSE
mean = 0.22878 PRE
mean = 0.319
-4 -2 0 2 4 6 8 10 12-30
-20
-10
0
10
20
30
Rig
ht O
uter
Aile
ron
[°]
time [sec] (synchronized, t=0 at large(st) induced rolling moment)
Piloted SimulationsCO Pilot Model
-150-100-50050100150200
-20
-10
0
10
20
30
40
50
60
wd_
zacv
txm
[m]
wd_yacvtxm [m]
Γ = 465-865 m 2/sec ΔΨWV = 15 ° ΔγWV = -3 °
WVE Start: Square WVE End: Dot
-30 -20 -10 0 10 20 30-3
-2
-1
0
1
2x 106
ΔL W
V [N
m]
time [sec] (synchronized, t=0 at large(st) induced rolling moment)
Slide 12D. Bieniek, R. Luckner 2 June 2010
40 45 50 55 60 65-1
-0.5
0
0.5
1
SS
RO
[-]
40 45 50 55 60 65
-20
0
20
Rig
ht In
ner A
il. [d
eg]
40 45 50 55 60 65-20
-10
0
10
20
Φ [d
eg]
40 45 50 55 60 65
-20
-10
0
10
20
p [d
eg/s
ec]
40 45 50 55 60 65-4
-2
0
2x 106
Δ LW
V [N
m]
time [sec]
CO
Deterministic PM Results - Online
Pilot-in-the-loop Simulator tests
PMPM
PM-in-the-loop Simulator tests
PM-in-the-Loop Tests
Test pilot model within dynamic loop (e.g. check for instability problems, offline vs. online behaviour)
Compare pilot model results to piloted simulations (only way to compare results for A/C response)
Validation of pilot model (are requirements fulfilled?)
Acceptable regions / limits as derived from statistical analysis of peaks
Respective peaks of PM results*
Fixed Encounter Simulation
Slide 13D. Bieniek, R. Luckner 2 June 2010
Probabilistic PM Optimizations
Considerations during OptimizationParameters correlations affect distribution results
Avoid correlation affects by reducing number of tuners
Gain sufficient to describe scatter in magnitude
Time delay sufficient to describe scatter in phase
Lead / Lag filter fixed (based on MC optimization)-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
SS
RO
[-]
b)
a)
1jT1jT
Lag
Lead
+⋅+⋅
ωω
Φ SSROPK eje τω⋅−
a) b)
Slide 14D. Bieniek, R. Luckner 2 June 2010
10-1 100 101-40
-30
-20
-10
0
10Bode Plot of CO PM
A(ω
) [dB
]
10-1 100 101
-150
-100
-50
0
50
ω [rad/s]
φ(ω
) [°]
FXE Multi-Case OptimizationFXE Single-Case Optimizations
0 1 2 3 4 5 60
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
KP
Pro
babi
lity
Den
sity
Optimization ResultsGamma Estimation
0 0.5 1 1.50
0.5
1
1.5
2
2.5
3
τe
Pro
babi
lity
Den
sity
Optimization ResultsWeibull Estimation
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6 1.80.5
1
1.5
2
2.5
3
3.5
4
τe
KP
Results modelled by continuous Gamma
probability distribution
Results modelled by continuous Weibull
probability distribution
Exemplary KP / τe optimization results (A330 Departure)
Optimization Results
Probabilistic parameters result in scatter of dynamic characteristics of pilot model describing function
Describing function behaviour is scattered around mean behaviour resulting from Multi-Case optimizations
No significant correlation between parameters KPand τe
Description of optimization results with continuous probability distributions (e.g. Gamma, Weibull)
Possible limitation of parameter values at 2.5 and 97.5 percentiles of cumulative distribution function
Slide 15D. Bieniek, R. Luckner 2 June 2010
Simulation Results
40 45 50 55
-1
-0.5
0
0.5
1
1 1.5 2 2.5 3 3.5 4 4.5 5 5.50
5
10
15
20
25
30
35
40
45
50
KP
Num
ber o
f Cas
es
Distribution of KP during Simulation
0.4 0.5 0.6 0.7 0.8 0.90
5
10
15
20
25
30
35
40
45
τe [sec]
Num
ber o
f Cas
es
Distribution of τe during Simulation
Deterministic PM
Probabilistic PMPeaks
Pilot Side-Stick Roll Input [-]
Exemplary A330 encounters during ILS approach
Bank angle inputs taken from PM-in-the-Loop simulations of these fixed encounter cases
Examples show 500 simulation runs with parameters generated via identified probability distributions
Side-Stick roll inputs of probabilistic PM are scattered around inputs of deterministic (mean) model
Comparison of statistical scatter of peaks with standard deviations of piloted tests has been difficult because of side-stick saturation
Evaluation of resulting scatter in A/C response should provides a more realistic quality assessment (PM-in-the-loop desktop simulation needed) 40 45 50 55
-1
-0.5
0
0.5
1
ti [ ]
40 45 50 55
-1
-0.5
0
0.5
1
Offline Test
Pilot Model Parameters during Simulation runs
*
Slide 16D. Bieniek, R. Luckner 2 June 2010
Conclusions
Probabilistic pilot models may contribute to worst case analysis
Methodology to set up simple probabilistic roll axis WVE pilot models has been introduced
Fast-time capable (model parameters are determined before simulation, similar to other probabilistic parts of Monte-Carlo simulation)
Testing / validation of approach within dynamic simulation still needed
Application to pitch axis is difficult because scatter in pilot behaviour much higher
Is this approach feasible / should it be revised (e.g. other probability distributions)?
Is this approach useful for future Monte-Carlo simulations?
Slide 17D. Bieniek, R. Luckner 2 June 2010
Open Questions / Discussion
Are pilot model optimization / validation criteria accepted by authorities?
What is the necessary level of required pilot model quality to be accepted by authorities?
Are additional single-event simulator tests (one unexpected encounter vs. sessions of 30-40 encounters) necessary to see more realistic pilot behaviour?
Thank you for your attention!Thank you for your attention!Questions?Questions?
Slide 18D. Bieniek, R. Luckner 2 June 2010
This document and all information contained herein is the sole property of TU Berlin. No intellectual property rights are granted by the delivery of this document or the disclosure of its content. This document shall not be reproduced or disclosed to a third party without the express written consent of TU Berlin. This document and its content shall not be used for any purpose other than that for which it is supplied.
The statements made herein do not constitute an offer. They are based on the mentioned assumptions and are expressed in good faith. Where the supporting grounds for these statements are not shown, TU Berlin will be pleased to explain the basis thereof.