Professional Documents
Culture Documents
IVHM Project
www.nasa.gov
Outline
Aviation Safety Program IVHM Project
Sensor Placement Background / Motivation Systematic Sensor Selection Strategy (S4) Methodology Overview Turbofan Engine Application Example Discussion and Summary
www.nasa.gov
Physical Problems
Permitting correction of
Allowing isolation of
Background:
Developed under NASA Space IVHM efforts Provides a systematic evaluation of the available sensor suite relative to the diagnostic requirements Selects sensors (type/location) to optimize the fidelity and response of engine health diagnostics
Iterative Iterative Down Down -Select -Select Process Process Candidate Sensor Suites Candidate Selection Complete No Yes Final Final Selection Selection Collection of Nearly Optimal Sensor Suites Statistical Statistical Evaluation Evaluation Algorithm Algorithm
System System Diagnostic Diagnostic Model Model Sensor SensorSuite Suite Merit Merit Algorithm Algorithm
Architecture Functionality:
Knowledge Base: System simulation Health information Down-select process: Diagnostic model Merit function Down-select algorithm Statistical evaluation: Considers sensor response and system/signal noise characteristics
Down Down -Select -Select Algorithm Algorithm (Genetic (Genetic Algorithm) Algorithm)
Optimal Optimal Sensor Sensor Suite Suite Knowledge Knowledge Base Base
www.nasa.gov
IVHM Project
www.nasa.gov
Non-linear aero-thermodynamic component level model 5 Rotating Components (FAN, LPC, HPC, HPT, LPT) Candidate Sensors:
Baseline Sensors: N1, N2, T25, T3, Ps3, T49, Wf36 Optional COTS Sensors: P17, T17, P25, T5, P5
Diagnostic Module
Aviation Safety Program IVHM Project
#y i =
!i
d=
1 m 2 ( ) " y ! i m i =1
Example Gas Path Fault Signature
Fault Discrimination (Isolation): Applies inverse model approach. Fault hypothesis which produces smallest residual estimation error is inferred to be the fault condition.
Residual measurement agreement metric between actual fault k and hypothesized fault j :
D j ,k = 1 m ~ 2 ! ("y i ) m i =1
Residual measurement agreement metric, Dj,k, values for faults 1-10 (5%)
The down-select algorithm is a genetic algorithm (GA) which applies the following steps: 1. Randomly generates the initial population of sensor suites. 2. Calculates the merit value for each sensor suite. 3. Advances a given number of sensor suites with the highest merit values to the next generation. 4. Next generation is formed by selecting sensor suite pairs using roulette-wheel selection and determining if crossover occurs:
No Crossover Both parents advance to the next generation without modification. Crossover Apply single point crossover with the possibility of mutation as each gene is copied.
Crossover Point Parent1 = G1 1 G1 2 G13 G14 G1 5 G1 6 Parent2 = G2 1 G2 2 G23 G24 G2 5 G2 6 Child1 = G1 1 G1 2 G23 G24 G2 5 G2 6 Child2 = G2 1 G2 2 G13 G14 G1 5 G1 6
5. 6.
Send new generation of sensor suites to Step 2 to reinitiate the merit assignment process. Repeat until target number of generations is reached.
www.nasa.gov
10
A final statistical evaluation is performed to evaluate and rank the top performing sensor suites: Adds sensor noise Other system uncertainty factors
Residual measurement agreement metric ( Dj,k ) Baseline senor suite (blue) vs. Optimal sensor suite (red) Comparison Baseline 7 Sensor Suite 1. N1 2. N2 3. T25 4. T3 5. Ps3 6. T49 7. Wf36 S4 Optimal 10 Sensor Suite 1. N1 2. N2 3. T25 4. T3 5. Ps3 6. T49 7. Wf36 8. P17 9. P25 10. T5
www.nasa.gov
11
S4: Turbofan Engine Application Example Monte Carlo Simulation Confusion Matrix Results (2.5% to 5.0% Faults) Aviation Safety Program IVHM Project
Inferred Fault Condition
1 1 674 0 1 0 0 0 0 0 310 1 0 2 0 1000 0 6 0 0 0 0 0 0 0 3 0 0 627 0 0 19 0 0 0 0 0 4 0 0 0 126 0 0 0 0 0 0 0 5 0 0 0 2 998 0 0 0 0 0 0 6 0 0 25 0 0 843 0 0 0 0 0 7 0 0 0 6 2 0 1000 0 0 0 0 8 0 0 0 12 0 0 0 1000 0 0 0 9 325 0 4 0 0 0 0 0 690 0 0 10 1 0 0 0 0 0 0 0 0 999 0 No Fault 0 0 343 848 0 138 0 0 0 0 10000 Accuracy 67% 100% 63% 13% 100% 84% 100% 100% 69% 100% 100%
2 3 4 5 6 7 8 9 10 No Fault
7 Baseline Sensors
13% miss-detect rate 7% mis-classify rate
2 3 4 5 6 7 8 9 10 No Fault
10 Optimal Sensors
0.8% miss-detect rate 1.3% mis-classify rate
www.nasa.gov
Summary
Aviation Safety Program IVHM Project
An initial application of the Systematic Sensor Selection Strategy (S4) has been applied to a turbofan engine application Demonstrated improved diagnostic performance with selected optimal sensor suite Follow-on efforts will apply updated performance metrics and additional system functionality (fault types, operating scenarios, advanced sensors) Provides a systematic approach towards the evaluation and selection of candidate sensors and diagnostic algorithms
www.nasa.gov
14
IVHM Project
Backup Slides
www.nasa.gov
15
S4: Turbofan Engine Application Example Monte Carlo Simulation Confusion Matrix Results (2.5% to 5.0% Faults) Aviation Safety Program IVHM Project
Inferred Fault Condition
1 1 887 0 7 0 0 0 0 0 56 3 0 2 0 1000 0 54 0 0 0 0 0 0 0 3 0 0 896 0 0 23 0 0 0 0 0 4 0 0 0 618 0 0 0 0 0 0 0 5 0 0 1 18 996 0 0 0 0 0 0 6 0 0 31 2 0 948 0 0 0 0 0 7 0 0 0 23 4 0 1000 0 0 0 0 8 0 0 0 36 0 0 0 1000 0 0 0 9 112 0 7 0 0 0 0 0 944 1 0 10 1 0 4 0 0 0 0 0 0 996 0 No Fault 0 0 54 249 0 29 0 0 0 0 10000 Accuracy 89% 100% 90% 62% 100% 95% 100% 100% 94% 100% 100%
2 3 4 5 6 7 8 9 10 No Fault
10 SubOptimal Sensors
(adds P17, T17, P5)
3.3% miss-detect rate 3.8% mis-classify rate
2 3 4 5 6 7 8 9 10 No Fault
10 Optimal Sensors
0.8% miss-detect rate 1.3% mis-classify rate
www.nasa.gov
IVHM Project
HPT
LPT
T25
T3 PS3
WF N1 N2
PS13 T13
P25
Advanced Sensors
m2
m25
m3
P45 m45
m5
Optimal sensor suite (optional + advanced sensors) Optimal sensor suite (optional sensors) Baseline sensor suite
Optional Sensors Optional and Advanced Sensors