NON-LINEAR ENGINE COMPONENT FAULT DIAGNOSIS FROM A LIMITED NUMBER OF MEASUREMENTS USING A COMBINATORIAL APPROACH

Transcription

1 NON-LINEAR ENGINE COMPONENT FAULT DIAGNOSIS FROM A LIMITED NUMBER OF MEASUREMENTS USING A COMBINATORIAL APPROACH N. Aretakis Research associate K. Mathioudakis Associate Professor A. Stamatis Research associate Laboratory of Thermal Turbomachines National Technical University of Athens

2 NON-LINEAR ENGINE COMPONENT FAULT DIAGNOSIS FROM A LIMITED NUMBER OF MEASUREMENTS USING A COMBINATORIAL APPROACH Definition of the diagnostic problem The principle of the method Types of faulty situations The basis of the method: Adaptive Modeling Technique Engine layout Diagnostic approach Application Test Cases Summary - Conclusions 2

3 Definition Of The Diagnostic Problem Objective: Identify component faults, when few measured quantities are available (fewer measurements than unknown health indices ) Solution Approach Examined: Combinatorial approach based on isolating the actual fault identity from all possible solutions (Heuristic data evaluation). 3

4 The Diagnostic Problem Examined Types of Faulty Situations: I. Engine Deterioration: all individual engine components degrades with time II. III. Engine Component Faults: changes in one or more components leading to degradation of their performance Individual Engine Identification: determination of the performance of all individual components of a particular engine The present method is dealing with Problem II 4

5 The Basis of the Method Adaptive Modeling technique Measurements Component Parameters Measurements for monitoring Operating point definition Adaptive Engine Model 5

6 The Principle Of The Method The method will be presented through its implementation on a modern type jet engine: A partially mixed, high bypass ratio turbofan Layout of a turbofan engine and station numbering for positions of interest. 6

7 Typical Monitoring Sets Measurements Component Condition Parameters Measurements for Monitoring Symbol Component Health Parameter Symbol No 2 LP Shaft Rpm HP Shaft Rpm XNLP XNHP Outer Fan Flow Factor Efficiency Factor SW2 SE Fan Outer Pressure HP Compressor Outlet Pressure P 3 P 3 Fan Inner Flow Factor Efficiency Factor SW2 SE HP Compressor Outlet Temperature LP Turbine Outlet Temperature T 3 T 6 HPC Flow Factor Efficiency Factor SW26 SE Fan Outer Temperature Operation Point Definition T 3 HPT Flow Factor Efficiency Factor SW4 SE Ambient Pressure Total Inlet Pressure Total Inlet Temperature Pamb P T LPT Nozzle Flow Factor Efficiency Factor Exhaust Area SW49 SE49 A8 9 4 Fuel Flow Rate WFE SWi W i = Pi T i W i T i Pi ref SE i η = i ηiref 7

8 The Principle Of The Method Measurement Set 7 Measurements Adaptive Model Diagnosis Component Condition Parameters (7 Parameters out of ) Given Set Unique Solution for each Observable Set 8

9 Define the set of possible situations F Apply procedure for all different parameter combinations. (Present Example: All possible combinations = 33) 7 F Derive Fault Signatures for each parameter combination. F Combinations may or may not contain the examined fault. 9

10 Example Situations Actual Fault Combination SW2 SE2 SW2 SE2 SW26 SW4 SE4 Actual Fault Combination 2 Parameter (a) (b) -2 SW2 SE2 SE2 SW4 SE4 SW49 A8 Parameter

11 Statistical analysis of all possible solutions F Obtain parameter distribution from calculated values for different combinations (for every single parameter) SE2 a) Combination Nr Number of Combinations (Frequency) Mean Value SE

12 Produce most probable fault signature F Evaluate mean value and standard deviation for each individual parameter. F Combine them to produce the fault signature 2

13 Produce most probable fault signature Number of Combinations Number o f Combinations SW SE2. F Evaluate mean value and standard deviation for each individual parameter. F Combine them to produce the fault signature SW2 SE2 SW2 SE2 SW26 SE26 SW4 SE4 SW49 SE49 A8 Parameter 3

14 Criteria for Faulty Component Isolation ( st Pass) FA Diagnostic index is derived for each component parameter xi DI x = i σ x i FThe Diagnostic index is derived for each component. 2 2 Diag i = DI i + DI 2i FA faulty component is identified as the one that corresponds to the diagnostic index of maximum value 4

15 2 nd Pass: Improvement of Estimation of Deviations F The procedure is repeated, using only combinations containing the health parameters of the components found faulty from st pass. Best fit approach: F The magnitude of deviations of the parameters of the component found faulty are derived, minimizing a function of the following form. OF= 2 M model measured Yi Yi measured i = σy Y i i 5

16 Block Diagram Of The Method Measurements Isolate Faulty Component: The one with the Maximum DI ( st Pass results) Repeat procedure (-6) using only combinations of parameters of the components found faulty Select Combination of Health parameters Adaptive Modeling 2 3 Sort values Diagnostic Index (DI) for each component The mean values of the health parameters (2 nd Pass Results) Combinations finished 4 Diagnostic Index for each parameter Yes A number of possible deviations for each health parameter 5 6 Mean Value and Standard Deviation for each parameter 6

17 Application Test Cases Percent Deviations of health indices Values for simulating faults Outer Fan Fan Inner HPC HPT LPT Nozzle Fault ID SW2 SE2 SW2 SE2 SW26 SE26 SW4 SE4 SW49 SE49 A8 A B - C D - E - F G - - H - I - J K - L -.6 M N - O 2 7

18 Data sets : LABORATORY OF THERMAL TURBOMACHINES Application Test Cases F Two types of data sets were used: (a) Noise free data in order to examine how the method performs when an exact solution exist (b) Data to which noise was added for a more realistic application Sensor Pamb P T WFE XNLP XNHP P 3 P 3 T 3 T 6 T 3 3 sigma Pa Pa 2 K 2 gr/s 6 rpm 2 rpm 3 Pa 5 KPa 2 K 2 K 2 K F A sequence of different flight conditions (5 sets) simulating real operation was created using the engine model for each examined fault. 8

19 Application Test Cases Overall Identification Results: Fault A B C D E F G H I J K L M N O Noise Free Data Noisy Data : correct detection : incorrect detection FThe proposed method allows the identification of a significant percentage of the examined faults FFaults C (HPC fault) and J (LPT fault) seems to be non-identifiable (observabilityproblem) FThe correct detection of fault C and the failure for fault K for noisy data sets is due to information counterfeiting because of the presence of noise 9

20 Application Test Cases Comparison between st Pass and 2 nd Pass (Fault L, Noise Free Data) Actual Fault Estimated st Pass (a) st Pass Actual Fault Estimated 2nd Pass (b) 2 nd Pass SW2 SE2 SW2 SE2 SW26 SE26 SW4 SE4 SW49 SE49 A8 Parameter 2

21 Application Test Cases Fault A, Noisy Data, 2 nd Pass.5 Actual Fault Estimated (a) Diagnostic Index (b) Fault Signature Diagnostic Index SW2 SE2 SW2 SE2 SW26 SE26 SW4 SE4 SW49 SE49 A8 Parameter 2

22 Application Test Cases Fault G, Noisy Data, 2 nd Pass.8 Actual Fault Estimated (a) Fault Signature (b) Diagnostic Index Diagnostic Index SW2 SE2 SW2 SE2 SW26 SE26 SW4 SE4 SW49 SE49 A8 Parameter 22

23 Application Test Cases Comparison between 2 nd Pass and Best Fit (Fault A, Noisy Data) Actual Fault Best Fit Approach Combinatorial 2nd pass SW2 SE2 SW2 SE2 SW26 SE26 SW4 SE4 SW49 SE49 A8 Parameter 23

24 Application Test Cases Estimation of Deviations (Best Fit approach) Fault Fault A Act. Est. SW2: - - SE2: SW2: SE2: I Act. Est. SW49: SE49: - -. B Act. Est. SE2: SW2: -. L Act. Est. SW49: SE49: D Act. Est. SW26:.2 SE26: - -. M Act. Est. A8: E Act. Est. SW26: - - SE26: -. N Act. Est. A8: - - F Act. Est. SW4: SE4: O Act. Est. A8: 2 2 G Act. Est. SW4: - - SE4: - -. H Act. Est. SW4: SE4:

25 Application Test Cases Estimated fault signatures for Fault C and J when additional Measurements are used (2 nd Pass, Noise Free Data) (a) Adding pressure at HPC Inlet Actual Fault Estimated (b) Also adding pressure at LPT inlet -.2 Actual Fault Estimated SW2 SE2 SW2 SE2 SW26 SE26 SW4 SE4 SW49 SE49 A8 Parameter 25

26 Conclusions A method for diagnosing faults in the components of a jet engine was presented, suitable for cases that a limited instrumentation set is available. The method can both isolate and identify the fault in individual components. The method was tested on an extensive set of component faults, with an instrumentation set typical for today's engines. It was found that it can detect and identify correctly faults in all components, with some limitation for some HPC and LPT faults. It was shown that this weakness is eliminated with the addition of inter-spool measurements, while keeping the measurement set realistic. 26