Proceedings of ASME Turbo Expo 2014 June 16
Proceedings of ASME Turbo Expo 2014, June 16 – 20, 2014, Düsseldorf, Germany, GT 2014 -26443 TURBOFAN ENGINE HEALTH ASSESSMENT FROM FLIGHT DATA N. Aretakis – I. Roumeliotis – A. Alexiou – C. Romesis – K. Mathioudakis LABORATORY OF THERMAL TURBOMACHINES National Technical University of Athens, Greece
Contents q Introduction • Scope of the paper • Test case description q Measurement analysis for diagnostic purposes q Engine adaptive model • The PROOSIS platform • Model creation process q Diagnostic methods application • Data pre-processing • The Probabilistic Neural Network (PNN) • The Deterioration Tracking Method q Summary-Conclusions GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 2
Contents Ø Introduction • Scope of the paper • Test case description q Measurement analysis for diagnostic purposes q Engine adaptive model • The PROOSIS platform • Model creation process q Diagnostic methods application • Data pre-processing • The Probabilistic Neural Network (PNN) • The Deterioration Tracking Method q Summary-Conclusions GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 3
Scope of the paper v Over the years various GPA diagnostic methods have been proposed, ranging from simple trend analysis-based methods to more advanced model-based methods. v Although the potential of many of these methods has been demonstrated, real-world applications from on-wing data are sparse. v The efficiency of diagnostic methods, as a function to their complexity and/or the accuracy of the available models is also under question. v This work presents the application of several approaches -where engine model of different detail are considered- to engine health assessment, using on-wing data obtained from an aircraft engine. GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 4
Test Case Description High bypass ratio turbofan engine of a commercial short-range aircraft Focus on MCR data, since: Altitude ü significantly more populated ü absolute humidity is negligible ü closer to quasi-steady state P 2 P 125 P 3 P 49 T 25 T 3 EGT N 1 Cycles N 2 ACC pos. Wf SVA pos. ALT [km] MN OPR BPR FN [k. N] SFC [g/(k. N·s)] 11 0. 78 28. 5 5. 7 19. 5 16. 9 1100 cycles ~ 1 year of operation Time On-wing engine performance monitoring system provides snapshot data for T/O and MCR operating points GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 5
Test Case Description Breakdown of test case engine operational environment Category Airports 15% 14% A Dust producing regions B Regions with significant industrial pollution and chemicals C Common regions, not classified as A or B 71% The regions an engine operates have significant effect on its performance degradation Engines operating by more than 70% through category C airports are expected to have lower degradation rates compared with operation through other regions GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 6
Contents q Introduction • Scope of the paper • Test case description Ø Measurement analysis for diagnostic purposes q Engine adaptive model • The PROOSIS platform • Model creation process q Diagnostic methods application • Data pre-processing • The Probabilistic Neural Network (PNN) • The Deterioration Tracking Method q Summary-Conclusions GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 7
EGT[K] Measurement analysis for diagnostic purposes T 2=259 K T 2=245 K T 2=240 K Exp. Avg. a=0. 85 0 200 400 600 Cycles 800 1000 1200 EGT variation with engine cycles (Raw Data) GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 8
Measurement analysis for diagnostic purposes Corrected Measurements EGT[K] 95 K T 2=259 K T 2=245 K T 2=240 K Raw Measurements Exp. Avg. a=0. 85 0 200 400 600 Cycles 800 1000 1200 Raw and corrected EGT variation with engine cycles GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 9
Measurement analysis for diagnostic purposes EGTcor[K] ‘Deltas’ of measurements Y is the measured quantity 10 K 85 90 N 1 cor[%] 95 Y 0 is the reference value of Y measured quantity 100 Operating conditions effect is diminished by using the changes of measurements against their reference value GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 10
EGT[K] Measurement analysis for diagnostic purposes EGTcor[K] Linear(EGTcor[K]) 0 200 400 600 Cycles 800 1000 1200 Corrected EGT variation with engine cycles GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 11
Measurement analysis for diagnostic purposes 6 5 ΔEGT[%] 4 3 2 shift 3 1 shift 2 0 ΔEGT[%] Exp. Aver. a=0. 85 -1 -2 Linear 0 200 400 600 Cycles 800 1000 1200 Corrected ΔEGT variation with engine cycles GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 12
Measurement analysis for diagnostic purposes 6 5 4 ΔWf[%] 3 2 shift 3 1 shift 1 0 shift 2 ΔWf[%] Exp. Aver. a=0. 85 Linear -1 -2 0 200 400 600 800 1000 1200 Cycles Corrected ΔWf variation with engine cycles GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 13
Measurement analysis for diagnostic purposes 1 0 ΔP 3[%] Exp. Aver. a=0. 85 Linear shift 1 ΔP 3[%] shift 2 -1 shift 3 -2 -3 -4 0 200 400 600 Cycles 800 1000 1200 Corrected ΔP 3 variation with engine cycles GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 14
Measurement analysis for diagnostic purposes 0. 5 ΔN 2[%] shift 1 ΔN 2[%] 0 Exp. Aver. a=0. 85 shift 2 shift 3 Linear -0. 5 -1 -1. 5 0 200 400 600 800 1000 1200 Cycles Corrected ΔN 2 variation with engine cycles GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 15
Measurement analysis for diagnostic purposes 4 3 ΔEGT[%] 2 1 0 ΔEGT[%] -1 -2 Exp. Aver. a=0. 85 0 200 400 600 Cycles 800 1000 1200 No observable shifts exist in the T/O ΔEGT or other parameters GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 16
Measurement analysis for diagnostic purposes Diagnostic Verdict v Limited diagnostic information can be extracted from raw and corrected measurements. v Parameter deltas analysis, gives an indication of engine performance shifts v For sudden shift detection, the use of more advanced diagnostic methods may not be necessary v Concerning deterioration, a clear trend is observable in all measurements. GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 17
Measurement analysis for diagnostic purposes Diagnostic Verdict v EGT increase coincides with Wf increase and P 3, N 2 decrease. A simultaneous fault in all sensors is highly unlikely to occur. v No observable shifts in T/O ΔEGT means fault is related to CR operation. v The occurrence of two consecutive shifts indicates that this is probably not a permanent internal gas-path component fault. v Although a potential component fault is detectable, neither the cause of performance shifts can be identified nor information about the deteriorated components can be determined. This information is crucial for planning correcting actions such as engine inspection or compressor washing and applying prognostic methods. GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 18
Contents q Introduction • Scope of the paper • Test case description q Measurement analysis for diagnostic purposes Ø Engine adaptive model • The PROOSIS platform • Model creation process q Diagnostic methods application • Data pre-processing • The Probabilistic Neural Network (PNN) • The Deterioration Tracking Method q Summary-Conclusions GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 19
Engine Adaptive Model The PROOSIS platform Ø Object-Oriented Ø Steady State Ø Transient Ø Mixed-Fidelity Ø Multi-Disciplinary Ø Distributed Ø Multi-point Design Ø Off-Design Ø Test Analysis Ø Diagnostics Ø Sensitivity Ø Optimisation Ø Deck Generation Ø Connection with Excel & Matlab Ø Integration of FORTRAN, C, C++ GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 20
The PROOSIS platform TURBO library of gas turbine components q Industry-accepted performance modelling techniques q Respects international standards in nomenclature, interface & OO programming Compressor map Turbine map Fan map GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 21
Engine Adaptive Model creation process GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 22
Automatic Deck Generation Procedure Robust Mathematical Model DP & OD Performance Data GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data Generic model adaptation procedure Customer Deck Generation 23
Engine Adaptive Model Predictions lie within: EGT Measured ICAO Adapted 45 deg Two performance models considered: P 3 Model EGT Model ± 2. 0% for the ICAO model ± 0. 5% for the ADAPTED model P 3 Measured v ICAO – generic model based on information available in the ICAO aircraft engine emissions data bank v ADAPTED – engine specific model using additional off-design points from the first 50 cycles. GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 24
Contents q Introduction • Scope of the paper • Test case description q Measurement analysis for diagnostic purposes q Engine adaptive model • The PROOSIS platform • Model creation process Ø Diagnostic methods application • Data pre-processing • The Probabilistic Neural Network (PNN) • The Deterioration Tracking Method q Summary-Conclusions GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 25
Data pre-processing 5 4 ΔEGT[%] 3 2 1 ΔEGT[%] Exp. Aver. a=0. 85 ΔEGT[%] Smoothed Deltas Exp. Aver. a=0. 85 0 -1 -2 0 200 400 600 800 1000 1200 Cycles Measurement deltas smoothing procedure involves: involves 1. exponential moving average to reveal step changes and/or sudden shifts. 2. these regions are excluded and a best fit polynomial is applied to the remaining data. 3. smoothed deltas are formed from the combination of the detected step changes, if any, and the best fit line GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 26
Diagnostic Methods Application Two model-based diagnostic methods are applied: v The Probabilistic Neural Network (PNN) method – a classification method allowing diagnosis of faulty or deteriorated engine components v The Deterioration Tracking method – that allows estimation of health parameters deviation, through an appropriate optimization approach. Component LP Compressor HP Turbine LP Turbine Health Parameter Flow factor Efficiency factor Symbol SW 2 SE 2 SW 25 SE 25 SW 4 SE 4 SW 45 SE 45 Set of components health parameters used for diagnosis GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 27
The Probabilistic Neural Network method PNN architecture Three layer feed-forward network allowing statistical pattern recognition based on Bayes’ decision rule GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 28
The Probabilistic Neural Network method Considered engine health conditions Considered Health Condition Symbol Deviated Parameters No. of Training patterns 12 LP Compressor fault LPC SW 2, SE 2 HP Compressor fault HPC SW 25, SE 25 12 HP Turbine fault LP Turbine fault HPT LPT SW 4, SE 4 SW 45, SE 45 Compressor fault C SW 2, SE 2, SW 25, SE 25 21 21 9 Turbine fault T SW 4, SE 4, SW 45, SE 45 HP system fault HP SW 25, SE 25, SW 4, SE 4 LP system fault LP SW 2, SE 2, SW 45, SE 45 Healthy operation OK NONE Total no. of training patterns: GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 18 18 18 36 165 29
PNN estimated probabilities 5. 67% 3. 9%6. 70% 0% 10% 48. 42% 20% 30% 40% 14. 54% 50% 60% 70% 9. 08% 80% 7. 86% 90% 2. 8% 100% Estimated probabilities are exhaustive and mutually exclusive, among considered classes GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 30
The Probabilistic Neural Network method PNN results using the ADAPTED model A fault located in HPT is detected in all points but two regions GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 31
The Probabilistic Neural Network method PNN results using ICAO model Healthy Probability (%) 100 HPT other 75 50 25 0 0 150 300 450 600 Cycles 750 900 1050 1200 The HPT fault is detected with lower probabilities GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 32
The deterioration tracking method f from previous step Yg from Engine Y Is OF minimum? Calculate OF Yes Output f No Performance Model Choose new f f Optimization Algorithm Main advantages of this method: v. Applicable in the case of limited number of available measurements v. Robustness against high levels of measurement scattering and noise. GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 33
The deterioration tracking method Estimated deviations of HPT efficiency 3 ICAO 2 Adapted SE 4[%] 1 0 -1 -2 -3 0 200 GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 400 600 Cycles 800 1000 1200 34
The deterioration tracking method Estimated deviations of HPT flow capacity 3 ICAO 2 Adapted SW 4[%] 1 0 -1 -2 -3 0 200 GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 400 600 Cycles 800 1000 1200 35
The deterioration tracking method Estimated deviations of HPC flow capacity 3 ICAO 2 Adapted SW 25[%] 1 0 -1 -2 -3 0 200 GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 400 600 Cycles 800 1000 1200 36
Diagnostic Methods Application Diagnostic Verdict v The fault is connected with the HP turbine system, according to the findings of both PNN and the Deterioration Tracking methods (when using the engine specific model). v The effect of the fault on the component efficiency is dominant. v The fault does not concern degradation of a turbo machinery internal gas-path component such as due to erosion, fouling or wear, but a fault that can be intermittent. From the above it can be concluded that the fault is probably connected to an HP turbine sub-system. GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 37
Diagnostic Methods Application Active Clearance Control (ACC) subsystem fault 6 Open 5 3 ACC ΔEGT[%] 4 2 1 ΔEGT[%] 0 ΔEGT[%] Exp. Aver. a=0. 85 -1 ACC Exp. Aver. a=0. 85 -2 0 200 400 600 Cycles 800 1000 Close 1200 The specific engine is equipped with ACC A failure of the bleed valve is expected to cause increased tip clearances, thus decreasing HP turbine efficiency, as detected by the diagnostic methods GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 38
Contents q Introduction • Scope of the paper • Test case description q Measurement analysis for diagnostic purposes q Engine adaptive model • The PROOSIS platform • Model creation process q Diagnostic methods application • Data pre-processing • The Probabilistic Neural Network (PNN) • The Deterioration Tracking Method Ø Summary-Conclusions GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 39
Summary v On-wing data obtained over a year from an engine of a commercial short-range aircraft have been analyzed using different approaches. v First, a trend analysis was performed using a measurements-derived model leading to the identification of engine deterioration and an engine fault. v Next, a diagnostic process consisting of two advanced model-based diagnostic methods has been applied. v Engine performance models of different adaptation quality have been developed using a semi-automated adaptation procedure. v Both PNN and the deterioration tracking method identify HP turbine as the faulty component. v Since performance shifts are detectable only in mid-cruise data, the fault is attributed to ACC operation, a finding confirmed from the recordings of the ACC valve position. GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 40
Conclusions v Using a typical set of on-wing measurements, valuable diagnostic information can be obtained. v Trend analysis is sufficient to detect deterioration and sudden changes in operation e. g. performance shifts. v Advanced diagnostic methods can effectively identify and quantify engine deterioration and component faults. v This information can further be used to support a decision making procedure e. g. regarding safety or maintenance planning. v Using a generic rather an engine specific adapted model affects the quality of the diagnostic information obtained. GT 2014 - 26443 Turbofan Engine Health Assessment From Flight Data 41
Proceedings of ASME Turbo Expo 2014, June 16 – 20, 2014, Düsseldorf, Germany, GT 2014 -26443 TURBOFAN ENGINE HEALTH ASSESSMENT FROM FLIGHT DATA LABORATORY OF THERMAL TURBOMACHINES National Technical University of Athens, Greece
- Slides: 42