Power Setting Sensor Fault Detection and Accommodation for Gas Turbine Engines Using Artificial Neural Networks

Abstract

Sensors installed on gas turbine gas path are used to obtain gas path measurement parameters for control and condition monitoring purpose. These sensors are prone to degradation or failure due to hostile working environment around them. Most gas path sensor diagnostic research is based on an assumption that the power setting sensor, a sensor used by engine control system to control engine power output, has no fault so engine measurement data can always be obtained at desired operating conditions. However in practice, power setting sensor may also be faulty, which may result in misleading measurement data and diagnostic results. In this paper, an artificial neural network based gas path diagnostic approach for engine power setting sensor fault detection and quantification has been introduced. Nested artificial neural networks (ANN) are used to detect power setting sensor fault and ensure prediction accuracy. Measurement noise is also considered in the training and testing samples to ensure the robustness of the diagnostic system. The developed power setting sensor diagnostic approach has been applied to a model 2-shafts industrial gas turbine engine similar to a GE LM2500+G4 engine to test the effectiveness of the approach. The selected power setting parameter is the shaft power output measured by a power setting sensor. An engine performance model is produced using Cranfield University’s gas turbine performance and diagnostics software, Pythia. Training samples with the consideration of sensor faults were simulated with the engine model assuming one of the sensors, either the power setting sensor or other gas path sensors may be faulty. In the nested neural network for sensor fault diagnostics, the system separately performs sensor fault detection, sensor fault identification and sensor fault quantification. Results show that the developed nested neural network diagnostic system is able to identify the power setting sensor fault and correctly predict the magnitude of the fault. This would allow the engine control system correct its control schedule and accommodate the power setting sensor fault.