Application of a Hybrid EKF-Bayesian Approach for Technical Condition Diagnostics of Gas Turbine Auxiliary Power Units
Keywords:
Gas Turbine APU, Extended Kalman Filter (EKF), Bayesian Inference, Fault Diagnosis, Flight Safety, Root Mean Square Error (RMSE), ROC Curve, Condition MonitoringAbstract
Ensuring the technical reliability and flight safety of gas turbine Auxiliary Power Units (APUs) is a critical task in aviation engineering. Traditional diagnostic methods, often reliant on visual inspections and fixed thresholds, are limited in detecting internal or early-stage faults. This study proposes a proactive hybrid diagnostic approach integrating the Extended Kalman Filter (EKF) and Bayesian Inference for real-time monitoring and fault forecasting of the APU. The EKF is employed to linearize nonlinear engine dynamics and filter stochastic sensor noise, while the Bayesian framework provides a recursive mechanism for fault probability estimation and classification.
Experimental results based on 120 test cycles demonstrate that the EKF algorithm reduces the Root Mean Square Error (RMSE) of critical parameters (rotor speed, exhaust gas temperature, and oil pressure) by an average of 65–67% compared to traditional measurement methods. The subsequent Bayesian classification achieved an overall diagnostic accuracy of 95.83% and a sensitivity of 95.0%. Furthermore, Receiver Operating Characteristic (ROC) analysis yielded an Area Under the Curve (AUC) of 0.956, confirming the high discriminatory power of the proposed model. The findings indicate that the integrated EKF-Bayesian approach offers a mathematically rigorous and effective solution for the early detection and trend forecasting of malfunctions, significantly enhancing aviation safety standards.
