State Baseline Modeling and Fault Detection for Aero-Engines With Unknown Performance Degradation Based on Deep Neural Networks

This paper is concerned with the data-driven state baseline modeling and fault detection problem for aero-engines. Long-term operations can result in performance degradation of aero-engines, which naturally causes the state baseline being varied at different stages of performance degradation and further increases false alarm and missed detection rate of fault detection. To tackle this problem, a deep neural network is first designed to learn the state evolution behavior and generate state baseline. Moreover, a multi-step reciprocal prediction method is developed to characterize the state prediction error and enhance the accuracy of state baseline modeling. Then a fault detector is designed by means of the state baseline to perform fault detection. Furthermore, a variable period state baseline update mechanism combining fuzzy logic system and periodic update is proposed to match the performance degradation so as to ensure the accuracy of fault detector and reduce false alarm. Finally, the effectiveness of the proposed state baseline modeling and fault detection method is verified through an aero-engine example. Numerical results show that the proposed method reduces the prediction error by an average of 42.8% compared with other advanced methods, while achieving a fault detection accuracy of 98.0% without performance degradation and 96.0% with 2% performance degradation.