Data-Driven Prognostics for Anomalous Conditions in Aircraft Hydraulic System

This paper systematically investigates the performance of data-driven algorithms for fault diagnosis in aircraft hydraulic systems. Firstly, the hydraulic system of an aircraft is modeled in AMESim software, and five typical faults are artificially injected. The pressure and flow curves from different position sensors are extracted to construct the fault diagnosis dataset. Then, a multi-level feature extraction method based on deep learning algorithms, including 1DFFCNN, stacked LSTM, and improved CNN-LSTM-Attention, is designed to identify the sensitive features of potential abnormal behaviors. Finally, we study the sensitivity of multi-source heterogeneous response data of the hydraulic system to the degradation of the hydraulic system’s state, and establish the correlation between the evolution of the hydraulic system’s working state and the multi-source heterogeneous response data, achieving the early prognostics of abnormal states of the hydraulic system. Numerical experiments demonstrate that the accuracy rate of the aircraft fault diagnosis based on the data-driven algorithm presented in this paper exceeds 98%.