Dynamic Fault Diagnosis of Aeroengine Control System Sensors Based on LSTM-CNN

To address the problem that it is difficult to diagnose sensor faults when the operating state of the aeroengine control system changes dynamically, an LSTM-CNN based aeroengine sensor fault diagnosis method is established in this paper. First, a nonlinear dynamic prediction model of the engine is constructed by using Long short-term memory (LSTM) network. The prediction model generates residual signals with each sensor measurement value to achieve decoupling between each sensor and between system state change and fault. Based on the constructed fault residual signal dataset, the classification of sensor faults is implemented based on Convolutional Neural Network (CNN). The simulation results show that the LSTM prediction network has high prediction accuracy, and the designed CNN classification network has high diagnosis accuracy with 91.33%.