一种数据驱动的涡轴发动机气路故障诊断研究OA

To improve the accuracy of gas path fault diagnosis for turboshaft engines and ensure the safe and reliable operation of helicopter/engine systems,this paper proposes a data-driven gas path fault diagnosis method.First,engine-related parameters are extracted from flight data.An engine parameter prediction model is built using a Long Short-Term Memory(LSTM)network to address the strong temporal dependency of the data.Then,by subtracting the predicted values of LSTM parameters from the measured values of engine parameters,a residual feature space is generated to amplify the changes in engine air path parameter characteristics before and after degradation.Finally,considering the high dimensionality and complexity of the residual feature space,a Deep Neural Network(DNN)is employed to build a degradation estimation model for fault diagnosis.Simulation results demonstrate the residual feature space based on LSTM-DNN framework markedly improves the accuracy of fault diagnosis and degradation identification compared with that directly using the collected data.