Aero-engine Gas-path Fault Diagnosis by CNN Based on QAR Data

Civil aero-engine gas-path fault diagnosis is challenging due to its complicated parametric variation mechanism and the nonlinear relationship between fault performance and parameter variation. There still lacks effective approaches to provide reliable fault detection results with the massive Quick Access Recorder(QAR) data which has been used to monitor the gas-path condition by expert experience. In this paper, we propose a novel fault diagnosis methodology using Convolutional Neural Network(CNN). We use C-MAPSS data to train the neural network. Using the theory of transfer learning, the trained neural network is used to diagnose the fault of QAR data. The diagnosis result shows the method is able to reliably monitor the aero-engine condition and detects the gas-path fault automatically.