Gas turbine engine gas path anomaly detection using deep learning with Gaussian distribution

Gas turbine engine anomaly detection is a critical means to ensure the safety and economic efficiency of a flight. As gas path faults make up a sizeable proportion of all the engine faults, an engine gas path anomaly detection method was proposed in the present article. Inspired by recent progress in deep learning, we explored a method that combined deep learning with traditional anomaly detection to improve the accuracy of engine gas path anomaly detection. Firstly a stacked denoising autoencoders model was built to learn robust features from datasets without labels. Then, we used learned features as the input to an anomaly detection algorithm based on Gaussian distribution to identify anomalies. To assure the engineering practicability of the proposed method, an experiment was performed to analyze real quick access recorder data of a certain type of turbofan gas turbine engine. Results demonstrated that this method could improve anomaly detection accuracy compared with traditional methods. The method could have the potential to be effectively applied in the engineering practice of engine health management.