Implementation of Artificial Intelligence for Aircraft Engine Health Monitoring and Prognostics

Abstract Improving the availability and reliability of aircraft engines is paramount in managing the aircraft fleet’s efficiency. While previous efforts have primarily focused on condition-based monitoring and Remaining Useful Life (RUL) prediction based on physics-based models, this paper introduces a novel approach to Engine Health Monitoring (EHM) using deep learning models. In particular, this work leverages critical engine parameters such as surge margin and exhaust gas temperature margin for interpretable EHM and prognostics. We present three deep learning models, namely, the Multi-Layer Perceptron (MLP), Convolutional Neural Network (CNN), and Long-Short Term Memory (LSTM), optimized for these tasks. The models were trained using synthetic data of 100 CFM 56 5B Turbofan engine-inspired models, simulating various flight cycles at steady-state cruise conditions using TURBOMATCH software (Cranfield University in-house aircraft engine performance simulation tool). The degradation in each engine was based on mass flow capacity and efficiency variation in the fan, compressor, and turbine, which is the effect of fouling, erosion, corrosion, etc. Unlike the existing datasets, this study deployed full factorial degradation of engine components and a wide range of degradation scenarios. Results demonstrate the competitiveness of the proposed models, as evidenced by low Root Mean Square Error (RMSE) values. The CNN model performs well in health monitoring, achieving an RMSE of 0.0148 health margin prediction. In contrast, the LSTM model proves most effective in predicting Remaining Useful Life, with an RMSE of 53.64 flight cycles. In conclusion, deep CNN and LSTM models showed a promising method for accurate engine condition monitoring and RUL predictions.