Enhancing the Prediction of Aircraft Engine’s Remaining Useful Life by Integrating One-Dimensional Convolutional Autoencoder and Regression Model

Purpose: Aircraft engine failures contribute significantly to overall aircraft failures, making up approximately 60% of incidents. In intricate systems like aircraft, the failure of a single component can affect the entire system, thereby posing operational disruptions and safety hazards. Therefore, the accurate prediction of the remaining useful life (RUL) of aircraft engines is crucial for ensuring safety and optimizing operational efficiency.BRMethods: To address this challenge, we propose a novel approach that integrates a one-dimensional convolutional autoencoder (1D-CAE) and a regression model. The 1D-CAE is trained using sensor data collected from specific aircraft engines. Subsequently, sensor data from other engines undergo encoding through the trained 1D-CAE to extract hidden features. These features are then utilized to train a regression model. The combined use of the 1D-CAE's encoder and the regression model enables the prediction of the RUL of aircraft engines.BRResults: Combined with the XGBoost regression model, the proposed approach demonstrated exceptional performance, as evidenced by a root mean squared error of 16.49. This approach outperformed conventional machine learning methodologies that did not incorporate the 1D-CAE. Additionally, our methodology exhibited superior performance compared to those reported in previous studies, resulting in an enhancement of approximately 6%.BRConclusion: The integration of the 1D-CAE and regression model provides an effective means of predicting RUL and extends its applicability to diverse fields while mitigating risks and reducing costs associated with system failures.