An online learning framework for aero-engine sensor fault detection isolation and recovery

Accurate and reliable sensor data are critical for the safe operation of modern aero-engine control systems. However, maintaining the accuracy and robustness of fault diagnosis models throughout the engine lifecycle is particularly challenging, especially under conditions of gradual degradation. To address these challenges, this paper proposes a novel Fault Detection, Isolation, and Recovery (FDIR) framework. The framework utilizes a Deep Denoising Autoencoder (DDAE) for fault detection, a multi-model strategy for fault isolation, and a dual-task learning framework for fault signal recovery, ensuring system integrity and continuous operation. Additionally, an online update mechanism based on distribution mean shifts is introduced, integrating parameter regularization and memory replay to prevent catastrophic forgetting and enhance adaptability. Experimental results demonstrate that the proposed framework achieves high-precision FDIR under both non-degraded and degraded conditions, exhibiting superior robustness and adaptability. By combining data-driven methods with adaptive online learning mechanisms, this work provides a scalable and reliable solution for aero-engine sensor fault diagnosis. It not only enhances the operational safety and efficiency of complex, data-intensive systems but also contributes to advancing the state of the art in this field.