Remaining useful life prediction across different operational conditions based on domain adaptation feature transfer

This paper proposes a domain adaptation feature transfer framework to address the challenge of data distribution discrepancies in cross-condition remaining useful life (RUL) prediction. The method integrates signal processing techniques (Hilbert-Huang transform) with deep denoising autoencoders (DAE) to extract degradation-sensitive features. Transfer Component Analysis (TCA) is subsequently applied to align feature distributions between source and target domains. Experimental results on the IEEE PHM 2012 bearing dataset demonstrate the superior performance of the proposed method over traditional statistical features and existing domain adaptation approaches, with root mean square error (RMSE) and mean absolute percentage error (MAPE) reduced by 21.3% and 39.1%, respectively. Further analysis validates the robustness of the method under noisy and small-sample scenarios, highlighting its potential for industrial applications such as wind turbine gearbox monitoring and aircraft engine maintenance.