FEMamba: A Feature-Enhanced Mamba Framework with Degradation-Stage Global Regularization for Bearing Remaining Useful Life Prediction

Accurate prediction of bearing Remaining Useful Life (RUL) is crucial for implementing predictive maintenance and ensuring safe operation of industrial equipment. However, existing methods still face challenges in handling multi-source heterogeneous time series data and modeling long-term degradation processes. This paper proposes a novel Feature Enhanced Mamba framework (FEMamba) that achieves high-precision bearing RUL prediction by integrating a Multi-Feature Enhancement Selection module (MFES) and a Degradation Stage-based Global Regularization strategy (DSGR). Specifically, the MFES module effectively processes heterogeneous sensor data through an adaptive feature selection mechanism, while the DSGR strategy optimizes the physical significance of feature representations by constructing feature space constraints that consider degradation dynamics. Experiments on the PHM2012 dataset demonstrate that our proposed method outperforms existing approaches across all evaluation metrics, achieving an R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> value of 0.9601 and a Mean Square Error (MSE) of 3.22×10<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">−3</sup>. Ablation studies further verify the effectiveness of the MFES module and DSGR strategy, showing MSE reductions of 7.9% and 10.6% respectively compared to the baseline model. These results indicate that our proposed method provides an accurate and reliable solution for bearing RUL prediction.