Hybrid framework for Remaining Useful Life (RUL) prediction of rolling bearing faults

Remaining Useful Life (RUL) prediction is critical for preventing catastrophic failures in industrial systems, enabling efficient maintenance scheduling and resource optimization. This paper presents a novel hybrid framework for RUL prediction that integrates advanced feature extraction with state-of-the-art deep learning methods. The proposed framework employs Modified Multiscale Permutation Entropy (MMPE) to compute a robust Health Indicator (HI) representing the system’s degradation behavior. A regression transformer model, trained on the FEMTO dataset and validated on the CWRU dataset, leverages the HI to capture complex temporal dependencies and predict RUL with high accuracy. Experimental results demonstrate the superior performance of the proposed approach, achieving a Mean Squared Error (MSE) of $3.6\times 10^{-5}$ and Mean Absolute Error (MAE) of $5.48\times 10^{-3}$, significantly outperforming existing methods. The framework’s ability to generalize across datasets and operating conditions highlights its applicability for real-world industrial settings, offering a robust and interpretable solution for predictive maintenance.