CARLE: A Hybrid Deep-Shallow Learning Framework for Robust and Explainable RUL Estimation of Rolling Element Bearings

TL;DR

Prognostic accuracy for rolling-element bearings is challenged by changing operating conditions and data noise. CARLE addresses this with a hybrid deep–shallow framework that stacks Res-CNN and Res-LSTM blocks, augmented by multi-head attention and residual connections, followed by a Linear block and a Random Forest Regressor, all fed by a compact time-frequency feature extractor that uses Gaussian filtering and Continuous Wavelet Transform to derive seven physically meaningful features. Across the XJTU-SY and PRONOSTIA datasets, CARLE achieves superior or competitive $MSE$, $MAE$, and Score, with ablation studies confirming the importance of ensemble learning, residual pathways, and attention. Interpretability is demonstrated via LIME and SHAP, showing predictions align with physically meaningful degradation cues such as $\sigma$, $f_d$, and $E$, enhancing trust and practical deployment potential in PHM.

Abstract

Prognostic Health Management (PHM) systems monitor and predict equipment health. A key task is Remaining Useful Life (RUL) estimation, which predicts how long a component, such as a rolling element bearing, will operate before failure. Many RUL methods exist but often lack generalizability and robustness under changing operating conditions. This paper introduces CARLE, a hybrid AI framework that combines deep and shallow learning to address these challenges. CARLE uses Res-CNN and Res-LSTM blocks with multi-head attention and residual connections to capture spatial and temporal degradation patterns, and a Random Forest Regressor (RFR) for stable, accurate RUL prediction. A compact preprocessing pipeline applies Gaussian filtering for noise reduction and Continuous Wavelet Transform (CWT) for time-frequency feature extraction. We evaluate CARLE on the XJTU-SY and PRONOSTIA bearing datasets. Ablation studies measure each component's contribution, while noise and cross-domain experiments test robustness and generalization. Comparative results show CARLE outperforms several state-of-the-art methods, especially under dynamic conditions. Finally, we analyze model interpretability with LIME and SHAP to assess transparency and trustworthiness.

Figures (18)

• Figure 1: Signal-to-Noise ratio (SNR) analysis w.r.t $\sigma_g$. The analysis shows the balance between noise reduction and signal preservation. When no smoothing is applied, the SNR remains high due to preservation of the original signal’s fidelity. As the smoothing parameter increases, the filtering mechanism effectively reduces high-frequency noise. However, the process simultaneously diminishes the finer details and dynamic components of the signal, resulting in a rapid decline in the SNR. After a certain smoothing intensity, noise reduction occurs at the cost of negligible signal distortion ($\sigma_g \approx 0.75$ for XJTU-SY and $\sigma_g \approx 1.1$ for PRONOSTIA). This stabilization point signifies an optimal parameter range where the balance between noise suppression and signal integrity is achieved.
• Figure 2: Schematic diagram of the compact feature extractor framework.
• Figure 3: The schematic diagram of the CARLE AI system.
• Figure 4: Schematic diagram of the Res-CNN block.
• Figure 5: Schematic diagram of the Res-RNN block.