SARNet: A Spike-Aware consecutive validation Framework for Accurate Remaining Useful Life Prediction
Junhao Fan, Wenrui Liang, Wei-Qiang Zhang
TL;DR
This work tackles remaining-useful-life (RUL) prediction for bearings under fault onset, where fixed spike rules and opaque models hinder reliability and deployment. It introduces SARNet, a spike-aware framework that fuses a ModernTCN forecaster with adaptive consecutive spike validation—thresholding at $\theta = \mu_{\text{ref}} + k\,\sigma_{\text{ref}}$ with $k=2$ and requiring $d_{\min}$ consecutive exceedances—and a post-onset RF–LGBM ensemble to deliver calibrated RUL via $RUL_{\text{seg}}(t) = (T_f - t)/(T_f - t_s)$. On the XJTU-SY bearing dataset, SARNet achieves RMSE ≈ 0.036 and MAE ≈ 0.020 with $R^2$ near 0.99, outperforming recent baselines and maintaining a lightweight, interpretable pipeline. The approach emphasizes robustness to noise, physics-informed triggers, and practical deployability, offering clear feature attributions and efficient CPU-friendly inference suitable for real-world prognostics.
Abstract
Accurate prediction of remaining useful life (RUL) is essential to enhance system reliability and reduce maintenance risk. Yet many strong contemporary models are fragile around fault onset and opaque to engineers: short, high-energy spikes are smoothed away or misread, fixed thresholds blunt sensitivity, and physics-based explanations are scarce. To remedy this, we introduce SARNet (Spike-Aware Consecutive Validation Framework), which builds on a Modern Temporal Convolutional Network (ModernTCN) and adds spike-aware detection to provide physics-informed interpretability. ModernTCN forecasts degradation-sensitive indicators; an adaptive consecutive threshold validates true spikes while suppressing noise. Failure-prone segments then receive targeted feature engineering (spectral slopes, statistical derivatives, energy ratios), and the final RUL is produced by a stacked RF--LGBM regressor. Across benchmark-ported datasets under an event-triggered protocol, SARNet consistently lowers error compared to recent baselines (RMSE 0.0365, MAE 0.0204) while remaining lightweight, robust, and easy to deploy.