Enhanced bearing health indicator extraction using slope adaptive signal decomposition for predictive maintenance.

This study introduces the Slope Adaptive Signal Decomposition (SASD) algorithm, a novel method for extracting enhanced intrinsic machine health indicators from vibration data. Leveraging advanced signal processing techniques such as dynamic Savitzky-Golay filtering, segmentation, and trend-based recalibration, SASD achieves superior noise attenuation while preserving critical trends. Applied to the PRONOSTIA platform's bearing datasets, SASD produces refined health indicators suitable for predictive maintenance tasks, including Remaining Useful Life (RUL) estimation. The extracted features are evaluated using deep learning models like GRU, LSTM, and hybrid architectures, as well as conventional regression approaches, demonstrating SASD's effectiveness in improving prediction accuracy. Among these, GRU exhibits best performance with R2 score above 0.96 across all 3 operating conditions for various bearings. This method bridges the gap between signal processing and data-driven prognostics, enabling robust bearing health monitoring under varying operational conditions. In the future, the SASD framework will be extended to other industrial datasets to benchmark its generalizability across different operating conditions. Additionally, integrating real-time data streaming capabilities and edge computing deployments can further improve the scalability for real-world predictive maintenance applications.•Enhanced trend extraction via dynamic signal smoothing & segmentation.•SASD-based indicators surpass traditional methods.•RUL prediction validated using statistical & deep learning.