Recent Advances In Bearing Remaining Useful Life Prediction: A Review Of Machine Learning And Deep Learning Approaches

Abstract

Rolling element bearings are critical components of rotating machinery and are highly prone to degradation and failure during prolonged operation, making accurate Remaining Useful Life (RUL) prediction essential for predictive maintenance, reducing unexpected downtime, and improving system reliability and safety. In recent years, bearing prognostics has evolved from conventional machine learning methods toward advanced deep learning and hybrid physics-data-driven approaches. This review presents a comprehensive analysis of recent developments in bearing RUL prediction, with emphasis on vibration signal processing, health indicator construction, feature extraction, temporal degradation modeling, transfer learning, attention mechanisms, and physics-informed learning. Representative studies are compared based on bearing type, datasets, preprocessing and feature extraction methods, prediction architectures, and reported performance. Widely used datasets, including PRONOSTIA/FEMTO, XJTU-SY, PHM2012, CWRU, and IMS, are examined to highlight current research trends and validation practices. The review further discusses the transition from conventional techniques such as Support Vector Machines, Random Forest, and shallow neural networks to CNN-, LSTM-, BiLSTM-, Transformer-, graph-based, and hybrid models. Particular attention is given to approaches that improve the representation of degradation, capture temporal dependencies, handle varying operating conditions, and incorporate physical degradation information. Overall, hybrid deep learning and physics-data-driven methods demonstrate strong potential for improving RUL prediction accuracy, robustness, and generalization; however, challenges remain in data availability, cross-domain generalization, noise sensitivity, interpretability, computational cost, and real-time deployment.