A REVIEW OF BEARING FAULT DETECTION, DIAGNOSIS AND PROGNOSTICS

This review explores advancements in bearing fault detection, diagnosis, and prognostics, highlighting the transition from traditional to AI-driven approaches. Time-domain, frequency-domain, and time-frequency domain methods are discussed alongside advanced signal decomposition like empirical mode decomposition and wavelet transform. The role of machine learning and deep learning in improving diagnostic accuracy, robustness, and adaptability is also discussed, supported by evaluations on public datasets for benchmarking. Additionally, it emphasizes prognostics and health management for remaining useful life estimation, linking it to degradation modelling and predictive analytics. The potential of digital twin technology for real-time monitoring, adaptive maintenance, and fault simulation is evaluated in bridging the gap between simulated and empirical data. Future research should focus on federated learning, edge AI, and digital twin to enhance predictive accuracy and industrial applicability.