Remaining Useful Life ( RUL ) Prediction Methods for Machine Health Estimation and Fault Diagnosis: A Comprehensive Review of Latest Techniques and Future Prospects

ABSTRACT This paper aims to provide a state‐of‐the‐art review of the most recent Remaining Useful Life (RUL) prediction methods, starting from statistical methods, machine learning (ML), deep learning (DL), and their ensemble methods. The limitations and strengths of the earlier techniques, data‐driven techniques, and combined model‐based and data‐driven solutions were discussed. The study focuses on RUL estimates in preventive maintenance policies in different industrial fields where failure prediction and maintenance plans are crucial. While model‐based methods provide high accuracy, they are highly dependent on system knowledge and are likely to be restricted by more comprehensive datasets and various types of degradation features. In contrast, data‐driven methods are becoming more popular and flexible in solving large‐scale problems and complicated degradation features. The paper also highlights the importance of different AI learning models for achieving higher accuracy in predictions, with dependency on the time aspect and hierarchical feature space. Both hybrid and ensemble methods are discussed to have promising applications in integrating the advantages of model‐based and data‐driven approaches to improve prediction reliability. Real‐life examples and applications are described to demonstrate how RUL prediction can help in increasing performance and decreasing expenses. Finally, future research directions for further work are identified to address these challenges.