A systematic review of fault diagnosis techniques specifically for ship rolling bearings operating within the demanding marine environment is still lacking. This gap becomes increasingly critical as the Fourth Industrial Revolution, characterized by the pervasive influence of artificial intelligence (AI), continues to deepen. This article presents a comprehensive review of fault diagnosis methods for ship rolling bearings, covering research achievements up to and including 2026. The review begins by emphasizing the critical role of rolling bearings in marine propulsion systems and outlines their common failure modes. Subsequently, it categorizes and analyzes the predominant diagnostic approaches for shipboard bearings, identifying three primary categories: vibration signal analysis, acoustic emission technology, and wear debris detection techniques. For each category, the review elaborates on: the underlying principles, the current state of research, and the application of key enabling technologies such as sensor systems, signal processing methods, and data analysis algorithms in the context of bearing condition monitoring. By synthesizing these findings and outlining future research trajectories, this review aims not only to advance the field of intelligent fault diagnosis for marine applications but also to stimulate the development of emerging research directions within the broader AI domain.
Current deep learning fault diagnosis networks for mechanical equipment may fail to reliably identify out-of-distribution (OOD) faults when samples from previously unseen fault categories are unavailable during model development, leading to the misclassification of such OOD samples as known categories. To address this issue, a deep learning fault diagnosis framework based on the generalized extreme value theory (GEVT) is proposed. First, this framework replaces the traditional label with one sample, providing waveform-level evidence for the resulting decisions, enabling users to intuitively confirm results during both training and testing. Then, classification is performed based on the mean absolute distance (MAD) between network predictions and sample labels to enhance interpretability. Finally, the GEVT is employed to fit the extreme values of the MAD, and the cumulative distribution function probability is calculated. Based on this probability, a threshold is determined to detect OOD faults. Validation on the PHM2009 gearbox dataset and BJTU-RAO bogie dataset demonstrates that this framework not only effectively identifies previously unseen OOD samples but also maintains high detection accuracy for in-distribution (InD) data.