Machine learning approaches for improving load-dependent bearing fault diagnosis of nuclear facility components

The nuclear industry is exploring applications of machine learning, including autonomous control and management of reactors and nuclear power plant and their components. The accurate diagnosis and classification of motor bearing faults under diverse operating conditions remain a significant challenge due to the complex nature of signal patterns, overlapping features, and dynamic environments. This study presents a comprehensive comparative analysis of multiple machine learning algorithms, including Random Forest, Gradient Boosting Machine, Decision Tree, Support Vector Machine (SVM), and K-Nearest Neighbours (KNN), applied to the HUST bearing dataset. This work also underscores the importance of load-dependent analysis, as fault signatures in bearings vary significantly with operating conditions. Experimental results indicate that ensemble models, particularly Random Forest, deliver superior performance across both binary and multi-class classification tasks, achieving up to 99.70% accuracy for 7-class cases and 99.37% for 21-class scenarios. Performance metrics further highlight the Random Forest model’s robustness, achieving 99.37% precision, recall, and F1-score with 15 features, confirming its suitability for real-time predictive maintenance applications. This study emphasises the importance of appropriate segmentation strategies and model selection, offering a reliable and scalable framework for industrial fault diagnostics and condition monitoring systems.