Research on high-speed train wheelset bearing fault diagnosis based on IDBA-CNN

Abstract

Abstract To address the challenges in high-speed train wheelset bearing fault diagnosis, including difficulties in optimizing Convolutional Neural Network (CNN) hyperparameters and the tendency of traditional metaheuristic algorithms to get trapped in local optima with imbalanced exploration–exploitation capabilities, this study proposes an intelligent fault diagnosis method that integrates an improved Detective Behavior Algorithm (IDBA) with CNN. First, a Refraction Opposition-Based Learning (ROBL) strategy is designed. By introducing a stochastic refraction coefficient and perturbation terms, the symmetry constraints of the original opposition-based learning are broken, enhancing population diversity. Next, an elite-guided adaptive information-sharing development strategy is implemented, which dynamically adjusts the strength of information sharing based on population fitness variance, achieving an adaptive balance between exploration and exploitation. The IDBA is coupled with the CNN model to automatically optimize critical hyperparameters, constructing an end-to-end fault diagnosis framework. Experimental validation on the self-built HST-WBSet dataset of high-speed train wheelset bearings demonstrates that the proposed method achieves a diagnosis accuracy of 98.96% and a macro-average F1 score of 98.97%, significantly outperforming mainstream comparative models including DBA-CNN, Vision Transformer, Swin Transformer, and CNN-Transformer. Ablation studies further confirm that both the refracted reverse learning strategy and the elite-guided adaptive information-sharing development strategy contribute effectively and synergistically to performance enhancement. This approach provides a systematic and high-precision solution for adaptive fault diagnosis of high-speed train wheelset bearings, improving robustness, accuracy, and practical applicability in industrial monitoring scenarios.