An intelligent fault diagnosis method for rolling bearings under class-imbalance conditions using numerical simulation

Abstract Data-driven artificial intelligence models have been widely applied in mechanical fault diagnosis owing to their efficiency and convenience. However, because mechanical equipment typically operates under normal conditions, fault samples are scarce, leading to a severe data imbalance and significantly restricting the performance of AI models. To address this issue, an intelligent fault-diagnosis method based on numerical simulations was introduced. First, a dynamic model of defective rolling bearings was developed. Then, fault samples were generated by solving the model and subsequently used to supplement the training dataset of the AI-based models. Finally, convolutional neural networks, residual networks, and visual geometry group networks were selected as representative AI models, and the trained networks were subsequently used to classify unlabeled fault samples collected from real-world machinery. The experimental results indicate that the proposed method effectively addresses the issue of low classification accuracy caused by class-imbalanced fault samples in the rolling bearings.