Research on Mechanical Fault Diagnosis under Complex Working Conditions Based on Improved YOLOv8

Abstract With the rapid development of artificial intelligence and big data technology, deep learning-based bearing fault diagnosis methods have demonstrated excellent diagnostic performance. However, most neural networks require a significant amount of prior knowledge during construction and training, and a lot of time is needed to build the optimal model to achieve high classification accuracy. In addition, when it comes to fault data across operating conditions and equipment, the network's generalization ability is often limited. In addition, the scarcity of labelled fault data and the large differences in data sample distribution further increase the difficulty of achieving high-precision fault diagnosis. To address this challenge, this article focuses on rolling bearings as the main research object proposes for the first time the application of YOLOv8 in bearing fault diagnosis and proposes an improved cross condition and cross equipment fault diagnosis method RSD-YOLOv8 based on YOLOv8. Firstly, the DSConv module is introduced in the YOLOv8 network to replace the original convolutional layers, enhancing the network's ability to capture vibration signal features. Secondly, replacing the C2f module with the C2f_Rep_SimAm module reduces the number of parameters while maintaining computational efficiency. This article simulates the fault states of rolling bearings under cross-operating conditions and cross-equipment conditions using two datasets, CWRU and XJTU-SY. It verifies the effectiveness of the model improvement. In addition, transfer learning experiments were conducted on the gearbox fault dataset of Jiangsu Qianpeng Diagnostic Engineering Co., Ltd. to verify the universality of the improved model. The successful application of the YOLOv8 network in bearing fault diagnosis has provided an effective solution for diagnosing bearings under different operating conditions and equipment.