Intelligent fault diagnosis of rolling bearing based on time–frequency processing and DLKA-YOLO

Traditional fault diagnosis methods are difficult to accurately identify faults of rolling bearings under strong noise and cross working conditions. To address this problem, an intelligent diagnosis model based on time–frequency processing and improved YOLOv8 algorithm was proposed. First, aiming at the insufficiency of the YOLOv8 algorithm in processing multi-scale information and feature extraction, the C2f module of the YOLOv8 algorithm was improved by utilizing the deformable large kernel attention (DLKA). It enhances the network’s recognition ability of the fault features. An intelligent diagnostic model for bearing faults was constructed. Then, for the non-smoothness of the vibration signal, the two-dimensional time–frequency diagram that can characterize the fault information was generated by the time–frequency analysis methods, such as variational mode extraction (VME) optimized by shuffled frog leaping algorithm (SFLA) and continuous wavelet transform (CWT). Finally, time–frequency images were used to train the intelligent diagnosis model with the Lion optimizer accelerating the training process. The experimental results show that the proposed method not only has better diagnostic accuracy, but also has strong noise robustness and certain generalization ability, and can accurately diagnose and identify the bearing faults. • The DLKA improves the recognition ability of multi-scale features. • The feature expression ability of time-frequency diagram is improved by the SFLA-VME. • The Lion optimizer reduces the training time and improves the diagnosis accuracy.