SVM bearing fault diagnosis based on fast inter-class distance in the feature space and DMPSO algorithm

Abstract

Abstract Support vector machines (SVMs) have good processing performance for small sample datasets. The giant search space for kernel parameters and the tendency of parameter optimization to fall into local optima are two essential factors that affect the generalization ability of SVM models and, thus, affect the accuracy of fault diagnosis results. Propose using fast inter-class distance (FICDF) in the feature space to reduce the search space for kernel function parameters and then use differential mutation particle swarm optimization (DMPSO) to optimize kernel function parameters to improve the generalization ability and classification accuracy of the SVM model. Firstly, the FICDF algorithm is used to calculate the Euclidean distance between classes, and a fast segmentation idea is proposed for fast operations to obtain a smaller kernel parameter search space. Then, the global search ability of the DMPSO algorithm is used to obtain the optimal parameter combination of the SVM model. Finally, the fault diagnosis model of the SVM is applied to the fault diagnosis of rolling bearings. The experimental results show that compared with other fault diagnosis methods, this model method has higher classification accuracy and verifies its better classification speed.