Fault classification method based on unsupervised transfer component analysis and support vector machines

To solve the problem of low identification rate of fault caused by lacking of fault samples and discrepancy in data distribution between source domain and target domain, the bearing fault detection method based on Unsupervised Transfer Component Analysis-Support Vector Machine (UTCA-SVM) was proposed. The sample features under different working conditions were mapped into the Hilbert Kernel space, and then the transferring sample data of source domain was measured by Maximum Mean Difference (MMD) to achieve the goal of transferring cross-domain feature information from source domain to target domain. The effectiveness of the proposed fault diagnosis method was verified by experiments. Compared with principal component analysis-support vector machines and SVM, the results showed that the proposed model could reduce the influence of domain distribution discrepancy and classify sample data more correctly and effectively. The fault status of rolling bearing could be detected precisely by the proposed methods. © 2023 CIMS. All rights reserved.