A Hybrid Transfer Learning Method for Fault Diagnosis of Machinery under Variable Operating Conditions

Intelligent fault diagnosis has been a research hotspot in recent years. However, most of the works are conducted based on the hypothesis that training and testing data subject to the same distribution. In engineering scenarios, machines usually work under variable operation conditions, which results in the data from different conditions subject to distribution discrepancy. Since transfer learning is able to reuse the related knowledge across different domains, a hybrid transfer learning method (HTLM) is proposed to utilize the diagnosis knowledge obtained from one operation condition to complete the diagnosis tasks under other conditions. In the method, transfer component analysis is firstly used to extract fault features with small distribution discrepancy from the cross-domain samples. After that, the features learned from the source domain help train a classifier by Tradaboost algorithm to improve its diagnosis accuracy on the target domain. The effectiveness of the proposed method is verified by a set of laboratory bearing datasets, in which the data under one operation condition are used to help identify the health states of bearings under another condition. The results indicate that the proposed HTLM is able to achieve a higher diagnosis accuracy than other diagnosis methods.