Compound fault recognition based on wavelet packet transform and the transferable combination of ResNet50 and multi-label classifier

The rolling bearing is the important part in a motor. Traditional studies in fault diagnosis domain mainly focus on the multi-class classification. However, due to the various and complex situation in engineering application, one signal generally contains several fault feature information. Now ResNet50 is a popular basic network and generally provides a good performance. Based on wavelet packet transform, this paper proposes a transferable combination of ResNet50 and multi-label classifier (WPT-MLRN). In detail, this paper selects the wavelet packet transform as the data processing technology. Furthermore, to demonstrate that the proposed method could extract useful fault feature, the extracted features of proposed method are input into the multi-class model. In the experiment, we compare the proposed method with traditional multi-label model. And the results show that the proposed method has higher accuracy. Meanwhile, here gives the better performance when inputting the extracted feature of proposed method into the multi-class model.