Unbalanced fault diagnosis of rolling bearings using transfer adaptive boosting with squeeze-and-excitation attention convolutional neural network

For practical fault diagnosis issues, normal data are always much more numerous than fault data, so this paper focuses on how to accurately classify the unbalanced datasets. Compared to individual models, the ensemble model can combine multiple models together to achieve higher identification accuracy. In this paper, a transfer adaptive boosting method (AdaBoost) with a squeeze-and-excitation attention convolutional neural network (SEACNN) is proposed to tackle the unbalanced fault diagnosis issues of rolling bearings. Firstly, an SEACNN is designed to extract representative fault features and improve identification performance. Secondly, a new AdaBoost is designed for the SEACNN to efficiently handle unbalanced fault datasets. Thirdly, transfer learning is adopted to sequentially transfer the learned knowledge of one SEACNN estimator to the next estimator, and update the weights in the process. Substantial experiments are conducted to sufficiently evaluate the effectiveness of the proposed method.