A rolling bearing fault diagnosis method using novel lightweight neural network

As an important part of rotating machinery, rolling bearing fault will lead to equipment fault, resulting in loss of property and personal safety. To overcome the deficiency of traditional methods, such as low recognition accuracy, slow diagnosis speed, and relying on manual extraction of features, a novel bearing fault diagnosis method based on inverted residual convolutional neural network embedded with squeeze-and-excitation block (SE-IRCNN) is proposed. This method adopts a lightweight concept to reduce the calculation amount significantly. The body of the model is built with inverted residual blocks to reduce the feature loss in the dimensional reduction. Squeeze-and-excitation block is embedded to recalibrate the features. The universality and robustness of the method are verified by changing the ratio of the train set and test set under two experimental datasets. Compared with the commonly used methods, SE-IRCNN has a smaller calculation amount, faster diagnosis speed, and higher accuracy.