MSWF-A-ResNet Rolling Bearing Fault Diagnosis Method

To address the problem of incomplete feature extraction for rolling bearing fault data mining using a single method, this paper proposes a fault diagnosis method based on the MSWF-A-ResNet algorithm. Since the Gram's angle field and the Markov variation field are different from the features mined from the data, the use of wavelet fusion enables the data to contain the information from using two different data mining methods. An attention mechanism is introduced into the model to achieve more efficient extraction of features, and Log-Softmax is used to replace Softmax to improve the numerical performance and make it more conducive to achieve gradient optimization. Through simulation experiments, it is verified that: (1) the diagnostic accuracy of the algorithm reaches 98.80%, which has good results, and it can effectively solve the problem of a single data containing few fault features under different working conditions and different datasets. (2) The algorithm has stable and accurate diagnostic performance and strong generalization ability.