A Bearing Fault Diagnosis Model based on Minimum Average Composite Entropy and Parallel Attention Mechanism Convolutional Neural Network

This work addresses the issues of low diagnostic accuracy and weak generalization in rotating machinery bearing fault diagnosis, especially under complex noise conditions. In this paper, a novel bearing fault diagnosis method is proposed. This method, known as MACE + PFACNN, combines the minimum average composite entropy (MACE) with a parallel fusion attention convolutional neural network (PFACNN). In MACE, the minimum average composite entropy, which is composed of the Renyi entropy and the sample entropy, is used as a fitness function to guide the dung beetle optimization algorithm for fault feature extraction. Then, the extracted signal features are converted into angle and field and angular difference fields by Gramian angle field transformation. Finally, a PFACNN is used for fault diagnosis. Experimental data and bench tests show that the proposed model achieves a classification accuracy of 99.93 %. Compared with the baseline model, the noise resistance under complex noise conditions has improved by more than 15 %, and the generalization ability has increased by 3.68 %.