An Improved CNN Employing SSA and Its Application in Bearing Fault Diagnosis

Aspiring to the problems of inadequate self-adaptive ability in classical feature extraction methods and weak generalization ability in single classifier under big data, an internal improved Convolutional neural network (CNN) method founded on Sparrow Search Algorithm (SSA) is proposed. First, to process bearing data, two-dimensional wavelet time-spectral transform is performed with bearing data, and some normalized time-frequency diagrams are obtained. The highest correct rate of the test set in the training network is used as the fitness function, in which SSA is used to hunt for the optimum parameter combination of CNN. Then correctly select the learning rate and batch learning times of SSA in the SSA that have a large impact on the training error. Simultaneously, the ideal structure distribution of CNN is given through comparison. Finally, combined with deep learning with powerful adaptive feature extraction and nonlinear mapping capabilities, the obtained samples are fed into CNN for training, and a load-bearing fault diagnosis model based on CNN is established. By testing the remaining samples multiple times, the diagnostic rate of the model can reach more than 99.5%, which is far superior to the traditional fault diagnosis method based upon pattern recognition and feature extraction. Experimental results show that the proposed method can effectively enhance the adaptive feature extraction ability and the accuracy of fault diagnosis of the model, so as to have better generalization performance.