An Intelligent Fault Diagnosis Method Based on Optimized Parallel Convolutional Neural Network

Abstract

Addressing the limitations in feature extraction and model optimization complexity of convolutional neural network (CNN), an intelligent fault diagnosis method based on the Beluga whale optimization (BWO) algorithm optimized parallel CNN (PCNN) is proposed. First, the preprocessed vibration signal of the rolling bearing is converted into a 2-D time-frequency image by continuous wavelet transform (CWT). Second, the PCNN model is constructed, wherein the two branches independently learn distinct image weight values. This approach enhances deep-space feature expression by complementing high-dimensional features. Then, the BWO algorithm is used to optimize the hyperparameters of PCNN, thereby enhancing the model’s feature extraction and classification performance. Finally, multihead self-attention (MSA) is introduced into the PCNN framework to further improve the quality of feature representation and realize fault identification. The effectiveness and superiority of the method are verified by experimental datasets of rolling bearing and field test datasets of reciprocating compressor, the results of which show that the proposed model is significantly superior to the other models, exhibiting higher accuracy and better noise resistance, which can provide reliable technical support for practical industrial applications.