Fault diagnosis of rolling bearing based on feature fusion of multi-scale deep convolutional network

Aiming for the limitations of traditional fault diagnosis model, such as, strong dependence on engineering priori knowledge, incomplete feature extraction, difficulties in selection of classifiers, a fault diagnosis model of rolling bearing based on feature fusion of multi-scale deep convolutional neutral network is proposed. First, a convolutional neural network model that integrates feature extraction and pattern recognition is constructed, the vibration signals of rolling bearing are converted into two-dimensional images by wavelet transform and used as the input sample set. Second, a multi-scale feature fusion module is built up in the network structure for the purpose of adaptive extraction of features at different levels of fault samples, aiming to extract different-scale features completely. Finally, fault samples are input into the network to realize adaptive features extract of bearing signals and end-to-end diagnosis. According to the experimental analysis results, the proposed fault diagnosis model based on feature fusion of multi-scale deep convolutional network can extract features of the signal at all levels and achieves higher diagnosis accuracy and robustness under interferences of different noises. It provides a theoretical basis to realize fault diagnosis of rolling bearings. © 2022, Solar Energy Periodical Office Co., Ltd. All right reserved.