An Improved Convolutional-Neural-Network-Based Fault Diagnosis Method for the Rotor–Journal Bearings System

More layers in a convolution neural network (CNN) means more computational burden and longer training time, resulting poor performance of pattern recognition. In this work, simplified global information fusion (SGIF-CNN) is proposed to improve efficiency diagnostic accuracy. the improved CNN architecture, feature maps all convolutional pooling are globally convoluted into corresponding one-dimensional sequence, then sequences concatenated fully connected layer. On basis, paper further proposes novel fault diagnosis method for rotor–journal bearing system based on SGIF-CNN. Firstly, time-frequency distributions samples obtained using Adaptive Optimal-Kernel Time–Frequency Representation algorithm (AOK-TFR). Secondly, time–frequency diagrams utilized train SGIF-CNN model shallow method, trained can be tested testing samples. Finally, transplanted equipment’s online monitoring monitor operating conditions real time. The verified data from rotor test rig an ultra-scale air separator, analysis results show that improves computing compared traditional while ensuring accuracy diagnosis.