TIGCNet: A Novel Lightweight and High Precision Intelligent Diagnosis Method

Rolling bearings are widely used in industrial equipment, and their reliability is essential to ensuring production safety and efficiency. However, most current research primarily emphasizes enhancing model accuracy, overlooking development costs and efficiency. To address this issue, this paper proposes a novel framework of rolling bearing fault diagnosis method based on an optimized convolutional neural network. To reduce parameter count while preserving robust feature extraction capabilities, two improved structure are proposed: (1) a reverse residual structure utilizing depthwise separable convolution along with an improved squeeze-and-excitation attention mechanism is implemented; (2) a dilated convolution module with variable channel numbers is designed to extract data features of varying dimensions at the bottom of the model and expand the receptive field of the model; Finally, transfer learning leverages pre-trained parameters to enhance model accuracy and training speed. Experimental results show that the proposed method excels in rolling bearing fault diagnosis and with high classification accuracy.