Rotating Machinery Fault Diagnosis Based on Residual Dense Network with Multi-branch Channel Attention Mechanism

A diagnosis model for rotating machinery faults, called RDB-MBCAM-CNN, based on a residual dense network with a multi-branch channel attention mechanism is proposed to address the poor fault recognition rate on a dataset of ten types of rolling bearing and gear vibration data due to the inability of shallow machine learning models to extract deep features from vibration signal data. This method is based on convolutional neural networks, with a lightweight multi-channel attention mechanism designed to reduce the computation parameter and increase the expression ability of key features. Additionally, it merges deep and shallow features and introduces the idea of a residual dense network by designing residual dense modules to enhance the expression ability of convolutional features. Experimental results show that compared to traditional lightweight CNN and attention mechanism CNN, this model has significantly improved fault recognition accuracy.