Research on bearing fault diagnosis method based on one-dimensional parallel three-channel convolutional neural network

As an important component of mechanical equipment, bearings are widely used in various industries and play a pivotal role. Therefore, it is essential to diagnose bearing problems in real time, accurately and efficiently. Existing methods for intelligent rolling bearing diagnosis seldom make full use of the information contained in the input signals. Another issue is that most approaches are evaluated on publicly available datasets, thus limiting their universal applicability. A one-dimensional parallel three-channel convolutional neural network (OPTCNN) model is proposed, consisting of three parallel channels, each containing three basic unit modules. The first basic unit module used for the convolutional layer convolutional kernel is larger in size, while the last two are smaller. The model can deeply extract local information from the signal, obtaining more original signal fault features at larger scales, thereby reducing the semantic gap between the features and increasing network recognition accuracy. The model was validated using both laboratory and public datasets. Comparative experiments demonstrated that the proposed model notably improved feature extraction capabilities and fault diagnosis accuracy.