Fault diagnosis method for rolling bearings based on BICNN under complex operating conditions

Abstract

To address the issues of poor noise resistance and insufficient generalization performance in traditional fault diagnosis methods, an end-to-end rolling bearing fault diagnosis method based on bidirectional interactive convolutional neural network (BICNN) is proposed. Firstly, the bearing vibration signal is directly input into the wide convolutional kernel for rapid feature extraction, reducing the interference of high-frequency noise. Secondly, a modified rectified linear unit (M-ReLU) activation function is designed to solve the problem of “neuron death” in the ReLU activation function. Then, a bidirectional interactive feature extraction module is constructed, and the features extracted are input into the bidirectional interactive feature extraction module to capture the channel and spatial feature information simultaneously. Next, the extracted information is imported the presented feature enhancement module to achieve more valuable information transmission and accumulation. Finally, a small convolutional kernel is applied to further extract feature information, and a global average pooling layer is used to replace the fully connected layer, reducing the number of parameters while avoiding the problem of model overfitting. The softmax is utilized to classify the types of bearing faults. Two different data sets are adopted to validate the fault diagnosis performance of BICNN model under − 4 dB and variable operating conditions. Experimental results show that the average recognition accuracy reaches 97.71% under variable load and 98.25% under variable speeds, which is higher than other comparison methods. It is verified that the proposed method has higher diagnostic accuracy and generalization ability in complex working conditions.