A fault diagnosis method based on improved convolutional neural network for bearings under variable working conditions

The fault diagnosis of rolling bearing will be negatively reduced because of variable working conditions, large environmental noise interference and insufficient effective data sample. To solve the problem, this paper proposes an improved convolutional neural network (CNN) method. The method firstly constructs a new network, multi-mode CNN (MMCNN) by using multiple parallel convolutional layers to effectively extract rich and complementary fault features, then transforms the 1D time-domain signal of the rolling bearing acquired under different frequency variable conditions to the 2D time-frequency grayscale by continuous wavelet transform (CWT) and put the grayscale into the MMCNN for training. Besides, the method combines the pseudo-label learning method with MMCNN, which can expands the labeled data set by pseudo-label processing of unlabeled data. The experimental results show that the proposed method can effectively improve the fault detection accuracy of rolling bearings under variable working conditions, which is superior to the existing methods.