Machinery fault diagnosis based on time-frequency images and label consistent K-SVD

Wavelet time-frequency analysis has been widely used for machinery fault diagnosis. Mechanical vibration signals can be converted to time-frequency images using wavelet transform, so machinery fault diagnosis can be transformed to the problem of image classification. Label consistent K-SVD algorithm has been proven to be effective in image classification, which incorporates a label consistent term namely discriminative sparse code error into the objective function. Therefore, in this paper, a novel bearing fault diagnosis method based on wavelet time-frequency image and label consistent K-SVD is proposed. Firstly, continuous wavelet transform is utilized to generate wavelet time-frequency images that can fully reflect bearing fault characteristics. Then texture feature extraction based on gray level co-occurrence matrix is implemented on the wavelet time-frequency images. Finally, label consistent K-SVD is conducted for classification of the time-frequency images, and thus bearing fault diagnosis is realized. The experiment results show that the texture features based on gray level co-occurrence matrix of wavelet time-frequency images can effectively extract the fault characteristics of rolling bearings, and label consistent K-SVD performs better than other classification methods based on dictionary learning under the same parameters.