Sparse feature extraction based on sparse representation and dictionary learning for rolling bearing fault diagnosis

The feature vector is composed of multiple characteristics which can reflect fault information of the rolling bearing. In order to quantify the sensitivity of features for fault diagnosis, the quantitative problem is transformed into the sparse representation problem based on the sparse representation theory. Since the feature vector sparseness is unknown, a sparse dictionary is constructed based on data training. The new method is proposed to extract sparse features from the sparse coefficient and used for fault classification and recognition. The results of experiments have verified that the method is effectiveness.