In order to judge the fault types more accurately and improve the clustering accuracy, the analysis of the fault diagnosis accuracy that is the accuracy of the clustering was performed, and then the wavelet clustering algorithm based on the grid refinement was proposed. The grid resolution technology was applied to the wavelet clustering algorithm, and the grid refinement was performed on the grid unit of the feature space to extraction of boundary point. The clustering accuracy of the algorithm before and after improvement was verified. The result shows that the improved wavelet clustering algorithm get a more accurate clustering result than the original wavelet clustering algorithm.