In order to identify and classify the rotor faults quickly and accurately, a method of improved wave cluster is presented in this paper. Firstly, three feature vectors are extracted from the vibration signal of the rotor, which are kurtosis, power spectral centroid and wavelet energy entropy, respectively. Secondly, the feature space is quantized and significant grid cell information is extracted before the wavelet transform is performed on data to remove noise. Finally, based on the breadth first search (BFS) algorithm, adjacent significant grid units is connected to achieve clustering. In the process of improved wavelet clustering, the space complexity can be reduced due to the establishment of the information storage table, and the mapping relationship between the original data and the clustering results can be established. BFS algorithm is applied to search the connected unit, therefore, the complexity of clustering algorithm is reduced. The improved wave cluster method can be extended to high dimensional data space. Through experimental verification and comparison, the results show that the algorithm complexity of high dimensional data space is reduced, and the efficiency and accuracy of rotor fault diagnosis are improved.