Abstract:A complex environment with many factors affects the bearing fault diagnosis, which leads to the high dimensional feature becoming a technical problem. Kernel principal component analysis (KPCA) is used to reduce the dimension of high dimensional features, and some results are obtained, but in KPCA,the influence of similarity among features on the computational complexity and separation effect does not considered, which restrictes the improvement of realtime and effectiveness of the calculation and the improvement of classification effect. Therefore, a clustering KPCA method based on K-means clustering algorithm and KPCA method was proposed. The idea of mean clustering algorithm was used to cluster the similar features of the extracted features in the time and frequency domains, so as to reduce the complexity of subsequent KPCA calculation. Then, KPCA was used to reduce the dimension of the features after clustering, and the higher dimensional features were mapped to a feature space with a higher classification degree. Four kinds of bearing state signal features, namely normal, inner ring fault, outer ring fault and rolling body fault, were used to test the clustering KPCA method. The results show that compared with the KPCA method, the proposed clustering KPCA method has better dimension reduction separation effect and stronger robustness.