Aiming at the problems of insufficient fault data and low accuracy of fault diagnosis in the health monitoring of high-speed train axle box bearing

a sparse classification fusion diagnosis method based on K-SVD reconstruction residual was proposed. In this method

the reconstruction error of K-SVD was used to characterize the classification trend of samples in each fault mode. The confusion matrix of each K-SVD classifier was estimated and the corresponding reliability matrix was calculated according to the distribution of the reconstruction residual of the samples. Then the fault identification results of the tested samples using each K-SVD classifier are fused based on the D-S evidence fusion theory to get the final diagnosis results. The experimental results of EMU axle box bearing showed that the recognition accuracy of proposed method is significantly improved compared with the single feature classifier and the traditional voting fusion method. Even in the case of small sample

its fault tolerance and stability are better. It solves the problems of low detection accuracy of single sensor

insufficient feature information of single domain and difficulty to effectively integrate the classification information in the fault diagnosis of the axle box bearing of the high-speed train.