In order to monitor the operation state of each wheel motor in distributed drive electric vehicle and ensure the safety of the vehicle，a fault diagnosis method of in-wheel motor based on improved multi-class support vector data description （MCSVDD） is proposed. The method incorporates two major improvements. First，a classification judgment rule based on the minimum distance to the cluster center within the class is proposed using the affinity propagation （AP） clustering algorithm to enhance MCSVDD. Second，a Weibull kernel function is constructed from the Weibull distribution to optimize data description model. Meanwhile，a dimensionality reduction method based on minimum-distance propagation discriminant projection （MPDP） is proposed for the multi-dimensional feature set of in-wheel motor operating state，which improves the separability of in-wheel motor fault states under different working conditions. Finally，in-wheel motors with typical bearing faults are customized respectively to collect vibration signals under 7 rotating speeds for verifying the effectiveness of the proposed method. The results show that the reduced dimension data's separability of observed samples of in-wheel motor operating state based on MPDP is better than that of linear discriminant analysis （LDA），minimum-distance discriminant projection （MDP） and locality preserving projection （LPP），and the recognition accuracy of MCSVDD's state recognition system based on Weibull kernel function is higher than that of polynomial and Gaussian kernel function.