In order to effectively extract the fault characteristics of the high-speed train bogie vibration signal and aim at the problem of the information of single-channel acquisition is difficult to fully reflect the running state of the train, a new fault feature extraction method based on full vector sample entropy (FVSE) algorithm is proposed. Firstly, the noise assisted multiempirical mode decomposition (NAMEMD) method is used to decompose the vibration signal to obtain a series of multi-intrinsic mode functions. Then, according to the correlation coefficient method, the most relevant intrinsic mode functions to the original signal are selected to calculate the sample entropy and the full vector sample entropy. Finally, the support vector machine is used to identify the train status. The experimental results show that the recognition rate of the train using the FVSE algorithm is generally 6% higher than the sample entropy algorithm, and the highest is above 98%, the validity of noise assisted multivariate empirical mode decomposition and full vector sample entropy for fault diagnosis of high speed trains is verified.