Abstract: In order to reveal the multi-parameter fault characteristics of rolling bearings, a method combining variational mode decomposition and multi-scale permutation entropy was proposed for feature extraction and fault diagnosis by different algorithms.The rolling bearing fault signal was decomposed by variational modes. Secondly, the fault characteristics of each modal component were quantified by multi-scale permutation entropy, and finally, the calculated entropy was composed of feature vector set, which was introduced into probabilistic neural network, limit learning machine and support vector machine for diagnosis, and the test time and correct probability were compared and analyzed. The results show that this method can extract fault features effectively and realize the classification and recognition of fault modes accurately, the probability of fault identification improves.