In order to extract the vibration signal characteristics of non-stationary and complex rolling bearings,a rolling bearing fault diagnosis method based on variable modal decomposition (VMD) and improved fireworks algorithm (IFWA) optimization support vector machine (SVM) was proposed.The original signal was decomposed by using VMD,the sample entropies of each IMF were calculated,and the time domain indexes (TDI) of the original signal were combined with them to form a feature matrix.In order to improve fault diagnosis efficiency,IFWA was used to optimize SVM,and the IFWA-SVM model was established.The feature matrix was trained and tested to achieve fault diagnosis of rolling bearings.The method was verified by using measured signals,and compared with the particle swarm optimization (PSO) algorithm.The results show that by using the method,the accuracy is increased by 3.33% and the training time is shortened by 21.55 s,which verifies the feasibility of the method.