The kernel selection of twin support vector machine (TWSVM) has an important influence on its classification performance. The kernel of TWSVM is generally local or global one，and the generalization ability and classification performance of the two kernels cannot be achieved simultaneously. Gauss kernel and polynomial kernel are combined in TWSVM to improve its generalization ability and classification performance, and simple particle swarm optimization (SPSO) is used to optimize the weights and parameters. Therefore, a classification model based on SPSO optimization multiple Kernel-TWSVM is proposed, and it is applied to the pattern recognition of rolling bearing fault diagnosis. The experimental results show that multiple Kernel TWSVM has a higher classification accuracy than the single kernel TWSVM and back propagation (BP) neural network.