In light of existing problems of the support vector machine, such as the difficulties of selecting its kernel function and control parameters, a dual-scale radial basis kernel support vector machine model is proposed that is optimized by hybrid particle swarm optimization based on parallel directional turbulence. The parametric intervals of the kernel have been actively divided. By utilizing the standard data set, statistical quantities such as mean value were acquired by multiple dependable and repeatable trials, and have tested and validated the support vector machine model. The data imbalance between difference classes had been considered during experiments. The test results show that in most cases, the performance of dual-scale radial basis kernel functions is better than single ones, and hybrid particle swarm optimization based on parallel directional turbulence is better than the common particle swarm optimization, because it can effectively restrain premature convergence and be beneficial in searching for better control parameters of the support vector machine. This method has been applied well to the fault diagnosis of engines.