Stochastic resonance (SR) phenomenon in nonlinear dynamic is widely concerned and applied in mechanical signal processing. In this paper, rich dynamic characteristics of high dimensional nonlinear system are investigated, and chaos as well as SR phenomenon are observed in the high dimensional system. Focusing on the SR phenomenon, theoretical threshold and system output driven by different input signal are studied. Since the performance of SR depends on the system state, which is directly related to system parameters, the adjustable parameters of SR system are optimized combined with particle swarm optimization (PSO) algorithm. Finally, Case Western Reserve University (CWRU) bearing fault datasets are used for method verification, and the bearing fault characteristic frequency component and signal to noise ratio (SNR) are enhanced greatly according to the experimental results. The ideal results in fault detection are obtained based on the parameters optimized SR, which provides a new idea for bearing fault detection.