Abstract: Suitable structure parameters determine the performance of parameter-induced stochastic resonance detection system. Considering the requirements of real-time detection, the structure parameters are optimized by knowledge-based particle swarm optimization(KPSO), which takes the mean signal-noise-ratio of the output as the fitness function and the property that stochastic resonance system produces the best resonance effect just when the intensity of noise approximately equals potential barrier as the knowledge. Compared with the PSO, this algorithm can obtain the optimal structure parameters more quickly and adaptively realize optimal matching among the nonlinear system, input signal and noise. Therefore the noise of multi-frequency noisy signal is weakened and signal-noise-ratio of the output is improved. Both simulated experiments and the application of extracting real turbine vibration signals show that the proposed detection method is efficient and feasible. It enables to detect the multi-frequency weak signal submerged in strong noise in case of less sampling points and extract early fault characteristic signal.

Key words: multi-frequency weak signal;adaptive stochastic resonance;knowledge-based particle swarm optimization;multiple parameter optimization