Abstract: Stochastic resonance (SR) is a kind of nonlinear phenomenon. Through the enhancement of the transmission of weak input signal through nonlinear system, SR can detect signals with lower signal-to-noise ratio (SNR) than linear methods by converting partial energy of noise into signal energy. In order to catch the characteristic signal of mechanical faults as early and accurately as possible, a new method is presented to detect weak signal buried in heavy noise based on stochastic resonance (SR) theory. Taking the SNR of input signals as variable and SNR improvement as the estimating criterion of signal amplifying degree, this method can adjust system parameters adaptively to detect the weak signal with different frequency. The principle and approach of this method are simply illustrated. This method is applied to detecting the weak frequency component signals characterizing the inception of rub-impact fault of rotor system. The result shows that the weak sinusoid signal of lower signal-to-noise ratio can be reliably extracted from heavy noise when the data length is shorting, and this method is simple and robust.

Key words: Adaptive, Mechanical fault diagnosis, Parameter-tuning stochastic resonance, Rotor rub-impact, Weak signal detection