Weak fault detection is crucial to incipient mechanical fault diagnosis. In order to extract weak fault signals, a method named improved Re-scaling Frequency Stochastic Resonance (IRFSR) is proposed in this paper. The method consists of four steps: (i) Frequency Information Exchange (FIE); (ii) Amplitude Coefficient Adjustment; (iii) Re-scaling Frequency Stochastic Resonance (RFSR); and (iv) Frequency Information Recovery. By means of the exchange of frequency information, the high-frequency information of the target signal is accordingly transferred to the low frequency band which can be rescaled to satisfy the small-parameter limits of classical stochastic resonance. Then IRFSR is able to overcome the limitation of RFSR, which is that the sampling frequency of RFSR is at least 50 times greater than the frequency of the target signal. Numerical results are reported to evaluate the effectiveness of IRFSR to detect the target signal with higher frequency at a low sampling frequency as compared with RFSR. And the feasibility of the IRFSR in incipient fault detection is demonstrated using case history data obtained from a sliding bearing test rig.
