In fault diagnostics, the valuable information of rotating machinery via vibration analysis are always masked in heavy noise, so it is challanging to extract useful attributes. Stochastic resonancep (SR) is a technique to detect weak signal which is swamped in dense noise. However the output results of convential SR methods driven by Gaussian noise are still unsatisfactory to detect mutli-frequency signals. We study multi-stable SR system in the presence of alpha stable noise instead of Gaussian noise to detect faults induced in rotating machines. The alpha stable noise distribution has heavy-tailed fluctuations and infinite variance. It can increase the performance of SR phenomenon when the noise deviates from the ideal Gaussian conditions. The appropiate alpha stable noise level can amplify the weak signal and improves the signal to noise ratio (SNR) of the output signal. The simulation results show that the proposed method efficiently detects the weak signals which are submerged in strong background noise. The present method is usable in different scientific and engineering applications.