Study of stochastic resonance recovery based on cosine fitting

The conventional stochastic resonance recovery techniques are facing some troubles, such as the complex process, too many parameters, wrong estimated amplitudes, and especially the difficulty for processing the multifrequency signals with frequencies staying far away from one another. A new strategy is proposed, where frequency-shifted and re-scaling stochastic resonance are adopted to detect frequencies, the input signals are least square fitted with cosines designed based on the acquired frequencies, and no special postprocessing is required for the distorted points, the waveform distortion has no influence on the amplitude quantization. Almost all the multifrequency signals can be processed. The effectiveness is validated by a simulation and a fault quantitative diagnostic case of electric locomotive running parts. The amplitude of periodic signals can be reduced by non-periodic sampling, the accuracy is thus developed by involving the spectrum correction technology, and more accurate signal frequencies and amplitudes can be detected.