Acoustic Signals Recovering for Rubbing From Arbitrarily Structured Noise With Joint Iterative Probabilistic Sampling

This article aims to provide insights into a challenging rubbing signal-recovering problem that arises when analyzing acoustic signals caused by rubbing in multirotor systems. Since acoustic data measured from actual industrial sites are seriously disturbed by arbitrarily structured surrounding noise, it is difficult to directly extract target features from measured data for analyzing the operating conditions of multirotor systems. So, this article provides a joint iterative probabilistic sampling (JIPS) method to recover acoustic signals from distorted ones instead of direct target feature extraction to support condition diagnosis in realistic applications. Specifically, a diffusion probabilistic process-based JIPS strategy is proposed to provide an analytical framework to recover target signals from distorted ones. Since the proposed JIPS method pays more attention to data distributions, it can still maintain a good effect in the face of high-power noise occasions and ameliorate the performance degradation of traditional data-driven approaches in the face of domain shift. The effectiveness of the proposed JIPS method is validated on signals obtained from a realistic dual-rotor test rig as well as a public dataset. Experimental results show that the proposed JIPS not only boosts the performance of scale-invariant signal-to-noise ratio improvement (SI-SNRi) in recovered target signals than other contributions in acoustic signal processing, but also retains cognizable representative features for condition diagnosis, resulting in an enhancement of the efficiency and accuracy of condition diagnosis in practical applications.