Models And Applications Of Stochastic Resonance-Based Weak Signal Detection

Machine condition monitoring and fault diagnosis are of significant for guarantee of machines’safety running, for reducing of breakdown loss and for avoiding casualties and environmental disasters. To realize condition monitoring and fault diagnosis, the vibration or acoustic signals that can reflect the health of machine components should be acquired for further analysis. However, machine running noise and/or background noise are always involved in the acquired signals, which reduces accuracy of signal processing and even leads to wrong results. In this regard, weak signal detection (WSD) technique is necessary for signal filtering and improving signal-to-noise ratio (SNR) of the useful signal.This study investigates a kind of noise-enhanced WSD technique, termed as stochastic resonance (SR), with its applications in condition monitoring and fault diagnosis. In the view of signal processing, SR can be regarded as a specific nonlinear filter. Traditional filters are based on noise suppression, namely, the filters eliminate or attenuate the noise components while retain the useful signal frequency components. However, when the bandwidth of the useful signal is overlapped with that of the noise, attenuating the noise can result in attenuation of the useful signal too, which further causes SNR degradation or signal distortion. Different from the traditional filters, the SR filters can utilize the noise to enchance the weak useful signal. This distince filtering mechanism is benificail to weak signal extraction, especially when the signal bandwithd is overlapped with the noise bandwidth.This study firstly reviews the state-of-the-art SR-based WSD techniques and algorithms, and then obtains a conclusion as:SR output is affected by three aspects as:1) SR potential,2) type of input signal and3) order of SR model. Subsequently, several new or modified techniques and algorithms are proposed to improve the performance of SR-based WSD, in accordance with the aboved three aspects. Specially, in the study of SR potential, two results are obtained as:1) a tristable SR method is proposed, and this method is realized via a nonlinear mechanical cantilever beam; and2) a Woods-Saxon potential SR method is proposed. In the study of interplay between SR output and input signal’s type, a sequential multiscale noise tuning SR method is proposed and implemented in an embedded system. In the study of order of SR model, an underdamped step-varying second-order SR method is proposed. The above proposed methods improve the effect of WSD, and have been successfully applied to fault diagnosis of gearbox and bearing.Meanwhile, this study also investigated a time-delayed SR model with its applications on WSD, and then proposed a new method that combines the merits of the time-delayed SR and parameter tuning SR to address engineering signal processing issues. Besides, considering that SR is a parameterized model, and the filtering effect is influenced by multiple parameters. Thus, a criterion is needed to guide the parameter tuning procedure of SR to obtain the optimal parameters and corresponding optimal filtered signal. Focusing on this issue, a new criterion that based on evaluation of regularity of zero crossing points, together with an adaptive SR algorithm, are proposed and implemented in an embedded system for online WSD.In summary, this study investigates the SR-based noise-enhanced WSD techniques and their applications in machinery condition monitoring and fault diagnosis. The methods proposed have distinct merits including good performance, high efficiency, low computational cost, easy to implement, etc., as compared with the traditional methods. Finally, the practicability and superiority of the proposed method are verified by analyzing the practical machinery signals carrying with fault information.