Online Signal Denoising Using Adaptive Stochastic Resonance in Parallel Array and Its Application to Acoustic Emission Signals

Abstract

Signal denoising has been significantly explored in various engineering disciplines. In particular, structural health monitoring applications generally aim to detect weak anomaly responses (including acoustic emission) generated by incipient damage, which are easily buried in noise. Among various approaches, stochastic resonance (SR) has been widely adopted for weak signal detection. While many advancements have been focused on identifying useful information from the frequency domain by optimizing parameters in a post-processing environment to activate SR, it often requires detailed information about the original signal a priori, which is hardly assessed from signals overwhelmed by noise. This research presents a novel online signal denoising strategy by utilizing SR in a parallel array of bistable systems. The original noisy input with additionally applied noise is adaptively scaled, so that the total noise level matches the optimal level that is analytically predicted from a generalized model to robustly enhance signal denoising performance for a wide range of input amplitudes that are often not known in advance. Thus, without sophisticated post-processing procedures, the scaling factor is straightforwardly determined by the analytically estimated optimal noise level and the ambient noise level, which is one of the few quantities that can be reliably assessed from noisy signals in practice. Along with numerical investigations that demonstrate the operational principle and the effectiveness of the proposed strategy, experimental validation of denoising acoustic emission signals by employing a bistable Duffing circuit system exemplifies the promising potential of implementing the new approach for enhancing online signal denoising in practice.

Adaptive unsaturated stochastic resonance in parallel array and its application to acoustic emission signal denoising

“…The block diagram in Fig. 1 illustrates the operational principle of utilizing unsaturated SR for denoising and signal recovery in a parallel array of N bistable systems inspired by suprathreshold SR approaches [30][31][32][33] . First, additional random noise 𝜂 𝑗 (𝑡) for 𝑗 = 1, … , 𝑁, which are independent Gaussian white noise with a common intensity 𝐷 𝜂 , are applied to the original signal of interest 𝑠(𝜏) contaminated by measurement noise 𝜉(𝜏).…”

“…This section explicates the denoising strategy considering the practical concerns for implementation including the original ambient noise and gain adjustments to realize the optimal noise level discussed in Section 2.2. The core principle of utilizing the unsaturated SR in the parallel array of bistable systems for signal denoising is to average out the noisegenerated incoherent interwell transitions while enhancing the stochastically coherent interwell transitions that are activated by the original signal 31,33 . As a result, it is crucial that the total noise (𝜉(𝜏) + 𝜂 𝑗 (𝜏) of Fig.…”

Low SNR Multi-Emitter Signal Sorting and Recognition Method Based on Low-Order Cyclic Statistics CWD Time-Frequency Images and the YOLOv5 Deep Learning Model

“…This paper proposes a multi-data source deep learning Object Detection network (MS-Yolo) based on millimeter wave radar and vision fusion. The authors of [ 23 , 24 , 25 , 26 , 27 ] used different deep learning models and preprocessing algorithms to classify and recognize radar signals, ships, and intracellular molecules. In [ 28 , 29 , 30 ], generative adversarial network and adaptive noise suppression are used to sort and identify modulation features for radar signals, and good results are obtained.…”

Wavelet sparsity enhancement for extracting transient vibration signatures of bearing structural damages

“…A variety of effective methods have emerged, such as mode decomposition methods, 4–6 spectral kurtosis methods, 7–10 wavelet analysis methods, 11–13 blind filter methods, 14,15 and stochastic resonance methods. 16–18…”

“…A variety of effective methods have emerged, such as mode decomposition methods, [4][5][6] spectral kurtosis methods, [7][8][9][10] wavelet analysis methods, [11][12][13] blind filter methods, 14,15 and stochastic resonance methods. [16][17][18] Among these methods, wavelet analysis is undoubtedly one of the most popular methods, as it is a new time-frequency analysis method with multi-resolution analysis performance which is ideally suitable for characterizing the transient signatures of bearing damages. [19][20][21][22] When localized damages exist in bearing basic components (e.g., outer race, ball, cage, and inner race), the fault signatures are usually impulsive and periodic.…”