Affiliations: College of Mechanical Engineering, Shenyang Jianzhu University, Shenyang, Liaoning 110168, China
Correspondence:
[*]
Corresponding author: Dong An, College of Mechanical Engineering, Shenyang Jianzhu University, Shenyang, Liaoning 110168, China. E-mail: andong@sjzu.edu.cn.
Abstract: It is always a hot research issue to get weak fault information from severely contaminated data before further analysis. Wavelet denoising methods have been widely applied by researchers in many different applications. This paper presents the application of the Bayesian wavelet packet method on mechanical equipment monitoring with an approach for selecting the wavelet decomposition levels based on the statistical autocorrelation function (ACF). Firstly, a reasonable decomposition level is selected; then, the degree of noise pollution of each wavelet coefficient can be got based on the Bayesian rule; finally, the estimated value of each wavelet coefficient can be obtained using the Bayesian hypothesis test. The denoising approach is compared with the conventional soft and hard threshold methods using signals simulated and obtained from track vibration test. The comparison result reveals the proposed method shows better denoising performance for weak information recognition.
