Symplectic transformation based Variational Bayesian Learning and its applications to gear fault diagnosis. (December 2019)

• Record Type: Journal Article
• Authors: Pan, Haiyang; Yang, Yu; Zheng, Jinde; Cheng, Junsheng
• Is Part Of: Measurement. Volume 147(2019)
• Publication Date: 2019-12
• Subjects: ST-VBL -- Symplectic geometry similarity transformation -- Mixture of Gaussians -- Variational Bayesian Learning -- De-noising; Weights and measures -- Periodicals; Measurement -- Periodicals; Measurement; Weights and measures; Periodicals; 530.8
• DOI: 10.1016/j.measurement.2019.07.055
• Languages: English
• ISSNs: 0263-2241

Abstract

Highlights: A de-noising method of symplectic transformation based Variational Bayesian Learning (ST-VBL) is proposed. The symplectic geometry similarity transformation is used to protect system structure. Mixture of Gaussians and variational Bayesian are used to deal with many kinds of noise. TS-VBL has obvious de-noising effect in gear fault diagnosis. Abstract: The feature enhancement is one of the most crucial roles in the fault diagnosis, but the fault information is often contaminated by noise and interference harmonics. Symplectic geometry spectrum analysis (SGSA), as an effective de-noising method, has been attracting great attention in recent years. Unfortunately, SGSA cannot take full account of various noise problem. Therefore, this paper presents a Symplectic transformation based Variational Bayesian Learning (ST-VBL) de-noising method. In this method, the initial de-noising matrix is constructed by using symplectic geometry similarity transformation and the contribution rate method, and most of the noise is removed. Then, using the Variational Bayesian Learning (VBL) and Mixture of Gaussians (MOG), the probability distribution of initial de-noising component matrix is obtained to further de-noise the signal. Moreover, the properties of ST-VBL are demonstrated by simulations and experiment, showing its superiority over the traditional de-noising methods and SGSA de-noising method.