Affiliations: [a]
School of Mechanical and Electrical Engineering, University of Electronic Science and Technology of China, Chengdu, China
| [b]
State Key Laboratory of Traction Power, Southwest Jiaotong University, Chengdu, China
Correspondence:
[*]
Corresponding author. Zhiliang Liu, School of Mechanical and Electrical Engineering, University of Electronic Science and Technology of China, Chengdu 611731, China. Tel.: +86 13438832072; E-mail: Zhiliang_Liu@uestc.edu.cn.
Abstract: Local defects of rotating machinery give rise to periodic impulses in vibration. To acquire this fault information, many diagnostic methods have been reported in the past decades. Among them, the envelope spectrum analysis is usually used as the final diagnostic tool; however, its success highly depends on the correct informative frequency band selection. The key problem is how to find the correct centre frequency and its related bandwidth associated to the fault. In this paper, a novel method is proposed for selection of the optimal frequency band parameters. This method improves the informative frequency band selection performance with two aspects. One is that it incorporates the normal data as a health reference, and the other is that an objective indicator that could fuse multidimensional information is proposed. An optimal frequency band can be obtained through this algorithm, and fault mode is then determined via comparing the squared envelope spectrum between the test and normal signals. At the end of this paper, the proposed method is validated on two diagnosis cases and is compared with two of the other diagnostic methods: the conventional envelope analysis and the kurtogram. Though comparison of the results, the validity and superiority of the proposed method have been proven.
Keywords: Frequency band selection, classification, health reference, envelope
analysis, fault diagnosis
