Fault diagnostics of rolling bearings using feature fusion based BP, RBF and PNN neural networks

Issue title: Proceedings from the 17th International Symposium on Applied Electromagnetics and Mechanics (ISEM 2015)

Guest editors: Fumio Kojima, Futoshi Kobayashi and Hiroyuki Nakamoto

Article type: Research Article

Authors: Cheng, Yuhua^a | Yin, Chun^{a; *} | Bai, Libing^a | Bai, Qiuju^a | Huang, Xuegang^b | Chen, Kai^a

Affiliations: [a] School of Automation Engineering, University of Electronic Science and Technology of China, Chengdu, Sichuan, China | [b] China Aerodynamics Research & Development Center, Mianyang, Sichuan, China

Correspondence: [*] Corresponding author: Chun Yin, School of Automation Engineering, University of Electronic Science and Technology of China, Chengdu 611731, Sichuan, China. E-mail:chunyin@uestc.edu.cn,yinchun.86416@163.com

Abstract: This paper employs three neural networks that are BP, RBF and PNN for rolling bearing fault diagnosis and compares their performance. The preprocessed vibration signals of rolling bearing provide fused feature vectors after the process of wavelet package decomposition and feature fusion. Then the fused feature vectors serve as the inputs of networks. The fault diagnostic aims to recognize health condition, fault types and fault severity of rolling bearings. The simulation results demonstrate that BP has the best accuracy and very complex computation efforts, and RBF has the fastest classification with the lowest precision. Meanwhile, PNN achieves perfect accuracy and speed that can be received.

Keywords: Fault diagnostics, rolling bearings, feature fusion

DOI: 10.3233/JAE-162080

Journal: International Journal of Applied Electromagnetics and Mechanics, vol. 52, no. 1-2, pp. 95-102, 2016