An online belt wear monitoring method for abrasive belt grinding under varying grinding parameters. (February 2020)

• Record Type: Journal Article
• Authors: Cheng, Can; Li, Jianyong; Liu, Yueming; Nie, Meng; Wang, Wenxi
• Is Part Of: Journal of manufacturing processes. Volume 50(2020)
• Page Start: 80
• Page End: 89
• Publication Date: 2020-02
• Subjects: Abrasive belt grinding -- Tool condition monitoring -- Multi-sensor -- Bayesian network -- Various parameters; Production management -- Data processing -- Periodicals; Manufacturing processes -- Periodicals; Procestechnologie; Productietechniek; Production -- Gestion -- Informatique -- Périodiques; Fabrication -- Périodiques; Manufacturing processes; Production management -- Data processing; Periodicals; 670.5
• DOI: 10.1016/j.jmapro.2019.12.034
• Languages: English
• ISSNs: 1526-6125

Abstract

Abstract: Abrasive belt grinding has attracted attention in recent years in both industry and academia due to the rapid development of abrasive belts; however, online monitoring of abrasive belt wear under varying grinding parameters is challenging. In this paper, the multi-sensor fusion of sound and current signals is used to resolve the abovementioned problem. First, the characteristics of the grinding sound and current are investigated in the time-domain and frequency-domain, and then their differences under various belt wear states and grinding parameters are discussed; based on the investigation, several features are extracted that indicate the belt wear state. The abovementioned discussion demonstrated that grinding sound signals have an abundance of information at high frequencies (6–20 kHz); in contrast, grinding current signals contain information at low frequencies. Furthermore, the grinding sound and current signals have different sensitivities to the grinding parameters and belt wear. Finally, a Bayesian network is proposed to identify the wear state; moreover, its adaptability under changing parameters and the effect of multi-sensor fusion are discussed. The results show that the accuracy reaches 100% with enough training data; additionally, when the training data only covers a limited range of grinding parameters, the fusion of the sound and current substantially improves the accuracy of the prediction results from 86% to 95%.