Novel monitoring method for material removal rate considering quantitative wear of abrasive belts based on LightGBM learning algorithm

• Published in: International journal of advanced manufacturing technology Vol. 114; no. 11-12; pp. 3241 - 3253
• Main Authors: Wang, Nina, Zhang, Guangpeng, Pang, Wanjing, Ren, Lijuan, Wang, Yupeng
• Format: Journal Article
• Language: English
• Published: London Springer London 01.06.2021; Springer Nature B.V
• Subjects: Abrasive belts; Abrasive wear; Algorithms; Belt grinding; CAE) and Design; Computer-Aided Engineering (CAD; Correlation coefficients; Engineering; Grinding; Industrial and Production Engineering; Machine learning; Material removal rate (machining); Mechanical Engineering; Media Management; Model accuracy; Model testing; Monitoring; Original Article; Prediction models; Signal processing; Sound; Wear rate; Material removal rate monitoring; LightGBM learning algorithm; Abrasive belt wear state monitoring; Abrasive belt grinding
• ISSN: 0268-3768; 1433-3015
• DOI: 10.1007/s00170-021-06988-6

Wear is an inevitable problem in abrasive belt grinding, and the material removal rate decreases with continuous wear of the abrasive belt. This indicates that the grinding control force is affected by two dynamic factors, namely the actual material removal and abrasive belt wear state. To obtain an accurate force-control model to achieve uniform material removal, a new method for online monitoring of abrasive belt material removal rates and their corresponding wear statuses is proposed herein using only the grinding sound signals. By performing material removal rate and abrasive belt wear experiments, the grinding sound signals during processing are obtained. The wear states of the abrasive belt are quantified using the newly defined gray-mean values of the topographical images of the belt into different levels. The grinding sound signals are quantitatively described via the statistical features of their sound wavelet signals. The statistical features related to material removal rates or belt wear states are selected on the basis of the Pearson correlation coefficients. The prediction models for material removal rate and wear levels of the abrasive based on the selected features are then established using the LightGBM learning algorithm. Experimental datasets are used to train and validate the established model. The test results show that the evaluation parameters of the prediction model of the material removal rate are all within 5%. Further, the accuracy of the wear levels of the abrasive belt can exceed 91%. Compared with other prediction models, the new LightGBM models exhibit superiority in terms of time factor without loss of accuracy of the model. It is thus proved that the proposed method can provide a good basis for monitoring the material removal rate and belt wear in the abrasive belt grinding process.