Modeling and monitoring the material removal rate of abrasive belt grinding based on vision measurement and the gene expression programming (GEP) algorithm

• Published in: International journal of advanced manufacturing technology 2022-05, Vol.120 (1-2), p.385-401
• Main Authors: Ren, Lijuan, Wang, Nina, Pang, Wanjing, Li, Yongchang, Zhang, Guangpeng
• Format: Article
• Language: English
• Subjects: Abrasive belts; Abrasive cutting; Abrasive wear; Algorithms; Artificial neural networks; Back propagation networks; Belt grinding; Big Data; CAE) and Design; Coefficient of variation; Computer-Aided Engineering (CAD; Continuity (mathematics); Cutting wear; Engineering; Feature extraction; Gene expression; Grinding; Identification methods; Image segmentation; Industrial and Production Engineering; Material removal rate (machining); Mechanical Engineering; Media Management; Monitoring; Neural networks; Original Article; Parameters; Prediction models; Support vector machines; Testing time
• ISSN: 0268-3768; 1433-3015
• DOI: 10.1007/s00170-022-08822-z

Accurately predicting the material removal rate (MRR) in belt grinding is challenging because of the randomly distributed multiple cutting edges, flexible contact, and continuous wear of the abrasive grains, undermining the ability to achieve the expected machining requirements for belt grinding using the planned parameters. With the development of sensing technology, big data, and intelligent algorithms, online identification methods for material removal through sensing signals have gained traction. A vision-based material removal monitoring method in the belt grinding process was investigated by adopting the gene expression programming (GEP) algorithm. First, the relationship between the grinding parameters and MRR was investigated through a series of experiments. Second, methods of image shooting distance calibration and automatic image segmentation were established. Furthermore, the definition and quantification method of 11 features related to the color, texture, and energy of spark images are described, based on which the features are extracted. Then, the optimal feature subset was determined by analyzing the fluctuation degree and correlation with MRR by computing the coefficient of variation of the features and Pearson’s coefficient of features and MRR, respectively. Finally, a continuous function model including the selected features was obtained using the GEP method. The predicted results and testing time were compared with those of other methods such as LightGBM, convolutional neural network (CNN), support vector regression (SVR), and BP neural network. The results show that the MRR prediction model based on the GEP algorithm can obtain explicit function expressions and is highly effective in predicting accuracy and test time, which is of utmost significance for accurate and efficient acquisition of MRR data online.