Improving the accuracy of TIF in bonnet polishing based on Gaussian process regression

• Published in: International journal of advanced manufacturing technology Vol. 110; no. 7-8; pp. 1941 - 1953
• Main Authors: Feng, Jingbo, Zhang, Yifan, Lin, Shuo, Yin, Yuehong
• Format: Journal Article
• Language: English
• Published: London Springer London 01.09.2020; Springer Nature B.V
• Subjects: Accuracy; CAE) and Design; Coefficient of friction; Computer-Aided Engineering (CAD; Contact pressure; Covariance; Engineering; Experiments; Gaussian process; Industrial and Production Engineering; Influence functions; Kernel functions; Kinematics; Mathematical models; Mechanical Engineering; Media Management; Model accuracy; Original Article; Parameter modification; Polishing; Pressure distribution; Pressure effects; Process parameters; Stress concentration; Gaussian process regression; Bonnet polishing; Preston equation; TIF model
• ISSN: 0268-3768; 1433-3015
• DOI: 10.1007/s00170-020-05917-3

Based on the Gaussian process regression (GPR), this paper aims to provide a method to improve the accuracy of the classic tool influence function (TIF) model in bonnet polishing (BP). Firstly, we build the velocity and pressure distribution models in TIF according to kinematics relations and Hertz contact theory. And our investigating experiments about contacting forces indicate that the constant K in the Preston equation is actually not a constant and depends on the interfacial friction coefficient μ . According to the experimental results, several main processing parameters (i.e., rotation speed, polishing depth, inflated pressure) have dramatic effects on μ . Thus, the classic model of TIF based on Preston equation needs to be revised. Relevant experiments and researches are conducted to search a more accurate linkage between TIF model and the main processing parameters. By designing composite covariance kernel functions for μ , we apply GPR method in TIF model and promote the accuracy of TIF acquired in experiments. Two groups of experiments are conducted, and the prediction performance of the new modified TIF model in our researches is verified to surely improve the accuracy of TIF comparing with the classic model. Since few researches are focusing on this aspect, our work is to find out the relation between μ and processing parameters and provide a method of modelling μ to modify K .