Research on Intelligent Prediction of Surface Roughness in Cutting 42CrMo Steel by using Particle Swarm Optimization-based Support Vector Machine

Abstract

42CrMo high-strength steel is a material that is difficult to machine and has difficulties controlling the quality of the machined surface. To ensure the stability of surface quality during cutting, lead the adjustment of cutting parameters to accurately predict the 42CrMo steel's machined surface roughness (Ra). Single factor cutting, orthogonal cutting, and response surface cutting experiments were conducted based on the experimental platform, and single factor, range, and grey correlation analyses were performed on the surface roughness measurement results. It can be concluded that within a given range, the feed per tooth has the greatest impact on surface roughness, and the cutting depth has the least impact on surface roughness. The PSO-SVM surface roughness prediction model was developed and compared with other widely used surface roughness prediction models (BP, SVM, GA-BP, PSO-BP) by using experimental data on the machined surface roughness of cutting 42CrMo steel. It can be concluded that the PSO-SVM training set prediction model has an average relative prediction error of 4.76% and a goodness of fit R² = 0.87198, which is quite near to 1. The PSO-SVM testing set prediction model has an average relative prediction error of 12.65% and a goodness of fit of R² = 0.86406, which is quite near to 1. Since it can effectively guide the selection and adjustment of cutting parameters, the PSO-SVM surface roughness prediction model has high prediction accuracy, good fitting degree, and stability. It also has a specific reference value for the study of the cutting process and surface quality of 42CrMo steel.