• An intelligent parameters determination method is proposed based on multi-algorithm. • An improved GRNN with high accuracy is proposed for small batch of experiments. • An improved NSGA-II is proposed to generate the Pareto frontier with good uniformity. Process parameters have a significant effect on surface integrity, which determines the service performance of the parts. To improve surface integrity, the process parameters are determined: 1) by experienced engineers directly, 2) based on the Pareto frontier automatically constructed by swarm intelligence algorithms. However, as the Pareto frontier contains many non-dominated solutions, the final parameters are still determined by experienced engineers, which reduces the intelligence level. Therefore, an intelligent process parameters determination method based on multi-algorithm fusion is proposed towards minimal surface residual stress in feed or transverse direction ( Rs f , Rs t ) and surface roughness ( Ra ) in five-axis milling. Firstly, the Improved Generalized Regression Neural Network ( IGRNN ), which enhances the nonlinear mapping capability even in dealing with a small batch of experiments, is proposed to predict the Rs f , Rs t , and Ra with certain inputs (including lead angle, tilt angle, cutting depth, feed speed, and spindle speed). Then based on the proposed model, the Improved Non-dominated Sorted Genetic Algorithm-II ( INSGA-II ), which improves the uniformity of the Pareto frontier, is used to obtain a series of non-dominated process parameters. Finally, the optimal parameters are determined by the Principal Component Analysis ( PCA ) without manual weight assignment for Rs and Ra . By comparing with the second-best one, although the Rs f decreases by 0.33%, which is still able to obtain negative residual stress, the Rs t and Ra are greatly improved by 9.3% and 47.94%, respectively. The proposed method could improve the intelligent level of process parameters determination and the service performance of the parts. Furthermore, it lays a foundation for the realization of intelligent manufacturing.
Read full abstract