A multi-objective optimization approach for selecting key features of machining processes

This paper presents a multi-objective optimization approach to select key machining features for improving the predictive accuracy of virtual metrology. Along increasing of complicated machining features, supervised optimization methods can be applied to select the significant features; however, these methods are inapplicable when the number of selected features are far greater than the number of training samples for modeling. Based on a novel unsupervised two-stage clustering procedure, this paper proposes a clustering non-dominated sorting genetic algorithm (CNSGA) to minimize objectives of selecting key features, e.g., the feature number and the clustering ratios. According to the selected features, a virtual metrology system was adopted to predict the machining quality of a machining process in a CNC lathe. The results show that precision and robustness of using the features selected by the proposed CNSGA for predicting machining accuracies of wheel rims are better than that of using the other selection approach.