The Strategy of Datum Reference Frame Selection Based on Statistical Learning

Abstract

A datum selection strategy based on statistical learning is proposed. The datum selection is an important part of tolerance specification which is the base of geometric tolerance selection and tolerance principle selection. The problem of datum selection is to deduce the datum reference frame (DRF) based on geometrical, contact, and positioning characteristics. Currently, heuristic rules are used for DRF selection, leading to suboptimal choice of DRF in many cases. The proposed strategy formulates normalized vectors computed from the geometric, contact, and positioning characteristics of surfaces. The surfaces of different parts can be compared by their normalized vectors. Then the statistical learning method is used for building a classifier which can discriminate datum feature vectors based on training samples. Finally, a case study is given to verify the strategy and the different algorithms are compared and discussed.