Adapting the coherent point drift algorithm to the fixtureless dimensional inspection of compliant parts

Abstract

In a free-state condition, deformable bodies (or compliant parts) have a significantly different shape than their nominal geometry (CAD model) due to gravity loads and residual stress. Typically, the dimensional metrology of such parts requires a particular approach where expensive and specialized jigs are needed to constrain and follow the component during inspection. This paper proposes a new method to fixtureless inspect deformable bodies by adapting the coherent point drift (CPD) algorithm. This new approach combines the optimization of the smoothness regularization parameters of the CPD powerful non-rigid registration method alongside the Thompson-Biweight statistical test as an identification method to distinguish profile and localization process manufacturing defects from a part’s deformation. In other words, a new “flexible” registration approach that meets the specifications of compliant parts is proposed. The CAD model is smoothly modified to fit the scanned part by minimizing two criteria that belong to deformable bodies. The first criterion is the conservation of the curvilinear distance (isometric transformation). That is to say, the condition that the stretch difference between the original CAD model and the modified one should be very small. The second criterion is the usual minimization of the Euclidean distance between the modified CAD model and its corresponding scanned part. Many case studies were performed on a typical industrial sheet metal which presented profile and localization defects. The estimated values of the defects, as well as their positions, are much closer to their reference ones in most cases, thus reflecting the effectiveness of the proposed method.