Comparison of Algorithms for Recognition of Cylindrical Features in a Voxel-Based Approach for Tolerance Inspection

Abstract

In injection molding production, automatic inspections are needed to control defects and evaluate the assigned functional tolerances of components and dies. With the "Smart Manufacturing" approach as a point of view, this paper resumes part of a wider research aiming the integration and the automation of a Reverse Engineering inspection process in components and die set-up. The paper compares two fitting approaches for recognition of portions of cylindrical surfaces. Therefore, they are evaluated in the respect of an automatic voxel-based feature recognition of 3D dense cloud of points for tolerance inspection of injection-molded parts. The first approach is a 2D Levenberg Marquardt algorithm coupled with a first guess evaluation made by the Kasa algebraic form. The second one is a 3D fitting based on the RANdom SAmple Consensus algorithm (RANSAC). The evaluation has been made according to the ability of the approaches of working on points associated to the voxel structure that locally divides the cloud to characterize planar and curved surfaces. After the presentation of the overall automatic recognition, the cylindrical surface algorithms are presented and compared trough test cases.