Automatic Feature-Based Point Cloud Alignment and Inspection

Abstract 3D scanning is a non-contact geometric inspection technique that has potential applications in many industrial sectors. Compared to coordinate measuring machines, 3D scanning is able to acquire high-density sampling of an entire part in a relatively short time without fixturing. In practice, point clouds, which are the output of 3D scanning, must be aligned with the base-truth CAD model before calculating the point-wise deviation. Existing methods accomplish this task by determining the optimal transformation that maximizes the overlap between the point cloud and base-truth CAD model. However, these methods do not consider the problem from a geometric dimensioning and tolerancing (GD&T) perspective. Specifically, dimensional errors are not quantified in reference to the established datum features. In this work, a new scheme is proposed to automatically align the datum features of 2D shapes based on point clouds via a polar coordinate representation. From the result of the new alignment scheme, the point-wise deviation is evaluated through the deviation distribution to differentiate manufacturing and measurement errors. The proposed scheme is applied in three case studies, in which the sample shapes to be inspected are generated by adding random errors to their nominal models to mimic measurement errors in physical experiments. The result illustrates the effectiveness and efficiency of the proposed scheme in automatic shape alignment based on datum features and in subsequent detection of point-wise deviation.