Korean Journal of Computational Design and Engineering

To automatically inspect machined components, including automotive and mechanical equipment parts, it is essential to identify the specific part. Although vision and AI technologies have advanced and are widely used for part recognition, errors still occur when identifying symmetrical or similarly shaped components. Additionally, to assess dimensional and shape deviations of machined parts, the use of 3D point cloud data is necessary. This study proposes a method for recognizing thin-shaped components using 3D point cloud data. Typically, part identification is performed by comparing the similarity between measured data and CAD data. In this approach, distances are compared using spheres, cylinders, and planes generated through linear and planar fitting of the measured data. However, point cloud data often contain significant noise, and the point distribution is not uniform. To address this issue, this study employs a grid-based structure to extract point cloud data in a form consistent with CAD data, which is then used for similarity evaluation. For symmetrical components, simple distance comparisons fail to distinguish between different parts. To overcome this limitation, the proposed method divides the region and evaluates similarity within each section, enabling accurate differentiation.

키워드 Part recognition, Point cloud, Similarity evaluation, Symmetrical part