Reconstruction of 3D surfaces from incomplete digitisations using statistical shape models for manufacturing processes

Abstract

Digitization of large parts with tight geometric tolerances is a time-consuming process that requires a detailed scan of the outer surface and the acquisition and processing of massive data. In this work, we propose a methodology for fast digitization using a partial scan in which large regions remain unmeasured. Our approach capitalizes on a database of fully scanned parts from which we extract a low-dimensional description of the shape variability using Statistical Shape Analysis. This low-dimensional description allows an accurate representation of any sample in the database with few independent parameters. Therefore, we propose a reconstruction algorithm that takes as input an incomplete measurement (faster than a complete digitization), identifies the statistical shape parameters and outputs a full scan reconstruction. We showcase an application to the digitization of large aeronautical fuselage panels. A statistical shape model is constructed from a database of 793 shapes that were completely digitized, with a point cloud of about 16 million points for each shape. Tests carried out at the manufacturing facility showed an overall reduction in the digitization time by 80% (using a partial digitization of 3 million points per shape) while keeping a high accuracy (reconstruction precision of 0.1 mm) on the reconstructed surface.