Obtaining as-built models of manufacturing plants from point clouds

The capability to adapt a manufacturing plant to changing requirements gains increasing importance in industrial production environments, e.g., triggered by Industrie 4.0 scenarios. A virtual as-built model of a manufacturing plant and its surrounding factory building provides important decision support and relevant information for digital twins, e.g., to trace assets and asset types across their whole lifetime, planning of renovations, plant and machine topology changes, or the simulation-based analysis of production processes. Based on point clouds obtained by terrestrial laser scanning or photogrammetric acquisition, reverse engineering can be applied to extract and to reconstruct relevant objects in a form suitable for CAD programs. In this article, we review approaches to capture a scene by point measurements and to reconstruct the geometry of its components given specific object models. This comprises the discussion of various representation schemes for objects and their relations, strategies for object recognition, and the explication of methods for model instantiation. Furthermore, depending on the requirements for specific tasks, we identify technology gaps and specify the degree of maturity of the related techniques.