FROM POINT CLOUD TO CAD-MODEL BASED ON AI

Abstract

Most vessels undergo modernization and overhaul during their lifecycle. A model representing the as-is status is a prerequite to plan these activities. Especially when it comes to overhauling a piping system, it is crucial to know their exact spatial layout. Otherwise, the existing parts and the newly delivered parts may not fit together. Capturing the surfaces of the parts using a 3D scanner and generating a point cloud representing the as-is situation is state of the art. The point cloud is then loaded into a CAD system to represent the existing parts before designing the new parts and ensure they fit together seamlessly. This requires remodelling at least the connection points of the existing parts manually. Remodelling is both time-consuming as well as error-prone, so we’d like to avoid it. In the DigiTwin research project [1], we developed a solution for this. It features a fully automatic generation of a CAD model based on recognizing piping components from the point cloud using artificial intelligence. The main task is running object recognition from the point cloud. It is crucial to find the correct components and detect their position. Based on that, we generate native CAD models for typical shipbuilding tools. This enables using the automatically generated CAD models for the as-is status as reference geometry for the updated design. In our paper, we’d like to present the steps needed to do so and also the result we achieved.