Deep-learning-based retrieval of piping component catalogs for plant 3D CAD model reconstruction

Abstract

In process plants, 3D computer-aided design (CAD) plant models are frequently generated via reverse engineering using point cloud data. Generating a 3D model from scan data consists of point cloud collection, preprocessing, and modeling. The process of 3D modeling to create a plant 3D CAD model from a registered point cloud consists of grouping similar point clouds into several segmented point clouds, identifying the components represented by each segment, selecting catalogs for the components, and placing them into 3D design space. The core of a 3D modeling process is to identify components represented by the segmented point clouds. This study proposes a deep learning-based method to retrieve catalogs for piping components to support the reconstruction of a plant 3D CAD model from point clouds. A prototype catalog retrieval system is implemented based on the proposed method, and the retrieval system is evaluated experimentally using point clouds obtained from a process plant. The results demonstrate that, in terms of accuracy, the proposed method outperformed a conventional shape descriptor-based retrieval method.