Automated Modeling by Recognition

Abstract

Abstract Facility 3D Computer Aided Design (CAD) models are not updated regularly, or at all, once a facility goes into operations. In many brownfield assets, there is no 3D facility model to begin with. As-built models are required as a foundational technology that, through the virtual and visual experience of dynamic 3D models, will enable several business functions such as training and simulation, operations monitoring and surveillance, design modifications and project execution planning support, and hazard analysis and system reviews. In addition, the regulatory environment is changing and could result in the use of immersive 3D environments for operator competency assurance and various other compliance requirements, similar to the airline industry. While there are existing model update processes, such as automated pipe recognition, the creation of the as-built model is a very time-consuming and labor-intensive activity that prohibits broader adoption and use, especially on large offshore Major Capital Projects. The current objective of our R&D effort is to build an innovative modeling system, which combines data-driven and primitive-based modeling techniques, to automate the creation of 3D geometry models and texture maps from LiDAR (Light Detection and Ranging) point cloud data. In essence, this system acts as a highly automated bridge between "dumb" laser scan data and "intelligent" 3D models, enabling pervasive use of as-built models throughout an asset’s lifecycle.