About us
Espace utilisateur
Education
INSTN offers more than 40 diplomas from operator level to post-graduate degree level. 30% of our students are international students.
Professionnal development
Professionnal development
Find a training course
INSTN delivers off-the-self or tailor-made training courses to support the operational excellence of your talents.
Human capital solutions
At INSTN, we are committed to providing our partners with the best human capital solutions to develop and deliver safe & sustainable projects.
Thesis
Home   /   Thesis   /   Automatic reverse-engineering of BIM models using Machine Learning

Automatic reverse-engineering of BIM models using Machine Learning

Artificial intelligence & Data intelligence Computer science and software Engineering sciences Technological challenges

Abstract

Today, BIM (Building Information Modelling) has become a standard for information management in factories, buildings and industrial facilities. The BIM approach is particularly well-suited to complex industrial environments, and nuclear facilities in particular, as it proves to be a relevant tool throughout the facility's lifecycle, from design to construction, operation and decommissioning, by offering shared, smart and structured modelling.
This approach is centered on a 3D model, generally produced from a point cloud obtained by lasergrammetry. In most cases, panoramic photos can be acquired at the same time. From this point cloud, a 3D model is generally rebuilt to represent the equipment present as set of single solid objects. This 3D model reconstruction stage is often long and tedious, and is currently carried out manually by a CAD designer.
This thesis proposes to develop an automatic method for reconstructing BIM models from point clouds using machine learning and image analysis, exploiting both the point cloud and available panoramic photos. The environments of nuclear facilities are composed of very specific steel or alloy processes, and mainly include piping equipment. By combining machine learning and computer vision, using both clustering and classification methods on the one hand, and shape and image recognition on the other, the work consists in directly identifying in the point cloud objects belonging to business object families such as pipes, elbows, valves, supports, fittings, tanks, etc., as well as some of their metadata: the material they are made of, their geometric properties (diameter, thickness, length), their volume and mass.

Laboratory

Département de recherche sur les Procédés et Matériaux pour les Environnements complexes
Service d’Etudes des Matériaux et de l’Etanchéité
Laboratoire d’étude des technologies du Numérique et des Procédés Avancés
Top envelopegraduation-hatlicensebookuserusersmap-markercalendar-fullbubblecrossmenuarrow-down