An Industrial Dataset for Scene Acquisitions and Functional Schematics Alignment
Authors: Flavien Armangeon, Thibaud Ehret, Enric Meinhardt-Llopis, Rafael Grompone von Gioi, Guillaume Thibault, Marc Petit, Gabriele Facciolo
Organizations: Universit´e Paris-Saclay, ENS Paris-Saclay, CNRS, Centre Borelli, 91190, Gif-sur-Yvette, France · EDF R&D (Electricit´e de France), France · AMIAD, Pˆole Recherche, France · Institut Universitaire de France
Aligning functional schematics with 2D and 3D scene acquisitions is crucial for building digital twins, especially for old industrial facilities that lack native digital models. Current manual alignment using images and LiDAR data does not scale due to tediousness and complexity of industrial sites. Inconsistencies between schematics and reality, and the scarcity of public industrial datasets, make the problem both challenging and underexplored. This paper introduces IRIS-v2, a comprehensive dataset to support further research. It includes images, point clouds, 2D annotated boxes and segmentation masks, a CAD model, 3D pipe routing information, and the P&ID (Piping and Instrumentation Diagram). The alignment is experimented on a practical case study, aiming at reducing the time required for this task by combining segmentation and graph matching.
Figures & tables
Fig. 2: Data modalities of the IRIS-v2 dataset. (a) point cloud, (b) CAD model reconstructed close to the points, (c) 300 spherical images and the P&ID, (d) pipelines routing, (e) annotated 2D masks, (f) annotated 2D boxes.
Fig. 3: Example of alignment between scene acquisitions and functional schematics. The problem consists in locating pieces of equipment from the functional schematics in the 3D scene. Since information on distances is not available in the schematics, the matching is done based on structure relationships between objects. The filter and vibration isolators cannot be segmented as they are hidden by heating isolation.
Fig. 4: 2D detection and 3D Segmentation results using our fine-tuned Grounding DINO model [ 9 ] and SAM [ 8 ] for 2D segmentation and standard 2D-to-3D masks projection and fusion for 3D segmentation.
Fig. 5: Scene graph construction steps. P nodes are pipe elements and A, B and C nodes denote equipment types.
Fig. 6: Graph matching results of SLOTAlign [ 21 ] on the scene illustrated in Fig. 3 . The filter in the schematics is hidden in the scene but kept in the schematics. The matching is perfect.
Fig. 7: Robust alignment: S is paired with F and the inconsistencies are detected and given to a human to correct the graphs. The process is repeated until no inconsistency is found.