Recognition method for sub-assemblies in shipbuilding based on three-dimensional point cloud data and lightweight models

초록

This study proposes a method for automatically recognizing target sub-assemblies in shipbuilding sub-assembly processes and mapping CAD-generated weld lines onto their measured positions. The method combines 3D point clouds acquired by a scanner with lightweight models widely used in shipyards, enabling discrimination of highly similar sub-assemblies without a large training dataset. From the lightweight model, plate surfaces, boundary edges, stiffener edges, and weld lines are extracted, and plate area, aspect ratio, and distance distribution are computed for candidate selection. From the point cloud, floor removal, noise filtering, and region segmentation are performed to isolate individual sub-assemblies and extract boundary and stiffener points. Initial candidates are screened using area difference, aspect ratio, and distance similarity, and the final sub-assembly is identified by combining ICP-based registration with hole and stiffener similarity. Experiments on three shipyard datasets and 469 CAD models showed that the method effectively eliminated most candidates at the initial stage and reliably distinguished sub-assemblies with similar boundaries by considering hole locations and stiffener layouts. The recognized CAD model was then aligned with measured data to automatically map weld lines for welding automation.

키워드