Skeleton-based discrepancy feedback for automated realignment of industrial assemblies

• This paper presents an algorithm for automated discrepancy quantification of construction components.
• Skeletonization is used for abstraction of massive laser scans of industrial assemblies.
• Orientation of straight lines is the key descriptor for automatic segmentation of straight members.
• The discrepancies are quantified by comparing the corresponding lines from skeletonized 3D models and laser scans.
• Accurate and reliable discrepancy results are captured within a significantly faster time frame.

Automated and timely detection, characterization, and quantification of fabrication discrepancies and errors are fundamental problems in construction engineering. Despite the fact that the precision of manufacturing machines is continually improving, there are inevitable discrepancies between the designed and built assemblies because of construction realities. Such non-compliant assemblies should be detected early, and the required corrective actions should be planned accordingly. This paper presents an algorithm for automated quantification of discrepancies for components of assemblies. Rather than using dense point clouds, the geometric skeleton (wireframe) of assemblies is extracted for further manipulation, once the as-built status is captured using the appropriate method. The extracted skeletons, which abstractly represent the designed and built states, are registered using a constrained iterative closest point (ICP) algorithm. In order to identify the points making up each straight segment, the skeletons are clustered, and a straight line is fit to each resulting cluster. The corresponding segments in both states are then compared and investigated for quantifying the incurred discrepancy in the form of a rigid transformation. Experimental results show that the accuracy and speed of the new framework are superior to a previously developed method (3D sliding cube).