Uncertainty-Aware Reverse Engineering from 3D Scan: Reporting Confidence on Maximum-Likelihood Reconstructed Geometries

Reverse engineering from 3D scans is a common practice for reconstructing CAD geometry used for design recovery and verification. However, the results are usually presented as deterministic models (surfaces) and if at all, uncertainty is only reported for a few derived scalar dimensions. This paper contends that from a metrological standpoint, such practice is incomplete: the main result of scan-based reverse engineering is an estimated geometry, and thus the proper delivery should be a Maximum-Likelihood (ML) geometry along with an uncertainty description. Here, the authors propose an overall, uncertainty-aware 3D scan reverse engineering framework, which starts with a spatial point cloud captures and integrates global uncertainty throughout the entire chain of operations, including sensor performance, point sampling, registration, segmentation, meshing/discretization, filtering-induced bias, model selection, and fitting strategy. Surface uncertainty is communicated via confidence information (such as confidence bands/regions) obtained from the residual pattern of the fitted model (e.g., through resampling/bootstrapping or local linearization) and then propagated to any measure and of interest necessary for decision-making. The method is illustrated by an edge reconstruction problem as a case study revealing how uncertainty associated with the reconstructed geometry can significantly change downstream parameters (such as local curvature radii) and compliance determinations near specification limits. The suggested framework facilitates traceable and transparent reverse engineering by advocating for uncertainty disclosure at the level of reconstructed geometries, thus allowing consistent propagation to any subsequent CAD-based measurement.