Predicting Sheet-Metal Part Form Deviation Using Discrete Inspection Points

Abstract

In high-volume automotive production there is a need to describe the geometry of a part more accurately in early product-realization phases, when comprehensive measurement data are not yet available at scale, so that large-scale statistical variation analyses can be run quickly with simplified yet credible models. This paper proposes a method to predict part shape from sparse information by optimizing both the number and placement of inspection points subject to a defined accuracy threshold. Two production body-in-white parts, a compact inner d-pillar and a large outer fender, were 3D scanned with a Nikon H120 blue-laser high-precision scanner and compared against finite-element (FE) simulation models. The surface is reconstructed by combining the mesh and stiffness matrix of an FE-based material model implemented in a computer-aided tolerancing (CAT) software, RD&T, and the method is commissioned/validated against the scan. The study is experimental: a high-density scan is used as reference to evaluate how accurately sparse point constraints can reconstruct the full-field deformation. An error-driven greedy algorithm is used to iteratively select additional inspection points at locations where the current reconstruction deviates most from the scanned geometry, until the accuracy criterion is satisfied while minimizing model size and computation time.

Results indicate a correlation between the 3D-scanned geometry and the simulation-based reconstructions, demonstrating that the approach can reproduce part shape with reasonable precision for early-phase analyses. The method enables rapid simulations on large populations of parts with a reduced model that remains accurate enough to support decision-making in tolerancing and variation analysis. The reduced model is compatible with standard non-rigid Monte Carlo variation simulations in RD&T, enabling fast early-phase tolerance studies on large virtual populations of parts.