Developing a Method for Digital Twin-Based Geometry Assurance of Sheet Metal Assemblies

Abstract

This thesis investigates how a digital twin-based geometry assurance workflow can be developed for compliant sheet metal assemblies. Current industrial workflows are limited by fragmented data, insufficient traceability between physical and virtual processes, and process adjustments that are not consistently stored or reused in simulation models. The study was conducted in collaboration with Mercedes-Benz AG and combined an exploratory case study, data mapping, compliant variation simulation, sensitivity analyses, machine-learning-based calibration, and a case-based evaluation of the proposed workflow. A rear roof frame assembly was used to identify the information required for non-rigid simulation and to investigate the influence of process parameters. The study identified data breaks related particularly to shimming and welding sequence. The investigated assembly showed a shimming sensitivity of 80.07% and a weldingsequence sensitivity of 31.58% for the scan-based model. A quantile-based regression model was selected for simulation-to-measurement calibration because the available simulation and measurement datasets lacked one-to-one case traceability. The model improved the agreement between distributions by an average of 90.22% and reduced the RMSE of the evaluated scanned case by 29.66%. Applying shimming optimization reproduced a deformation pattern similar to that observed in the physical assembly, although differences in magnitude remained. The results indicate that the proposed workflow can provide a foundation for digital twinbased geometry assurance. However, stronger evaluation and case-specific prediction require additional traceable physical cases, recorded process parameters, and reusable data formats.