Statistical Analysis of Simulation-to-Reality Deviation in Deep Drawing with a Benchmark Dataset

Abstract Despite widespread research on surrogate modelling in sheet metal forming processes based on finite element simulation data, validating such simulation-based surrogate models against real production data remains a critical challenge for their industrial deployment. Therefore, this paper introduces a benchmark dataset specifically designed to enable systematic investigation of the simulation-to-reality (Sim2Real) gap in deep-drawing processes. The dataset comprises synchronised measurements determined in real-world experiments and corresponding simulation data. Here, force, part geometry, oil film, and initial sheet metal thickness were recorded for two geometries at varied process parameters, resulting in 9,000 matched simulation-reality instances. A hierarchical variance decomposition revealed that part geometry accounts for 77 % to 92 % of the total deviation variance, while blank-holder force explains up to 33 % of the geometry-adjusted variance. A gradient-boosted ensemble captured non-linear interactions with cross-validated R 2 values of 81 % to 92 %, identifying maximum forming force and oil application consistency as additional predictive factors. The dataset and baseline implementations are publicly released to accelerate research in data-driven sheet metal forming.