Data-driven parameter identification for processing of Sheet Molding Compounds

Abstract

The reliable simulation of Sheet Molding Compound (SMC) compression molding requires an accurate description of its rheology. Two aspects are particularly critical: nonNewtonian viscosity and wall-slip behaviour. These factors, combined with anisotropic fiber orientation within the polymer, make a complex flow process. Classical rheological methods, such as plate-plate or capillary rheometry, fail to capture this complexity, while modern in-situ rheometers, though more representative, demand elaborate and costly experimental setups. To identify this complex structure with a simpler approach this thesis presents a data-driven parameter identification framework for SMC rheology. The method uses Bayesian Optimization (BO) with Gaussian Process (GP) surrogate modeling, combined with Autodesk Moldflow through a python script to automate the optimization. Experimental force-time curves serve as the reference data, enabling direct calibration of a modified Cross viscosity model and wall slip model under process conditions. The optimization loop was first validated on a reduced two-parameter slip model before being extended to the complete parameter set, including viscosity and slip. The results demonstrate that the framework converges towards parameter sets that minimize the error between simulations and experiments. Overall, the proposed framework provides a scalable methodology for data-driven rheology identification in composite processing, bridging the gap between industrial experiments and process simulations.