New preprint on Fourier-based neural operators for simulating mold filling processes
The Chair of Casting Technology presents a new preprint: “Fourier Neural Operators for Two-Phase, 2D Mold-Filling Problems Related to Metal Casting.” In this article, we demonstrate how mold filling processes can be predicted extremely quickly and accurately using Fourier-based neural operators. The data-driven model achieves errors of around 5 % and is two to three orders of magnitude faster than classic CFD simulations. This makes it ideal for integration into the ongoing design and optimization of mold filling processes. The method is not limited to metal casting. It can be applied to a wide range of filling problems involving complex multiphase flows and enables rapid variant studies without complete CFD calculations. The preprint is freely available at this link: https://arxiv.org/abs/2510.25697.