In the case of structures made of complex structured materials, such as fiber composites, the manufacturing process has a major influence on the local properties of the component. Although these effects can be predicted virtually using process simulations, such simulations are too computationally intensive for algorithmic optimization of the component in many iterations. We are therefore developing data-driven methods for the rapid approximation of such effects.?

?

• Fast approximation of mold filling processes of discontinuously reinforced fiber-reinforced composites with machine learning ( Greif, Lechner & Meyer 2024, Greif & Meyer 2026)
• Bayesian optimization of printing parameters in ceramic 3D printing

Conventional structural optimization methods are usually iterative procedures with high computing times. We utilize machine learning methods and automatic differentiation frameworks to accelerate the optimization process. A key focus is placed on the optimization of sustainable structures.

For instance, hybrid composite structures are designed to be not only lightweight and durable but also specifically reusable at the end of their first life cycle.

?

• Patch Optimization?with differentiable FEM ( Riesner, Koch & Meyer 2026)
• Design for Remanufacturing of hybrid materials (DFG project ReHyb in collaboration with Prof. Weidemann's group)

Using differentiable finite element models and experimental measurement data, we reconstruct local material properties.

• Detection of defects in composite structures ( CfP project "From Effect to Defect" in cooperation with Prof. Peterseim's group)

• Neural Fields as material properties ( torch-fem example)