Fiber-reinforced polymers are composite materials made of fibers and a surrounding matrix. Due to their outstanding lightweight potential, they are applied in numerous areas of everyday life, from aerospace, automotive and construction to energy and medical technology. This wide range of applications seconds the importance of constantly improving the materials and the processes used to produce them. The project aims to achieve both by using machine learning methods in order to enable more efficient and accurate component or process simulations.
In this project, developing a surrogate model of the Liquid Composite Molding (LCM) process of manufacturing carbon fiber-reinforced polymers is of interest. In this process, a liquid polymer flows through the preformed fiber structures and solidifies to finally produce the fiber-reinforced composite. This process is governed by flow phenomena of the polymer in different spatial scales spanning across six orders of magnitude from micrometers to meters. Current conventional methods to simulate these flows require high computational effort and time. Hence, the goal is to accelerate the process simulation workflow using hybrid machine learning techniques learning from data and physics. The challenges lie in learning to accurately emulate these flow phenomena (Physics-informed Machine Learning) and efficiently bridging the different spatial scales involved (Multiscale Bridging).
Leibniz-Institut für Verbundwerkstoffe GmbH (IVW)
Leibniz-Institut für Polymerforschung Dresden e.V. (IPF)
Fraunhofer-Institut für Techno- und Wirtschaftsmathematik (Fraunhofer ITWM)
Forschungsverbund Berlin e.V. (Weierstraß-Institut für Angewandte Analysis und Stochastik)