AI for Microstructure Modelling
Accelerating Phase Field models via Physics-informed Machine Learning
Phase Field Modelling pertains to be an impactful method to gain insight into the evolution of structures at the microscopic scale. However, the fine grid sizing necessary to accurately track a solidifying phase front leads considerable computational efforts.
Classic numerical schemes used such as the Finite Difference and Finite Volume method capture the physics well, but are not without their drawbacks with regards to computational efficiency. Especially solving problems on a large scale in parallel remains challenging.
In this cooperation with the University of Bayreuth, we try to implement Phase Field methods together with the relatively new field of physics-informed machine learning. There are techniques within this scope that can accelerate solution time and enable solving problems on massively parallel architectures, i.e. GPUs.
You can follow our work at my GitHub Repo.