Data-driven Reduced Order Approaches for Micromagnetism

This project is dedicated to improve computational studies for magnetic materials widely used in green energy applications. In fact, many green technologies use permanent magnets, for instance, the generator of a direct drive wind mill or motors of hybrid and electric vehicles. The transition to green energy, therefore, goes hand in hand with a strong increase in demand for permanent magnets and the critical materials contained therein, particularly resource critical rare earth elements. However, apart from the supply chain unpredictability and price fluctuations linked to these materials, the environmentally taxing process of open-pit mining results in significant issues like water contamination, the generation of radioactive waste, and disruption to ecosystems. These issues have fueled significant interest in finding sustainable alternatives such as rare-earth-free or reduced-content magnets alternatives. Nevertheless, developing viable alternatives is challenging because the best performing conventional magnets offer exceptional magnetic properties that are difficult to replicate. This is where computer simulations based on novel and effective (data-driven) methods comes into play. Realistic large-scale 3D models are difficult to investigate computationally with classical approaches since multiple levels of length scale and theories are involved, from quantum mechanics to Maxwells equations. Therefore, the objective of this project is to extend machine learning methods to facilitate large-scale full 3D micromagnetic simulations for magnetic materials and grain structures in order to promote the search for less critical materials with equally good material properties. In order to achieve this goal, the team will develop and build upon innovative numerical techniques and low-parametric descriptions to reduce computational burdens while bridging length-scales effectively. Several mathematical challenges need to be addressed to combine conventional techniques with contemporary data-driven approaches, from magnetic field computation and optimizing algorithms for minimizing total energies and solving time-dependent differential equations to physics- aware machine learning concepts. The project team intends to develop a comprehensive 3D micromagnetic physics-informed open-source simulation package with modular structure also applicable for development in neighboring fields of computational science. The research is conducted by the Principal Investigator Doz. Lukas Exl and his research team at Wolfgang Pauli Institute with national collaboration at Danube University Krems and additional international collaboration with the Department of Energy Conversion and Storage at Technical University of Denmark.