Regensburg 2025 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MA: Fachverband Magnetismus
MA 35: PhD Focus Session: Using Artificial Intelligence Tools in Magnetism
MA 35.3: Hauptvortrag
Donnerstag, 20. März 2025, 10:35–11:05, H20
AI used for micromagnetic simulations — •Thomas Schrefl1, Felix Lasthofer1, Qais Ali1, Heisam Moustafa2, Harald Oezelt2, Alexander Kovacs2, Masao Yano3, Noritsugu Sakuma3, Akihito Kinoshita3, Tetsuya Shoji3, and Akira Kato3 — 1Christian Doppler Laboratory for magnet design through physics informed machine learning, Wiener Neustadt, Austria — 2University for Continuing Education Krems, Wiener Neustadt, Austria — 3Advanced Materials Engineering Division, Toyota Motor Corporation, Susono, Japan
Micromagnetic simulations are an excellent means for prediction of magnetic properties. However, the required computational resources limit the use of micromagnetics for materials design. Machine learning models can serve as surrogate for evaluating target properties during optimization. Artificial intelligence can sort pictures based on content or create new images given keywords. Treating the magnetization distribution as an image, methodologies from image processing can be applied in magnetism. We used this approach to predict the magnetization dynamics of thin film elements. The magnetic states are encoded by a convolutional neural network. For bulk magnets a different approach is required. Their three-dimensional grain structure can be represented by a graph. The regular pixels are replaced by the nodes and edges of a graph. We applied graph neural networks to predict hysteresis properties of permanent magnets. Trained machine learning models can be used for inverse design. Given certain targets, optimized magnets are suggested.
Keywords: Micromagnetics; Physics informed machine learning; Graph neural networks