SKM 2023 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
MA: Fachverband Magnetismus
MA 23: Poster Magnetism I
MA 23.67: Poster
Dienstag, 28. März 2023, 17:00–19:00, P1
AI-based Recognition of Numerically Generated Multimode Dispersions — •Paul Schreier1, Milan Ender1,2, Pascal Frey1,2, and Philipp Pirro1 — 1Fachbereich Physik and Landesforschungszentrum OPTIMAS, RPTU Kaiserslautern-Landau, Germany — 2Aithericon, Kaiserslautern, Germany
The collective excitation of a magnetic system can lead to the excitation of spin waves. The relation between their wave vector and frequency is described by their dispersion relation. This can be determined by means of micromagnetic simulations. The resulting numerical data give an intensity distribution of all frequency-wave vector combinations. To extract the dispersion relations from these data, peak detection algorithms must be used. Compared to, for example, optical dispersion relations, spin wave dispersion relations are much more complex and non-linear, which makes their extraction from the raw data difficult. For this reason, a Convolutional Neural Network is trained with synthetic data to obtain a robust analysis tool. The network is specialized to classify segments in images. The training data consists of intensity distributions and dispersion relations calculated from micromagnetic simulations. Conventional peak detection algorithms were first applied to simple dispersion relations to generate a base set of training data. Recombination of this data produces a more complex and larger data set that requires very few resources. Compared to conventional peak detection algorithms, the robustness and fault tolerance can be increased by appropriate training. Evaluations based on dispersion relations can thus be automated to a large extent.