Analysis of the Excess Loss in High-Frequency Magnetization Process Through Machine Learning and Topological Data Analysis

In this study, we investigate the magnetization process at high frequencies based on the energy landscape outputted by machine learning. Employing a combination of topological data analysis (TDA) and machine learning, we analyze how microstructures influence magnetization at frequencies from 1 up to 100 kHz. Our approach uses a magneto-optical Kerr effect (MOKE) microscope for visualizing magnetic domains at various frequencies, revealing insights into their behavior and structure. Persistent homology (PH), a method within TDA, transforms complex topological features of these domains into analyzable vectors. These vectors are then processed through principal component analysis (PCA) to extract the significant information, focusing on the most impactful aspects of the data. This process allows for a detailed examination of the magnetic properties and their changes with frequency, offering an in-depth analysis of material properties under high-frequency conditions. By investigating the elements of PCA, we could analyze the energy loss and connect the topological elements to the anomalous eddy current loss. This study gives a step toward integrating advanced analytical techniques into material science, opening new pathways for innovation in high-frequency applications.