A novel theory of the magnetostriction mechanism using topological data analysis

去年，我从其他合作者那里获得了单晶Fe-Ga合金的Kerr显微镜图像数据集。与其他同事合作，对数据进行了预处理，去除图像数据中样品表面的噪声和划痕。采用鲁棒主成分分析和奇异值分解相结合的方法对数据进行预处理。对于分析，我已经能够使用无监督机器学习来区分磁化和磁致伸缩的贡献，仅基于图像数据。我使用PCA和快速傅里叶变换相结合的方法提取图像数据的主要特征并连接到物理参数。我观察到PCA可以有效地区分结构域壁的运动和类型，特别是90度和180度结构域。通过观察PCA分解，我可以观察到第一个分量（PC1）与180度域壁直接相关，而第二个分量可以获得180度和90度域壁的信息。综上所述，我可以观察并赋予实验图像数据的PCA特征物理意义。结果表明，该开发方法可以对多物理场进行分析。

Throughout last year, I have obtained a Kerr microscope image dataset of single crystal Fe-Ga alloy, from others collaborators. Working in cooperation with other colleagues, the data has been pre-processed to remove noise and scratches on the surface of the sample in the image data. The pre-processing has been done by a combination of robust principal component analysis (PCA) and singular value decomposition. For the analysis, I have been able to use unsupervised machine learning to separate the contributions of magnetization and magnetostriction based solely on the image data. I used a combination of PCA and fast Fourier transformation to extract the main features of the image data and connected to the physical parameters. I have observed that PCA could effectively distinguish the movements and types of domain walls, specifically 90 and 180-degree domains. By observing the PCA decomposition, I could observe that the first component (PC1) has a directly correlation with the 180-degree domain walls, while the second component could obtain the information from the 180 and 90-degree domain wall. In conclusion, I could observe and assign physical meaning to PCA features for experimental image data. Also, this results shows that multi-physics can be analyzed by the developing method.