[24p-E203-1~16] 23.1 Joint Session N "Informatics"

[24p-E203-4] Data-driven approach to XRD analysis: its application to magnetic alloys

〇Kenta Hongo¹, Ryo Maezono², Kosuke Nakano^2,3, Keishu Utimula⁴ (1.RCACI, JAIST, 2.Sch. I.S., JAIST, 3.SISSA, 4.Sch. M.S. JAIST)

Keywords:XRD analysis, clustering, auto-encoder

We have recently proposed two machine-learning models to recognize X-ray diffraction (XRD) patterns: one is a clustering based on Ward's method combined with dissimilarity by dynamical time-wrapping method, another is a neural-network-based auto-encoder model, half of which is used to map the XRD vectors onto the two-dimensional feature space. We applied them to recognize XRD patterns generated for SmFe₁₂-based magnetic alloys substituted partially with Ti and Zr elements. In this talk, we will demonstrate our machine learning models and their results. In particular, we found that our feature space is useful to identify how significant each XRD peak is.