# Fileset

[abstract.docx](https://mdr.nims.go.jp/filesets/18f0ae02-bab0-4ed5-9354-f0dc18effc38/download)

## Creator

[BOLYACHKIN Anton](https://orcid.org/0000-0003-0420-1806), [SUBAGJA Toni](https://orcid.org/0009-0003-8810-8808), LI Jiangnan, ZHANG Jiasheng, [ASHOK KRISHNASWAMY Srinithi](https://orcid.org/0000-0001-6209-3837), [ABE Taichi](https://orcid.org/0000-0002-5065-0939), [TANG Xin](https://orcid.org/0000-0001-6762-6145), TOZMAN KARANIKOLAS Pelin, [KULESH Nikita](https://orcid.org/0000-0001-7046-2671), [SEPEHRI AMIN Hossein](https://orcid.org/0000-0002-7856-7897), [OHKUBO Tadakatsu](https://orcid.org/0000-0003-3548-1951), [HONO Kazuhiro](https://orcid.org/0000-0001-7367-0193)

## Rights

[In Copyright](http://rightsstatements.org/vocab/InC/1.0/)

## Other metadata

[Machine Learning Assisted Optimization of SmFe12-based Alloys](https://mdr.nims.go.jp/datasets/8095a264-c064-49d1-8949-230b2fb172ff)

## Fulltext

Machine Learning Assisted Optimization of SmFe12-based AlloysA. Bolyachkin*, T. Subagja, J. Li, J.S. Zhang, A.K. Srinithi, T. Abe, Xin Tang, P. Tozman,N. Kulesh, H. Sepehri-Amin, T. Ohkubo, and K. HonoNational Institute for Materials Science, Tsukuba 305-0047, Japan*bolyachkin.anton@nims.go.jpSmFe12-based alloys with ThMn12-type crystal structure (1:12) are among the most intriguing hard magnetic materials [1], whose remarkable intrinsic magnetic properties (µ0Ms = 1.64 T, µ0Ha = 12 T, and Tc = 555 K for SmFe12 [2]) are difficult to convert into good extrinsic magnetic performance of the bulk magnets – high coercivity and remanence. The main obstacle comes from the instability of the SmFe12 phase, which requires multiple substitutions, e.g., (Sm,Zr)(Fe,Co)12-xMx, where M are nonmagnetic elements such as V, Ti, Mo, and others. Sophisticated optimization is needed to design a composition with minimized loss of magnetization and magnetic anisotropy. Machine learning on large DFT-based datasets has recently been employed to accelerate this optimization [3]. However, there are other critical factors that should also be considered during the optimization – phase composition, separation of 1:12 grains with an intergranular phase [4], crystallographic texture, etc. To control these synthesis-related factors, in this work we conducted machine learning on an experimental dataset. The dataset comprised 908 samples collected from articles and our own experiments on the SmFe12-based alloys which were obtained either by mechanical alloying or by melt-spinning followed by heat treatment. The descriptor of each sample consisted of the chemical composition and synthesis details. The importance of the features and their correlations were analyzed, then two gradient boosting regressors were trained to predict coercivity and remanence as the main targets. Next, a large feature space of (Sm,Zr)x(Fe,Ti,V)100-x melt spun samples with annealing temperatures ranging from 923 to 1373 K (≈57·106 candidates in total) was examined to define a Pareto front of the competing targets. Finally, we proposed a list of the most prospective alloys for an experimental validation, the results of which will be reported, as well as other details of the machine learning on the SmFe12-based isotropic alloys.References[1] P. Tozman, H. Sepehri-Amin, and K. Hono, Scr. Mater. 194 (2021) 113686.[2] Y. Hirayama, Y.K. Takahashi, S. Hirosawa, and K. Hono, Scr. Mater. 138 (2017) 62.[3] D.N. Nguyen, H. Kino, T. Miyake, and H.C. Dam, MRS Bull. 48 (2023) 31.[4] X. Tang, J. Li, A.K. Srinithi, H. Sepehri-Amin, T. Ohkubo, and K. Hono, Scr. Mater. 200 (2021) 113925.