Clustering over the XRD peak patterns (26 intotal) of [Sm_1−yZr_y] Fe₁₁Ti,performed by DTW (dynamical-time-wrapping) scoring and Ward linkage method.Putting the threshold around 1,000 for the dissimilarity (horizontal brokenline), the patterns are clustered into four groups, sharing almost the samenumber of the substitutions by Zr. The red arrows at the bottom show the errorswhere a ’zero substitution’ is wrongly sorted into the group with ’onesubstitution’ etc.

Abstract

SmFe12 is one of the candidate of the mainphase in rare-earth permanent magnets [1]. The origin of intrinsic propertiesemerging at high temperature as well as that of the phase stability has not yetbeen clarified well. Introducing Ti and Zr to substitute Fe and Sm is found toimprove the magnetic properties and the phase stability. To clarify themechanism how the substitutions improve the properties, it is desired toidentify substituted sites and its amount quantitatively, preferably with highthroughput efficiency for accelerating the 'materials tuning'. Motivated by theabove, we have developed [2] a machine learning clustering technique todistinguish powder XRD patterns to get such microscopic identifications aboutthe atomic substitutions. Ab initio calculations are used to generatesupervising references for the machine learning of XRD patterns: We preparedseveral possible model structures with substituents located on each differentsites over a range of substitution fractions. Geometrical optimizations foreach model give slight different structures each other. Then we generated manyXRD patterns calculated from each structure. We found that the DTW (dynamictime wrapping) analysis can capture slight shifts in XRD peak positionscorresponding to the differences of each relaxed structure, distinguishing thefractions and positions of substituents. We have established such a clusteringtechnique using Ward's analysis on top of the DTW, being capable to sort outsimulated XRD patterns based on the distinction. The established technique canhence learn the correspondence between XRD peak shifts and microscopicstructures with substitutions over many supervising simulated data. Since theab initio simulation can also give several properties such as magnetization foreach structure, the correspondence in the machine learning can further predictfunctional properties of materials when it is applied to the experimental XRDpatterns, not only being capable to distinguish the atomic substitutions. The'machine learning technique for XRD patterns' developed here has therefore thewider range of applications not limited only on magnets, but further on thosematerials which properties are tuned by the atomic substitutions.

举办单位

尊龙凯时物理学院

超硬材料国家重点实验室

计算物理方法与软件创新中心

尊龙凯时省物理学会