1-21. Machine learning prediction of alloying element occupation of superalloy based on the first-principles calculations

Tangyu chao2, Bin xiao1, Yi Liu1,2*

1 Institute of material genome engineering, Shanghai university 2 Department of physics, Shanghai university

Abstract: The preferential site occupancy of alloying elements is fundamental to understanding the strengthening mechanism for superalloy design. It is, however, a formidable task for first-principles (FP) calculations to explore the enormous potential doping configurations in the complex multi-component alloys. The data mining models were developed based on the preliminary FP calculated data for several hundred configurations of Nb-base superalloy, e.g. 8-10 alloying element substitution at multiple nonequivalent sites. The data mining models can be more efficiently used for further prediction that reduces the cost of expensive FP calculations while maintains the certain accuracy. In this work we proposed a “Center Environment” (CE) model to construct descriptive features by combining elemental properties and local structural/chemical environment information. It is shown that the CE features can be correlated well with both substitution energy and local geometry of alloying elements in Ni-base superalloy. Our work demonstrated that the machine learning prediction using feature construction with both composition and structure information is more accurate and robust than that with composition only. It is an effective strategy to combine first-principle calculations and data mining models to explore efficiently enormous configurations usually required in materials design.

Keywords: Machine learning; High-throughput computation; First-principles calculation; Multiple-component alloy; Superalloy

   基于第一性原理计算的高温合金元素占位的机器学习预测

唐宇超 2，肖斌 1，刘轶 1，2* 

1.上海大学材料基因组工程研究院，上海市南陈路 333 号(200444)

2.上海大学物理系，上海市上大路 99 号(200444) 

摘要:铌硅基高温合金由于其高熔点和良好的机械性能成为替代目前镍基高温合金的候选材料 之一。铌硅基合金中合金化元素达到 8-10 余种。理解合金化元素对体系的能量和几何结构的影 响对确定合金化元素在不同相和位点中的择优占据至关重要，是进一步研究合金元素强韧化机 制和成分优化设计的基础。由于其巨大的构型和元素变化多样性，合金化元素占位的第一性原 理计算研究变得极其昂贵而受限。本工作将高通量第一性原理计算和机器学习方法结合，研究 铌硅基单晶高温合金中合金化元素置换造成的局部能量和几何结构变化。构建具有第一性原理计算精度的机器学习预测模型，加速合金元素占位的计算研究。 首先我们通过第一性原理密度泛函 GGA-PBE 方法高通量计算了 Nb 和 Nb5Si3 相中多种元素在晶格双位点上置换时引起的能量变化-置换能及局部平均键长变化。基于上述第一性原理计算 产生的数据构建机器学习模型，预测铌硅单晶高温合金元素的置换能和几何结构。我们设计了 能够描述局部成分和结构特征的“中心-环境(center-environment，简称 CE)”模型，将合金元素的基本物性参数按照 CE 模型构建描述因子，通过随机森林和支持向量机方法进行机器学习建 模预测。然后对机器学习的数据集进行特征选择，确定影响合金元素置换能和局部平均键长的 重要因素。本研究表明机器学习方法可在一定精度内替代部分的第一性原理计算预测性质，从 而加速合金化元素占位研究。结合计算和数据的方法能够兼顾计算的精度和机器学习的效率， 已成为加速计算材料设计的有效手段。 关键词:机器学习;高通量计算;第一性原理计算;合金化元素;高温合金