粒子擴散數(shù)據(jù)庫的高效構建—自動評估與不確定度量化

擴散指的是粒子（即原子，分子或離子）的隨機運動。單個粒子的運動是隨機的，但它受系統(tǒng)的熱力學和動力學狀態(tài)控制。在復雜的系統(tǒng)中，不同類型的粒子間相互作用往往不同。隨著體系中組元數(shù)目的增加，這種粒子相互作用會越來越復雜。以高熵合金為例，其復雜的粒子相互作用不但有助于高熵合金的緩慢擴散效應，而且還賦予了高熵合金優(yōu)異的力學和服役性能。

為了表征復雜的粒子相互作用對擴散過程的影響，人們希望用精準的數(shù)學物理函數(shù)來描述系統(tǒng)的擴散速率。多組（主）元合金系統(tǒng)所覆蓋的成分范圍廣闊，而且擴散系數(shù)函數(shù)的參數(shù)評估需要以大量的實驗數(shù)據(jù)為基礎，因此建立精準的多元合金的擴散速率與成分以及溫度的關系是一項極具挑戰(zhàn)性的任務。

來自中南大學粉末冶金國家重點實驗室的張利軍教授團隊基于數(shù)據(jù)挖掘技術，搭建了一個基于大規(guī)模數(shù)據(jù)集對擴散動力學數(shù)據(jù)庫的參數(shù)進行自動化篩選、評估和不確定度量化的計算框架，并成功應用于CoCrFeMnNi高熵合金面心立方相擴散動力學數(shù)據(jù)庫的建立。

值得一提的是，該研究小組對CoCrFeMnNi高熵合金的擴散動力學數(shù)據(jù)庫進行深入分析發(fā)現(xiàn)，高熵合金的擴散速率與其配置熵并無明顯的相關性，而是呈現(xiàn)隨成分、溫度變化的復雜關系。他們的研究證明，自動化計算框架能夠提供一種自適應和自更新的高質(zhì)量數(shù)據(jù)庫參數(shù)評估機制，展示了數(shù)據(jù)挖掘技術在擴散動力學數(shù)據(jù)庫自動化計算和評估方面的巨大優(yōu)勢。

自動化的擴散動力學數(shù)據(jù)庫構建方法不僅可有效提高材料基因數(shù)據(jù)庫的評估效率和質(zhì)量，還能為揭示復雜的材料現(xiàn)象提供充分的數(shù)據(jù)支持。 該文近期發(fā)表于npj Computational Materials 7： 35 （2021），英文標題與摘要如下。

Automation of diffusion database development in multicomponent alloys from large number of experimental composition profiles

Jing Zhong， Li Chen & Lijun Zhang

Nowadays， the urgency for the high-quality interdiffusion coefficients and atomic mobilities with quantified uncertainties in multicomponent/multi-principal element alloys， which are indispensable for comprehensive understanding of the diffusion-controlled processes during their preparation and service periods， is merging as a momentous trending in materials community.

However， the traditional exploration approach for database development relies heavily on expertise and labor-intensive computation， and is thus intractable for complex systems. In this paper， we augmented the HitDIC （High-throughput Determination of Interdiffusion Coefficients， https://hitdic.com） software into a computation framework for automatic and efficient extraction of interdiffusion coefficients and development of atomic mobility database directly from large number of experimental composition profiles.

Such an efficient framework proceeds in a workflow of automation concerning techniques of data-cleaning， feature engineering， regularization， uncertainty quantification and parallelism， for sake of agilely establishing high-quality kinetic database for target alloy. Demonstration of the developed infrastructures was finally conducted in fcc CoCrFeMnNi high-entropy alloys with dataset of 170 diffusion couples and 34，000 composition points for verifying their reliability and efficiency.

Thorough investigation over the obtained kinetic descriptions indicated that the sluggish diffusion is merely unilateral interpretation over specific composition and temperature ranges affiliated to limited dataset. It is inferred that data-mining over large number of experimental data with the combinatorial infrastructures are superior to reveal extremely complex composition- and temperature-dependent thermal-physical properties.

編輯：jq