A data-driven surrogate model to rapidly predict microstructure morphology during physical vapor deposition
A data-driven surrogate model to rapidly predict microstructure morphology during physical vapor deposition
复制标题
一种数据驱动的替代模型用于快速预测物理气相沉积过程中的微结构形态
DOI：
10.1016/j.apm.2020.06.046
复制
发表时间：
2020-12-01
影响因子：
5
通讯作者：
Dingreville, Remi
中科院分区：
文献类型：
--
作者：
Herman, Elizabeth;Stewart, James A.;Dingreville, Remi
在这里，我们提出了一个替代模型，快速预测的二元合金薄膜的微观结构在物理气相沉积。该代理模型是从物理气相沉积相场模拟产生的数据集构建和训练的。它依赖于微观结构的统计表示，主成分分析，多项式混沌展开，和微观结构重建算法来估计作为沉积参数和被沉积材料的性能的函数的微观结构。该协议在简化的物理气相沉积模型上进行，证明了替代模型快速预测作为沉积条件函数的广泛一类微结构的功效，相对于高保真度模型具有良好的准确性。相当大的计算增益从代理模型相比，详细的相场的方法突出了追求这样的方法的重要性，特别是当用于生产参数-微观结构图的微观结构的快速和准确的预测。因此，该替代模型可用于指导沉积条件和沉积材料的选择，以制造具有目标微结构的功能薄膜。(C)2020作者（S）爱思唯尔公司出版
Here, we present a surrogate model that rapidly predicts the microstructures of a binary-alloy thin film during physical vapor deposition. This surrogate model is constructed and trained from a data set produced by phase-field simulations of physical vapor deposition. It relies on a statistical representation of the microstructure, principal component analysis, polynomial chaos expansion, and a microstructure-reconstruction algorithm to estimate the microstructure as a function of the deposition parameters and properties of the materials being deposited. This protocol, exercised on a simplified physical vapor deposition model, demonstrates the efficacy of the surrogate model to rapidly predict a broad class of microstructures as a function of deposition conditions with good accuracy relative to high-fidelity models. The considerable computational gain from the surrogate model compared to the detailed phase-field approach highlights the importance of pursuing such approaches, especially when used for producing parameter-microstructure maps for rapid and accurate predictions of the microstructure. As such, this surrogate model can be used to guide the choice of deposition conditions and materials being deposited to fabricate functional thin films with targeted microstructures. (C) 2020 The Author(s). Published by Elsevier Inc.