Designing thermal functional materials by coupling thermal transport calculations and machine learning
Designing thermal functional materials by coupling thermal transport calculations and machine learning
复制标题
DOI：
10.1063/5.0017042
复制
发表时间：
2020-10-28
影响因子：
3.2
通讯作者：
Shiomi, Junichiro
中科院分区：
文献类型：
--
作者：
Ju, Shenghong;Shimizu, Shuntaro;Shiomi, Junichiro
材料信息学（MI）结合了材料性能计算/测量和信息学算法，其进展已经实现了热功能材料纳米结构中的一些性能，这些性能是基于物理直觉和模型的经验方法所无法达到的。在本教程中，我们以超晶格结构的优化问题为例，介绍了结合热传输计算和机器学习的材料信息学的技术流程和基础知识（补充材料中有示例脚本）。为了提供关于如何使用材料信息学的基本指导，我们描述了有关描述符、目标函数、性能计算器、机器学习（贝叶斯优化）算法以及优化效率的实际细节。然后，我们简要回顾了材料信息学在设计热电和热辐射材料方面最近的成功应用。最后，我们对该主题进行了总结并提供了未来展望。
Advances in materials informatics (MI), which combines material property calculations/measurements and informatics algorithms, have realized properties in the nanostructures of thermal functional materials beyond what is accessible using empirical approaches based on physical instincts and models. In this Tutorial, we introduce technological procedures and underlying knowledge of MI combining thermal transport calculations and machine learning using an optimization problem of superlattice structures as an example (sample script available in the supplement). To provide fundamental guidance on how to use MI, we describe practical details about descriptors, objective functions, property calculators, machine learning (Bayesian optimization) algorithms, and optimization efficiencies. We then briefly review the recent successful applications of MI to design thermoelectric and thermal radiation materials. Finally, we summarize and provide future perspectives about the topic.