Data and code underlying the PhD thesis: Data-driven methods to design, learn, and interpret complex materials across scales

This repository contains code and data related to the underlying PhD thesis: Data-driven methods to design, learn, and interpret complex materials across scales. The repository is divided into the individual codes and datasets of each chapter. <strong>Chapter 2</strong> explores the inverse design of 2D metamaterials for elastic properties, utilizing machine learning techniques to optimize material structure and performance. <strong>Chapter 3</strong> focuses on learning hyperelastic material models without relying on stress data, employing data-driven approaches to predict material behavior under large strains. <strong>Chapter 4</strong> extends this by developing interpretable hyperelastic material models, ensuring both accuracy and physical consistency without stress data. <strong>Chapter 5</strong> explores the inverse design of 3D metamaterials under finite strains and applies novel ML frameworks to design these complex material structures. <strong>Chapter 6</strong> investigates the use of deep learning to uncover key predictors of thermal conductivity in covalent organic frameworks (COFs) and reveals new insights into the relationship between molecular structure and thermal transport. <strong>Chapter 7</strong> introduces a graph grammar-based approach for generating novel polymers in data-scarce settings, thus combines computational design with minimal data.

本仓库收录了与支撑其研发的博士学位论文《跨尺度复杂材料的设计、学习与可解释数据驱动方法》相关的代码与数据集。本仓库按研究章节划分为各章节对应的独立代码与数据集模块。第2章针对弹性性能需求，开展二维超材料的逆向设计研究，借助机器学习技术优化材料结构与性能表现。第3章聚焦于无需应力数据的超弹性材料模型学习任务，采用数据驱动方法预测大应变工况下的材料力学行为。第4章在此基础上研发了无需应力数据的可解释超弹性材料模型，在保证预测精度的同时满足物理一致性要求。第5章探索了有限应变工况下三维超材料的逆向设计，并应用新型机器学习（Machine Learning）框架开展复杂材料结构的设计工作。第6章研究了利用深度学习揭示共价有机框架（Covalent Organic Frameworks, COFs）导热系数关键影响因子的方法，并为理解分子结构与热输运之间的关联提供了全新视角。第7章提出了一种基于图语法的方法，可在数据稀缺场景下生成新型聚合物，实现了仅依赖少量数据的计算设计。