MicrographBank: Interfacial Intermetallics

A high-throughput synthetic dataset of heterogeneous microstructure morphologies generated via multi-phase-field simulations and machine learning analysis Description: This Interfacial Microstructure Atlas is a curated dataset consisting of 1672x25 or 41800 microstructure images derived from multi-phase-field simulations designed to explore the effects of uncertain interfacial energies on intermetallic compound (IMC) morphologies. Developed within the framework of Integrated Computational Materials Engineering (ICME), this dataset systematically samples the 3D interface energy space (σSI, σSL, σIL) using Sobol sequences and captures the transient evolution of the Cu6Sn5 phase during heterogeneous solidification in Cu/Sn systems. Each image encodes the phase-field order parameter of the IMC and has been analyzed using Fourier Descriptors, semi-supervised learning (label spreading), and variational autoencoders (VAE) to classify the shapes into wetting, invariant, and dewetting regimes. The dataset enables studies on: Microstructure classification Uncertainty propagation in PDE-based models Process-Structure-Property (PSP) linkages Deep learning applications in synthetic materials data Key Features: 1,672 phase-field simulations with distinct interfacial energy parameters, 1672 x 25 or 41800 microstructure images 388×343 px² raw images, preprocessed to 64×64 px² Corresponding Fourier Descriptors and interface energy labels Label spreading–based shape classification and VAE latent embeddings The data was interpreted in the following article: Machine Learning-Assisted High-Throughput Exploration of Interface Energy Space in Multi-Phase-FieldModel with CALPHAD potential Vahid Attari, Raymundo Arroyave Texas A&M University Link to article: https://doi.org/10.1186/s41313-021-00038-0 For more information please refer to Open Phase-field Microstructure Database (OPMD) curated at https://microstructures.net.基于多相场模拟与机器学习分析构建的高通量异质微观结构形貌合成数据集。 本界面微观结构图谱（Interfacial Microstructure Atlas）为精选整理的数据集，包含1672×25或41800张微观结构图像，均源自多相场模拟，旨在探究不确定界面能对金属间化合物（intermetallic compound, IMC）形貌的影响。本数据集在集成计算材料工程（Integrated Computational Materials Engineering, ICME）框架下开发，采用Sobol序列系统采样三维界面能空间（σ_SI、σ_SL、σ_IL），并捕捉Cu/Sn体系中异质凝固过程内Cu₆Sn₅相的瞬态演化行为。 每张图像均编码有金属间化合物的相场序参数，并通过傅里叶描述子（Fourier Descriptors）、半监督学习（标签传播，label spreading）及变分自编码器（variational autoencoders, VAE）开展分析，将形貌划分为润湿、稳恒及脱湿三类形貌类别。本数据集可支撑以下方向的研究： - 微观结构分类 - 基于偏微分方程（PDE）模型的不确定性传播 - 工艺-组织-性能（Process-Structure-Property, PSP）关联关系 - 深度学习在合成材料数据中的应用 核心特征： 1. 1672组具有差异化界面能参数的相场模拟，对应1672×25或41800张微观结构图像 2. 原始图像分辨率为388×343像素²，已预处理至64×64像素² 3. 配套傅里叶描述子与界面能标签 4. 基于标签传播的形貌分类结果与变分自编码器隐空间嵌入向量 本数据集对应的研究论文如下： 《基于CALPHAD势函数的多相场模型界面能空间机器学习辅助高通量探索》 作者：Vahid Attari、Raymundo Arroyave 单位：德克萨斯农工大学 文章链接：https://doi.org/10.1186/s41313-021-00038-0 更多信息可查阅托管于https://microstructures.net的开放相场微观结构数据库（Open Phase-field Microstructure Database, OPMD）。