Melt-Pool-Kinetics: A multi-source melt pool compilation for vision-based analytics applications in additive manufacturing

Additive manufacturing (AM) is a technique used to fabricate physical objects by layering materials from a 3D digital model. Melt pool represents the region of superheated molten material in metallic AM. Melt pool monitoring in AM processes has been shown to be effective for a range of downstream analytics tasks. However, metallic AM processes are characterized by a wide variety of process parameters, materials, and monitoring conditions. This diversity has driven the need for a comprehensive dataset that can standardize melt pool image data across process, monitoring, and application contexts. To this end, the current work presents a diverse collection of melt pool images sourced from a range of process monitoring applications. The dataset is systematically collected from research institutions, government organizations, scientific articles, and open databases. A total of 32 datasets across 23 sources, amounting to 1.9 terabytes of data were down sampled and processed to a 48.6 gigabytes general dataset in an HDF5 file format. We release this first compilation as Melt-Pool-Kinetics. This dataset aims to facilitate downstream analytics applications such as machine learning (ML) for in-situ monitoring, process optimization, and closed-loop feedback for AM quality prediction and process repeatability. The dataset is organized into structured levels, offering both raw and processed image data in the standardized HDF5 file format. Additionally, we provide a diverse subset for large datasets, which could enhance model training efficiency and boost performance on AM-related tasks. The current version can be extended in the future by adding latest datasets to enhance its applicability.

增材制造（Additive Manufacturing, AM）是一种通过基于三维数字模型逐层堆叠材料来制造实体物件的技术。熔池（Melt Pool）指金属增材制造中处于过热熔融状态的材料区域。现有研究表明，增材制造过程中的熔池监测可有效支撑一系列下游分析任务。然而，金属增材制造过程具有工艺参数、材料体系及监测条件多样复杂的特点，这种多样性催生了对标准化熔池图像数据集的需求——该数据集需能统一不同工艺、监测场景及应用场景下的熔池图像数据标准。为此，本研究构建了一套涵盖多类工艺监测场景的熔池图像数据集。该数据集的原始数据均经系统采集自科研机构、政府组织、学术文献及公开数据库。本研究从23个数据源中共整合了总计1.9TB的原始数据，并经降采样与标准化处理后，生成了容量为48.6GB的通用数据集，存储格式采用HDF5。我们将这套首个整合型数据集命名为Melt-Pool-Kinetics。本数据集旨在支撑各类下游分析应用，例如用于增材制造原位监测、工艺优化的机器学习（Machine Learning, ML）模型，以及用于质量预测与工艺稳定性提升的闭环反馈系统。该数据集采用分层结构化组织方式，以标准化HDF5格式存储原始图像与处理后图像两类数据。此外，本数据集还提供了面向大规模数据场景的多样化子集，可有效提升模型训练效率，并优化增材制造相关任务的模型性能。当前版本可通过后续新增数据集进行扩展，以进一步提升其应用适用性。

搜集汇总

背景与挑战

背景概述

Melt-Pool-Kinetics是一个多源熔池图像数据集，用于增材制造中的视觉分析应用。数据集包含来自23个来源的32个数据集，经过下采样和处理后以HDF5格式提供，总大小为48.66 GB，支持机器学习和过程监控等下游分析任务。