发表机构
Institute of Integrated Research, Institute of Science Tokyo; The Institute for Solid State Physics, The University of Tokyo; Department of Materials Science and Technology, Tokyo University of Science（东京科学大学综合研究学院; 东京大学固体物理研究所; 东京理科大学材料科学与技术系）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本文提出整合PIONEER系统、OMNES及机器学习分析的以数据为中心的框架，实现软X射线光束线数据的生成、积累与利用，保障AI辅助材料研究的可靠数据支撑。
AI 中文摘要
人工智能正快速变革材料探索领域，但可靠的人工智能辅助研究依赖于保留上下文、来源及机器可读结构的实验数据。因此，高通量同步辐射测量不仅需要自动化数据采集，还需将数据生成、积累与利用整合到连续工作流程中。本文提出一种以数据为中心的研究框架，整合PIONEER系统、OMNES及机器学习分析工具。PIONEER系统在NanoTerasu的BL08U软X射线光束线实现样品处理、真空转移及X射线吸收谱（XAS）扫描自动化，系统生成大量空间分辨光谱数据。测量与分析文件存储于基于云的ARIM-mdx数据系统，OMNES则管理描述样品、制备流程、测量条件、仪器设置、数据位置及分析历史的元数据。元数据结构采用表格与JavaScript对象表示法（JSON）格式，数值与单位分开存储以利机器处理。OMNES基于实验与分析过程间的因果关系组织这些记录，通过关系数据模型将样品、测量、原始数据、处理后数据及分析结果关联，并以图形式可视化其关系。这种基于因果性的组织保留了实验与分析来源，明确各结果如何由前期过程与数据推导而来。积累的光谱随后可通过机器学习方法（包括降维与材料分类）处理，所得数据可重新注册到OMNES中。本文以BN XAS光谱的机器学习分析为例，展示了可应用于该光束线采集数据的分析工作流程。
英文摘要
AI is rapidly transforming materials exploration, yet reliable AI-assisted research depends on experimental data that retains their context, provenance, and machine-readable structure. High-throughput synchrotron measurements therefore require more than automated data acquisition: data generation, accumulation, and utilization must be connected within a continuous workflow. Here, we present a data-centric research framework that integrates the PIONEER system, the OMNES, and ML analysis. The PIONEER system automates sample handling, vacuum transfer, and scanning XAS at the BL08U soft X-ray beamline of NanoTerasu, systematically generating large volumes of spatially resolved spectral data. Measurement and analysis files are stored in the cloud-based ARIM-mdx data system, while OMNES manages metadata describing samples, preparation procedures, measurement conditions, instrument settings, data locations, and analysis histories. Tabular and JavaScript Object Notation formats are used according to the metadata structure, with numerical values stored separately from units to facilitate machine processing. OMNES organizes these records based on causal relationships among experimental and analytical processes, linking samples, measurements, raw data, processed data, and analysis results through a relational data model and visualizing their relationships as a graph. This causality-based organization preserves experimental and analytical provenance and clarifies how each result was derived from the preceding processes and data. The accumulated spectra can subsequently be processed using ML methods, including dimensionality reduction and materials classification, and the resulting data can be registered back into OMNES. Previously reported ML analysis of BN XAS spectra is presented as an example of the analytical workflow that can be applied to data acquired at the beamline.
