发表机构
Helmholtz-Zentrum Dresden-Rossendorf（亥姆霍兹德累斯顿-罗斯多夫中心）
机构由 AI 辅助整理，请以论文原文为准。AI 中文总结
本文提出在ELBE加速器控制系统中集成检索增强生成框架，通过向量数据库索引文档与运行数据，利用大语言模型为操作员提供可追溯的实时知识访问支持。
AI 中文摘要
加速器设施的高效运行日益依赖于对异构运行知识的快速访问，这些知识包括日志簿、联锁报告、机器参数和历史归档数据。在ELBE，我们提出了一个检索增强生成（RAG）框架，将设施文档和运行记录集成到一个统一的AI辅助操作员支持工具中。该系统预期使用存储在向量数据库中的领域自适应嵌入来索引电子日志簿、机器归档时间序列数据和子系统手册。用户查询预期通过大型语言模型处理，该模型检索最相关的运行上下文，并生成带有可追溯来源引用的结构化、面向操作员的响应。本贡献介绍了系统架构、数据集成策略以及迈向实时AI辅助加速器运行所面临的挑战。
英文摘要
The efficient operation of accelerator facilities increas- ingly relies on rapid access to heterogeneous operational knowledge, including logbooks, interlock reports, machine parameters, and historical archive data. At ELBE, we pro- posed a Retrieval-Augmented Generation (RAG) frame- work that integrates facility documentation and operational records into a unified AI-assisted support tool for operators. The system is expected to index electronic logbooks, ma- chine archive time-series data, and subsystem manuals using domain-adapted embeddings stored in a vector database. User queries will be expected to be processed through a large language model that retrieves the most relevant oper- ational context and generates structured, operator-oriented responses with traceable source references. This contribu- tion presents the system architecture, data integration strat- egy, and challenges toward real-time AI-assisted accelerator operation
Commentshttps://jacow.org/ipac2026/pdf/WEP6008.pdf presented at IPAC'26, Deauville, France, May 2026, paper WEP6008, unpublished
Journal refpresented at IPAC'26, Deauville, France, May 2026, paper WEP6008, unpublished
