Speaker
Description
Beamline operation at the High Energy Photon Source (HEPS) requires the integration of control systems, experimental data, operational knowledge, and data-analysis tools. This work presents an intelligent agent-assisted framework for HEPS beamline operations based on locally deployed Large Language Models (LLMs) and the Osprey agent platform. The framework connects the language model with beamline data and control interfaces while keeping experimental data within the local HEPS computing environment.
A controlled Plan–Check–Execute workflow is adopted to separate LLM reasoning from direct device operations. The current system integrates internal technical documents, EPICS process variables, historical archive data, detector images, and scan-control interfaces. Implemented functions include source-referenced local document retrieval, visualization of historical EPICS data, detector image reconstruction and OpenCV-based image analysis, and one-dimensional motor scanning.
In a representative scanning experiment, detector images acquired at different motor positions are analyzed using image metrics, allowing the system to recommend a refined scan range for subsequent verification and execution. The results demonstrate that the proposed framework can assist users in experimental planning, data analysis, and scan optimization within a locally deployed and controlled environment. Human supervision and confirmation are retained for operations that modify device states, thereby improving operational efficiency while maintaining data security and experimental controllability.
