大学化学 >> 2026, Vol. 41 >> Issue (10): 364-369.doi: 10.12461/PKU.DXHX202510081

化学实验 上一篇    

基于AI智能体的实验室安全隐患闭环管理系统研究

杨国鑫, 张向荣, 朱敏, 梁军艳, 白艳红, 李银环   

  1. 西安交通大学化学学院, 化学实验教学中心, 陕西 西安 710049
  • 收稿日期:2025-10-23 录用日期:2025-12-19 发布日期:2026-10-13
  • 通讯作者: 杨国鑫, 李银环 E-mail:yangguoxin@xjtu.edu.cn；liyh@xjtu.edu.cn Guoxin Yang, Yinhuan Li
  • 基金资助:
    2024年西安交通大学本科实验实践与创新创业教育教学改革研究专项(24SJZX04)

Research on a closed-loop laboratory safety hazard management system based on AI agents

Guoxin Yang, Xiangrong Zhang, Min Zhu, Junyan Liang, Yanhong Bai, Yinhuan Li   

  1. Chemical Teaching Experiment Center, School of Chemistry, Xi'an Jiaotong University, Xi'an 710049, Shaanxi Province, China
  • Received:2025-10-23 Accepted:2025-12-19 Published:2026-10-13
  • Contact: Guoxin Yang, Yinhuan Li E-mail:yangguoxin@xjtu.edu.cn；liyh@xjtu.edu.cn

PDF

207

赞

摘要： 高校化学实验室安全管理是国家科技创新体系建设中的重要组成部分，但当前管理仍面临安全知识传递碎片化导致人员安全意识与知识储备不足、隐患识别过度依赖人工、管理流程难以形成有效闭环等系统性挑战。本文旨在探索人工智能技术与实验室安全管理深度融合的新路径，提出并实践了一种以“智能体”为核心的化学实验室安全管理模式。研究依托西安交通大学“交晓智”低代码开发平台，设计“隐患猎手HuntRisk”智能体，通过“意图解析-知识库增强-多工作流引擎”的架构体系，整合自然语言处理、计算机视觉及自动化工作流技术，构建了一个集智能问答、多模态隐患识别与自动化处理为一体的实验室安全隐患综合管理系统。

关键词: 实验室安全, 智能体, 隐患识别, 闭环管理

Abstract: Safety management in university chemistry laboratories constitutes a critical component of China’s national scientific and technological innovation system. Nevertheless, current management practices encounter systemic challenges, including fragmented safety knowledge dissemination leading to inadequate personnel awareness and knowledge retention, excessive reliance on manual hazard identification, and difficulties in establishing effective closed-loop management processes. This study explores novel approaches for the deep integration of artificial intelligence technologies with laboratory safety management by proposing and implementing an intelligent agent-centric governance model for chemical laboratory safety. Utilizing Xi’an Jiaotong University’s “Jiaoxiao Zhi” low-code development platform, the research developed the “HuntRisk” intelligent agent. Employing an architectural framework of “intent parsing-knowledge base enhancement-multi-workflow engine”, the system integrates natural language processing, computer vision, and automated workflow technologies to establish a comprehensive laboratory safety hazard management system that combines intelligent Q&A, multimodal hazard identification, and automated processing capabilities.

Key words: Laboratory safety, Intelligent agent, Hazard identification, Closed-loop management