세미나 소개
-----------------------------------------------------------------------------------------------------------------
주제: From Vision AI to LLM Agents: Closed-loop Industrial AI for Safety Operations
연사: 임창원 CEO (GlobeAI 물류인공지능 스타트업)
일시: 2026년 5월 11일 (수) 오후 4:30 - 6:00
장소: 한양대학교 제2공학관 502호
세미나 주제:
• Challenges of deploying AI models in real-world industrial environments
• Vision AI for detecting unsafe behaviors, hazardous events, and near-miss situations
• RAG-based integration of manuals, safety regulations, accident cases, and operational knowledge
• LLM agents for risk interpretation, action recommendations, and automated reporting
• Statistical evaluation and validation strategies for reliable industrial AI systems
Abstract:
Recent advances in computer vision, large language models, and retrieval-augmented generation are rapidly expanding the role of AI in industrial environments. However, deploying AI in real-world industrial fields requires more than achieving high benchmark accuracy. Industrial AI systems must operate reliably under complex visual conditions, noisy data, strict latency constraints, and safety-critical decision-making processes.
In this seminar, I will discuss how Vision AI and LLM agents can be integrated into a closed-loop industrial safety operation system. First, I will introduce Vision AI applications for detecting unsafe behaviors and hazardous events, such as personal protective equipment violations, fire, falls, restricted-area intrusion, and worker–equipment near-miss situations. I will then explain how RAG-based systems can connect AI models with domain-specific knowledge, including equipment manuals, safety regulations, accident reports, and internal operational guidelines.
The seminar will further explore how LLM agents can support risk interpretation, action recommendation, incident documentation, and audit-ready reporting. Rather than treating AI as a standalone prediction model, the proposed perspective emphasizes an end-to-end operational workflow: detection, interpretation, risk assessment, recommendation, human approval, documentation, and continuous improvement. Finally, I will discuss statistical evaluation strategies for trustworthy industrial AI, including event-level performance metrics, false alarm management, latency measurement, scenario-based validation, and field deployment challenges.
연사 소개
---------------------------------------------------------------------------------------------------------------------------------------
• (2023 ~ 현재) GlobeAI / 공동대표
• (2014 ~ 현재) 중앙대학교 / 교수
• (2011~ 2014) Loyola University Chicago / Assistant Professor
• (2009 ~ 2011) National Institute of Environmental Health Sciences / Postdoctoral Research Fellow
• (2009) University of North Carolina at Chapel Hill / PhD in Statistics
- 문의: 한양대학교 산업AI부트캠프 사업단 (yoonseonchoi@hanyang.ac.kr)