LLM-Driven Multi-Agent Inspection Planning via Semantically Enriched Knowledge Graphs for Non-Destructive Testing

This paper presents a multi-agent system for inspection planning in Non-Destructive Testing (NDT), extending our prior work on Knowledge Graph (KG) generation. We integrate a KG with agents—PlannerAgent, ToolSelectorAgent, and ForecasterAgent—that collaborate via LangChain to create context-aware inspection plans, select tools, and forecast timelines.

Key contributions include enriching the KG with new semantic relationships for infrastructure types, materials, and defects, as well as developing a real-time Streamlit-based UI for visualizing plans and reasoning subgraphs. Case studies demonstrate the system’s improved relevance, explainability, and precision in defect-to-NDT mapping. We also address interoperability with Linked Data standards and validate plan consistency using SHACL constraints. Overall, this work showcases a novel integration of semantic web technology and LLM-based reasoning for infrastructure maintenance.