SHM-Agents: AI system lets engineers inspect bridges via natural language

A new paper from Yuequan Bao and colleagues introduces SHM-Agents, a hybrid AI system that integrates the reasoning and planning capabilities of large language models (LLMs) with the accuracy of specialized engineering algorithms. Structural health monitoring (SHM) traditionally requires separate, hard-to-use tools for each task—anomaly detection, signal processing, modal analysis, damage identification, finite element model updating, and more. SHM-Agents wraps these into a single agent framework that users control through natural language, enabling an engineer to simply type a request and get results without writing code or training separate models.

The system uses a generalist agent powered by an LLM to interpret user intent and orchestrate work across a suite of specialist agents, each optimized for a specific SHM function. It supports deep learning pre-training to simplify deployment on new structures and allows flexible expansion via a modular plugin design. In experiments on a long-span cable-stayed bridge, SHM-Agents successfully performed 11 distinct tasks including data anomaly diagnosis and recovery, statistical analysis, vehicle load modeling, reliability assessment, and bridge knowledge Q&A. The work bridges the gap between cutting-edge AI and practical civil engineering, potentially lowering barriers for widespread adoption of smart infrastructure monitoring.