A new system for assessing building damage after wind events

This framework connects images and reports to predict building recovery and downtime.

Researchers developed a multimodal engineering Graph Retrieval-Augmented Generation (GraphRAG) framework. This system links observed physical damage to safety, habitability, and functionality. It also helps infer recovery actions and downtime.

The framework uses image-based and text-based evidence. It transforms this data into a component-level representation. This representation distinguishes confirmed damage, explicit negative evidence, unresolved conditions, and temporary measures.

Graph-aligned retrieval combines comparable building scenarios with engineering relationships. Compatibility, contradiction, evidence-precedence, and abstention mechanisms constrain inference. These mechanisms work under partial observability.

Recovery actions come from supported damage-consequence relationships. Downtime uncertainty is moved through an action-dependency network using Monte Carlo simulation. The system was tested on tornado-affected low-rise residential buildings.

The framework shows improved agreement for building-condition and recovery outcomes. It provides traceable engineering reasoning and preserved evidence provenance. It supports transparent, recovery-oriented assessment from incomplete multimodal post-event evidence.

Key facts

  • The framework uses a multimodal engineering Graph Retrieval-Augmented Generation (GraphRAG) approach.
  • It processes both image-based and text-based evidence from post-wind-event scenarios.
  • The system uses Monte Carlo simulation to propagate downtime uncertainty through an action-dependency network.

All sections

Edition