Towards Circular Disaster Management: A Foundational Multi-Agent Debris Life Cycle Simulation Framework for AI-Driven Digital Twins and Cyber-Physical-Social Systems

This paper presents a novel debris life cycle simulation framework driven by generative artificial intelligence (AI) agents, designed to support the development of digital twins (DTs) and cyber-physical-social systems (CPSS) for circular disaster management. Escalating climate-induced disasters produce vast quantities of construction and demolition (C&D) debris that pose significant challenges to urban resilience and sustainability by straining local capacities, disrupting essential services, and causing substantial environmental impacts. Conventional linear debris management approaches, i.e., dominated by rapid clearance and mass landfilling, exacerbate resource depletion, pollution, and greenhouse gas (GHG) emissions while foreclosing opportunities for material recovery. Circular economy strategies, emphasizing reuse, repair, and recycling, offer a more sustainable recovery pathway but remain underutilized for post-disaster contexts. To address this challenge, this study develops a multi-agent framework that connects cyber (virtual building stock models and simulations), physical (material stocks, debris flows, and environmental impacts), and social (a stakeholder engagement interface) dimensions, embodying CPSS-aligned DTs for community-scale circular recovery planning. The proposed framework integrates large language models (LLMs), building information modeling, geographic information systems, life cycle assessment, interactive visualization, and AI-generated audio narration to simulate and compare linear and circular debris management scenarios at the community scale. Using a bottom-up archetype-based approach, the framework profiles building stocks by typology and vintage, enabling detailed debris quantification and cradle-to-grave environmental impact assessment across three scenarios: linear landfilling, moderate circularity via recycling, and enhanced circularity combining reuse and recycling. Validated through a case study of the 2023 tornado in Sullivan, Indiana, the framework demonstrates that circular strategies can divert nearly 46% of debris from landfills and reduce embodied GHG emissions by approximately 28% compared to conventional methods. The LLMs, coupled with an interactive user interface, translate complex life cycle metrics into stakeholder-friendly insights, enabling transparent, participatory decision-making among emergency managers, planners, and policymakers. Overall, this research establishes a conceptual and technical foundation for advancing AI-driven DTs and CPSS that enable circular disaster management. Future work should focus on real-time data integration, expanded impact metrics, and enhanced multi-agent coordination to enable adaptive, scalable, and resilient urban recovery systems.