A Dynamic Infrastructure Resilience Assessment Framework Integrating Agentic AI and Causal Reasoning under Extreme Climate Conditions

As climate change increases the frequency and severity of extreme weather events, aging civil infrastructure requires risk assessment methods that can update engineering priorities as exposure and structural condition change. Conventional multi-criteria decision-making methods, particularly the Analytic Hierarchy Process (AHP), provide transparent weighting logic but remain sensitive to subjective judgment and static weight assignments.

This study proposes a dynamic AI-assisted bridge resilience assessment framework integrating fuzzy AHP, an operationalized Agentic AI calibration module, causal reasoning, and physics-informed constraints. In the proposed workflow, the AI agent receives normalized inputs from structural health monitoring or inspection records, climate and flood-exposure scenarios, traffic and network-importance data, emergency accessibility indicators, and maintenance history. Its objective is to recalibrate the decision-weight vector by minimizing the discrepancy between predicted Comprehensive Risk Index (CRI) values and reference damage or condition labels, while enforcing non-negative normalized weights, monotonic risk responses, and structural-mechanics constraints related to deterioration and exposure.

The framework is demonstrated through a computational case study of a representative 25-year-old reinforced concrete bridge with an average daily traffic volume exceeding 35,000 vehicles and increasing flood exposure. The CRI is defined on a 0-1 scale, with 0.60-0.80 interpreted as moderate-to-high risk. Using the calibrated indicator scores, the bridge obtains a CRI of 0.71. The largest contributions are associated with structural performance and environmental exposure, indicating that deterioration-flood interaction is the dominant risk driver in the demonstration case.

Compared with a conventional static AHP baseline, the calibrated framework reduces weight variability by approximately 18-25% in repeated perturbation tests and preserves the same risk class under +/-30% weight perturbations. The results should be interpreted as a framework-level computational validation rather than full field deployment. The main contribution is a transparent and adaptable workflow for translating bridge-condition, climate-exposure, and operational data into maintenance prioritization decisions, with potential relevance to sustainable infrastructure management under SDG 9 and SDG 13.