Integration of inspection and monitoring data for RL-enhanced sustainable life-cycle management of infrastructure networks

Abstract

Existing civil infrastructure may face challenges from deterioration and harsh environmental conditions, necessitating effective management strategies to ensure resilience and sustainability. The abundance of inspection and structural health monitoring (SHM) data provides a valuable opportunity for assessment and management using reinforcement learning (RL). This study presents a sustainable management framework for infrastructure network incorporating inspection and SHM data, aiming to establish more efficient life-cycle maintenance policies that prioritize safety and sustainability. Initial probabilistic deterioration models for grouped infrastructure are developed with Markov models, with subsequent Bayesian updating based on real-time SHM data. The sustainability assessment is conducted, incorporating safety, economic viability, and low-carbon factors, with the results transformed into a utility model to inform optimization rewards. The final sustainable management planning is achieved with RL techniques. Key findings highlight the framework’s ability to integrate various data sources, enabling accurate predictions of structural performance and sustainable maintenance needs. The optimal management policy derived from RL ensures good sustainable performance while balancing regional budgetary constraints. Validation on a transportation infrastructure network demonstrates practical utility, with efficient maintenance practices leading to improvements in both efficiency and sustainability compared to traditional methods. Overall, this framework offers a promising approach to sustainable infrastructure management.