VDInM: A versatile decision-making agent for infrastructure life-cycle management

The performance of infrastructure deteriorates over time during its service life cycle. Many structural components fail earlier than infrastructure designed life, making effective inspection and maintenance scheduling essential for ensuring safety and extending service life. However, current infrastructure management systems lack intelligence and cannot support proactive maintenance policies. Traditional passive maintenance strategies are unable to rationally allocate non-destructive testing and preventive maintenance, leading to inefficient use of maintenance resources. To tackle this, this study develops a deep reinforcement learning–based decision-support tool that trains an intelligent agent to recommend maintenance actions for multiple components. The agent takes component state, deterioration behavior, and service age as inputs and outputs the most suitable maintenance decision, while also generating future maintenance schedules and estimating the frequency of different actions over the life cycle to support budget planning. Comparative case studies show that, relative to traditional strategies, the proposed intelligent policy can reduce life-cycle maintenance costs by approximately 5–30% across different component types while satisfying safety requirements. To facilitate practical use, all functionalities are integrated into a Python-based graphical user interface that enables maintenance personnel to generate life-cycle maintenance plans by providing only basic infrastructure information.