Sustainability-oriented maintenance management of highway bridge networks based on Q-learning

Current bridge management systems provide maintenance strategies that balance costs and structural safety of bridge networks. However, they are cost-oriented rather than sustainability-oriented. Moreover, the existing literature lacks research on reducing the environmental impacts of a bridge network during operation and maintenance periods. This paper develops a condition-based maintenance approach for highway bridges, aiming to minimize the total carbon emissions of bridge networks subject to maintenance budget constraints. A twodimensional Markov chain model is applied to predict the deterioration processes of bridges, and a Q-learning algorithm is proposed to determine the maintenance strategy for a single bridge. Then, integer programming optimizes the budget allocation process for a bridge network. The proposed method is demonstrated using the concrete bridges in New York state. Sensitivity analyses indicate the impacts of budget levels on maintenance plans and total carbon emissions. 10% drop in budget level results in 30 additional bridge repairs delayed as well as 39 tonne extra annual carbon emissions. The proposed optimization framework would contribute to intelligent infrastructure asset management and sustainable society.