Machine Learning–Based Bridge Maintenance Optimization Model for Maximizing Performance within Available Annual Budgets

Effective maintenance planning for bridges is crucial for maintaining their performance, safety, and minimizing maintenance costs. Timely implementation of interventions can improve the performance of bridges and avoid the need for costly interventions. However, bridge maintenance is often delayed because of inadequate planning and budget allocation, as well as resource constraints such as funding. With the availability of historical condition data of bridges in databases such as the National Bridge Inventory (NBI) and National Bridge Elements (NBE), there is an opportunity to use data-driven methods to predict deterioration of bridge elements and optimize their maintenance interventions to maximize the performance of bridges. This paper presents the development of a novel system that uses machine learning (ML) techniques, to predict the condition of concrete bridge elements, and binary linear programming optimization method, to identify the optimal selection of maintenance interventions and their timing, to maximize the performance of bridges while complying with available annual budgets. Four ML methods are explored: decision tree, random forest, gradient boosting, and support vector machines. The results of the ML evaluation show that, while the values of the predictive performance metrics varied for different elements, random forest method had the best performance for all elements. A case study of a concrete bridge is analyzed to evaluate the performance of the system and demonstrate its new capabilities. The case study results show that the developed model identifies optimal maintenance interventions for various annual budgets over a 50-year study period. The primary contributions of this research to the body of knowledge are as follows: (1) the development of a novel system that integrates machine learning techniques and linear programming for predicting bridge element conditions and optimizing maintenance interventions; (2) modeling and predicting the deterioration of bridge elements based on health index metric; and (3) generating long-term maintenance plans for each of the bridge elements to maximize the performance of bridges within available annual budgets. The present system is expected to support decision makers, such as highway agencies, in allocating the limited financial resources for bridge maintenance more efficiently and cost-effectively.