Link to Latest Report : Coming Soon.
Background :
Bridges are fundamental to transportation infrastructure, supporting economic activity and public mobility. However, maintaining these structures presents significant challenges due to limited budgets, aging materials, and the complexity of predicting deterioration patterns. Traditional bridge maintenance strategies primarily rely on periodic visual inspections and rule-based scheduling, which have been shown to be inefficient in accurately predicting bridge deterioration trends (Liu & Zhang, 2020). Existing bridge management systems lack the capability to integrate heterogeneous data sources, including textual inspection reports, sensor data, and historical maintenance records, leading to suboptimal resource allocation and delayed interventions (Trach et al., 2022).
A critical challenge is the uncertainty in deterioration models, which often fail to account for the variability in bridge materials, construction methods, and environmental conditions (Karaaslan, 2019) . Current decision-making frameworks heavily depend on expert judgment, leading to inconsistencies and inefficiencies. Additionally, scalability remains a concern, as existing approaches may not be easily adaptable across different bridge types and geographic locations due to the lack of standardized data collection and processing methods (Feng et al., 2023).
Despite the promise of AI and machine learning for predictive maintenance, challenges such as data availability, integration with legacy systems, and interpretability of AI-driven recommendations hinder their widespread adoption (Liu & Zhang, 2020). Addressing these challenges requires a comprehensive approach that combines advanced predictive modeling, optimization techniques, and seamless integration with existing bridge management frameworks to ensure cost-effective and timely maintenance interventions.
Objective :
The objectives of this research include:
- AI Model Development: Develop machine learning models capable of accurately predicting the condition of bridge elements using historical and real-time data sources.
- Optimization of Maintenance Interventions: Integrate binary linear programming techniques to optimize maintenance interventions under budget constraints.
- Decision-Support Framework: Implement a data-driven decision-support framework that assists transportation agencies in planning and resource allocation efficiently.
Scope :
The proposed project includes several tasks:
Task 1 – Literature Review:
- In this task, A comprehensive literature review will be conducted focusing on the application of artificial intelligence (AI) in bridge maintenance, with a particular emphasis on predictive analytics, optimization modeling, and decision-support frameworks to understand current challenges, methodologies, and advancements in bridge maintenance optimization using AI. The review will cover traditional maintenance strategies, their limitations, and recent developments in machine learning and optimization techniques. The literature review will also aim to explore key areas such as machine learning (ML) techniques for condition assessment, deep learning (DL) for complex data processing, and optimization techniques for resource allocation.
Task 2 – Data Collection.
- Reliable and comprehensive data will be collected to develop an accurate AI model for bridge maintenance optimization. The raw data from existing bridge inventories often contain missing values and non-relevant information, making them unsuitable for direct application in machine learning models. Therefore, based on a thorough literature review and the availability of relevant data, a set of input variables has been carefully selected for developing the proposed AI models in this study. These selected features aim to capture the critical aspects influencing bridge deterioration and maintenance planning.
Task 3 – Data Processing.
- Raw NBI and NBE database data, as well as other traffic and environmental data, often contain missing values and redundant data and are not directly usable as input to machine learning models. The data will, therefore, be reorganized, cleaned, and restructured to form structured data that can be utilized for AI analysis. The raw data files, which include bridge records from the NBI database, will be reorganized into a structured data repository containing the selected features that will be used for training the predictive model. The original NBI data is organized based on bridge structure numbers, NBI Item ID, and calendar year, which increases data retrieval times when processing historical trends of specific features. To optimize data handling, the reorganization will involve constructing tables where rows represent individual bridges, identified by their unique structure number, and columns represent the historical feature values spanning multiple inspection years.
Task 4 – AI Model Developement
- Selecting the right model is essential to ensure optimal performance in predicting bridge deterioration and assisting in proactive maintenance planning. Traditional corrective maintenance strategies often rely on manual inspections and reactive interventions, which are not only time-consuming but also costly and inefficient. Predictive maintenance, enabled by advanced AI and machine learning (ML) techniques, offers a proactive approach to identifying potential bridge deterioration early, optimizing resource allocation, and improving overall infrastructure management. The proposed AI framework leverages three machine learning models—Decision Trees (DT), Random Forest (RF), and Artificial Neural Networks (ANN)—each providing unique strengths in analyzing bridge inspection data, categorizing deterioration levels, and prioritizing maintenance actions.
Task 5 – Optimization Modeling and Integration.
- Traditional methods are nevertheless insufficient to resolve complex, multi-dimensional problems during bridge maintenance, such as budget limitations, diverse performance objectives, and environmental considerations. To ease these challenges, this study proposes the use of one of the latest evolutionary optimization methods, Genetic Algorithm, to develop optimal maintenance strategies that balance cost-effectiveness and improvement in performance.
Task 6 – Cast Study Analysis and Result Evaluation
- IntelliBridge system will be tested through real-world case studies to validate its performance and effectiveness in optimizing bridge maintenance.
Research Team:
Principal Investigator: Qianwen (Vivian) Guo, Ph.D.
Co-Principal Investigator: Eren Erman Ozguven, Ph.D.