Deep learning approaches in data-driven pavement performance analysis and asset management

Thumbnail Image
CitirRazavi_iastate_0097E_21248.pdf (10.24 MB)

File Embargoed Until: (2024-07-25)
Citir Razavi, Nazik
Major Professor
Ceylan, Halil
Alipour, Alice
Kim, Sunghwan
Cho, In-Ho
Avci, Onur
Zhang, Wensheng
Committee Member
Journal Title
Journal ISSN
Volume Title
Research Projects
Organizational Units
Journal Issue
Is Version Of
Civil, Construction, and Environmental Engineering
Pavement management systems (PMS) play a vital role in cost-effective road network management, optimizing pavement performance over its predicted service life. The Moving Ahead for Progress in the 21st Century program (MAP-21) mandates US state highway agencies (SHAs) to adopt performance-based approaches in their pavement management processes, and utilizing a remaining service life (RSL) model is one such performance-based approach to facilitate pavement management decisions for SHAs. Accurate performance prediction modeling is vital for successful PMS implementation to plan future maintenance and rehabilitation strategies. In Iowa, secondary roads constitute over 70% of the entire road network, with more than 20% being paved and hard-surfaced. These roads are significant for public and private property access, requiring constant maintenance and reconstruction, often with substantial budgets allocated to them. Many of Iowa's county roads have multi-layered structures due to numerous construction and renewal cycles, posing challenges for county engineers in predicting both functional performance and structural capacities. Collecting distress and condition data for secondary roads is also labor-intensive and time-consuming, possibility limiting the data available for robust computational models. To address these complexities, this study introduces Artificial Intelligence-based deep learning approaches in data-driven pavement analysis and management. The comprehensive framework includes a detailed methodology for database generation and artificial neural network (ANN)-based pavement performance prediction models for connecting pavement age, material properties, and traffic inputs to pavement condition and distresses in rigid, flexible, and composite county road sections. Structural performance prediction models also establish connections between material and structural features and pavement responses associated with structural capacities. This study delves into the multifaceted realm of pavement engineering, seeking to enhance our understanding of pavement performance, predictive modeling, and effective asset management. By addressing the complexities of Iowa's county road network and providing user-friendly tools for decision-making, these models, supplemented with remaining service life predictions, offer a data-driven comprehensive approach to pavement engineering and management.