Degree
Doctor of Philosophy (PhD)
Department
Engineering Science
Document Type
Dissertation
Abstract
The performance of pavement plays a critical role in Maintenance, Rehabilitation and Reconstruction (MR&R) for highway agencies. Reliable and accurate estimation of pavement performance can be instrumental in prioritization of the limited resources and funding for highway agencies. The latter requires robust prediction models that can handle large-scale, real-world data and can forecast pavement performance in the long run. Unfortunately, the traditional performance prediction models have raised concerns regarding their efficiency and accuracy. It is because these models were based on a limited number of explanatory variables, were not up-to-date and were designed for forecasting short-term (up to five years) pavement conditions. Therefore, the ultimate objective of this study was to develop a robust decision-making tool, using advanced machine learning techniques, for predicting the long-term field performance (up to 11 years) of Asphalt Concrete (AC) overlays placed on asphalt pavement in Southern states in the United States (US) using only one pavement condition measurement at the pre-treatment stage and other readily available key project conditions.
The proposed tool was the result of assessing the prediction accuracy of six machine learning algorithms including Decision-Tree (DT), Random Forest (RF), eXtreme Gradient Boosting (XGBoost), Categorical Boosting (CatBoost), Artificial Neural Network (ANN), and ensemble-learning method. Pavement Condition Index (PCI) was used as the pavement performance indicator. For each algorithm, six models were developed sequentially based on historical pavement condition data collected from Louisiana Department of Transportation and Development (LaDOTD) Pavement Management System (PMS) database. A total of 1,115 log miles of randomly placed AC overlay sections in Louisiana were studied. The output of these models was the future PCI of AC overlays at a biannual rate from one to 11 years.
Findings showed that the ensemble-learning method yielded the highest accuracy among the six algorithms in Python. As expected, the prediction accuracy decreased as the prediction horizon increased from year 1 to 11. At the training stage, the prediction accuracy in terms of coefficient of determination (R2) decreased from 0.98 at age 1 to 0.95 at age 11, the Root Mean Squared Error (RMSE) value increased from ±0.49 at age 1 to ±1.89 at age 11, the Mean Absolute Error (MAE) increased form 0.35 at age 1 to 1.10 at age 11. At the testing stage, R2 values decreased from 0.97 at age 1 to 0.80 at age 11, RMSE values increased from ±0.56 to ±3.62, and MAE increased from 0.31 to 2.11.
A decision-making tool, ready for implementation, was created by incorporating the ensemble-learning models. This tool seamlessly combined Python with Visual Basic for Applications (VBA) Macros within Microsoft Excel. Users can effortlessly apply the machine learning models proposed, even without expertise in coding software. Through the long-term prediction tool developed for the AC overlays field performance in hot and humid climate, state agencies could now allocate highway funds in a more robust and cost-effective way over a longer and a more stable time-frame, resulting in significant budget savings.
Date
11-17-2024
Recommended Citation
Mansour, Elise, "DEVELOPMENT OF MACHINE LEARNING-BASED TOOL FOR PREDICTION OF LONG-TERM FIELD PERFORMANCE OF ASPHALT CONCRETE OVERLAYS IN A HOT AND HUMID CLIMATE" (2024). LSU Doctoral Dissertations. 6613.
https://repository.lsu.edu/gradschool_dissertations/6613
Committee Chair
Hassan, Marwa
Included in
Computational Engineering Commons, Construction Engineering and Management Commons, Other Engineering Science and Materials Commons, Other Materials Science and Engineering Commons