Date of Award

3-26-2025

Document Type

Thesis

Degree Name

Master of Science in Engineering Management

Department

Department of Systems Engineering and Management

First Advisor

Willie F. Harper, Jr., PhD

Abstract

The United States Air Force maintains a $30 billion portfolio of airfield pavements essential for national security. Despite their importance, budget constraints have limited average annual repair funding to $260 million over the past decade, requiring pavements to last significantly longer than their typical life span. This project addresses the central question: What design features or characteristics contribute to the longevity of Portland Cement Concrete (PCC) airfield pavements? Utilizing data from AFCEC's Airfield Pavement Evaluation (APE) team and the PAVER sustainment management system, this study employed linear regression, logistic regression, and deep artificial neural network (ANN) modeling techniques to identify factors influencing pavement performance. Key explanatory variables included pavement section network type, rank, material properties, and deterioration rates. Data mining techniques were applied to extract nearly 3,800 pavement section records from APE reports and PAVER rollups, which were cleaned, engineered, and integrated into comprehensive datasets. Linear and ANN models predicted pavement life as a continuous variable, whereas the logistic regression model classified pavements as long-lived based on a 60-year threshold for a section to reach a pavement condition index of 50. Key findings indicated that pavement thickness design was crucial, with an optimal performance range observed in pavements at least 8 inches thick, and with minimal benefit past 16 inches. Pavements designed for aircraft over 500,000 lbs, especially those between 700,000-900,000 lbs showed a notable decrease in longevity, indicating sub-optimal design practices for pavements in these ranges. This study also highlighted the importance of well-graded gravels, especially in base layers, which significantly enhanced pavement durability. In contrast, poorly-graded gravels, sands, and fine-grained soils were negatively correlated to longevity, likely due to their inferior structural characteristics. The influence of increasing base layer soil strength, as indicated by k-values, was also recognized as enhancing pavement performance, though less so than soil type and gradation. Pavement flexural strength and slab dimensions had limited impact discerning longevity, likely because of adequate current design standards. Systemic factors, such as mission priorities and maintenance strategies, significantly influenced pavement outcomes, with lower resource allocation likely causing underperformance. Model performance metrics were validated using repeated k-fold cross-validation and bootstrapping to ensure robustness. Overall, the logistic regression model provided the best balance of interpretability and performance, with an area under the curve (AUC) of approximately 0.87. The Deep ANN model marginally outperformed the linear model in predictive capability, achieving 0.5% lower mean absolute error (MAE), 12.6% lower root mean squared error (RMSE), and a 0.1% higher R², but lacked interpretability. The linear model offered actionable insights but utilized a natural log transformation and various interaction terms, limiting its practical application.

AFIT Designator

AFIT-ENV-MS-25-M-093

Comments

An embargo was observed for posting this work.

Distribution Statement A: Distribution Unlimited. Approved for public release. PA case number: 88ABW-2025-0233

Download

Share

COinS