Graduation Semester and Year
Summer 2026
Language
English
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Civil Engineering
Department
Civil Engineering
First Advisor
Mohammad Najafi
Second Advisor
Vinayak Kaushal
Third Advisor
Himan Jalali
Fourth Advisor
Caroline Krejci
Sixth Advisor
Jinzhu Yu
Fifth Advisor
Nilo Tsung
Abstract
Aging wastewater infrastructure presents significant challenges for municipalities across the United States, with many sewer networks approaching or exceeding their design life. Conventional inspection methods, such as closed-circuit television (CCTV), are limited by subjectivity, cost, and inefficiency. To address these challenges, this study develops an ensemble machine learning framework for assessing the structural condition of sewer pipelines using inspection records enriched with geospatial attributes.
The primary objective of this research is to enhance predictive accuracy, interpretability, and decision-support for risk-based asset management. The scope of the study encompasses 4,802 CCTV inspection records from Dallas, TX, and Tampa, FL, integrating physical, environmental, and operational variables such as pipe age, material, slope, soil pH, and roadway conditions.
The methodology integrates five supervised classifiers Decision Tree, Logistic Regression, Support Vector Machine, k-Nearest Neighbors, and Multi-Layer Perceptron within a soft-voting ensemble to improve generalizability and robustness. Rigorous data preprocessing steps, including imputation, normalization, and one-hot encoding, ensured consistency across heterogeneous datasets. Geospatial integration using GIS enabled alignment of pipe segments with soil and roadway conditions.
Results indicate that the ensemble model outperformed individual classifiers, achieving 74.2% accuracy and an F1-score of 70.6%. The model exhibited over 95% sensitivity in detecting severely deteriorated pipes (Class 3), supporting proactive inspection and rehabilitation prioritization. Interpretability analyses using SHAP values and partial dependence plots identified pipe age, material, and soil pH as the dominant predictors of deterioration.
The conclusions demonstrate that ensemble machine learning offers a robust, transparent, and transferable framework for predictive sewer condition assessment.
Recommendations for future research include extending the model to additional municipalities, incorporating real-time inspection and environmental data, and integrating the predictive framework into GIS-based decision-support systems for long-term infrastructure planning.
Keywords
Wastewater Infrastructure, Sewer Condition Assessment, Ensemble Machine Learning, Predictive Asset Management, Geographic Information Systems (GIS), CCTV Inspection, Infrastructure Deterioration Prediction.
Disciplines
Engineering Education
License
This work is licensed under a Creative Commons Attribution-NonCommercial-No Derivative Works 4.0 International License.
Recommended Citation
Rouhi, Mahnaz, "Sewer Pipe Condition Assessment Using an Ensemble Machine Learning Framework for Infrastructure Decision Support" (2026). Civil Engineering Dissertations. 9.
https://mavmatrix.uta.edu/civilengineering_dissertations2/9