Graduation Semester and Year
Fall 2025
Language
English
Document Type
Dissertation
Degree Name
Doctor of Philosophy in Civil Engineering
Department
Civil Engineering
First Advisor
Vinayak Kaushal, Ph.D., P.E.
Abstract
Aging wastewater infrastructure in the United States demands renewal strategies that balance cost-effectiveness, sustainability, and minimal social disruption. Conventional open-cut pipeline installation (OCPI) involves extensive excavation and restoration, whereas trenchless alternatives such as Cured-in-Place Pipe (CIPP) Renewal Technology offer reduced surface disturbance and environmental impacts. However, existing studies primarily focus on Construction Costs (CC), neglecting integrated life-cycle evaluations that also account for Environmental Costs (EC) and Social Costs (SC).
The primary objectives of this research are to develop supervised machine-learning (ML) models to estimate and predict CC in 2025 US dollars (USD) per linear foot of OCPI and CIPP Renewal Technology across small, medium, large, and very large sewer diameters (6–84 in.) and to establish an integrated LCCA–ML framework that unifies CC, EC, and SC modules for holistic cost prediction and comparative assessment. Secondary objectives include: (1) developing, training, and validating four ML models—Multiple Linear Regression (MLR), k-Nearest Neighbors (KNN), Decision Tree Regressor (DTR), and Gradient Boosting Regressor (GBR), using cross-validation and hyperparameter tuning; (2) quantifying EC using SimaPro 2024 with the ReCiPe 2016 Midpoint (H) method and monetizing emissions with CE Delft environmental prices; (3) evaluating SC deterministically through established equations for detour delay, fuel consumption, sales tax loss, productivity loss, dust control, and restoration; and (4) integrating CC, EC, and SC outputs into a unified LCCA framework to compare OCPI and CIPP Renewal Technology across multiple diameter categories.
A multi-state dataset of more than 750 pipeline installations across the United States was compiled and normalized to 2025 USD. Rigorous preprocessing included treatment of missing data using Multiple Imputation by Chained Equations, outlier detection and removal, and stratified partitioning into training and testing subsets using an 80 to 20 split. Exploratory data analysis and visualization were employed to identify cost drivers and assess variable distributions. Model performance was evaluated using the coefficient of determination (R2), cross-validation coefficient of determination (CV R2), root mean square error (RMSE), mean absolute percentage error (MAPE), and accuracy within plus or minus 20 percent of observed construction cost. For OCPI, the tuned KNN model achieved the best performance, with a R2 of 0.79, a CV R2 of 0.65, an RMSE of 159.06, and a MAPE of 31.69%, with approximately 59% of predictions within plus or minus 20%. For CIPP Renewal Technology, the tuned GBR model performed best, with a R2 of 0.80, a CV R2 of 0.92, an RMSE of 160.22, and a MAPE of 15.8%, with about 80% of predictions within plus or minus 20%.
Feature-importance results further indicate that, for OCPI, KNN identifies pipe diameter as the dominant cost driver with an importance score of 0.891. For CIPP Renewal Technology, GBR highlights pipe diameter with an importance score of 0.659 and pipe length with an importance score of 0.237 as the most influential variables.
All SL models were fine-tuned using grid search with five-fold CV to minimize RMSE. For OCPI, KNN with 11 neighbors, distance-based weights, and the Manhattan metric achieved the lowest cross-validated RMSE, indicating that OCPI construction costs are best represented by a proximity-based learner that emphasizes closer observations and additive feature differences. DTR with a maximum depth of three, a minimum split of two, and a minimum leaf of one, as well as GBR with 200 estimators, a learning rate of 0.1, a maximum depth of five, and a subsample of 0.8, performed competitively but did not surpass the tuned KNN configuration. For CIPP Renewal Technology, GBR with 100 estimators, a learning rate of 0.2, a maximum depth of three, and a subsample of 0.8 delivered the strongest performance, supporting that boosted ensembles more effectively capture the nonlinearities and interactions governing CIPP Renewal Technology costs. The alternative tuned models, including KNN with five neighbors, distance-based weights, and the Manhattan metric, and DTR with no depth limit, no feature limit, a minimum split of two, and a minimum leaf of one, were less accurate than the selected GBR model under cross-validation.
For OCPI, life-cycle cost (LCC) increases sharply with diameter and is dominated by CC, which contributes more than 85% of total LCC. From small to very large diameters, OCPI LCC rises by roughly a factor of four, driven mainly by increased excavation, installation, shoring, and hauling. EC remains in the range of about 60–120 $/ft and SC near 4.7 $/ft, indicating that OCPI LCC is primarily controlled by construction activities rather than EC or SC. Across the full diameter range, CIPP Renewal Technology LCC values are, on average, about 56% lower than OCPI LCC. The greatest savings, up to approximately 69%, occur in the small to large diameters (6–60 in.), which represent most of the sewer network. For very large diameters (63–84 in.), the advantage decreases to approximately 29% because the demand for CIPP Renewal Technology resin and the energy required for curing increase with diameter. Within CIPP Renewal Technology LCC, CC accounts for roughly 56% of the total LCC for small diameters, increasing to about 80% at very large diameters, while SC remains nearly constant at approximately 2.26 $/ft (less than 2% of the LCC). When CC, EC, and SC are fully internalized, CIPP Renewal Technology enables approximately 2.3 times more rehabilitated length per dollar than OCPI, underscoring the importance of improving construction and material efficiency, especially for large and very large diameters, to further reduce CIPP Renewal Technology LCC.
Environmental Impact Assessment (EIA) across 18 ReCiPe 2016 Midpoint (H) categories shows that for small diameters, OCPI has higher emissions than CIPP Renewal Technology in 16 of 18 categories, with CIPP Renewal Technology higher only in Terrestrial Ecotoxicity and Marine Ecotoxicity. For medium diameters, OCPI again dominates 16 of 18 categories, with CIPP Renewal Technology higher only in Stratospheric Ozone Depletion and Marine Eutrophication. For large diameters, CIPP Renewal Technology exhibits higher emissions in 10 categories, and for very large diameters, in 12 categories, primarily due to increases in resin volume, curing energy, and transported materials with diameter. The characterized EIA indicates that excavation and transportation are the primary emission drivers for OCPI, whereas polyester resin production and curing energy are the dominant sources for CIPP Renewal Technology. Across all diameters, the largest burdens for both methods are associated with Global Warming, Terrestrial Ecotoxicity, and Human Non-carcinogenic Toxicity.
This research presents a practical, modular LCCA framework that integrates ML-based CC prediction with deterministic environmental and social cost modules for OCPI and CIPP Renewal Technology, applicable to a diameter range of 6–84 inches. The framework allows municipal engineers, utility owners, and consultants to compare open-cut and trenchless alternatives on a consistent LCC basis, identify diameter ranges where CIPP Renewal Technology provides the greatest economic and environmental advantages, and quantify trade-offs when OCPI remains necessary. The calibrated cost-prediction models and LCCA outputs can be embedded in capital planning, budgetary estimating, and alternatives analysis workflows to support more transparent, data-driven decisions on sewer renewal and rehabilitation programs.
Keywords
Trenchless pipeline technology, Sustainable underground infrastructure design and development and sewer pipeline rehabilitation, Cured-in-place pipe (CIPP), Open-cut pipeline installation, Life-cycle cost analysis (LCCA), Life-cycle assessment (LCA), Construction-environmental-social costs, Supervised machine learning and artificial intelligence prediction models and data visualization, Environmental impact assessment (EIA) and emissions, SimaPro
Disciplines
Civil Engineering | Construction Engineering and Management
License
This work is licensed under a Creative Commons Attribution-NonCommercial-Share Alike 4.0 International License.
Recommended Citation
Thakre, Gayatri R., "DEVELOPMENT OF SUPERVISED MACHINE LEARNING-BASED LIFE-CYCLE COST ESTIMATION AND PREDICTION MODELS FOR COMPARATIVE SEWER SYSTEM REHABILITATION USING OPEN-CUT AND TRENCHLESS CURED-IN-PLACE PIPE METHODS" (2025). Civil Engineering Dissertations. 534.
https://mavmatrix.uta.edu/civilengineering_dissertations/534