| Author: | Taiwo, Ridwan Ademola |
| Title: | Integrated intelligent models for understanding and predicting pipe failures in water distribution networks |
| Advisors: | Zayed, Tarek (BRE) |
| Degree: | Ph.D. |
| Year: | 2025 |
| Subject: | Water-pipes -- Maintenance and repair Water -- Distribution -- Management Water leakage Hong Kong Polytechnic University -- Dissertations |
| Department: | Department of Building and Real Estate |
| Pages: | xxix, 286 pages : color illustrations |
| Language: | English |
| Abstract: | Water distribution networks (WDNs) are critical infrastructure systems facing increasing challenges due to frequent failures, which have significant environmental, social, and economic consequences. The failure of water pipes and the subsequent breakdown of WDNs pose significant obstacles to the sustainability and functionality of these vital systems. To mitigate this challenge, it is crucial to gain a comprehensive understanding of the factors contributing to pipe failure and develop predictive models capable of forecasting the probability of failure (POF) of water pipes and their associated causes. Therefore, the primary aim of this study is to improve the current understanding of water pipe failure factors and develop predictive models to enhance the management of WDNs considering four objectives: 1) identify failure factors and failure modes of water pipes, 2) model, rank, and investigate the relationship between water pipe failure factors and failure modes, 3) develop and automate optimized models to predict the POF, probability of leak (POL), and probability of burst (POB) of water pipes, and 4) develop and automate an optimized model to predict the causes of water pipe failure (COF). To address these objectives, this study employs a rigorous multi-method approach, combining innovative techniques in data analysis and machine learning. A scientometric and systematic review identifies failure factors and modes, while Fault Tree Logic (FTL) establishes a structured framework for analyzing complex factor relationships. The application of Partial Least Squares Structural Equation Modeling (PLS-SEM) investigates the relationship between failure factors and modes using global expert survey data. For predictive modeling, the study introduces novel advanced machine learning techniques. A synergistic integration of Logistic Regression and Genetic Algorithm optimizes POF prediction. For POL and POB prediction, state-of-the-art deep learning algorithms (Deep Neural Networks, Convolutional Neural Networks, and TabNet) are enhanced through Bayesian Optimization, representing a pioneering approach in WDN management. The COF prediction model employs cutting-edge ensemble learning algorithms (AdaBoost, Random Forest, XGBoost, LightGBM, and CatBoost), optimized using the Tree-structured Parzen Estimator (TPE) algorithm. This comprehensive methodological framework demonstrates the study's rigor and innovative approach to addressing complex WDN challenges. The systematic review identifies 30 failure factors, categorized into pipe-related, operation-related, soil-related, and external-related factors, along with five distinct failure modes of water pipes. The PLS-SEM model reveals 19 critical failure factors and confirms the hypothesis that factors influencing water pipe failure significantly impact failure modes, as evidenced by p-values less than 0.05 and a path coefficient (β) of 0.567. Using historical data from the Hong Kong (HK) WDN, the POF prediction model achieves remarkable accuracy, with an F1 score of 0.868 and an AUC of 0.944 The POL model achieves high accuracy (0.994) and F1 score (0.924), while the POB model shows similarly impressive results with accuracy of 0.999 and F1 score of 0.872. Both models exhibit strong performance in precision, recall, Matthews Correlation Coefficient (MCC), and Cohen's Kappa, indicating their robust predictive capabilities for leak and burst probabilities in WDNs. The COF model, optimized using TPE, shows significant improvements with macro F1 scores increasing by up to 13%. The optimized XGBoost model achieves the highest accuracy (0.82) and macro precision (0.65), while LightGBM excels in AUC (0.87) and computational efficiency. SHapley Additive exPlanations and feature importance analyses identify water type, material, age, and diameter as key predictive factors for water pipe failure causes. This research contributes both theoretically and practically to the field of WDNs, providing valuable insights for sustainable management alongside web-based applications for implementing the POF, POL, POB, and COF models developed in this study. By understanding the underlying failure factors, accurately predicting failure probabilities, and forecasting causes of water pipe failure, stakeholders and decision-makers can effectively allocate resources, prioritize inspections, and implement preventive measures. Ultimately, this study contributes to the performance, reliability, and sustainability of WDNs, ensuring the consistent delivery of clean water to communities. |
| Rights: | All rights reserved |
| Access: | open access |
Copyright Undertaking
As a bona fide Library user, I declare that:
- I will abide by the rules and legal ordinances governing copyright regarding the use of the Database.
- I will use the Database for the purpose of my research or private study only and not for circulation or further reproduction or any other purpose.
- I agree to indemnify and hold the University harmless from and against any loss, damage, cost, liability or expenses arising from copyright infringement or unauthorized usage.
By downloading any item(s) listed above, you acknowledge that you have read and understood the copyright undertaking as stated above, and agree to be bound by all of its terms.
Please use this identifier to cite or link to this item:
https://theses.lib.polyu.edu.hk/handle/200/13669