Mzembegwa, Takudzwa Sikumbuzo2025-10-032025-10-032025https://hdl.handle.net/10566/21017Pipe bursts cause a considerable loss of treated water, increase the risks of environmental contamination and are a health hazard for the end-user as they can create a passage for contaminants to enter water distribution networks (WDN). Identifying pipe burst locations will help water service providers repair pipe bursts in a timely manner. Given the ever-increasing importance of water, a great number of methods to locate pipe bursts have been proposed. But none have proved to produce results accurate enough for water service providers to heavily rely on. Therefore, this thesis presents a comprehensive investigation addressing two critical challenges in pipe burst localisation: optimising fully-linear deep learning (FL-DL) architectures for accurate detection and developing real-time localisation methods using Change point detection (CPD) algorithms. The research is structured in two main phases to tackle these challenges. The first phase conducts a comparative analysis of hyperparameter optimisation techniques, Particle Swarm Optimisation (PSO) and Population-Based Training (PBT), for FL-DL architectures. This investigation addresses the limitations of traditional detection methods, which are often costly, labour-intensive, and limited in scalability. Results demonstrate PSO’s superior performance, with PSO-optimised models consistently achieving higher accuracy and lower variance compared to PBT implementations. Notably, PSOFL-ResNet achieved a mean accuracy of 98.92% and PSOFL- DenseNet reached 98.78%, significantly outperforming their PBT counterparts at 96.70% and 97.22% respectively.enChange point detection (CPD)Cumulative Sum (CUSUM)Water Distribution Network (WDN)Shewhart Control ChartsReal-time Pipe Burst LocalisationThesis