Leakage in Water Distribution Networks (WDNs) remains a persistent challenge for utilities, causing significant Non‑Revenue Water (NRW) losses and reducing operational efficiency. Contemporary reviews emphasize the importance of combining hydraulic modelling with data‑driven methods to better support detection and localization of leaks in operational systems [1], [2].
This thesis presents an integrated hybrid framework for leakage detection and localization within a District Metered Area (DMA). A calibrated hydraulic model, developed using widely adopted tools such as EPANET and WNTR, forms the foundation for understanding system behaviour under both normal and leak conditions [3], [4].
For leak detection, SCADA pressure and flow measurements are processed to produce pressure signals from which statistical indicators are derived. These indicators are used to identify periods where leakage is likely to be present. This step establishes when a leak may have occurred in the network.
For leak localization, the hydraulic model is used to simulate representative leak scenarios. The DMA is partitioned into hydraulically coherent zones using established graph‑based clustering approaches [5]. Simulated responses are then used to characterize each zone’s leak
behaviour. A machine‑learning‑based zone classifier provides an estimate of the most likely affected zone, after which a prototype‑based similarity comparison is applied to determine a prioritized set of pipes for investigation [6], aided by rank‑aggregation principles that support consistent prioritization across multiple indicators [7].
The following aspects are covered within the scope of this Master thesis:
- Reviewing literature on leakage management and hybrid modelling approaches for detection and localization in WDNs [1], [2].
- Developing and calibrating a hydraulic model using EPANET and WNTR tools, incorporating real-world data from a Danish utility.
- Implementing a detection workflow based on SCADA pressure signals and statistical indicators to identify periods of likely leakage.
- Partitioning the DMA into hydraulic zones using spectral clustering to support structured and interpretable localization [5].
- Generating simulated leak scenarios to produce zone‑level behavioural signatures for localization.
- Applying a machine‑learning‑based zone classifier and a prototype‑based pipe ranking method, supported by rank‑aggregation concepts to produce a list of candidate leak locations [6], [7].
Together, this two‑stage hybrid method unifies statistical detection, physics‑based simulation, and machine learning to determine both when a leak occurs and where it is likely located, with deliverables that are directly actionable for field inspections.
References
[1] Puust, R., Kapelan, Z., Savić, D. A., & Koppel, T. (2010). A review of methods for leakage management in pipe networks. Urban Water Journal.
[2] Adedeji, K. B., Hamam, Y., Abe, B. T., & Abu‑Mahfouz, A. M. (2017). Towards achieving a reliable leakage detection and localization algorithm for application in water piping networks: An overview. IEEE Access.
[3] Rossman, L. A. (2000). EPANET 2 Users Manual. U.S. EPA.
[4] Klise, K. A., Murray, R., & Haxton, T. (2020). Water Network Tool for Resilience (WNTR). U.S. EPA.
[5] Von Luxburg, U. (2007). A tutorial on spectral clustering. Statistics and Computing.
[6] Snell, J., Swersky, K., & Zemel, R. (2017). Prototypical Networks for Few‑Shot Learning. NeurIPS.
[7] Pihur, V., Datta, S., & Datta, S. (2009). RankAggreg: Weighted rank aggregation. BMC Bioinformatics.