Strategic sensor placement using centrality metrics and machine learning techniques for leak localization in water distribution networks

ABSTRACT A significant volume of water is lost in water distribution networks due to undetected underground leaks. Traditional leak detection methods are often inefficient, labor-intensive, and time-consuming. To address these challenges, this study investigates the utility of machine learning (ML) techniques for leak localization. However, the effectiveness of ML-based methods is highly dependent on the strategic placement of sensors. The proposed methodology employs spectral clustering to form clusters and integrates network topology with sensitivity-based centrality metrics to identify sensor locations within each cluster based on centrality scores. A multi-layer perceptron neural network framework is trained using pressure and flow data from these sensor locations to localize leaks. Incorporating 3D coordinates led to more spatially diverse and topographically optimized sensor placements, improving network coverage compared to 2D-based clustering. Leak localization accuracy varied significantly, from 31.62 to 84.79% at zero tolerance and from 47.27 to 92.82% at 100 m tolerance, highlighting the critical role of optimal sensor placement and data quality in determining the effectiveness of ML-based leak localization. The study concludes that zone-based leak localization maintains high accuracy even under noisy conditions, highlighting its field applicability.