Enhance the Effectiveness of Leak Detection in Water Distribution Networks (WDNs) Using Machine Learning (ML) and Pressure Transducers

Water distribution systems (WDS) are critical infrastructure that supply water to communities, but leaks often occur in these systems and can lead to significant water losses and operational inefficiencies. Traditional leak detection methods often struggle with environmental noise and limited scalability. This research proposes a data-driven approach for leak detection using pressure transducers and machine learning (ML). The research contribution is the design of a low-cost, noise-resistant leak detection framework and a comparative analysis of multiple ML classifiers based on experimental data. To facilitate data acquisition, we constructed a prototype PEHD hydraulic circuit that measures 100 meters in length and 40 mm in diameter, on which two pressure transmitters were installed. Data were collected via a dSPACE acquisition system during both normal and leak-induced conditions. Six ML models—support vector machine (SVM), decision tree (DT), random forest (RF), naïve Bayes (NB), logistic regression (LR), and k-nearest neighbors (KNN)—were trained and evaluated using standard classification metrics. LR outperformed all other models, achieving 100% across accuracy, precision, recall, and F1-score. This result may be attributed to the linear separability of the leak signatures in the experimental setup. However, further validation is necessary to assess model generalizability under real-world conditions with varying pipe materials, flow rates, and noise levels. The study demonstrates that integrating pressure transducers with ML can enable reliable leak detection in WDNs, offering a scalable, hardware-efficient monitoring solution. Future work will focus on expanding the dataset and evaluating system performance in live environments.