Sensor-Efficient Machine Learning for Leak Detection, Localization, and Loss Estimation in an Experimental Water Distribution Network

Water losses in distribution networks are difficult to identify because pressure and flow variations caused by leakage can be confounded with changes in normal hydraulic operating conditions. This study develops and experimentally evaluates a machine-learning framework for detecting, localizing, and quantifying water leakage under variable valve configurations, including both single and simultaneous leaks. Experiments were conducted on a laboratory-scale hydraulic network instrumented with pressure and flow sensors, generating 2,447 steady-state observations comprising 373 no-leak observations, 576 observations with leakage at location L1, 760 at location L2, and 738 with simultaneous L1 + L2 leakage. Hydraulic measurements and actuator states were used to classify the four operating conditions and estimate total water loss. To reduce information leakage between hydraulically similar observations and provide a more rigorous assessment of generalization, model performance was evaluated using grouped cross-validation. Random Forest achieved a leak-state localization accuracy of 0.974, while Extra Trees estimated total leakage with R² = 0.974, MAE = 0.159 L min⁻¹, and RMSE = 0.287 L min⁻¹. A reduced-input analysis using only four pressure measurements retained substantial diagnostic capability, achieving a localization accuracy of 0.859. Matched operating-condition analysis further demonstrated that leaks produced spatially differentiated pressure responses across the network, supporting the physical basis of the learned classification patterns. The results demonstrate that single and simultaneous leaks can be distinguished from normal hydraulic variations and that leakage magnitude can be estimated accurately using a relatively small sensor set. The proposed approach provides an experimentally validated basis for sensor-efficient, data-driven monitoring of water losses and supports the development of scalable leak-detection systems for water distribution networks.

Authors

Institutions

Publication Details

Journal
Research Square
Published
2026-11-20
DOI
https://doi.org/10.21203/rs.3.rs-11156716/v1
Primary Topic
Water Systems and Optimization
Type
preprint
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
preprint

Sensor-Efficient Machine Learning for Leak Detection, Localization, and Loss Estimation in an Experimental Water Distribution Network

Awni Y. Al-Otoom
Research Square
Water Systems and Optimization
preprint

Sensor-Efficient Machine Learning for Leak Detection, Localization, and Loss Estimation in an Experimental Water Distribution Network

Awni Y. Al-Otoom
preprint en

Abstract

Water losses in distribution networks are difficult to identify because pressure and flow variations caused by leakage can be confounded with changes in normal hydraulic operating conditions. This study develops and experimentally evaluates a machine-learning framework for detecting, localizing, and quantifying water leakage under variable valve configurations, including both single and simultaneous leaks. Experiments were conducted on a laboratory-scale hydraulic network instrumented with pressure and flow sensors, generating 2,447 steady-state observations comprising 373 no-leak observations, 576 observations with leakage at location L1, 760 at location L2, and 738 with simultaneous L1 + L2 leakage. Hydraulic measurements and actuator states were used to classify the four operating conditions and estimate total water loss. To reduce information leakage between hydraulically similar observations and provide a more rigorous assessment of generalization, model performance was evaluated using grouped cross-validation. Random Forest achieved a leak-state localization accuracy of 0.974, while Extra Trees estimated total leakage with R² = 0.974, MAE = 0.159 L min⁻¹, and RMSE = 0.287 L min⁻¹. A reduced-input analysis using only four pressure measurements retained substantial diagnostic capability, achieving a localization accuracy of 0.859. Matched operating-condition analysis further demonstrated that leaks produced spatially differentiated pressure responses across the network, supporting the physical basis of the learned classification patterns. The results demonstrate that single and simultaneous leaks can be distinguished from normal hydraulic variations and that leakage magnitude can be estimated accurately using a relatively small sensor set. The proposed approach provides an experimentally validated basis for sensor-efficient, data-driven monitoring of water losses and supports the development of scalable leak-detection systems for water distribution networks.

Research Square
Jordan University of Science and Technology (JO)
Clean water and sanitation, Industry, innovation and infrastructure
Water Systems and Optimization
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon
Sensor-Efficient Machine Learning for Leak Detection, Localization, and Loss Estimation in an Experimental Water Distribution Network — Awni Y. Al-Otoom · Research Square (2026) | TGRS Research Map | TGRS