Machine learning-based leak detection and predefined leak-location classification in water distribution systems using experimental pressure and vibration data
Water Supply · 14.11.2026
DOI: 10.2166/ws.2026.218 ↗Kurz gefasst
Fachartikel in der Zeitschrift Water Supply, online erschienen am 14.11.2026. Verfasst von Ranganathan S und 2 weiteren Autor:innen. Die Arbeit nennt 30 Literaturangaben.
Abstract (Originalsprache)
Leakage in water distribution systems can cause substantial water loss, infrastructure deterioration, and operational challenges, motivating data-driven detection and localization methods. This study investigates experimentally acquired pressure and vibration measurements from a laboratory water-distribution pipeline using a unified machine-learning framework. The approximately 6.0-m test loop used nominal 1-inch pipeline sections, a 1.3-kW, 2,700-rpm circulation pump, an inlet valve maintained at approximately 70% open, and an observed no-leak operating pressure of approximately 1 bar. Three artificial leakage locations used 8-mm openings. Three sensing configurations – pressure-only, vibration-only, and pressure–vibration fusion – were evaluated using Random Forest, XGBoost, LightGBM, and Multi-Layer Perceptron models with engineered time- and frequency-domain features. Binary leak detection and predefined single-leak-location classification were evaluated separately using fixed held-out test partitions. The majority-class baseline achieved 85.0% accuracy, demonstrating that accuracy alone can be misleading under class imbalance. Localization was substantially more challenging, with a maximum accuracy of 40.0% and macro-F1 of 38.89%. Statistical comparisons did not identify significant pairwise differences after correction. The study provides an offline experimental proof-of-concept requiring independent-run and field-scale validation.
Zitieren
APA
S, R., Vinothkumar, N., & Saravanabalaji, M. (2026). Machine learning-based leak detection and predefined leak-location classification in water distribution systems using experimental pressure and vibration data. Water Supply. https://doi.org/10.2166/ws.2026.218
Vancouver
S R, Vinothkumar N, Saravanabalaji M. Machine learning-based leak detection and predefined leak-location classification in water distribution systems using experimental pressure and vibration data. Water Supply. 2026. doi:10.2166/ws.2026.218
BibTeX
@article{S2026,
author = {S, Ranganathan and Vinothkumar, N and Saravanabalaji, M},
title = {Machine learning-based leak detection and predefined leak-location classification in water distribution systems using experimental pressure and vibration data},
journal = {Water Supply},
year = {2026},
doi = {10.2166/ws.2026.218}
}
EUVIMED