Hydraulic system monitoring: Review of analysis tools and widely used machine learning for anomaly detection

Abstract

The efficiency and robustness of managing a valuable resource such as water, which is scarce relative to the growing number of consumers, has become one of the fundamental levers for improving drinking water distribution services. With the aim of conducting an analytical study of defects and anomalies in a drinking water supply system (DWS), the process of acquiring information from a real network represents a complex task in terms of time, cost, and the implementation of data collection mechanisms and methods, given the vast number of consumption nodes connected to a dense network of pipelines and reservoirs that constitute distribution systems. Overall efficiency improves significantly, using: AI and machine learning to analyze massive amounts of data (flow rates, pressures, acoustics, etc.) and identify abnormal behavior, with an accuracy of up to 95% in some models. IoT and smart sensors for continuous monitoring, and detection of leaks.Hybrid models combining ML algorithms (CNN, neural networks… ) with software (e.g. EPANET) offering results exceeding 97% accuracy. International statistics on problems related to hydraulic networks confirm that losses caused by leaks represent the most significant threat. This article proposes a comparison between hydraulic detection and analysis tools for problems related to drinking water supply networks, as well as artificial learning methods used for the prediction of these anomalies, particularly the most frequent ones such as potential leaks, pressure drops, air pockets, and water quality issues.