Enhanced Water Network Leak Detection Methods

Abstract

Water is an essential element and source of life. It is considered a scarce element that goes through several procedures for it be consumable. Once the water is treated and is deemed potable or clean, it is then distributed to the consumers through the water distribution network (WDN) from the generation plant or source. However, efficient water distribution is not always the case, as leakages in the (WDN) is inevitable. Leakages occur due to several reasons such as breakages in the pipelines, bad excavations, poor network maintenance and high water pressure. In addition to reducing the efficiency of supplying water, leakages in the WDN creates economic, social and quality/health problems. As a result, many methods are introduced in the water industry to identify leakages in the WDN such as acoustics detection, vibration utilization, underground imaging, modelling, and pressure-based methods. Although acoustic methods are costly and are one of the main detection methods, they suffer major drawbacks including long durations to detect leaks, inaccuracies in detecting leaks in non-metallic pipelines, and manpower dependencies. This research aims at enhancing the reliability and accuracy of acoustic methods through use of machine learning (ML). Most of the surveyed literature that used ML used experimental data or is not applied in an actual WDN to develop the models. In this research, real data collected from Dubai Electricity and Water Authority’s (DEWA) sound sensors that is used to develop and train a ML model. This model predicts the availability of a leak where the sensors are installed at an accuracy of 89% as per the tests conducted on a collection of data from several communities in Dubai. Finally, the economical applicability of using the ML model using the current system and a proposed smart system is conducted.