Faster detection of bursts saves water, minimises the inconvenience of interruption to customers and decreases the damaging consequences to infrastructure. Flow monitoring techniques are used by water service providers to monitor leakage, generally through offline application of mass balance type calculations and manual observations of change in night line values. This paper presents the combination of real-time data collection (using cello loggers with General Packet Radio Service communications) and a self-learning, online Artificial Intelligence system for detection of bursts at the District Meter Area level. The system components consist of communications software, a data warehouse and a MATLAB application. The online system continuously analysed a set of 146 DMAs in a case study area every hour generating automated alerts in response to abnormal flow. Specific examples are given, including a validation field test, and overall results are presented for a one year period. 36% of alerts were found to correspond to bursts confirmed by repair data or customer burst reports with only 18% ghosts. The results indicate that the software tool has the potential to reduce lost water and improve customer service hence enhancing water service provider's reputations.
Keywords: artificial neural networks, data analysis, leakage, on-line monitoring, water distribution systems