Real-time Leak and Burst Detection using Wireless Smart Sensors for Monitoring water distribution System.

---

Wenyan Wu, Mohamed Ibrahim M. Mohamed

Thursday 2 july 2015

9:00 - 9:15h at North America (level 0)

Themes: (T) Special session, (ST) FP7 ICT and water

Parallel session: 10I. Special session: FP7 ICT and water

---

Continuous monitoring of water infrastructure using wireless smart sensors has the potential to save annual operational costs by predicting the behaviour of the assets in the long run and detecting any changes in the behaviour of the system infrastructure. Detecting leak and burst detection is a very important issue for the water industry, not only because of the cost of losses but also for maintaining national water resources. The current technology for detecting leak and burst events relies on offline techniques by collecting loggers’ data from multiple locations. We believe that detecting such events in real time, with smart sensors nodes, could improve monitoring operations and save operational costs. In this paper we analyse different machine learning techniques and investigate the ability to use them as real time, intelligent event detection tools to run using the limited processing capability in the wireless sensor nodes. There are many challenges related to use machine learning techniques for detecting events in real time. In general, existing machine learning techniques are computationally expensive and therefore need to be modified to be light-weight enough to fit within the limited processing capability of the sensor nodes. Additionally, most machine learning techniques are designed to run offline. However, in our application these conventional techniques are modified to run in real-time. Regardless, the machine learning technique must be able to distinguish between outliers and events. There is a trade-off between the accuracy and latency needing to be investigated. We proposed a novel lightweight event detection technique to detect leak and burst events. Additionally, we evaluate the proposed technique against exciting ones in terms of accuracy, and computational complexity