Anomaly-detection software has been developed to determine the probability of a burst pipe in a network.
Many UK water supplies have been segmented into district metering areas (DMAs), which have defined and permanent boundaries. These allow water flow into and around each district to be monitored, enabling the location of breaks and leakages to be identified.
Despite this, the UK’s water utilities are still losing millions of litres of water through such networks each year. For every litre of water lost, another litre of water has to be treated and pumped through the network to compensate, increasing the cost of production, reducing the available water supply and contributing to carbon emissions.
Although water utilities use sensors to measure pressure and flow into and around the DMAs, the real-time data acquired and sent via their supervisory control and data-acquisition (SCADA) systems is not being evaluated adequately to allow the utilities to determine the behaviour of the networks.
Now, Project Neptune, a venture between Yorkshire Water, United Utilities, ABB, the EPSRC and academics at seven UK universities, has developed an integrated decision-support system that can analyse the data generated by such sensors, allowing water operators to prioritise, detect, isolate and mitigate the consequences of failures in the water network more efficiently.
According to Exeter University’s Prof Dragan Savic, technical project manager of the Neptune project, many of the anomalous signals received in water utility control rooms must still be manually examined by operators to determine whether they are an indication of a likely break or whether they are caused by ghosts (or inaccurate data received by the system).
To automate the analysis of the data, the team developed anomaly-detection software comprising an artificial neural network that works in tandem with a fuzzy classifier. After spurious signals such as ghosts have been eliminated, the software analyses the data from the flow and pressure sensors to determine the probability of a burst in the network and how much water would be lost through it.
Once the probability of a burst has been determined, a further set of models is deployed in the decision-support system to provide an indication of where the burst pipe may have occurred.
UK water utilities are losing millions of litres of water through district metering areas each year
This set of models comprises three software modules that work in tandem a hydraulic model of the network of the pipes in the system, a pipe burst propensity model and a customer contact model. The system looks at evidence from all three before estimating where a burst may have occurred.
By simulating a burst occurring at each pipe in turn in the hydraulic model of the DMA, the system can compare the differences between the theoretical (modelled) pressure estimates and real measurement data acquired at the entrance to the DMA and at specific points around it. Owing to the small number of measurement points, however, this provides only the first set of evidential information for estimating options for where the break may have occurred.
Information on the physical attributes of the pipes, as well as historical data that indicates where bursts have occurred in the past and are likely to occur again the pipe burst propensity model are used to refine the possible point of failure.
These are constrained further by the customer contact model, which allows the system to consider the location of customers from the water supply when they report a change in the pressure of their supply or another possible consequence of the pipe break to the water company.
While the system is capable of determining if a burst of a particular size may occur and where it is likely to be found, it can also automatically prioritise any alarms that would indicate disruption to critical parts of the water distribution network, as well as providing the operators in the utility control rooms with an indication of the consequences of a burst, highlighting which customers would be affected.
A prototype system has been successfully integrated into a SCADA system in a Yorkshire Water facility that is responsible for controlling and monitoring the water supply to the Harrogate and Dales region.
According to Savic, Yorkshire Water now plans to roll out the system across the whole of its distribution system. He said that the system would enable the company’s personnel to identify the potential problem and make repairs to leaking pipes in much shorter periods on average (between six and 12 hours).