Around the world, close to 20 per cent of drinking water is lost to leaks in water networks, and that figure doubles in some developing countries, where resources to repair outdated infrastructure are lacking.
Given the scarcity of drinking water in many parts of the world and the overall costs of water treatment, finding leaks and fixing them has been a long-standing challenge to public utilities.
It is also the topic of one of our most recent research projects, funded by EPFL Middle East.
Detecting and locating leaks in water networks is difficult. Leaks can happen anywhere, and because the water network is buried underground, they often go unnoticed for years. But thanks to today's cheap sensors and computational power, that tide could be poised to turn.
Water utilities companies know approximately how much water they are losing. They find out by comparing the amount of water they collect to the total volume of water consumed, although sometimes this is only approximate due to unmetered consumption and evaporation. Determining where they are losing it, on the other hand, demands an invasive approach involving the deployment of sensors.
Acoustic sensors, for example, have been used to detect leaks based on the way sound propagates along the pipes.
They are well adapted for detecting local leaks but since their range and accuracy depend on the type of pipe and its surrounding geology, their application to entire water distribution networks is limited.
This limitation has been overcome with some success by using data interpretation techniques. Sensors distributed throughout the underground water network can be used to measure the flow rate or pressure in the pipes. Computer models of the city's water network are then used to evaluate these measurements and to attempt to pinpoint the location of the leaks.
But if we have learnt anything about computer models of complex systems, it is that they are always subject to errors.
Furthermore, systems change over time, as anyone who has ever looked into a broken water conduit knows. Minerals accumulate along their walls, altering the flow rate of the water. Models that are designed for systems assuming that they are well characterised can, therefore, be way off the mark.
At EPFL's Applied Computing and Mechanics Lab in Switzerland, we are using new sensing and data analysis techniques to come up with a cost-effective and efficient procedure to locate leaks, so they can be narrowed in on and repaired with minimal intervention.
Obviously, inaccurate models do a poor job at analysing accurate measurements. To avoid falling into this trap, we pursue an alternative approach: out of a large number of leak scenarios, we look for those that can be falsified because they are not compatible with the data that our sensors provide.
To do this, we first generate a multitude of computer models of the distribution network we are interested in, each with slightly different pipe properties and leak locations.
Then we simulate the flow through each of these virtual water networks - from the waterworks to households, shops and factories - and find the winning population of candidates by elimination, or falsification, of the models that perform poorly.
In the coming years, we will focus on developing tools to help us select locations for sensors that are optimal with regard to the information they provide. A great deal of time will be spent on the nitty gritty of making the computational work tractable on today's computers. And once all of the development work is completed, we will test our approach on real cities, potentially in Switzerland and in the UAE.
Our focus goes beyond urban water distribution networks. Pipelines that carry pressurised liquids such as oil and toxic chemicals can also leak and pose a threat to health, safety and the surrounding environment. Detecting and fixing leaks more quickly will therefore allow us to make better use of, and protect, the natural resources we rely on.
Professor Ian Smith is director of the Applied Computing and Mechanics Laboratory at EPFL in Lausanne, Switzerland.