In 2018, the UK’s National Infrastructure Commission called on water companies in England and Wales to halve leakage to protect long term water supplies.
That means reducing leakage from 20% to 10% by 2050.
With their world-leading place in leak detection, companies are fairly confident they’ll get to 17% by 2025 and triple leak detection.
So, sounds eminently doable. Why not be even more ambitious?
Could it be because for much of the last 15 years reducing leakage seems to have been a game of diminishing returns?
It’s getting harder to match the reductions of the past. Much of the low hanging fruit, so to speak, has been picked just as the stakes are getting higher.
UK Water Industry Research (UKWIR) did some pretty exceptional research on the prospects of achieving zero leakage as part of its impressive ‘Leakage Big Question Programme (How do we Achieve Zero Leakage in a Sustainable Way by 2050?)’.
As World Water-Tech comes back to London for its annual look of the sector’s burgeoning innovation culture later this month, we looked back over some of the problem areas that were identified with acoustic leak detection back in 2017, and examined what’s changed.
1 Detecting leaks on plastic pipes can be very difficult unless the leak is close to the sensor
Detecting leaks in plastic water pipes has historically posed particular problems for the water industry.
In plastic, noise does not propagate far, often less than 50m, making extensive coverage by proprietary sensors very expensive.
The good news is that AI is just as accurate (over 92%) on plastic pipes as it is on other materials. And with FIDO’s semi-permanent sensors included at no additional cost, covering the network is much more cost effective.
2 Noisy environments, such as city centres can pose problems
The human ear can only really pick up the loudest sound in any environment, so things like traffic, electrical noise and sudden high water demand can mask the sound of a leak.
Again, not a problem for AI.
Having been trained on verified data, FIDO can pick out the unique noise signature of a leak no matter what else is going on in the background. No need for delaying city centre investigations until night-time.
3 Cost can be prohibitive, in terms of deploying systems/sensors at sufficient spatial resolution
Proprietary sensors have traditionally been expensive to install and maintain, which has been a barrier for many smaller water networks around the world. But the advent of high quality, low-cost robust components should mean they are cheaper and more effective than ever.
FIDO’s capex-free version, FIDO Bugs, need no special training or calibration. Plus, they can be used to correlate exact leak location. Use them for cost-effective expansion into non-sensor areas, or even to create instant smart sensor estates from scratch. No need for additional pressure or flow data, although this can be incorporated into the FIDO platform.
4 The noise from smaller leaks can sometimes mask those from larger leaks
UKWIR noted that smaller leaks were frequently louder than larger ones and could mask the sound of their larger neighbours.
In fact, in developing and training FIDO AI we now know that some of the largest leaks are below the threshold of human hearing. This means that no-one would pick them up even without any background noise. Not a problem for FIDO AI. There’s no threshold or frequency beyond the reach of its deep-learning algorithms.
5 The sum of all the small leaks over the network is considerable; how to better detect such leaks needs to be addressed
One of the game-changing developments from AI leak detection is that algorithms can be trained to do things humans could never do – providing insight into water leaks that’s never been available before.
Leak sizing is one example of this.
FIDO AI maps an area’s leakage profile overnight to generate a visual heatmap by location and size. With this information, leak teams can prioritise the largest leaks, saving more water faster. And, because FIDO gives each leak a unique ID number, it tracks degradation over time so the small leaks can be monitored.
6 Areas of network comprising pipes made from different materials pose a particular challenge, due to the differing acoustic signal propagation characteristics
Another problem which has been largely overcome with AI. With our latest update, engineers can now do correlation on pipe segments made up of three different materials.
FIDO AI is accurate because it was trained on verified data. This means we physically verified its early decisions, enabling it to hone its accuracy. This works on any material.
Through open data and collaboration we continue to add to FIDO’s growing library of data, which now contains more than 2 million files. Beware of AI not trained on verified data.
7 Leaks in large diameter pipes can be difficult to locate
A bit like large leaks, leaks on larger pipes can be lower in noise level and frequency. This means they don’t travel so well along the pipe wall or through the water. Also, just like large leaks, this is not a problem for AI in any material.
Another benefit of FIDO is that it’s sensor agnostic. It works with any logger for instant additional ROI on your existing estate.
To sum up, FIDO AI has removed barriers to smart leak detection using a deliberately collaborative, open data approach. The addition of leak-sizing is truly game-changing and transformative. It helps clients halve leak runtime, the cost of saving water, and approach not just net zero but absolute zero.