Sorting out water leaks is hard. It’s no wonder that the idea of prevention is so enticing. Ahead of her panel session on Leak Forecasting at the World Water-Tech Innovation Summit next week, FIDO CEO Victoria Edwards looks at the ongoing evolution of leak prediction. Holy grail, or unattainable dream?
Renewing water pipes is expensive and, until recently, prioritising the right ones for investment has been a game of educated guesswork.
One of the ultimate red flags any planner looks for is leakage history. Frequent leaks and bursts are the very definition of a developing pipeline failure.
Every leak is a unique entity, with its own antecedents, characteristics and behaviours. So, providing you understand these, you should be able to predict them, right?
Well, yes and no.
Actionable insight from data
Leak prediction is the ultimate evolution of actionable insight from data. You need AI to extract this level of benefit from vast amounts of data. In fact, technology can already do this to a limited degree, by spotting patterns and trends to aid near-term decision-making.
But true, long-term forecasting? The type you could bet your mortgage on? Not so much.
I’ll explain why in a moment. But first, it’s worth remembering why this matters so much right now.
Around 30% of the world’s water is lost to leaks.
We can’t take more from the environment to supply cities facing Day Zero when one in every three glasses of expensively and carbon-intensively sourced, treated and pumped water just seeps away.
Reducing leakage is a no-brain response to climate change.
The problem with leak data
Ask any leak technician and they’ll tell you, finding leaks is tough work. Around 90% never show above ground so we don’t have much data on where to find them. In places, even information on where the pipes are, let alone their material, age and condition, is incomplete.
As for burst and repair history, much of what we know about an individual pipe’s behaviour is stored in the heads of the engineers who look after it. It gets lost as they move jobs or retire.
Technologies which do generate leak data – like pressure valves, smart meters and acoustic loggers – often store it in proprietary silos and it still needs too much human analysis.
Therein lies the problem for leak forecasting. It’s only as good as the data you have.
What are the prospects for leak prediction?
I recently asked people on Linked-in how soon they thought leak prediction will happen. The results were optimistic. Over 75% think it will happen within five years, and of those more than half believe within the year.
Technology is certainly moving apace. The development of digital twins and artificial intelligence is breaking down the barriers between data silos to enable smarter decision-making.
But the group I found most intriguing were the ‘nevers’ at 13%. Who are these people? I have some sympathy with them.
It’s not that I lack faith in technology, or man’s resourcefulness. It’s just that in terms of leak forecasting we are running before we can walk. And, more importantly, I fear that it deflects us from the urgent need to tackle the present.
Advances in prediction using verified data
Accurate prediction comes down to accurate data. Even if you had the world’s most accurate leak forecasting robot, feeding in duff data is only going to yield results which are misleading.
Yet that’s exactly what we’re contemplating doing. Our current data about water networks is far too assumptive and anecdotal to make predictive analysis viable in the short term, never mind the basis for long-term investment decisions.
Thankfully, we are on the way to fixing this. The smarter technologies like FIDO AI use only verified leak data so we can extract more and more value from it. Some clients are already using FIDO to clean up their historical leak data, even to the point of prioritising the backlogs in their repair books. This type of data validation and insight is a step towards better certainty in leak prediction.
Collaboration through open data
Next, we need to share. FIDO is open data. We are sensor and platform agnostic. We’ll take any leak data from any source and analyse it and share the results with other platforms. Technologies which do this advance collectively and for the benefit of all.
Finally, more data. You can never have too much data grist for the machine-learning mill. As FIDO gathers leak data from around the world, we are careful that only verified results are fed back into the learning algorithm. The most valuable results come from the rare occasions where the AI gets it wrong. These exceptions enable FIDO to perfect its decision-making and from there we are seeing definite patterns and trends which are the embryo of genuine leak prediction.
The future for leak prediction
Despite all these advances, technology is only beginning to scratch the surface of the knowledge we need to predict leaks accurately for true certainty in long-term planning.
The best action we can carry out now is to sort out the leaks we have got. It remains and always will be our main goal. But in doing it, we’ll build up that repository of verified data about leaks and their origins. We’ll spot more patterns and trends and develop more machine-learning insight to aid leak teams with their longer-term resource and investment decisions.
Maybe one day we’ll eventually be able to predict leaks years in advance with a degree of comfort. But in the meantime, we’ll be fixing the problem now. And that’s what matters.
As is so often with tomorrow, it looks so much brighter if you do things right today.