Powerful leak detecting AI has advanced to the stage where it now calculates a leak’s size as well – providing a powerful new tool to transform leakage performance and water loss from public networks, say its developers.
The breakthrough by FIDO Tech, an Oxford-based firm, in collaboration with United Utilities, for the first time allows water companies to prioritise larger leaks for investigation and repair, and tackles the global issue of non-revenue water more effectively.
FIDO’s leak-detecting AI combines several unique processes to analyse the files generated by acoustic loggers and hydrophones. Unlike human processing, FIDO’s algorithms can filter out non-leak noises like pumps, sudden high customer usage, nearby generators or traffic, which can trigger false alarms. It is already 92% accurate.
Now developers have used the growing amount of data in FIDO’s system to train its neural network algorithm to extract even more information from sound and kinetic sensor files.
The first of these – leak sizing – went live in December after trials in the North West with leakage teams working for United Utilities and is now in day to day use, with each leakage file being assigned a unique FIDO ID to track through the whole workflow.
FIDO not only predicts whether an alarm is a leak or non-leak with relentless accuracy, it can estimate whether the leak is large, medium or small, allowing leakage managers to send technicians straight to the largest leaks first.
FIDO CEO Victoria Edwards said the development was already allowing water companies to prioritise and boost the effectiveness of resources to tackle water leaks more effectively than ever before.
“Our leak-sizing function officially went live on December 4 after three months of validation and testing and is now the business as usual approach for United Utilities. It’s being rolled out as we speak in Australia and is already attracting lots of interest in the US. FIDO is a SaaS offering and needs no integration so can be operational and delivering results within 24 hours.
“Being able to grade leaks by size is a real breakthrough in leak detection, because it’s providing teams with that missing piece of the puzzle to get them right on the spot of their largest leaks first time with fewer wasted visits or expensive dry digs,” she said.
FIDO leak sizing is just another aspect of the machine-learning capabilities of its algorithm, one of the most successful alumni of United Utilities’ much-heralded Innovation Lab.
The ongoing collaboration between FIDO and United Utilities continues to yield real-world results. Access to United Utilities’ DMA (district metered area) data enabled FIDO developers to train the algorithm to recognise and evaluate never before identified patterns in leak noise which indicate the size of the leak.
United Utilities’ leakage manager Hannah Wardle said: “The ability to rank the leaks by size will allow us to target resources more efficiently. Leakage is one of those issues our customers care deeply about and working with FIDO is one way of the many innovative ways we are tackling it.”
Over the course of three months, FIDO’s rapidly-learning brain began to home in on actual volumes of water being lost in individual leaks, by applying deep-learning principles to the raw data in sound files submitted by sensor alarms. Now, the size range within each of the small, medium and large denominations can be adjusted according to an individual network’s characteristics.
“We were able to teach FIDO how to size leaks because the North West, like much of the UK, has such a mature approach to leak detection, with a large sensor estate and defined district metered zones where we could accurately validate leak volumes against observed flow and pressure data,” explained physicist and FIDO Chief Technical Officer Neil Edwards.
“But the truly brilliant thing is that now FIDO knows what to look for it no longer needs that proof-check, meaning it can provide that same granular detail on leaks even on networks without DMAs, which is the case in most of the countries in the world.”
And according to Victoria, leak sizing is the tip of the iceberg of possibilities that FIDO is bringing to leak detection as its elegant algorithm is trained to identify other patterns.
“It could not be coming at a more pressing time. As the world grapples with the effects of a changing climate water is becoming an ever more precious commodity. Our aim is to save it one megalitre at a time,” she said.