December 21, 2023
ai data innovation-in-water-challenge water-innovation winnersLed by: United Utilities
Partners: Dŵr Cymru (Welsh Water), Scottish Water, Severn Trent Water plc, Thames Water Utilities Limited, Water Research Centre, and Yorkshire Water Services Ltd.
Challenge: Innovation in Water Challenge
Sewers are vital to how we function as a society but to perform properly, they must undergo regular maintenance checks. Historically, they have been assessed by small cameras that are sent into the pipework to capture footage, which is then reviewed by inspectors. This is a repetitive and time-consuming process requiring manual identification of defects in pipes.
The AI and sewer defect analysis project, funded by the Ofwat Innovation Fund, used artificial intelligence and machine-learning software to remove the need for this job. The AI automatically recognises features in camara footage to give a better – and more accurate – understanding of sewer deterioration. This also increases efficiency and reduces the cost of inspections, ultimately saving water companies, and therefore customers, money.
The project collected data (including video footage and imagery of various defects such as pipe cracks and invasive tree roots) from six different water companies. This was used to teach the software how to recognise the defects. The software was then trained to categorise the defects in terms of severity, and make recommendations as to how the water company should respond in light of local guidelines. The resultant software was found to be over 90% accurate in identifying the defects, and over 80% accurate in its response recommendations, all while saving time.
During the project, over 187,000 files were processed, involving 726,000 images, 27,000 of which were used in the final data-set. This data-set will now be published for other water companies – and other organisations – to use, to train their own algorithms, paving the way for future innovation and development across the water sector and beyond. An adoption plan will also be published to help show water companies how best to use the data.
One of the project’s key challenges was collecting and handling significant volumes of large files. In order to work with varying security systems across different water companies, the team had to use multiple file storage systems. These files also had to be collected in different file formats so needed a variety of conversion software for analysis. The large volume of data presents a challenge for where to host the information in future.
We’re really pleased with the level of accuracy in defect detection achieved using this data set. It was a hugely ambitious project – shown by the sheer volume of data we were able to process – and we hope that water companies and suppliers will take this data set to improve on the accuracy of their models. We also hope that this will be valuable to people outside of the sector, who may able to use this for innovation we haven’t even considered. The achievement should also inspire and accelerate the use of AI within the water sector more generally.
