top of page

WRc provides sewer image library to train AI to prevent pipe bursts and spills

by Karma Loveday

The Water Research Centre (WRc) has created a world first, free-to-download library of sewer images that can be used to instruct artificial intelligence (AI) software solutions to assess the condition of drains and sewers.


WRc, part of the RSK Group, worked on the Ofwat Innovation Funded project with United Utilities as the lead water company partner and Spring, which hosts the library on its platform.


The open data project was designed to accelerate the development of AI software and to assist in the inspection of sewers, helping to prevent sewer collapses, flooding and overflow spills.


The collaborative project involved seven water companies (United Utilities, Thames Water, South West Water, Dŵr Cymru Welsh Water, Scottish Water, Severn Trent Water and Yorkshire Water), which operate 72% of the UK’s sewers or 560,000km. WRc used previously coded CCTV survey footage with accompanying metadata, which included a video reference, the sewer material and diameter, a defect code, a defect description and the location of a defect in the frame.


WRc principal consultant Peter Henley explained: “This process gave us a total of 726,290 images, and from this we successfully developed a library of 27,262 images to act as the single benchmark dataset by checking and categorising each of the images for accuracy and clarity. For each defect, an optimum target of 1,000 images was sought from the original total of 726,290 images. This target was selected after discussion with current AI software providers concluded that it was the ideal number to train an AI solution. However, providing even a limited number of images for a defect type is beneficial in the improvement of AI accuracy.


“We were able to identify sufficient images for 17 defect codes (with 1000 images each) and a further 55 defects having images identified and classified, which will still be useful in the development of AI solutions. The library now contains images of 72 defect codes – significantly more than the expected 60 defect codes identified at the onset of the project.”


The new library supplements WRc’s 1980 industry essential guide, the Manual of Sewer Condition Classification.

Comments


bottom of page