Sewer inspection technicians watch hours of CCTV footage to manually tag cracks, blockages, and other pipe defects—a tedious task that this project replaces with machine learning software that does the annotation automatically. The problem is that manual inspection is slow, subjective, and inconsistent. The project has already uncovered surprisingly large discrepancies in how different technicians annotate the same footage, raising questions about the reliability of human-led surveys. Without consistent data, water companies cannot prioritise repairs or predict which pipes will fail next. If the software is deployed commercially, it could transform how the UK’s ageing wastewater networks are maintained. The algorithms are being optimised for low-power portable and robotic devices, meaning future inspections could be carried out by autonomous robots crawling through pipes. The system will also incorporate historical survey data to predict deterioration rates, and use large language models to automatically recommend repair scopes for detected defects. This is applied research with a clear commercial route. The work is already being implemented by South West Water and commercialised through project partner iTouch Systems, with parallel hardware development at the University of Sheffield’s Pipebots programme.
View original technical description
South West Water (SWW) and the whole UK water industry fight a constant battle to maintain and rehabilitate their ageing wastewater networks. To effectively do so, water companies regularly inspect their sewer pipes with CCTV, recording footage of a pipe’s interior. This enables a trained technician to assess a pipe’s condition, annotating faults and features within the recorded footage. Thanks to the monotonous and subjective nature of this task, the process is time consuming and prone to human error. Combating this, the award-winning work of this fellowship utilises machine learning and Artificial Intelligence (AI) techniques to improve the accuracy and efficiency of this existing surveying process. This project has developed software for the automatic annotation of faults and their associated meta-data including shape, location, and defect code in line with industry standards. Furthermore, in an effort to identify a “gold standard” dataset for benchmarking AI techniques, we have begun to objectively uncover the surprisingly large inconsistencies in surveying practices. This unanticipated outcome has already prompted further investigation into the reliability of human annotation, and the further scope for AI to yield accurate and consistent annotations. In parallel, the project has already secured routes to commercial exploitation via project partner iTouch Systems. SWW’s implementation of the tool conservatively predicts an annual efficiency saving of £250k per year. On the other hand, the licensing and sale of the product predicts a year one gross profit of £150k, growing to £700k by year five. Renewal of this fellowship, will ensure the impact of this technology, further accelerating its current trajectory and enabling its seamless integration with current industry practices. This is reflected by the proposal’s renewed objectives: 1) Optimise the developed algorithms for use on low power portable and robotic devices. 2) Incorporate further historic information, from previous surveys, to improve the reliability of automated decisions and predict the rate of pipe deterioration. 3) Use large language models to further support the decision-making process for remedial works, automatically identifying a scope of works for rehabilitating detected defects. 4) Investigate the reliability and capacity of human surveying practices, highlighting existing challenges and inconsistencies for the application of AI. These objectives supplement the project’s original goals improving the developed software, producing a high-quality dataset for benchmarking AI technologies, and embarking on the journey of robotic implementation. This renewal also aligns development with new project partners at the University of Sheffield, who are developing hardware for robotic sewer inspection devices as part of the “Pipebots” and subsequently “Pipeon” programmes. Most importantly these renewed objectives will support the development and deployment of a commercial product for use by South West Water, and exploitation by project partner iTouch Systems. This parallel development is not only a testament to the research of the project so far, but the best identified route to commercial impact, a self-sustaining research and development process and the realisation of the technology’s true potential.
Plain English summaries and category classifications on this site are generated by AI and may not perfectly reflect the original research.
Is something wrong? Let us know