The KAI-Vision platform is being used at National Grid to support analysts to rapidly identify the objects of interest in recorded footage. Using deep learning, KAI-V is able to identify the key components automatically and present these to the operator for review.
National Grid were able to achieve an over 60% reduction in the time taken to review over head line footage for condition assessment. As part of a follow on project the input from the users assessment is being used to train a machine learning algorithm to allow the model to automatically identify defects.
The problem statement
- National Grid captures video footage of Overhead Lines to identify any potential faults with towers, fittings or conductors.
- Analysts spend thousands of hours reviewing footage looking for problems and reporting on condition.
- The process is laborious, has significant bottlenecks and quality is dependant on the engineer reviewing the footage.
THE MAGIC NUMBER
% Reduction in time taken to process footage
- OPEX Saving: Reduced the time taken to process footage by 66%.
- Self Sufficiency: Analysts are able to train AI models to identify and extract relevant components (Spacers, Joins, Dampers, Insulators).
- Risk Reduction: Reduced the risk of asset failure by removing the backlog of footage to be processed.
- Improved Asset Management: Allows the cataloguing of historic footage to support and maintain asset inventory.
- Rapid ROI: Entire capability deployed (with models and user training) in two months.
"It took just one meeting! Keen AI went way and built a working prototype. We’ve spent months with some external suppliers and not got anywhere."
- Mark Simmons OHL Condition Monitoring Team Leader National Grid ETO
We worked closely with National Grid teams to fully understand the end to end process.
A mix of images, video and infra-red footage is now collected and analysed.
Risk reduction, opex saving and rapid ROI achieved with the KAI-V platform.
Phase 2 was to identify multiple objects within a video. Probabilities of an object's detection is shown.