Register for free to
view the full innovation profile.
Computer vision deep-learning approach to overcome asset conditional ambiguity.
About
Automating imagery analysis of electricity network assets
Distribution network operators (DNOs) own and operate the electrical networks that deliver energy from the national grid to households and businesses. They need to get the best performance and reliability from their network while reducing their operating costs. They carry out periodic surveys to determine the state of their assets so they can plan repair and maintenance.
For much of their network, they do this by taking photographs from helicopters, and then engineers assess the condition by manually analysing the images. This is costly both in terms of time and money. There has been a move towards using drones to capture the images; however, this does not address the lengthy and error-prone task of manual condition assessment.
We wanted to determine whether it is technically possible to perform automatic analysis of the imagery using machine learning and whether this would be a commercially viable prospect.
Key Benefits
A great new way for the asset condition assessment: faster, safer, easier and more accurate
SEIA is a proof of concept for a computer-vision asset condition assessment model utilising aerial data. It analyses photographs of electrical towers, automatically tagging their geographical location and which part of the asset it is from. It then uses a suite of deep learning models to determine the asset’s condition and labels the image accordingly. Issues are identified in close to real-time. That means our solution can be applied to imagery streamed directly from a drone, enabling the engineer to identify and focus on potential problems and take more photographs, providing greater accuracy.
Applications
A trained computer vision replaces monotonous and timely human analysis.
SEIA uses state-of-the-art, deep-learning computer vision. Northern Powergrid provided photographs collected from their past helicopter surveys together with the results of their engineers’ analysis. We trained a suite of deep learning models to predict the correct condition score for each image for each of several condition factors. The images are automatically assigned to the asset using Global Navigation Satellite System (GNSS) data embedded in the images, and the prediction condition scores are then combined to produce a report for each asset. SEIA shows how the time taken to complete the monotonous and repetitive task of a human analysing the photography can be greatly reduced, thereby saving money as well as improving assessment quality.
A solution endorsed by ESA for different infrastructure types
SEIA has a huge potential to be used by companies that have remote assets and would benefit from the efficient and effective reading of their condition. As well as power networks, this could be wind turbines, 5G masts, oil rigs, storage tanks and many other infrastructure types. SEIA has been included in the ESA directory of business applications. We also have the ambition to integrate SEIA with our CHARM asset condition and investment tool, which will provide a powerful combined capability for remote assessment and investment planning.