Learnings

Outcomes

The system has been installed and tested at DCI. Evidence has been captured in support of the BVLOS OSC which has been delivered to NGET ready for submission to the CAA. The project has successfully shown that AI and ML techniques can be used to analyse thermal data captured of high voltage substation assets.

Recommendations for further work

Further refineme nt of the AI model using novel techniques to quickly and efficiently generate large training data sets is recommended to further realise the benefits of the technology.

Lessons Learnt

Deliverable 8: A key learning from the thermal data capture flights include highlighting the importance of collecting a wide variety of data to support the AI model as many of the assets at Thorpe Marsh are of different ages so can look visually different even though they are the same asset type.

Deliverable 11: Key learnings from the refined prototype AI model demonstration presentation include:

• The need to capture large amounts of training data. The data captured in Deliverable 8 was supplemented using data augmentation techniques to generate more training data.
• Large data sets are needed to reduce bias when training the AI models.
• Further research is needed into recovering the true temperature data of an asset as the data collected by the drone is affected by factors including reflectivity, emissivity, distance, background temperature, and air temperature.
• Machine learning (ML) techniques are feasible to use for processing thermal data from substations.

Dissemination 

Project AATIS was disseminated at Utility Week Live 2023 in Birmingham on the 16th and 17th May 2023.

Project AATIS featured in the following LinkedIn post on how National Grid are utilising AI. https://www.linkedin.com/pulse/6-ways-ai-technology-revolutionising-our-energy-networks-awcke/