Harnessing AI for safer power networks: Northpower's intelligent asset inspection
Location: Northland
Project date: 2023
Project partners: Amazon Web Services
Key people involved:
Head of Technology & Innovation – Andreas Åström
Asset Engineer – Carlos Ulloa-Canales
The challenge
In 2023, a tragic incident involving a cow’s electrocution exposed a critical safety risk in Northpower’s network – stay wires lacking essential insulators. Identifying which of the 57,230 poles had this issue was a daunting task, as the necessary data was missing from the databases.
The solution
Assembling a powerhouse of talent, Northpower brought together a diverse mix of technology experts, asset engineers, and business representatives from their Network and Digital teams. This multidisciplinary team, in collaboration with AWS, a cloud technology provider, and Sculpt, a computer vision specialist, embarked on a transformative “data archaeology” journey.
Leveraging the latest advancements in AI, computer vision, and data science, the team analysed 800,000 inspection photos and 70,000 scanned inspection protocols spanning the last 20 years. Through Optical Character Recognition (OCR), they extracted structured data from inspection protocols to pinpoint poles with stay wires. They then deployed sophisticated Computer Vision models, trained to detect insulators, eyebolts, and stay wires in the inspection photos.
Tackling the challenge of false positives head-on, the team devised an innovative scoring system. This intelligent algorithm considered the presence of key components across all photos of an asset, flagging those with low scores for human review.
The results
This pioneering approach yielded impressive results, accurately identifying 31 defective stay wires, which were swiftly rectified. The initiative not only bolstered the network’s safety but also delivered cost savings and carbon emission reductions compared to manual inspection methods.
The project’s impact extended beyond the immediate stay wire issue. The team’s solution has already been successfully applied to identify potheads, an aging network component requiring replacement.
The future
This project demonstrates the value of collaboration, bringing together various areas of expertise to work towards a shared objective. It aims to enhance data-driven asset management practices, exploring how the combination of human knowledge and artificial intelligence can provide new insights from previously underutilised data. Through this work, Northpower has not only made its network safer and reduced the risk of similar incidents but also developed a blueprint for proactive asset management.
Northpower would like to acknowledge the invaluable contributions of Anthony Miles, Carlos Ulloa-Canales, and Mike Gibbs from the Network team; Chris Wang and Scott Patterson from AWS; Andreas Astrom, Ed Overy, Nigel Stowe, and Kayla Williams from Digital; and Mac Jones, a computer vision expert from Sculpt.
Watch the presentation from the EEA, Electricity Engineers’ Association Conference 2024 below to learn more. Northpower extends its gratitude to the EEA for the opportunity to share this work.