Learnings

Outcomes

The project took a major step towards the goal of developing automated drone inspections of NGET towers to enable remote identification and mapping of corrosion to a tower schematic.

The project successfully completed all milestones - in particular, all data collection flights were completed; algorithms for identifying, assessing, and mapping corrosion to a tower schematic were successfully developed; levels of corrosion on surveyed towers were mapped, enabling targeted intervention from NGET or its suppliers; and an updated version of the data processing platform is available for use in business-as-usual steelwork condition monitoring activities.

The following are the key outcomes from this project:

1. Demonstrated that fully automated analysis for corrosion is achievable provided data is collected appropriately.
2. Compared to current data capture operations the approach improved cost effectiveness (by reducing both direct and indirect costs); reduced environmental impact (by reducing CO2 emissions and noise); and improved quality of data captured (by delivering a more structured data set that’s more suited to analysis using AI)
3. Developed and integrated multiple technologies (Wifi, cellular, satellite and advanced radio) and successfully managed to ensure communications with the drone in a wide range of conditions out to a range of c.1.5km
4. Secured routine non-segregated BVLoS permission to fly in the vicinity of the high voltage electricity transmission network - initially a short section of NGETs assets. This was a huge result as it is the first time the UK CAA has ever given a company this permission.
5. By the end of the project, complex flight paths around NGET towers took place and high-quality inspection data was captured consistently.
6. Demonstrated the feasibility of deploying autonomous drones for automated data collection
7. Used AI and deep learning to automatically detect corrosion on a tower and completing a tower schematic for reporting.

At project start the anticipated benefit was expected to be £1,284,000 over a 10-year period. On completion of the project another CBA has been completed and benefit is now expected to be £655k over the same time period. In the initial CBA an assumption was made that we'd be able to apply the analysis techniques to a greater proportion of the assessments that we undertake than we are able to at the end of the project. The reason that assumption was incorrect is that we can only use this technique when we have lidar data for the towers we are imaging. As our helicopter fleet still flies the bulk of the network, we intended to source this data from lidar scanners that we would install on our helicopters. Funds are sanctioned for these helicopter upgrades however this is still pending completion. Once the helicopters are upgraded, we will be able to realise this benefit. 

Lessons Learnt

Key learnings from this project related to data capture and delivery are:

• Significant learning was generated around the process for applying for advanced aviation permissions in general, and in specific about permissions related to flying on the grid.
• Made significant progress during this project in developing better communications to the drone. A combination of Wi-Fi; cellular comms; satellite comms (Starlink); and specialised long-range high bandwidth meshed radio comms (doodle-labs) were used.
• The iterative flights around our OHL assets generated learning around precisely flying near the high voltage assets - an extremely challenging electromagnetic environment. During the project, problems were experienced with the drones Global Navigation Satellite System (GNSS), magnetometer and accelerometer sensors. Challenges were also faced when controlling the drone in windy conditions. This is a particularly significant challenge given the low centre of gravity (CoG) on the drone from using a heavy underslung data capture payload.
• The drone height offset needs to be adjusted to avoid occlusion of far steelwork by near steelwork. The conclusion is that it is not possible to avoid all occlusion but overlapping images at various positions yields a decent result.

Key learnings from this project related to data processing, analysis and reporting are:

• It was found the magnetometer on the flight controller experienced significant interference when flying near power lines.  A clear 50Hz signal was visible of approximately the same magnitude as the earth’s magnetic field ~0.5-1 Gauss. This was resolved by implementing a 10Hz filter on the magnetometer signal, which increased the logging rate of the magnetometer to prevent aliasing. The change reduced the noise by approx factor 100.
• We were able to process data not only from the autonomous drone but also any suitably equipped traditional drones or helicopter. Provided the following data was collected during data collection then we can map 2D images to the 3D digital twin:
  • the drone's x,y,z coordinates
  • the LIDAR point cloud file - note: The LIDAR file and drone position need to be in the same coordinate system or be able to be mapped from one to the other.
• Three approaches to optimising the corrosion assessment pipeline were attempted:
  • Corrosion detection directly
  • Steel segmentation followed by corrosion detection
  • Fine tuning existing model architectures vs. creating custom architecture.

We found a 2-step pipeline was the most effective, where steel was segmented first and then corrosion detection applied. For best results we also needed to create a custom model architecture. Finally, like many deep learning projects, results improved significantly with more manual labelled data.

The project was successfully able to identify the presence and extent of corrosion on towers. Accurately assessing extent is difficult for a human but falls naturally out of the computer-based approach and to a much higher degree of accuracy.

Dissemination

• Article & video published on Nationalgrid.com in May 2022: “The age of AI: National Grid to trial futuristic automated corrosion inspection of electricity transmission pylons”.  https://www.nationalgrid.com/age-ai-national-grid-trial-futuristic-automated-corrosion-inspection-electricity-transmission
• Video created and shared by one of the innovation partners on LinkedIn and Twitter summarising the project. https://www.youtube.com/watch?v=GoNRItp4nnI
• Article published in Energy, Oil & Gas Magazine in Feb 2023 - “Using autonomous drones to inspect pylons”, by Kathryn Fairhurst. https://energy-oil-gas.com/news/using-autonomous-drones-to-inspect-pylons-by-kathryn-fairhurst/
• Project overview included in NGET’s Innovation Annual Summary 2022.
• Project showcased at the Energy Innovation Summit (EIS) in September 2022 in Glasgow.
• Project presented to Energy Innovation Centre (EIC) DNO Forum in March 2023.
• Presented at Utility Week Live in May 2023.