Learnings

Outcomes

Conclusions

1. Replacing current inspection process with drone/virtual inspection process not feasible both operationally and commercially
2. Number of items that would need to be resolved/mitigated to increase productivity and address gaps
3. Potential improvement path over time which would eventually see the costs come down to today’s levels plus gain many other benefits not seen with traditional inspection methods
    

Replacing the current inspections programme with a virtual inspections programmes right now based on findings from the trial would result in an increase in Opex for the programme of approx. 44% however there are multiple enablers over time to remove/mitigate the issues detailed in the lessons learned section 

Lessons Learnt

Image Capture Trial – lessons learned/challenges

UK Drone Aviation Laws

1.    Unable to fly within 50m of people, buildings, cars or trains – in more urban areas this limited the number of poles that could have images taken. Would result in a split process where poles not able to be flown would need to be inspected by a traditional manual inspection

2.    VLOS – currently require the drone to be within the pilot’s line of sight. Limits the inspector to taking images of 4 or 5 spans at most at one time then needs to pack up the drone and walk along the line to the next point – pretty much the same distance needed to be walked as with a foot patrol

3.    BVLOS – current restrictions prevent the ability to pilot the drone along a full circuit from a single static position or even programme a flight path along a circuit for an autonomous drone

Site Conditions

1.    Drones currently used by SPEN are not waterproof and cannot be flown in the rain. There are higher quality models which are waterproof however are much more expensive than the current DJI Mavic 2 model used as BAU.

2.    Image quality has a big impact on both the AI being able to detect defects present in the image and the inspector’s ability to spot defects. If images are taken in the rain, bright sunlight, too much shade, overexposure etc then it reduces the ability to use these images

3.    Drone impacted by high winds and get high wind alert advising the pilot to land in a safe place

4.    If there were trees or other types of vegetation were close to the line, then this would occasionally obscure the view of the poles and prevent the ability to take images from all required angles and therefore risk that a defect would be missed

5.    GPS signal – when flying the drone within a rural valley the drone pilot got an error message advising GPS signal had been lost and was unable to control the drone while still in flight and the camera froze

Battery Life

1.    Drone batteries do not last for a full working day and have to be regularly replaced. Used on average 6 batteries per day when the drone pilots only have two as standard. These 6 also needed to be charged in the evening to be available for use the next day

2.    The battery in the drone remote control do not last for a full working day. There is no way to swap batteries in the control and can only be charged so the drone pilots had to swap to a second controller for the second half of the day and also needed to re-connect the drone to the new controller

3.    Charging in the company van was very slow and also not enough ports to be able to charge all batteries, spare remote and all other standard items like company mobile phone, sat nav, tough pad etc. Also the engine had to be kept running so that the power would not cut off which has a negative environment impact and fuel cost

Memory Cards

1.    Approx. 6 images from various angles are required for each pole to ensure all possible defects could be picked up. Even with memory cards with increased capacity this would require a process for the images to uploaded to the virtual inspections application

Pole Residual Strength (Hammer Test)

1.    A mandatory requirement as part of a statutory inspection is to measure the strength of the pole by physical hammer test assessment. As the drone is unable to do this then a separate programme with associated FTE/cost would be required just to hammer test the poles so this would still involve the same level of travel and walking between poles as during a current inspection

Conductor Heights & Proximity Hazards

1.    The height of a conductor from the ground is unable to be determined from an image. Also identifying the lowest point on the point on a span would be not be possible which is a mandatory part of a statutory inspection

2.    Similarly, identifying all instances where there is a proximity hazard as per the ESQCR minimum required distances would not be possible from an image

Productivity

1.    The drone pilots attempted to fly and take images of LV, HV and EHV lines. Due to a combination of aviation restrictions and weather conditions they were only able to take images of 93 out of 243 (35%) poles on the LV & HV circuits and 160 out of 162 (98%) poles on the EHV circuit

Virtual Inspections Trial – lessons learned/challenges

Productivity – 208 desk based virtual inspections were completed over a 6-day period so an average of 35 poles per day

Software Costs – on top of the cost for internal resources for flying drones and virtual inspections, additional cost per pole is required for access to the systems and image storage. The more you use the system, the more it costs 

GIS Co-ordinates – details in GIS system not always 100% correct to actual location on site resulting in only 67 of the 93 LV/HV poles being available for virtual inspection and 140 of the 160 EHV poles available for inspection. 

GIS Co-ordinates – incorrect GIS data also resulted in images from multiple poles being aligned to the one pole ID in grid vision and also occasionally only 2/3 images associated to other poles which is not enough to carry out a full inspection

Artificial Intelligence – limited number of distribution network associated defect types currently fully trained and available within grid vision. The majority of defects had to be raised manually which takes a similar amount of time as it would to raise defects in the field in FWMS

Artificial Intelligence – suggestions were not always accurate. Often what was perceived to be cracks in pole where just the natural markings on the wood or suggestions of a flashed/contaminated insulated was just bird droppings when the images were inspected manually

Multiple images of same pole – if there were 6 images of a pole and the danger plate could only be seen in one of them, the system still suggested 5 danger plate missing defects for the others as it was unable to look across all images and not raise a suggested defect if it was present in any of them. The danger plate model tested during the pilot is however only an MVP version since it’s a new model and the workflow is planned to be improved in future versions.

Multiple images of same pole – if a defect was raised by the inspector as it was visible in one image, the onus was on the inspector to determine that it is the same defect present in the other images and not raise multiple of the same defects types for a single defect. This is even harder when the images are from different angles making it difficult to determine if it is the same defect e.g. a cracked insulator or a second defect on another insulator

iHazards – as the drone pilots were just tasked with taking images of the poles and did not have an OHL background, they would not necessarily have the ability to spot a critical defect/iHazard which should be reported immediately once spotted. A number of iHazards were spotted by the virtual inspector when reviewing the images later on in the process however this was weeks after the images were taken. 

Pole Heights/Scarf Mark/Pole off Plumb – it is not possible to determine the height of a pole from an image which is a mandatory piece of information to be collected on every inspection. Similarly, it would not be possible to determine of the scarf mark was at the minimum required height and therefore if the planting depth was adequate or not. Also the lean of a pole would be difficult/impossible to determine as this would be impacted by the angle the images are taken from

Improved defect detection – Virtual inspection provides the possibility to detect some types of defects in the top part of the pole that would be difficult/impossible to detect from the ground, e.g. some insulator damages, cracks, pole top rot etc. In the trial 104 defects were detected that would likely have been missed during traditional inspection from ground level

Full documentation and transparency – In a virtual inspection all results are accessible and transparent for any person that could benefit from it. It’s e.g. possible to seek support/second opinion from colleagues which is especially suitable when supporting less experienced staff. It also provides the possibility for maintenance planners to review images of defects before sending out maintenance crews decreasing the risk of requesting unnecessary maintenance.

User experience – It was only one inspector using Grid Vision during the trial and the overall user experience was good. The inspector was experienced in conducting traditional statutory inspection and therefore well suited to evaluate the ability of Grid Vision to meet the requirements. It was the first time the inspector used Grid Vision and had not been using similar tools before but still experienced the tool as intuitive and could conduct the inspection independently after only two training sessions. Main improvements areas included updating the workflow for the danger plate prototype to reduce the number of clicks. Main impression that the software supports the process and lived up to the requirements and the inspector saw benefits of being able to spot defects at the top part of the pole that was not possible to see in the field.