To conduct a thermal survey of solar pv by UAV/drone the data can be collected in a variety of formats (video/stitched images/individual image) and processed in different ways, all of which have pros and cons.
We thought it would be interesting to compare thermal orthographic (stitched images) versus individual images, to examine flight times of data collection and resulting deliverables.
A small field of approx. 1.5MW with a selection of anomalies was chosen for this exercise.
The following results are an illustration of the types of differences that can be observed between methods and heights. Not all solar farms are the same so results will vary.
Thermal Orthographic (6GSD)
By collecting many overlapping images and processing them in some very clever software, thermal images can be stitched together to create a thermal orthographic or ‘map’ of the solar farm.
This flight was programmed to collect images with a 62% sidelap and a 75% frontlap. The imagery is at 6cm GSD (click here for previous post regards GSD). There were 389 images and the flight took 16m 11s.
As the modules need to be inspected during sunny conditions there is a constant battle to avoid the suns glare reflecting from the modules. This can disrupt the software trying to match and stitch images.
The video below shows the flight route and the resulting ‘map’. There’s no denying it looks really good, most of the anomalies can be seen and it is easy to overlay a layout diagram to report the location easily, however when we try to take a closer look the data is very low resolution and difficult to analyse.
The RGB data that is needed to be able to confirm if a hotspot is worthy of a visit by an engineer (or just the side effect of bird soiling for example), would also need to be collected, often with another flight, processed and analysed in tandem with the thermal map.
It should also be noted that this ortho was processed using the Rjpegs and is not radiometric. To be able to analyse temperature deltas in the ortho the data would have to be processed using radiometric TIFF files and suitable software.
Individual images
3 rows (6cm GSD)
By spending a little bit more effort on the flight planning we can program our UAV to take precisely positioned individual images. The flight route needs to be precise, but the payoff is delivered when you compare the amount of MW you can survey per flight.
The image below shows the flight plan of the thermal ortho compared to that for the ‘3 row’ flight.
This flight took just 5min 44s and created 45 images.
As there is no need to try and stitch the images, no overlap was required (sidelap) and the focus is on keeping the rows squarely framed in each shot. Each image overlaps slightly (frontlap) at around 20% just to ensure that no modules are missed. This can be increased to 50% if necessary, to negate glare.
As the flight was at the same height and therefore the same 6cm GSD, the results were very similar.
The RGB data was recorded in the same flight, which is adequate at best.
2 rows (4 GSD)
With such a good time saving achieved, we repeated the inspection flight but changed the parameters to survey 2 rows of modules per sweep.
This flight still only took 7min 57s and recorded 76 images.
A good workflow is needed to report the position of the modules.
The improved GSD of over 30% gives much better resolution and clarity on both the thermal and RGB data. Each jpeg is radiometric so can simply be worked on for in-depth thermal analysis.
1 row (2 GSD)
Encouraged by the time savings and improved results of the 2-row survey we then ran another mission for 1 row.
This flight took over 24minutes and did not cover the whole sample site. Obviously, it is double the length of flying when compared to the 2-row route, but another consideration is this level of flight must be performed a lot slower airspeed to avoid motion blur in the data.
The data looks great, but localising the fault is challenging.
This type of close inspection flight is ideally used to inspect of an area of modules which are known to have subtle latent defects such as snail trails or PID for example.
Conclusion
Thermal orthographics look nice and are easy to fly, but the amount of data and flying that is required is quite substantial. There are some obvious restrictions on the achievable quality and post processing. Not only does it become unfeasible financially to collect and process high resolution data, but stitching softwares tend struggle to process images that have been flown from a low altitude, particularly without ground control points. In my opinion this workflow is suitable where more obvious faults need to be analysed and flights are conducted at a high level.
For clients who require more detailed results, then significant improvements to the data quality can be gained by flying lower and using an individual image inspection workflow. Additionally, substantial time savings can be made in the field and on the post processing.
When compared to the orthographic method, the 2-row route:
· improved the data quality by over 30%
· reduced flight times by over 50%
· required 80% less data
· was processed considerably quicker
For more information please get in touch [email protected]
