Inspection of a solar plant with thermal orthomosaic vs. digital twin with individual thermal image analysis

The use of drones in the inspection of photovoltaic plants is already known for its many advantages, including the speed of data collection compared to manual methods. When performing a thermal inspection of PV panels with drones, data recording can be done in different ways, two of which are the thermal orthomosaic and the digital twin with individual analysis of the thermal images.

We have done a flight simulation with each methodology on a 2MW photovoltaic installation to present a real time-of-flight comparison of each, number of images needed, overlaps and results obtained. Finally highlighting the best of each methodology in a breakdown of advantages and disadvantages.

In the following we will explain what is thermal orthomosaic and individual analysis of thermal images for digital twin with their respective simulations.

What is a thermal orthomosaic?

Image 1. Thermal orthomosaic of a 2 MWp solar installation and a GSD of 6 cm/px.

It is the set of images taken by aerial equipment, in this case a drone, and processed by specialised software that "stitches" them together geometrically corrected and georeferenced, thus creating a sort of detailed map of the photographed place, equivalent to a geographical chart. This particularity of the thermal orthomosaic process, of eliminating the perspective of each individual image by creating a single image at the same level, i.e. with an orthogonal view, is called "orthorectification" and is the function performed by such software, joining all the corresponding images together. To create an orthomosaic, a large overlap between one image and another is necessary, by this we mean overlaps approaching 70-70 and even 85-85, frontal and lateral.

Each photo taken for the production of the thermal orthomosaic does not have much value because its level of detail is not very high, but the combination of these photos creates a valuable product that helps us to detect certain anomalies and allows us to know the general state of the plant. 

In turn, the images for the thermal orthomosaic can be taken at a GSD of 3-3.5 cm/px with which we can access a good level of detail. However, the RGB images that complement the thermal orthomosaic to access a higher accuracy of the anomalies are usually taken at a higher GSD than the IEC standard, otherwise, too many very high quality images would be taken, which would lead to a very heavy, costly and inefficient processing of these images.

For the realisation of the thermal orthomosaic, an average of 7-8 times more images will be needed than for the individual thermal imaging inspection, increasing the flight time by at least 7 times. We will expand on this later.

As an example, let's look at the following flight simulation we performed on a 2MW PV installation with thermal orthomosaic at a GSD of 3 cm/px and overlaps of 85-85 front and side respectively.

Figure 2. Flight simulation made with DJI GS Pro for a 2 MWp solar installation and 85-85% overlap.

What is individual analysis of thermal images based on Digital Twin?

Picture 3. Digital Twin of a 2 MWp solar installation and a GSD of 3 cm/px.

Individual thermal imaging is the second methodology we propose for the inspection of solar plants. They are also a set of images photographed by aerial equipment (in this case, a drone) but unlike the orthomosaic, these images do not need to be uploaded to any software that "stitches" them together (stitched images), but are analysed individually one by one.

This means significantly fewer images because less overlap is needed. Likewise, IR and RGB imaging can be done at lower GSD which allows us to identify smaller anomalies more easily. 

Flying to a standard GSD gives us the possibility to analyse the individual image in more detail so that we can clearly recognise the type of anomaly and take appropriate action. For example, if we can recognise that a "hot spot" that is reflected in the thermal image is really a "hot spot" when contrasted with the RGB image (colour image) we will send the O&M manager to replace the module, but if we find that the "hot spot" is caused by vegetation the measures would be different, e.g. call the company in charge of the pruning.

The individual images require less overlap and therefore fewer images, so the flight time is shorter than the flight time for imaging a thermal orthomosaic. 

We are going to see the second simulation that we did on the same 2MW photovoltaic installation and in this case we carried out with the digital twin method and individual analysis of the images.

Figure 4. Flight simulation made with DJI GS Pro for a 2 MWp solar installation and 80-20 overlaps.

Comparison of the results between thermal orthomosaic vs. digital twin with individual image analysis.

Thermal orthomosaicDigital twin with individual images
Flight duration time: 264 minutes 44 seconds

Number of images: 6535 photos

Battery used: approx. 17 sets

Front overlap between images: 85%

Side overlap between images: 85% 
Flight duration time: 41 minutes 1 second

Number of images: 920 photos

Battery used: 3 sets approx.

Front overlap between images: 80%

Front overlap between images: 20%.

As we can see from the results, the flight time is multiplied for the thermal orthomosaic as well as the number of images, which leads to a much higher workload compared to the digital twin method with single image analysis.

Taking a closer look at the results we have separated the advantages and disadvantages of each method.

Thermal orthomosaicDigital twin with individual thermal imaging
- This is a "map" that allows us to see the entire plant.
- View groups of disconnected CTs or Inverters.
- Chronological evolution of the anomalies.
- Fly at a lower GSD for RGB images, comply with international standards (IEC TS 62446-3:2017).
- Higher accuracy: Better resolution of images makes it easier to identify damage.
- Faster flight time.
- Fewer images.
- Faster processing.
- Chronological evolution of anomylies.
- The reflection of the sun on the panels may make it difficult for the software to create the orthomosaic.
- Flying time is longer.
- Image processing may be cumbersome.
- May be more expensive.
- Depending on the GSD, difficulty in identifying entire groups of disconnected Inverters or CT's which will require specific image processing protocols to locate these faults.

In this way, we would like to conclude by saying that, as we have seen above, each methodology has its pros and cons, but both have the same objective: to know the state of the solar plant and identify the anomalies present.