An award-winning South African drone services start-up says it has found a novel formula to inspect concentrated solar power plants.

Cape Town-based Integrated Aerial Systems said it had to think on its feet after it was recently contracted to inspect a concentrated solar power plant just outside Upington in the Northern Cape province.

That is because concentrated power plants are different from conventional photovoltaic power plants and according to IAS, drone inspection on these surfaces is not the same.

“When it comes to aerial surveys of solar farms, drones are commonly used to detect anomalies on Photovoltaic panels,” the company explains.

“When sunlight is absorbed by a module, it is converted into electricity. However, if cells on the module are not optimally converting the irradiance from sunlight into electricity, they heat up and produce ‘hot spots’. These hot spots can be picked up by the drone’s thermal camera, which is highly sensitive.

“(Therefore) a typical Solar PV site inspection involves a drone with a thermal and visual camera. The drone flies in a grid pattern collecting thousands of images, which are then processed using machine learning systems to accurately identify and categorise anomalies across the site.

“These inspections have become standard practice, with specific data capturing methodologies dictating the type of reporting outputs.”

Concentrated power plants, however, are a different kind of fish altogether. They use mirrors to concentrate thermal irradiance to generate electricity. These mirrors reflect and concentrate sunlight onto a specific point (in this case, a Heat collecting element (HCE)) to create heat.

This heat is then stored in a medium, such as molten salt, and then run through a heat exchanger, which in turn creates steam. This steam drives a turbine, which creates electricity. The plant runs in a ‘closed loop’ system, which basically means that water and molten salt are recycled.

IAS discovered that this setup would mean that this kind of solar power generating system would need a health check up system of its own; which is different from the conventional thermal inspection system.

“Inspecting this kind of site required a methodology to be developed from the ground up, as no industry-established methodology existed. This lack of guidelines meant that Integrated Aerial Systems (IAS) developed its own methodology.”

As explained by the IAS CEO Warren Witte, concentrated solar power generation systems use mirrors; and the reflective nature of the mirrors meant that drone operators on the inspection mission had to keep adjusting the drone’s camera angles to reduce glare and distortion in RGB imagery.

That then ruled out the possibility of automated flights, as a pilot was supposed to stay on active control to be on the lookout for mirror reflections and other changes that require camera adjustments.  

Charting a new path was not without its challenges, however.

“Several complications were encountered during this process, including the reflective nature of the mirrors, the distance between the strings, and the oscillating movement of the Parabolic trough.

“Due to these complications, flights had to be done manually, which was far slower than an automated inspection. In total, approximately 38,000 images were produced from the inspection using the DJI M210 v2 drone.”

Given the sheer volume of data that had to be collected, it took the drone company nearly a month to cover the whole plant, and even then, they had to rope in a second drone to help the first one.

But they got the job done, and after succeeding in collecting the data, the next challenge was to process it.

“As a bespoke solution like this had not been conducted before, IAS had to develop a customised processing solution through a geospatial software house,” the company said.

“The goal was to identify partially- and fully-dislocated internal shields (which are responsible for creating a vacuum inside the parabolic tube), as well as broken tubes. A machine learning model was developed and trained to analyse each image and identify these shields and broken tubes.

“However, the model had difficulty quantifying the data. To solve this, a separate algorithm was run to provide a minimum and maximum number of possible broken elements, as an exact number could only be determined by visually inspecting each image.”

IAS added that while identifying dislocated shields can only be done with visual inspection and high-resolution cameras, broken elements and those that have vacuum loss are easily identifiable with thermal imagery coupled with the RGB data using temperature delta.

“A loss of vacuum in the HCE does not allow heat to dissipate, and a significant deviation in temperature is observed. By applying a second RGB layer, elements that are broken and those that are not can be separated.”

The trouble IAS went through was worth it however, as the company says the results of the survey were exceptionally accurate and saved their client significant time and money over the alternative of a manual inspection.

The client was able to use IAS’s data as proof for warranty claims on certain components as well as fast-track targeted repairs and maintenance.