The efficiency of solar power plants is being improved using machine vision systems. eSolar, a company based in Pasadena, California and currently building Concentrated Solar Power (CSP) plants in the Southern Californian desert, are using vision systems to maintain the alignment of mirrors focussing the sun's rays.
Where other solar thermal power plants use a relatively small number of expensive and large curved mirrors to reflect the sunlight, eSolar uses thousands of small flat mirrors, or heliostats. By utilising a sophisticated tracking system mirrors continuously focus sunlight onto a thermal receiver. Such concentrated light is used to boil the water within the receiver into steam. The steam is aggregated to a turbine and powers a generator.
A key ingredient of the system is image processing based on a set of cameras installed around the field. Prosilica supplied the cameras used by eSolar including the 752 x 480 monochrome GC750 and the GigE Vision-compliant GC750.
Mirrors are programmed to track and focus the sun to the centre of a target. Random mirrors are sampled throughout the day and individual frames are taken at 30-second intervals. When an image is received, its background is automatically removed to show the image of the mirror reflection. The basic analysis is a centroid of the reflected sunlight compared to the centre of the target. A perfectly aligned mirror will consistently reflect sunlight to the centre of the target throughout the day. When a divergence is spotted, eSolar engineers update the proprietary sun-tracking algorithms to correct the mirror alignment. By allowing heliostat tolerances to be corrected via the tracking system, eSolar achieves lower overall costs by cutting down on manufacturing and reducing the individual alignment of mirrors.
Camera functions controlled by software include triggering, exposure, and gain. All of these are set to provide the best image of the reflected sun. The application software was developed by eSolar's engineering team using the Prosilica SDK for Windows. The programming was done using C#, C++ and Microsoft Visual Studio tools.
