Dust InSMS: Intelligent soiling measurement system for dust detection on solar mirrors using computer vision methods

Abstract

The dust accumulation strongly impacts the optical efficiency of solar concentrators, in particular the reflectivity of solar mirrors. Therefore, reducing the impact of reflectivity losses due to soiling and optimizing cleaning strategy are key factors. In this paper, the impact of dust accumulation on the reflectivity parameter of Fresnel mirrors is studied at the GEP research platform during the dry period. Based on the collected data, a new system for dust detection is proposed based on the classification approach using the convolutional neural networks and image processing algorithms in which no similar work is presented in the literature that uses the same approach to quantify the soiling phenomenon on CSP mirrors. The test loss and accuracy obtained by the proposed model are respectively 0.28 and 0.96. The outdoor validation results obtained so far suggest that the Dust InSMS concept and method could be a promising efficient and low-cost sensor. As the proposed system performs the GPS coordinates for each measurement, an optimal cleaning scenario is developed based on genetic algorithms to optimize the cleaning scenario and to come up with the shortest cleaning path.