Machine learning-based prediction and augmentation of dish solar distiller performance using an innovative convex stepped absorber and phase change material with nanoadditives

Abstract

As well known, the solar distiller is one of the introduced solutions to the freshwater shortage problem, but it is demerited by the low freshwater output. In this paper, a design modification that includes the use of a convex dish absorber instead of a flat absorber liner was proposed. Also, a circular stepped surface was used instead of the flat absorber surface. The modified solar distiller is nominated by dish solar distiller and abbreviated by DSD. In addition, a cotton wick was used as a wetting material for facilitating the evaporation process inside the distiller. Besides, the effect of different water heights in the steps was investigated for 0.50, 1, 1.50, 2, and 3 cm. Finally, the space under the dish absorber is filled with a phase change material (PCM) of paraffin wax mixed with CuO nanoparticles. Experimental results revealed that the best water depth over the steps absorber of DSD that provided the highest freshwater productivity was 1.50 cm, where the average daily yields of DSD (at 1.50 cm) and conventional distillers were reported as 6525 and 2800 mL/m².day, respectively. Then, the productivity of DSD was improved by around 133% over that of the conventional distiller. In addition, when using the phase change material, the average daily distillate of stepped DSD was improved by approximately 178% compared to that of the conventional solar still, where the distillate of conventional still and DSD with PCM at 1.50 cm water depth over the steps absorber of DSD was 2950 and 8200 mL/m².day, respectively. The water productivity of the three established solar distillers has been predicted using machine learning algorithms. Besides, the maximum thermal efficiency of DSD was obtained when using PCM at 1.5 cm water depth over the steps absorber of DSD, where it was 67.62% compared to 31.71% for the conventional distiller. The proposed machine learning algorithms succeeded in predicting water productivity with a high correlation coefficient of 0.99.