Performance estimation of a mini-passive solar still via machine learning

Abstract

Achieving high water productivity in single-basin solar stills remains a challenge and may require efficient still insulation and downscaling to ease experimentation. Here, mini-passive polystyrene (PS)-based single-slope solar still is designed for brackish water desalination. Supervised machine learning regressions are applied to create trained models from experimental results. The proposed method aims to develop accurate predictive models via dimensional analysis and datasets expansion from in-between randomization. Built models predicted the still performance (η) when replacing PS with another wall-insulating material. We correlated the water-glass temperature (Tw–Tg) and evaporative coefficients (hewg) to the still outputs using the stepwise linear regression (SLR) showing minimum statistical errors (R2≈1) and RMSE<0.016. A good agreement between theoretical, numerical, and experimental results is observed; while decreasing feed rates boosts evaporation/condensation. The still achieved a maximum η = 18.33% corresponding to F = 30 mL/day, Tw–Tg = 4.6 °C, hewg = 21.11 W/m2°C, and radiative water-glass coefficient (qrwg) = 0.188 W/m2 at 15:00 time. Hourly-measured still outputs fitted against NASA insolation followed similar patterns confirming the successful operation. Polyurethane (PU) and Silica are found to be promising wall-insulating candidates for maximizing the still output owing to their low kins. This work paves the way towards retaining the still absorbed radiation via thin-film foil-wrapped low-conductive insulators.