Optimizing operating parameters for the scalability of photothermal membrane distillation using machine learning techniques

Membrane distillation (MD) is emerging as a promising desalination technology capable of addressing global water shortages. However, it is limited by high energy requirements, with temperature polarization reducing thermal efficiency and flux, hindering its scalability and commercial viability. This challenge can be mitigated through the development of photothermal membranes, which enhance localized heating and reduce polarization effects, thereby improving the overall performance and scalability of MD systems. This study aims to optimize the operating parameters of a novel photothermal membrane, PVDF/h-MWCNTs, to enhance process efficiency as the membrane area is scaled up. A mixed-level factorial design (MLFD) experimental dataset from an air-gap membrane distillation (AGMD) setup was utilized to simulate various machine learning approaches, including regression, tree-based, and neural networks. An optimization algorithm was also employed to determine the optimal operating conditions for photothermal MD (PMD) technology. The influence of operating parameters (inlet feed velocity, solar intensity, and air gap width) on permeate flux, localized heating effect, and photothermal efficiency was investigated across different membrane areas. The results showed that the gradient boosting model best predicted permeate flux (R²: 0.943) and photothermal efficiency (R²: 0.893), while the random forest model excelled in predicting localized heating effects (R²: 0.977). Optimization identified that the membrane with the largest area of 536.1 cm², operating under conditions of 0.24 cm/s feed inlet velocity, 0.24 cm air gap width, and 0.7 kW/m² solar radiation intensity, achieved a permeate productivity of 1.95 L/m²h. This study provides a framework for further analysis and scale-up of PMD processes.