ANN-GA driven prediction and optimization for upscaling photothermal air-gap membrane distillation systems using Octy-Cu/PVDF membranes

With escalating global water scarcity, particularly in arid regions, there is a critical need for sustainable desalination technologies beyond conventional energy-intensive methods. Photothermal membrane distillation (PMD) has emerged as a promising solar-driven alternative, yet its scalability remains challenging due to complex heat and mass transfer dynamics. This quantitative study advances the upscaling of PMD by integrating machine learning (ML) models with experimental data of novel octylamine-functionalized copper oxide (Octy-Cu) incorporated into PVDF membranes. The smaller module's permeate flux increased with higher solar radiation and lower feed inlet velocity. In contrast, larger modules exhibited enhanced performance at higher feed inlet velocities due to improved turbulence and reduced temperature polarization. Several ML regression models, including Decision Trees, Random Forest, and Artificial Neural Networks (ANN), were employed to predict flux. The ML-based method demonstrated superior predictive capabilities and flexibility in capturing complex non-linear behaviors compared to traditional empirical or semi-empirical scale-up approaches. Although ANN initially showed signs of overfitting (R2 = 0.94 training, 0.64 testing), its architecture was optimized (2 layers, 160 neurons, 0.2 dropout), significantly improving predictive accuracy (R2 = 0.90 on test data). Furthermore, Genetic Algorithm (GA) optimization identified the optimal operating conditions to achieve 1 LMH flux at 0.1025 m2 membrane area, 0.0148 m/s feed velocity, and 1.9 kW/m2 solar radiation. The integration of ML-driven optimization of PMD enables a scalable and energy-efficient desalination approach, offering a viable pathway for the real-world deployment of sustainable water treatment systems.