Optimization of polyimide-based membrane distillation performance through machine learning parametric study

Membrane distillation (MD) is a prevalent separation technology employed in water treatment applications using porous membranes. This study presents a detailed performance evaluation and predictive modeling of an intrinsic porous polyimide membrane prepared from 4,4'-(hexafluoroisopropylidene) diphthalic anhydride and 2,3,5,6-tetramethyl-p-phenylenediamine (i.e., 6FDA-TMPD polymer). The influence of operational parameters, such as feed salinity, temperature, and flow rates, on membrane performance, particularly flux and gain output ratio (GOR), was analyzed. Results showed that increasing the synthetic feed solution temperature from 50 to 80 ?C led to a significant improvement in both flux and GOR. In contrast, replacing the synthetic feed solution (NaCl, 70,000 ppm) with real seawater (48,200 ppm) reduced the flux and GOR by 22.01 % and 26.71 %, respectively. The 6FDA-TMPD membrane demonstrated stable salt rejection > 99.94 % for both synthetic and real seawater. Furthermore, different data preprocessing techniques and machine learning (ML) models were utilized. ML models, specifically boosted trees (BT) and adaptive neuro-fuzzy inference systems (ANFIS), were developed to predict flux and GOR using experimental data. The ANFIS model demonstrated excellent performance, indicating high predictive accuracy with root mean square error (RMSE) values of 0.0529 and 0.0544 for flux and GOR, respectively. Analysis of operational parameters evaluated in this study identified feed temperature and concentration as dominant factors influencing membrane performance. This study contributes to advancing MD membrane optimization for sustainable high saline water treatment by integrating experimental and data-driven techniques.