Machine learning modeling of polycarbonate ultrafiltration membranes at different temperatures, Al2O3 nanoparticle volumes, and water ratios

Abstract

The efficacy of novel polycarbonate ultrafiltration, aluminum oxide nanoparticle (Al₂O₃-NPs) volume fraction, temperature, and water/ethylene glycol (EG) ratio were evaluated to determine the thermophysical properties of the membrane. 5%-10% of Al₂O₃-NPs have been added to the PC. A machine learning approach was used to compare the volume fraction of Al₂O₃-NPs, the temperature, and the water-to-ethylene glycol (EG) ratio. To determine the impact of Al2O3-NPs loading on the Response Surface Method (RSM), DOE, ANOVA, ANN, MLP, and NSGA-II, the number of aluminum oxide nanoparticles (Al₂O₃-NPs), temperature, and water/ethylene glycol (EG) on membranes in PC ultrafiltration are evaluated. Based on the Relative Thermal Conductivity Model (RSM), the regression coefficient of Al₂O₃ in water and EG was 0.9244 and 0.9170 with adjusted regression coefficients. A higher concentration of EG enhances the thermal conductivity of the membrane when the effective parameters are considered. The effect of temperature on the relative viscosity of the membrane led to the conclusion that Al₂O₃ water/EG can cool at high temperatures while providing no viscosity change. When Al₂O₃ is dissolved in water and EG, more EG is necessary to optimize the mode of reactivity. Using the MLP model, the calculated R-value is 0.9468, the MSE is 0.001752989 (mean square error), and the MAE is 0.01768558 (mean absolute error). RSM predicted the average thermal conductivity behavior of nanofluid better. The ANN model, however, has proven to be more effective than the RSM in simulating the relative viscosity of nanofluids. The NSGA-II optimized results showed that the minimum relative viscosity and maximum coefficient of thermal conductivity occurred at the lowest water ratio and maximum temperature.

Keywords: Aluminum oxide; NSGA-II; Optimization; Thermophysical characteristics.