Tailoring thin film composite membranes for clean water production: A study on structural variations and predictive insights using machine learning

Abstract

This study focuses on the fabrication of three thin film composite membranes by interfacial polymerization on polyacrylonitrile support. The objective is to explore the influence of different monomers on structural and functional features of resultant membranes for treating saline water and removing micropollutants. A new linear aliphatic amine 4A–2E was used as an aqueous monomer for the first time. Trimesoyl chloride, isophthaloyl chloride, and terephthaloyl chloride were used as organic phase crosslinkers. The resulting membranes exhibited distinctive structural features such as varied roughness and hydrophilicity, which correlated with their functional performance. The membrane roughness for 4A–2E-TPC@PAN/PET reached the highest value of Ra = 56 nm among the fabricated membranes. Moreover, permeate flux of 32, 56, and 49 L m−2 h−1 at 25 bar was measured for 4A–2E-TPC@PAN/PET, 4A–2E-IPC@PAN/PET, and 4A–2E-TMC@PAN/PET, respectively. Notably, 4A–2E-TPC@PAN/PET membrane exhibited 98% rejection of MgCl2 among the tested divalent salts. Furthermore, the membrane displayed high rejections of micropollutants, with 85% rejection for Amitriptyline. Upon chlorine exposure, the 4A–2E-TPC@PAN/PET membrane proved more resistant than both 4A–2E-IPC@PAN/PET and 4A–2E-TMC@PAN/PET membranes. Machine learning predictive insight was developed using Matérn Gaussian Process Regression model to simulate flux before and after chlorination of membranes. The results in both training and testing proved reliable predictive insight with 99% accuracy in terms of Nash Sutcliffe Efficiency. Hence, current work highlights the use of a linear amine 4A–2E and TPC crosslinker to fabricate a dense polyamide membrane and the application of ML algorithms to desalination data makes it possible to predict the membrane performance.