Application of Machine Learning for the Development of the Next Generation of Membrane Material for Water Treatment Purposes.

点击翻译按钮获取中文摘要

According to several reports, consequences of global warming and climate change including severe droughts, saltwater intrusion into groundwater, and shrinkage of glaciers, in addition to the increased water demand due to economic developments, fast urbanization, and population increase, have combined to result in an estimated 40% global water shortage by the year 2030. Hence, in addition to improvements in the conservation, distribution, and management of water resources, ensuring that new sources of fresh water can be readily available is essential in order to meet the increasing demand. As a result of its abundance, seawater desalination and wastewater treatment and reuse have been proposed as a solution to the problem of water deficit.Water treatment using membrane technologies is an attractive solution to water shortages since they consume little energy, operate efficiently, have a small footprint, and require little maintenance. Some of these solutions have been applied on an industrial scale including Reverse osmosis (RO), Nanofiltration (NF), Ultrafiltration (UF), Microfiltration (MF), Membrane distillation (MD), Electrodialysis, Forward osmosis (FO). Despite all the recent progress in this field, serious issues remain unaddressed yet. High energy requirement and fouling in RO membranes, concentration polarization, and reverse solute flux (RSF) for FO are among these. It is possible to alleviate the effect of some of the mentioned issues by optimizing the process parameters and conditions; however, some of them such as internal concentration polarization are intrinsic to the membrane itself and the limitations imposed by the membrane material. Consequently, in order to further improve the performance of such membranes it is necessary to consider different materials for membrane synthesis to overcome the current functional limitations. Although the idea of applying different combinations of polymers for membrane synthesis seems feasible it could be a rather long, tedious, and expensive process and there would always be the possibility of overlooking eligible candidates due to limited time and resources.Recently, A number of machine learning (ML) techniques, such as response surface methodologies (RSMs) and artificial neural networks (ANNs), have been increasingly applied to optimize process parameters as well as to predict, simulate, and/or model inputs and outputs. In machine learning, computational algorithms are used to develop these models based on data. It is possible to analyze the relationships between input variables and output variables using a learning mechanism, rather than using mathematical model equations, and to predict complex non-linear systems with high precision using this method. While most studies have focused on the optimization of processes by modifying operating conditions, to this date there are no comprehensive studies investigating the effect of the application of different membrane materials and polymers on the performance of the system.This study would aim to investigate the possible developments in membrane fabrication and material in order to overcome the limitations imposed on the membrane water treatment process performance due to the intrinsic characteristics of membranes. In this regard, a comprehensive database of different membranes used will be implemented to train the algorithm, and later on, simulations will be carried out to predict the performance of various possible combinations of material when used as a membrane. This process can significantly reduce the amount of time and resources related to testing all the candidate material in the laboratory and shortlisting the ones with desired performance. In the next step, potential candidates with noticeable performance regarding CP, fouling, rejection rate, water recovery rate, etc. will be fabricated and tested in real conditions to validate the simulation results.