Machine learning-based approach to identify the optimal design and operation condition of organic solvent nanofiltration (OSN)

Abstract

Abstract Organic solvent nanofiltration (OSN) is one of the most anticipated separation technologies that provides wide-ranged industrial applications such as solvent recovery, solute concentration, and diluent separation. Despite of technical merits of the OSN technology, the numerous characteristics and perplexing nonlinearity on the OSN system have been a critical obstacle for understanding the governing principles, thereby prohibiting practical deployments. Recently, machine learning (ML) based approaches have been widely used for the modelling, discovery and optimization of complex design problems in chemical engineering area such as catalysis, electrochemistry and physicochemical systems. Therefore, this study aims to develop a new ML-based approach for modelling and optimizing the design scheme and operating condition of the OSN system. By collecting commercial OSN data through literatures reviews, the major descriptors for the prediction of the OSN membrane, such as MWCO, solute mole weight, solute concentration, solvent parameter, temperature, pressure, flux, were defined. We then screened noises and outliers of the collected data to ensure a high and consistent density and uniqueness. Support vector machine (SVM) was implemented as a prediction models to simulate the OSN performance and identify the optimal conditions as well as the process scheme. As a result, the optimal operation strategies (i.e., pressure, temperature and solvent and solvent types) were analyzed to meet the targeted specification of the OSN system (mass flux and rejection rate). The proposed ML-based approach can promote a real-world OSN application by reducing a number of time-consuming and expensive experiments for establishing OSN design and operation strategy.