Machine Learning in Membrane Distillation for Pulp and Paper Wastewater Treatment: A Review

Abstract

Membrane distillation (MD) is a promising thermally driven separation technology for treating high-strength industrial wastewater, particularly from the pulp and paper industry. However, its performance is constrained by complex and nonlinear interactions among operating conditions, feed characteristics, and membrane properties, leading to challenges such as fouling, membrane wetting, and flux decline. Conventional modelling approaches are often inadequate in capturing these dynamic behaviours, resulting in limited predictive reliability. This review focuses on the application of machine learning (ML), with particular emphasis on artificial neural networks (ANNs), in enhancing MD performance prediction and optimization. ANN models have demonstrated strong capability in modelling nonlinear relationships and accurately predicting key performance indicators, including permeate flux, rejection efficiency, and fouling behaviour. Special attention is given to recent developments integrating ANN models with advanced membrane materials, such as polyvinylidene fluoride/cellulose nanocrystal (PVDF/CNC) composite membranes, to improve system efficiency and stability. Furthermore, probabilistic approaches, including Bayesian neural networks and Monte Carlo simulations, are discussed for uncertainty quantification and risk-informed decision-making. Key challenges, including data scarcity, model interpretability, and generalizability, are critically evaluated alongside emerging solutions such as explainable AI and digital twin systems. Overall, ANN-driven approaches offer significant potential to transform MD into a predictive, reliable, and intelligent wastewater treatment technology.