Sustainable Isopropyl Alcohol Recovery via Data-Driven, Active-Learning Optimization of Vacuum Membrane Distillation

The need for separation and recovering solvents from industrial byproducts is critical, given their environmental and economic impact. This study explores the feasibility of energy-efficient vacuum membrane distillation (VMD) using polypropylene hollow fiber membranes to recover and reuse isopropyl alcohol (IPA) from wastewater. Notably, this study pioneers the utilization of active learning-based Bayesian optimization (BO) to optimize the complex, multiobjective, and constrained VMD process. BO is an effective method for global optimization in black-box functions, particularly when data is limited and difficult to acquire. It creates a data-driven surrogate model (e.g., Gaussian process model) due to the shortage of data and the need for uncertainty quantification. The data-driven approach by BO efficiently reduces the number of necessary experiments, demonstrating the value of active learning in sustainable chemical processes. The research focuses on the data-driven optimization of key process variables in VMD required in scale-up operation, including the feed solution temperature, lumen-side membrane pressure, solution flow rate, and IPA concentration. To assess the long-term viability of the optimal conditions identified by BO, particularly given the risk of membrane pore wetting, we also evaluate the long-term stability of the optimized process. This study bridges the gap between fundamental research and practical application, offering a robust foundation for using advanced optimization techniques in the separations field. It holds promise for broad industrial applicability, providing both environmental and economic benefits.