(166c) Integrating Physics and Machine Learning: A Hybrid Modeling Approach for Electrochemical Separation

Electrochemical separation plays a pivotal role in a variety of industries, ranging from water treatment to oil refineries. Despite its efficiency, the challenge lies in the design and synthesis of targeted processes. Mathematical models are crucial in this context, as they facilitate the development of new technologies by enabling simulation, sensitivity analysis, techno-economic evaluations, and optimization. These models are instrumental in providing insightful information while curtailing the time and resources necessary for devising new designs. The integration of data-driven and knowledge-based modeling approaches can significantly hasten the innovation of new technologies.

In our study, we present a novel approach where physics-based models generate synthetic data imbued with specific material properties. This data is then used to train a machine learning-based surrogate model on which Transfer Learning are applied to fine-tune the model's performance, aligning it more closely with experimental measurements. Our modeling approach is tailored for systematic screening of material properties and selecting optimal parameters for guiding electrochemical separations. The effectiveness of this methodology is exemplified through its application in modeling two widely used water desalination technologies: electrodialysis and electrodeionization.