Hybrid Modeling Approaches for Pervaporation: Combining Physical Models with Machine Learning for Interpretable Process Predictions

Membrane-based separation processes such as pervaporation offer energy-efficient alternatives or complements to conventional thermal separation methods and can therefore make a significant contribution to the defossilization of the process industry. However, existing models for describing pervaporation are predominantly empirical and typically require system-specific calibration to experimental data, which is often only sparsely available in the literature. As a result, new mixtures or membranes usually necessitate dedicated experimental campaigns, limiting the applicability of existing models in the early stages of process development.

My research aims to develop hybrid modeling approaches that combine machine learning with physics-based transport models. The focus lies on the data-driven estimation and generalization of model parameters to significantly improve the transferability and predictive capability of pervaporation models. In addition, I am developing an experimental setup for the efficient and precise characterization of pervaporation membranes.

Based on this foundation, the resulting modeling approach can be seamlessly integrated into flowsheet based process simulations or used directly as a predictive tool, providing a level of flexibility comparable to classical thermal separation models and thereby substantially lowering the barriers to applying pervaporation in early stage process development.