A systematic screening of neural network-based hybrid models of adsorption in chromatography processes

In this study, various adsorption models using neural networks were developed and integrated into a mechanistic chromatography transport model, resulting in hybrid models. A systematic screening of 10 different hybrid model structures was performed to find the optimal balance between mechanistic and data-driven components in modeling adsorption for chromatography processes. The hybrid models were trained and tested on two different case studies of varying complexity and compared to a validated mechanistic model. The first case study investigated a nonreactive binary component system, whereas the second case study investigated a four-component reactive system. Most hybrid models performed well in the first case study, showing performance comparable to that of the mechanistic model. However, in the second case study, fewer hybrid models demonstrated performance comparable to that of the mechanistic model. Through the two case studies, advantages and limitations of different hybrid model structures for modeling adsorption were explored.