Estimation of general nonlinear adsorption isotherms from chromatograms

Adsorption isotherms are the most important parameters in rigorous models of chromatographic processes. In this paper, a numerical method to estimate them from chromatograms is proposed. In contrast to existing numerical methods, the isotherms are represented by neural networks. The weights and the biases of neural networks are optimized by a nonlinear least squares algorithm. Due to the universal approximating capability of neural networks, this method could theoretically retrieve any form of isotherm from the chromatograms. Several issues of the method are studied, which include the chosen type of neural networks, the choice of the mapping relationship, the initialization of neural networks, the importance of the experimental design used to generate the chromatograms and the consideration of the physical role of the isotherms. The potential of the new method is illustrated by the application to the isotherm estimation for EMD53986 on a chiral phase from measured batch elution chromatograms.