Computer-Aided Screening and Design of Solvents under Uncertainty

  • Rechnergestütztes Lösungsmittelscreening und -entwicklung unter Unsicherheit

Wicaksono, Danan Suryo; Marquardt, Wolfgang (Thesis advisor); Liauw, Marcel (Thesis advisor)

Aachen : Publikationsserver der RWTH Aachen University (2015)
Dissertation / PhD Thesis

Aachen, Techn. Hochsch., Diss., 2015

Abstract

The selection of proper solvents can be a key decision variable in engineering process attributes towards a certain objective. This contribution outlines a systematic framework for screening and/or design of solvents under both model parametric and structural uncertainty. The framework integrates model-based and data-driven methods. The model used is based on molecular descriptors, such as Kamlet-Taft parameters. The contribution addresses the selection of not only traditional organic solvents but also sophisticated novel solvents for chemical reaction engineering and biomass processing.This contribution demonstrates that the proposed framework is able to identify promising reaction solvents for a class of SN1 reactions amidst uncertainty in the data. The undesirable uncertainty propagation is treated using a combination of Tikhonov regularization and optimal design of experiments. The uncertainty propagation analysis employing Monte Carlo simulations demonstrates the advantages of employing the proposed framework over another method based on chemical insights. This contribution discusses the application of the framework on cellulose dissolution in ionic liquids which quantitatively reveals the contribution of each Kamlet-Taft parameters in this context. Hydrogen-bond acceptor basicity is dominant but not the sole contributor. By combining Kamlet-Taft parameters and some specific molecular structures, two separate regions of cellulose dissolving and non-cellulose dissolving ionic liquids can be characterized. This contribution discusses the application of the framework on the selection of the liquid solvent to be paired with the compressed CO2 and its composition in gas-expanded liquids for a Diels-Alder reaction. A mixed-integer nonlinear optimization model which incorporates Bayesian multimodel inference is proposed. Two reformulation strategies, tailored big-M and binary multiplication, are proposed in order to achieve better computational performance. Three CNIBS/R-K models are shown to be inferior to two preferential solvation models in predicting the Kamlet-Taft parameters of CO2-expanded liquids.