(71g) Deterministic and Stochastic Mass Transfer Models for CO2 Capture Processes  

For accelerating deployment of novel CO₂ capture technologies, rigorous process models can be very valuable to reduce the risk and uncertainty in scaling up the process by skipping intermediate scales. For such process models, it is critical that the uncertainties in the key submodels be quantified. For chemical solvent-based post-combustion CO₂ capture technologies, mass transfer models play an important role especially in estimating the performance of the absorber. With this motivation, comprehensive mass transfer models of a chemical solvent based CO₂ capture process with quantified uncertainties is developed in this work.

The typical approach in the literature to development of mass transfer models is sequential. In chemical solvent based CO₂ capture systems such as for the MEA-H₂O-CO₂ system, very fast reaction kinetics make it difficult to obtain measure of the mass transfer flux and identify the associated parameters with the mass transfer models. Assuming the diffusivity model to be satisfactory, mass transfer coefficients are typically calculated from the experimental data from the wetted wall column (WWC) experiments. The experimental data from packed columns are the used to obtain the interfacial area model assuming the mass transfer coefficient models are still valid. As the WWC and packed column operate in different flow and operating regimes, it is not necessary that the mass transfer coefficient models and their parameters are still optimal for the packed column. In addition, assumption of accuracy of the diffusivity model is also questionable. Overall, a suboptimal mass transfer model is obtained. To circumvent these issues, simultaneous regression of parameters for the mass transfer coefficient model, diffusivity model, and interfacial area model is carried out in this work for the MEA-H₂O-CO₂ system by simultaneously using the experimental data from the packed columns and wetted wall columns. Since such simultaneous regression is not feasible in existing state-of-the-art simulation software, a computation tool named Framework for Optimization, Quantification of Uncertainty and Surrogates (FOQUS) developed as part of U.S. DOEâ??s Carbon Capture Simulation Initiative is leveraged to accomplish this objective.

Following the deterministic parameter regression, rigorous uncertainty quantification is performed for the mass transfer models. The best guess for parametric uncertainty (priors) is then cast in the framework of a Bayesian inference methodology to obtain more informed posterior parametric uncertainty. The computational expense for using the Bayesian approach to these models can be prohibitive. Therefore, response surface models are generated as computationally inexpensive surrogate models with sufficient accuracy by using the input-output data. Finally, the posterior distributions in all parameters are propagated through the process model to obtain the uncertainty bounds in key operating variables. Our results show how the uncertainty in process models vary depending on the operating conditions. Our study also indicates possible approaches to reducing uncertainty in mass transfer models.