(211d) Thermodynamic Modeling and Uncertainty Quantification of CO2-Loaded Aqueous MEA Solutions | AIChE

(211d) Thermodynamic Modeling and Uncertainty Quantification of CO2-Loaded Aqueous MEA Solutions

Authors 

Soares Chinen, A., West Virginia University
Omell, B., National Energy Technology Laboratory
Bhattacharyya, D., West Virginia University
Tong, C., Lawrence Livermore National Laboratory
Miller, D. C., National Energy Technology Laboratory
The US DOEâ??s Carbon Capture Simulation Initiative (CCSI) has a strong focus on the development of state of the art process models to accelerate the development and commercialization of post-combustion carbon capture system technologies. One of CCSIâ??s goals is the development of a â??gold standardâ? process model that will serve as a definitive reference for benchmarking the performance of MEA-based CO2 capture systems. The focus of this specific work is the development of the thermodynamic framework, an essential element of the physical properties package for a reactive separation system, for the gold standard model. A stochastic model of the thermodynamic framework is also included in this work, in which a Bayesian inference methodology is used to quantify the parametric uncertainty.

 The thermodynamic framework is developed in Aspen Plus®, and the complete e-NRTL activity coefficient model is used as a starting point. The Aspen Plus® physical properties regression system is used to calibrate the parameter values, including data for VLE of binary MEA-H2O and ternary MEA-H2O-CO2 systems as well as heat of absorption and heat capacity of the ternary system. Due to the large number of parameters in the full e-NRTL model, a parameter screening methodology based on the Akaike Information Criterion (AIC) is developed. The use of a reduced model is not only useful in avoiding over parameterization in the deterministic model, but also improves the feasibility of the uncertainty quantification (UQ) procedure. A Bayesian inference methodology is used to develop the stochastic model, in which prior distributions estimated from the deterministic regression output and the experimental data are used to derive informative posterior distributions. To reduce the computational expense of the Bayesian approach, response surface models are generated as computationally inexpensive surrogate models with sufficient accuracy by using the input-output data. The Markov Chain Monte Carlo (MCMC) algorithm with Gibbs Sampling is used in the Bayesian inference procedure.

The thermodynamic framework is implemented into an Aspen Plus® model of the CO2 absorption process, consisting of both absorber and regenerator columns, which are modeled as rate-based columns with kinetic reactions. The reaction kinetics used in this work are written in terms of the activity-based reaction equilibrium constants in order to ensure consistency with the thermodynamic model. Case studies are performed, separately for the absorber and regenerator columns, in which prior and posterior parameter distributions are propagated through the process models. This results in estimates of the uncertainty in output variables, specifically CO2 capture efficiency in the absorber and outlet solvent CO2 loading from the regenerator, due to the thermodynamic model. The study shows that the uncertainty due to the VLE model can vary widely depending on the operating conditions, but does reduce considerably for the posterior distributions in comparison to the priors for all the cases studied. The study also helps to identify how the uncertainty due to the thermodynamic model can be reduced further.