Integration of High-Fidelity CO2 Sorbent Models at the Process Scale Using Dynamic Discrepancy

A high-fidelity model of a mesoporous silica supported, polyethylenimine (PEI)-impregnated solid sorbent for carbon capture has been incorporated into a model of a bubbling fluidized bed adsorber using Dynamic Discrepancy Reduced Modeling (DDRM). The sorbent model includes a detailed treatment of transport and amine-CO₂-H₂O interactions. Using a Bayesian approach, we calibrate the sorbent model to the Thermogravimetric (TGA) data. Discrepancy functions are included within the diffusion coefficients for diffusive species within the PEI bulk, enabling a 100-fold reduction in model order. The discrepancy functions are based on a Gaussian process in the Bayesian Smoothing Splines ANOVA framework, which provides a convenient parametric form for calibration and upscaling. The dynamic discrepancy method for scale-bridging produces probabilistic predictions at larger scales, quantifying uncertainty due to model reduction and the extrapolation inherent in model upscaling.  The dynamic discrepancy method is demonstrated using TGA data for a PEI-based sorbent and an Aspen Custom Modeler-based model of a bubbling fluidized bed adsorber.