Computational Fluid Dynamics (CFD) and Machine Learning (ML) modelling of CO2 removal using Rotating Packed Bed (RPB) Technology

Abstract

Computational fluid dynamics (CFD) and Machine-Learning (ML) techniques have been used to build an integrated multiscale model for CO2 capture in Rotating Packed Beds (RPB). The work aims, and for the first time, at developing standalone accurate models for RBPS and this based on multiscale CFD and ML modelling. This innovative model coupling offers a reliable alternative and/or supplement the uncertain semi-empirical models and the often-costly experimentation. First, CFD simulations using recalibrated porous media model have been implemented to predict the flow behaviour and pressure drop characteristics in RPB under dry packing conditions. The validation results showed that the maximum relative error in the predicted pressure drop against experimental measurements can reach up to 30%. After resorting to Machine-learning (ML) techniques to recalibrate the porous media CFD model, the errors on pressure drop predictions have dropped to less than 1%. These more accurate CFD models rely on machine-learned porous media coefficients rather than the well-known fixed Ergun’ coefficients. Second, a single-phase small-scale CFD model was established to simulate a full-scale RPB and resolve the typical flow details within its real packing structure. The gas flow characteristics in the different RPB parts: casing, packing and inner cavity have been analysed, and deeper understanding of gas maldistribution and free vortex flow have been gained. Validation results show a good agreement between small-scale model and experimental results within 9% root mean square error (RMSE). Moreover, a simplified analytical model has been derived to mimic the impact of wire mesh packing on the gas flow behaviour and verify the viscous and inertial resistance models. Third, volume-of-fluid (VOF) multiphase CFD model has been implemented in two-dimensional form (2D) to simulate liquid flow behaviour within the wire mesh packing under different operating conditions (wide combination of liquid flow rate and RPB rotating speed). Fourth, a recalibrated porous media coefficient that relies on multiscale simulations, a machine learning surrogate model, and least squares optimization technique has been proposed. The new coefficients have been implemented in a two-phase CFD simulations based on Eulerian-Eulerian multiphase approach in three-dimensional form (3D) to obtain the 3D-liquid holdup at different operating conditions. The mean droplet diameter and liquid holdup obtained from 2D-VOF model and recalibrated 3D large-scale model, respectively, have been employed together to estimate the gas-liquid interfacial area. Finally, the gas-liquid interfacial area at different operating conditions has been utilized in the mass transfer model in order to mimic CO2 capture process in RPB. The CO2 capture rate is related to the main RPB hydrodynamics parameters and guidelines on design improvement are proposed.

Date of Award	Aug 2023
Original language	American English
Supervisor	Abdallah Berrouk (Supervisor)