Comparison of rate-based and machine learning modeling for CO₂ capture in a rotating packed bed absorber

• Mathematical model to predict CO 2 absorption efficiency in RPB utilizing MDEA+PZ. • Employed ML approach – DKL, RELM for CO 2 absorption prediction. • Performed SHAP analysis to identify input features influencing CO 2 absorption. • Conducted multi-objective optimization to maximize CO 2 absorption and minimize energy consumption. The increasing concern regarding Carbon dioxide (CO₂) emissions from fuel combustion and industrial exhaust underscores the necessity for efficient CO₂ capture techniques to attain the net-zero ambition for hard-to-abate industries. One of the sustainable methods for reducing CO₂ emissions from such industries remains absorption-based post-combustion CO₂ capture (PCC). In this process, amine solvents are used in conjunction with packed columns and Rotating Packed Beds (RPBs). The RPB absorber is a powerful alternative to conventional packed columns due to its compact design and enhanced mass transfer efficiency. Increasing the effectiveness of CO₂ removal necessitates a comprehensive solvent development. The present study focuses on a blended amine solvent comprising piperazine (PZ) and methyldiethanolamine (MDEA) to investigate CO₂ absorption. The study incorporates both mathematical modelling and machine learning (ML) to predict CO₂ absorption efficiency. Additionally, this study aims to optimize the energy requirements. Experimental data from the literature is used to assess the prediction accuracy of mathematical models and two ML techniques, namely Deep Kernel Learning with Gaussian Process Regression (DKL-GP) and Regularized Extreme Learning Machine (RELM). Statistical analysis is employed to enhance the ML models. Predictive performance is observed to be suitable for both models, although the DKL-GP accuracy is greater, with R 2 , MAE, and RMSE values of 0.99, 0.006, and 0.006, respectively. Furthermore, the prediction results from the mathematical model and ML models are also in close agreement with each other. The results highlight the importance of integrating ML with conventional modelling to inform future design of increased carbon capture processes.