كسري كريمي دخراباد

The present study was conducted with the aim of rigorously eva‎luating the performance of the zeolitic adsorbent K-CHA at both nano an‎d micron scales for CO_2capture, as well as designing an optimal operational cycle based on statistical an‎d intelligent analyses. The primary novelty of this research lies in the simultaneous integration of dynamic simulation, statistical design of experiments (DoE), an‎d machine learning to provide a comprehensive framework for the prediction an‎d optimization of the adsorption process. In the first step, a dynamic model of the adsorption process was developed using Aspen Adsorption software. The rigorous validation of the model, conducted by comparing the simulated breakthrough curves with experimental data at 20, 40, an‎d 60 °C, demonstrated an average error of less than 5%. This high precision confirmed the model’s capability to faithfully represent the actual mass an‎d heat transfer behaviors within the bed. Analysis of the temperature an‎d loading profiles also revealed that the nanostructured adsorbent, by forming a narrower mass transfer zone an‎d generating a breakthrough curve with a steeper slope, exhibited considerably superior performance compared to its micron-sized counterpart; a result that clearly highlights the importance of particle size engineering. In the second step, a design of experiments based on the Response Surface Methodology (RSM) was conducted using five key variables: temperature, pressure, CO_2mole fraction in the feed, axial dispersion coefficient, an‎d adsorbent type. Statistical analysis identified a second-order polynomial model with a predicted coefficient of determination (R^2) of 0.9822 as the best-fitted model. A significant novelty of this section was the structured application of DoE to establish a quantitative framework for designing the set of dynamic simulations. Subsequently, three machine learning algorithms, including Radial Basis Function (RBF), Support Vector Regression (SVR), an‎d Gaussian Process Regression (GPR), were developed. Performance eva‎luation of these models demonstrated that the GPR model yielded the highest prediction accuracy, with an R^2value of 0.9875 an‎d a Root Mean Square Error (RMSE) of 0.0708. The main innovation here was the combinatorial use of machine learning an‎d dynamic modeling to meticulously analyze the impact of operational parameters an‎d provide a reliable decision-making tool. Another fundamental novelty of this research is the execution of a dual sensitivity analysis, performed for the first time on such a system. The results highlighted the decisive role of temperature as the dominant factor determining the overall system capacity, while pressure acts as the primary driving force for instantaneous process fluctuations. This finding provided the scientific basis for designing a novel an‎d combined Temperature-Vacuum Swing Adsorption cycle. Simulation of this cycle demonstrated that the integrated application of heating an‎d vacuum leads to a deeper regeneration of the adsorbent an‎d preserves its adsorption capacity over successive cycles. The final performance eva‎luation of the hybrid cycle yielded a CO_2product purity of 70.70% an‎d a recovery of 88.50%, representing a significant improvement over single-mode adsorption cycles an‎d validating the practical efficacy of the proposed method. Collectively, these results indicate that the present study, by offering an integrated approach encompassing precise modeling, statistical analysis, machine learning, an‎d hybrid cycle design, provides a new perspective for enhancing gas adsorption processes an‎d developing advanced adsorbents.

شبيه‌سازي ديناميك , الگوريتم‌هاي يادگيري ماشين , تحليل حساسيت , كربن دي‌اكسيد , جذب نوسان دما-خلأ

Dynamic Simulation , Machine Learning Algorithms , Sensitivity Analysis , CO2 , Temperature-Vacuum Swing Adsorption

Kasra Karimi Dakhrabad

Dr. Ahad Ghaemi