Theses and Dissertations
Advisor
Amirlatifi, Amin
Committee Member
Ramezani Bakhtiari, Somayeh
Committee Member
Heshmati, Mohammad
Committee Member
Elmore, Bill
Date of Degree
12-12-2025
Original embargo terms
Embargo 2 years
Document Type
Dissertation - Open Access
Major
Engineering (Chemical Engineering)
Degree Name
Doctor of Philosophy (Ph.D.)
College
James Worth Bagley College of Engineering
Department
Dave C. Swalm School of Chemical Engineering
Abstract
The efficiency of carbon dioxide (CO2) capture is strongly governed by the physical and chemical characteristics of the adsorbent, which determine its adsorption uptake, selectivity, and regeneration performance. As a performance benchmark, desorption kinetics were experimentally evaluated by thermogravimetric analysis (TGA) for activated carbon and zeolite, revealing desorption activation energies of 141.14 and 28.11 kJ mol−1, and CO2 uptakes of 5.79 and 2.90 mg g−1, respectively, at 0.04% CO2 concentration. These results motivate the exploration of metal–organic frameworks (MOFs) for their tunability in pore structures and surface functionality for CO2 adsorption. In this work, machine learning and molecular simulations were integrated to model and predict carbon dioxide/nitrogen (CO2/N2) selectivity in MOFs. A neural network trained on 400 ideal MOFs from the Topologically Based Crystal Constructor (ToBaCCo) dataset, using categorical descriptors of topology, node and edge building blocks, and functional groups, was developed to predict CO2 selectivity. The model was subsequently applied to an external dataset of 9,146 computation-ready experimental (CoRE) MOFs described by geometric descriptors, yielding an ��2 score of 0.999 for predicted selectivity. SHAP analysis identified chemical functionalization and network connectivity as the influential contributors to adsorption behavior. Monte Carlo perturbation and Bayesian calibration produced mean selectivity values of 6.25 and 6.33, confirming the model’s robustness, stability and interpretability. GCMC simulations under flue-gas conditions demonstrated that charged Cu-BTC and Zr-DMBD MOFs achieved CO2 uptakes of 1.499 and 2.985 mmol g−1, respectively, outperforming their neutral counterparts. Molecular dynamics (MD) simulations further demonstrated that the charged frameworks possessed enhanced electrostatic cohesion and structural stability, indicating that framework polarization strengthens CO2 binding without compromising rigidity. Together, these findings establish a unified computational framework for predicting adsorption behavior and guiding the design of next-generation MOFs for both Direct Air Capture (DAC) and post-combustion CO2 separation.
Recommended Citation
Do, Maria Thuy, "Optimization of metal organic frameworks design for CO2 capture through machine learning and molecular dynamic simulation" (2025). Theses and Dissertations. 6808.
https://scholarsjunction.msstate.edu/td/6808
