Accelerating Materials Discovery For Gas Capture And Separation Applications Combining High Throughput Simulation And Machine Learning

Machine learning (ML) methods have become increasingly popular across various domains. Particularly, scientific research appears to be adopting this technology rapidly. While topics such as medical sciences and climate change garner attention when ML is applied, its use in material science cannot be overlooked. Metal-organic frameworks (MOFs) stand out in this field as their reticular nature makes them suitable candidates for applying ML methods. This thesis focuses on using ML methods on MOFs for gas adsorption and separation purposes. Background is provided on adsorption mechanics and the broader porous materials domain in the introduction section. The state of the art is explored in the literature review and the methods through which ML is applied in this project is discussed in the methodology section. ML was first applied to study the O2/N2 selectivity properties of MOFs in Chapter 4 of the thesis. The model’s predictions displayed good agreement with the values obtained through simulations; design strategies were derived through trends in the data. The application of ML on predicting CO2 adsorption was then studied. As low partial-pressure conditions were considered, electrostatic interactions were the dominant force leading to adsorption. The commonly used descriptors did not capture information related to electrostatic interactions; to improve the models, descriptors that quantify the effects of framework charges were introduced. These were shown to improve the model and were orders of magnitude faster to obtain than other descriptors in the dataset. Finally, a classification model was successfully built to predict ammonia working capacity in temperature-swing conditions. This chapter also revised the charge-based descriptors from the previous chapter to make it more robust.<p></p>