Optimization and prediction of oxygen purity in pressure swing adsorption systems using zeolite adsorbents and machine learning models

This research investigates the optimization of oxygen purity in Pressure Swing Adsorption (PSA) systems using four different zeolite materials—3A, 4A, 5A, and 13X—for air separation applications. The experimental setup involved packing zeolite samples into PSA adsorption beds and systematically optimizing cycle parameters such as pressure, flow rate, and regeneration temperature. Ambient air containing 21% oxygen was used as the feed gas, and the system operated at a column pressure of 2 bars. Zeolite 13X consistently exhibited the highest oxygen purity, reaching up to 95.9% over multiple cycles, followed by 5A, 4A, and 3A. To complement the experimental work, a machine learning regression framework was applied using 10 models to predict oxygen purity based on key process variables. Linear Regression emerged as the most accurate model, achieving an R2 score of 0.9919. Feature importance analysis revealed that adsorption capacity, flow rate, and pore size were the most influential factors in determining oxygen purity. These findings highlight the critical role of material selection and data-driven modeling in enhancing PSA system performance, with direct implications for medical, industrial, and environmental applications requiring high-purity oxygen streams.