With the continuous emergence of new materials, efficiently screening suitable materials for hydrogen purification by pressure swing adsorption (PSA) technology has become a hot topic. Metal-organic frameworks (MOFs) hold significant promise in gas adsorption and separation due to their unique structure and versatility. Breakthrough curves depict the dynamic adsorption kinetics of multi-components in fixed-bed columns, offering a means to screen adsorbents. In this study, a deep neural network (DNN) model comprising 11 hidden layers is developed to predict the dynamic breakthrough behavior of MOF CuBTC and MIL-125(Ti)_NH2 adsorbents. The training data for the DNN were generated from a non-isothermal model established using Aspen Adsorption software. Results demonstrate that the DNN model effectively predicts the H2/CO2 breakthrough behavior on CuBTC and MIL-125(Ti)_NH2 adsorbents. Compared with empirical mathematical correlation models, the DNN can predict breakthrough profiles under varying operating conditions, exhibiting superior adaptability. Furthermore, parametric studies based on the DNN model reveal that initial temperature, adsorption pressure and feed flow rate play important roles in the dynamic adsorption process, significantly influencing the position and shape of the breakthrough curve. Decreasing the feed flow rate and initial temperature while increasing adsorption pressure can enhance breakthrough times and hydrogen purity. By comparing the breakthrough curves of the two adsorbents under different operating conditions for H2/CO2, the dimensionless breakthrough time of CO2 on the CuBTC adsorbent was longer. Additionally, simulation results of the PSA cycle suggest that CuBTC yields higher hydrogen purity than MIL-125(Ti)_NH2. Consequently, CuBTC is deemed more suitable for hydrogen purification of binary H2/CO2 mixtures than MIL-125(Ti)_NH2.
Read full abstract