Prediction of Gas-Phase Adsorption Isotherms Using Neural Nets

1 For virtually any physical adsorption process, the capacity of an adsorbent decreases as the temperature of the system increases. As the temperature increases the adsorbed molecules acquire sufficient energy to overcome the van der Waals’ attraction, holding them in the condensed-phase and migrating back to the gas-phase. Adsorption is an exothermic process. At low concentrations the heat release is minimal and is quickly dissipated by the airflow through the bed. At high concentrations considerable heating of the bed can occur, which if not removed can cause the adsorbent efficiency to decrease rapidly. In addition, in the case of granular activated carbon (GAC) beds, heat accumulation in the interior of the bed can even cause auto-ignition of the carbon bed. Thus, it becomes essential to monitor the bed temperature and to understand its relationship to the adsorption capacity. A large number of adsorption isotherm models presented in the literature are generally able to describe the relationship between the adsorbate concentrations in the fluid and adsorbed phases at a given temperature. This necessitates an experimental study at each anticipated temperature of the adsorption application. A few adsorption models have temperature as a variable. This paper describes some of these models and a newly proposed methodology for the prediction of gas-phase adsorption of isotherms at different temperatures using artificial neural networks (ANNs).