Predictive dynamic model of a small nonisothermal pressure swing air separation process

A predictive dynamic model of a small pressure swing adsorption air separation process was developed for the purposes of design, optimization, evaluation and control of On-Board Oxygen Generation Systems (OBOGS) for military aircraft. A mathematical model of the adsorption beds was formulated by application of fundamental mass and energy transport modeling techniques. These equations were discretized using the Galerkin finite element technique. The resulting ODE systems were coupled with ODE's describing rate of change of pressure in each bed and models of the feed and exhaust valves and purge orifice. The model was developed so that it is possible to predict the dynamic response of product oxygen composition and feed air consumption to step-changes in feed pressure, temperature of the feed and surroundings, product flow rate and cycle time.; All parameters in the model were estimated from literature sources or from literature correlations, with two exceptions. Heat transfer coefficients were estimated from single-bed steady-state temperature profiles. Estimates of the feed/exhaust valve and purge orifice discharge coefficients were determined from pressure drop data.; Several simplifications to the model were necessary to obtain reasonable predictions of two-bed experimental data. These simplifications were to use a constant axial diffusion coefficient, constant mass transfer coefficients, and a constant isosteric adsorption heat.; A laboratory PSA unit similar in size to an OBOGS was constructed to validate the model. The laboratory unit was constructed so that step-changes could be implemented and the responses observed for comparison with the model.; Reasonable predictions of bed pressure, cycle-averaged feed flow rate, and cycle-averaged bed temperature vs. time in response to step-changes in all input variables compared to two-bed PSA data were achieved without additional parameter estimation from two-bed data. Predictions of the cycle-averaged product oxygen composition vs. time were consistently low with the initial parameter values compared with two-bed data. However, most predictions were easily corrected for the initial steady-state by modification of the purge orifice coefficient. Subsequent predictions of dynamic response from the initial cyclic steady-state agreed well with the two-bed data.