Application Of Optimal Regularization Neural Network On Comparison Of Two Solvents For Absorption Of CO2 From Air In A Packed Column

The performances of Di-Ethanol Amine (DEA) and Methyl Di-Ethanol Amine (MDEA) solutions (both 40 wt%) for separation of carbon dioxide from air in a packed column is compared using an optimally trained regularization network (RN). The experimental data are collected from a pilot scale packed absorption column (11 cm diameter) filled with 20.9 m ceramic Raschig rings of ¼ inch size. The independent input variables are selected to be : concentration of carbon dioxide in inlet air, entering air flow rate and solvents (DEA and MDEA solutions) flow rate. To ensure that the RN performs adequately, both recall and generalization performances of a 3D test function are successfully computed by the regularization network using a bunch of noisy training data set. The RN armed with leave one out cross validation (LOOCV) technique performed very adequately by filtering the noise and extracting the true underlying hyper-surface embedded in the noisy outputs. Afterwards, the absorption performances of both solvents are investigated at constant temperature and pressure by resorting to trend analysis of the absorption percentage in various concentrations of carbon dioxide, different flow rates of entering air and solvent solutions. The four dimensional hypersurface constructed via the optimal RN predicted that the maximum carbon dioxide absorption percentages are approximately 45 and 100 percent for MDEA and DEA solutions, respectively

Keywords:

Absorption , Simulation , Packed column , Trend analysis , Regularization network