Neural Network Prediction of Air Stripping KLa

Abstract

Design of air stripping packed towers used to remove volatile organic compounds requires an estimate of the overall mass transfer coefficient $\left(K_{L}a\right)\text{,}$ which is frequently obtained via physically based parametric correlations. Parametric correlations have some shortcomings and produce predictions with relatively large deviations for full-scale, modern application of air stripping towers. In this study, neural network (NN) technology, a powerful new nonparametric approach, is used to analyze mass transfer characteristics in air stripping towers and to simulate $K_{L}a\text{.}$ A large database that is representative of current applications of air stripping towers was assembled for this analysis. The $K_{L}a$ predictions by neural networks were superior to both the Onda model [Onda, K., Takeuchi, H., and Okumoto, Y. (1968). “Mass transfer between gas and liquid phases in packed columns.” J. Chem. Eng. Jpn., 1, 56–62.] and an improved Onda model [Djebber, Y. and Narbaitz, R. M. (unpublished)], the best existing parametric models for air stripping applications. The average absolute error for the validation, as well as for the development data, were found to be less than 19%. The NN model was able to simulate the sudden increase in $K_{L}a$ at high gas loading rates. Also, it simulated more realistically the effect of the packing depth and liquid flow.