Volumetric mass-transfer coefficients (k{sub}La{sub}W, K{sub}La{sub}W, k{sub}Ga{sub}W, K{sub}Ga{sub}W) required for randomly dumped packed tower design were gathered from the literature to generate a working database comprehending 2675 measurements relevant to water and air pollution abatement processes. The cross-examination of two important correlations predicting mass-transfer coefficients was achieved through this database (Onda correlation, 1968; Billet and Schultes correlation, 1993). Some limitations regarding either the level of accuracy or the application range came to light with this investigation. Artificial neural network (ANN) modeling is then proposed allowing all four mass-transfer coefficients predictions. A single ANN correlation was built to predict the dimensionless gas (or liquid) film Sherwood number (Sh{sub}(L/G)) as a function of six dimensionless groups, namely, the liquid Reynolds (Re{sub}L), Froude (Fr{sub}L), Eotvos (Eo{sub}L) numbers, the gas (or liquid) Schmidt number(Sc{sub}(L/G)),the Lockhart-Martinelli parameter (x), and a bed-characterizing number (K). Using the ANN correlation and the two-film theory, a reconciliation procedure was further implemented resulting in better predictions of the gas (or liquid) overall volumetric mass-transfer coefficients. The resulting correlation yielded an absolute average relative error of 22.1 and a standard deviation of 21.1 based on whole database while the ANN predictions remain in accordance with the physical evidence reported in the literature.
展开▼
