Application of Machine Learning Methods in Estimating the Oxygenation Performance of Various Configurations of Plunging Hollow Jet Aerators

Plunging jet aerators are considered energetically attractive devices for oxygenation because of their good mixing characteristics and ease of construction and operation. In this mechanism of plunging jet aeration, the air-water interfacial area is increased by a free-falling jet impinging on the surface of a water pool. In this study, experimental data from various configurations of plunging hollow jet aerators are explored in formulating the correlations for predicting the values of volumetric oxygen transfer coefficient (KLa) with the jet variables (discharge, jet thickness, jet velocity, jet length, depth of water pool, pipe outlet diameter, number of jets). Nonlinear regression modeling equations derived from dimensional and nondimensional data sets are compared with the neuro-fuzzy (ANFIS), support vector regression (SVM), artificial neural network (ANN), M5 tree (M5), and random forests (RF) methods. SVM models calibrated with both types of data sets provided better results when tested on the unseen data sets. Regression equations are also useful and give acceptable results. The SVM models and regression equations are further checked for effectiveness on the data set of past study on plunging hollow jets. The nondimensional form of the regression equation derived in the current study fits reasonably well when tested on the oxygenation data from previous work as compared to the other regression models. The sensitivity of the jet variables is also tested, which showed jet velocity and jet thickness as major contributing factors in oxygenating the aeration pools.