Prediction of Rise Velocity of Taylor Bubbles in Pipes Using an Artificial Neural Network
AbstractMost of the drift velocity models that exist have limitations, as they were developed based on experiments that hardly considered the conjunction of liquid properties and pipe geometry effects. Additionally, some of the models are formed complexly and several parameters require to be optimized. This study focuses on the application of the artificial neural network (ANN) model in predicting the drift velocity of Taylor bubbles rising in stagnant fluids through horizontal and inclined pipes. A comprehensive experimental database including 364 data points from the open literature was applied to develop the ANN model. Inclination angle, pipe diameter, liquid density, liquid viscosity, and surface tension were used as input variables. Finally, a multilayer perceptron (MLP) feed-forward backpropagation neural network having 12 neurons in the hidden layer was selected as the optimized ANN model that showed the best performance for the prediction of drift velocity. The obtained ANN model showed superior performance in comparison with the support vector machine (SVM) model and four drift velocity correlations, with high accuracy for the training data set (MSE=0.000127, R2=0.9985, and MAPE= 5.85%), testing data set (MSE=0.00028, R2=0.9983, and MAPE= 5.45%), and all data (MSE=0.000137, R2=0.9982, and MAPE= 5.69%).