In designing a gas-liquid bubble column, an accurate estimation of the gas hold-up is important. A generalized Machine learning-based data-driven methodology is presented to predict the gas hold-up, with the industrial-scale application for designing a bubble column. The predictions by machine learning methods such as support vector regression (SVR), random forest (RF), extra trees (EXT), and artificial neural network (ANN) have been compared. The methodology adopted for the prediction of gas hold-up with the help of independent parameters such as column diameter, column height, sparger design, sparger location, percentage free area, superficial gas, liquid velocity, pressure, temperature, the density of gas & liquid, viscosity of gas & liquid, surface tension has been presented. An extensive set of experimental data (4042 data points) has been extracted from the literature covering various design and operating parameters. The performance of the machine learning methods has been compared using mean absolute percentage error (MAPE), mean square error (MSE), and determination coefficient/prediction accuracy (R-square). Based on these statistical parameters, with 97% of prediction accuracy (R-square), MSE = 0.00031, and MAPE = 7.9, the performance of the extra tree is found to be most suitable for the prediction of the gas hold-up.
Read full abstract