APPLICATION OF MACHINE LEARNING FOR PREDICTING PRESSURE DROP IN FLUIDIZED DENSE PHASE PNEUMATIC CONVEYING

Abstract

It is difficult to model the pressure drop that occurs in fluidized dense phase conveying (FDP) of powders because the flow involves several interactions among the solid, gas, and pipe wall. These interactions are challenging to include in a model. Pressure drop is influenced by geometrical, material, and flow properties. When used with different pipeline designs that have different pipeline lengths or diameters, the current models exhibit considerable inaccuracy. The current work explores how machine learning (ML) algorithms can estimate the pressure drop in the FDP conveying of particles. The network was trained using experimental data from pneumatic conveying, and it subsequently used that information to predict pressure drops. For estimating the pressure drop, four distinct ML algorithms-AdaBoost, CatBoost, gradient boosting, and random forest-were selected. AdaBoost, CatBoost, gradient boosting, and random forest models predicted the data of pressure drop with MAE of 20.72, 4.06, 4.68, and 3.0, respectively, for training as well as testing data. The AdaBoost model performed more poorly in predicting the pressure drop than other models considered for the study, with ± 10% error margin while training and evaluating the data and ± 10% error margin in validating the data.