Neural-network-based filtered drag model for gas-particle flows

Abstract

Filtered two-fluid model (fTFM) for gas-particle flows require closures for the sub-filter scale corrections to interphase drag force and stresses, the former being more significant. In this study, we have formulated a neural-network-based model to predict the sub-grid drift velocity, which is then used to estimate the drag correction. As a part of the neural network model development effort, we derived a transport equation for drift velocity and then performed a budget analysis to conclude that an algebraic model for drift velocity in terms of the filtered variables that are resolved in a fTFM simulation is adequate, and the model should include the filtered gas-phase pressure gradient as a marker in addition to the filtered particle volume fraction and the filtered gas-solid slip velocity. Both a priori and a posteriori analyses reveal that the present model for drift velocity when used in a fTFM simulation is able to capture the fine-grid simulation results quite well.