Machine Learning-Based Filtered Drag Model for Cohesive Gas-Particle Flows

Abstract

The accuracy of filtered two-fluid model simulations critically depends on constitutive models for corrections that account for the effects of inhomogeneous structures at the sub-grid level. The complexity of accounting these structures increases with cohesion. In the present study, a dataset from filtered Euler-Lagrange simulations with systematic variations of the cohesion level and the filter length was created to investigate the development of a machine learning-based drag correction model for liquid bridge-induced cohesive gas-particle flows. A-priori tests revealed that these models afford robust and accurate predictions of the drag correction and the actual drag force. Further it was demonstrated that an anisotropic drag correction model is more accurate than an isotropic model.