Deep Learning for Drag Force Modelling in Dilute, Poly-Dispersed Particle-Laden Flows with Irregular-Shaped Particles

This study employs machine learning-based approaches to develop a drag force model for particles of any irregular-shapes in incompressible gas-solid flows. The irregularity of the particle shape is described using the spherical harmonic model.
The gas-solid fluidization behavior represented bysparse, poly-dispersed particle-laden flows at low to intermediate particle Reynolds numbers is describedthrough in-house particle-resolved direct numerical simulations (PR-DNS). We utilize the PR-DNS to obtain drag force coefficients and flow fields of singleparticles. A variational auto-encoder model is appliedto obtain latent vectors to represent the geometrical features of the particles, with artificial neural networks (ANN) developed to predict drag force coefficients and flow fields of a single particle system. This study then applies a pairwise interaction extended point-particle (PIEP) model to obtain the drag coefficients of a single particle in multi-particle systems by assumingthe flow fields of individual neighboring particles can be linearly superposed over those of the single particle in consideration. The PIEP and ANN results show moderate correlation and accuracy based on the PR-DNS results. With the PIEP and ANN methods, this study provides a steady drag force model that does not require heavy data collection for the gas-solid fluidization systems with irregular-shaped particles.