Evaluation of an efficient data-driven ANN model to predict agglomerate collisions within Euler-Lagrange simulations

In this study, a recently developed data-driven model for the collision-induced agglomerate breakup (CHERD 195, 2023) is evaluated. It is especially intended for Euler-Lagrange simulations of flows with high mass loadings, where coupled CFD-DEM predictions are too expensive. Therefore, a surrogate model relying on the hard-sphere approach in which agglomerates are represented by effective spheres was developed. Based on a variety of DEM simulations, artificial neural networks were trained to predict the post-collision number of arising fragments, their size distribution and their velocities. In the present contribution, the agglomerate collision model is assessed using the particle-laden flow through a T-junction. Since two fluid streams with agglomerates are injected at both opposite ends, the setup is particularly suitable for investigating breakage caused by collisions. Two flow configurations (laminar flow at Re = 130 and turbulent flow at Re = 8000) and two different powders (primary particle diameter of 0.97 and 5.08 micrometers) are taken into account. The latter allows to study the influence of the strength of the agglomerates on the collision-induced breakage. The laminar case offers the possibility to evaluate the effect of the collision angle in detail. The collision-induced breakage proves to be the most dominant deagglomeration mechanism in both the laminar and turbulent flow scenario. Nevertheless, the role of the fluid stresses and especially the drag stress becomes more prominent in the turbulent case, while in the laminar flow their effects are negligible.

Automated summary

This study evaluates a recently developed data-driven model for collision-induced agglomerate breakup in high mass loading flows. The model uses artificial neural networks to predict the post-collision behavior of agglomerates, reducing computational costs compared to coupled CFD-DEM simulations.