A Finite Element Neural Network Approach for Modeling Particles in Non‐Newtonian Fluids

ABSTRACT We describe a hybrid simulation approach for the modeling of particle‐laden flows taking into account nonlinear effects of non‐Newtonian flows. We focus on the situation of small particles with low concentration, whose influence on the flow can be neglected and which can be modeled as point particles. While the case of spherical particles is in the Newtonian case well understood, the description of the interaction with non‐spherical particles is not straightforward, since the coefficients for the transmission of forces depend essentially on the shape of the particles. In a previous work, we have shown that these coefficients can be trained very efficiently in neural networks based on some detailed simulations. Here we deal with the case of non‐Newtonian flow and in particular address the question to what extent global effects, such as the clustering of particles or particle orientation, can be represented by local predictions based on neural networks.