Deep learning methods for predicting fluid forces in dense particle suspensions

Two deep learning methods, Multi-Layer Perceptron (MLP) network and Convolution Neural Network (CNN) are evaluated to predict drag forces in dense suspensions of ellipsoidal particles using data from Particle Resolved Simulations (PRS). The MLP is trained on the mean flow Reynolds number, solid fraction suspension, aspect ratio particle, orientation direction. CNN given an additional 3D spatial map particle interest its immediate neighborhood via a distance function. prediction capability networks tested at different levels complexity: unseen arrangement (Level 1), all arrangements numerical experiment 2), finally experiments or 3). shown perform better than for testing except when ratio.