A Deep Neural Network-Based Approach for the Force Predictions of Particulate Multiphase Flows

The hydrodynamics forces within random distributions of fixed monodisperse spherical particles are predicted using deep neural network-based technique. The test data are obtained from particle resolved-direct numerical simulations. The fully resolved simulations are based on direct forcing immersed boundary method. The methodology is examined for wide ranges of Reynolds number and mean volume fraction. A data augmentation strategy is implemented to achieve 3D rotation-invariant trained network and to improve the accuracy of the predictions. To evaluate the performance of the model, direct force predictions are compared with the corresponding fully resolved solutions.