Physics-informed neural network for modeling force and torque fluctuations in a random array of bidisperse spheres
Published in International Journal of Multiphase Flow, 2023
An interpretable shared-block neural network (NN) was established using TensorFlow in Python to predict the force and torque fluctuations in any random arrays of spheres with different sizes. The $R^2$ score of this NN can reach up to 0.91, which is better than any existing NN that can only predict the cases of spheres with the same size. In the meantime, the number of model parameters in the present model is only around 20% in comparison with that of a fully-connected NN introduced by other researchers, which is a result of the delicate design of my NN architecture based on the prior knowledge of physics, symmetries and other constraints.
Recommended citation: Zihao Cheng and Anthony Wachs. Physics-informed neural network for modeling force and torque fluctuations in a random array of bidisperse spheres. International Journal of Multiphase Flow, 169:104603, 2023. https://www.sciencedirect.com/science/article/pii/S0301932223002239?via%3Dihub