Application of machine learning to viscoplastic flow modeling

Yield stress fluid flows play an important role in the oil and gas industry. There are many numerical methods for modeling such flows, and they often are computationally expensive. We propose to reduce the computational complexity of parametric studies of Bingham fluid flows by utilizing machine learning techniques. The idea is as follows: instead of solving the PDE for each parameter value, we first do several simulations for a few scenarios (build a training dataset), construct a surrogate model to predict the solution for any parameter value. In this case, we have a much faster model. We apply this approach to a well-known Mosolov problem with Bingham number as a parameter.