Artificial neural networks (ANNs) can be trained to recognize patterns within data. In this work, we train an ANN to detect cells that have a shock, so that it can be used as a troubled cell indicator with high-order hydrodynamic methods that must either apply limiters to polynomials or activate an artificial viscosity scheme near a shock depending on the type of numerical method. The overarching goal is to enable high-order solutions on smooth flows and to reduce the accuracy towards first-order near shocks. We train an ANN on a dataset consisting of shocks and smooth field variations. The ANN returns a scalar value in the range of zero to one, where the scalar value is very close to zero in a smooth region and is very close to one at a shock. The ANN shock detector is used with a high-order residual distribution (RD) Lagrangian hydrodynamic method to simulate multidimensional shock driven flows. The details on the ANN shock detector are presented along with simulation results from test problems.
展开▼
