Modelling of 2D steady flow fields with artificial neural networks and integration of physical knowledge

The application of artificial neural networks (ANN) for the prediction of twodimensional steady flow fields is described. It could be shown that a properly trained ANN is able to predict whether wakes in a bodies outflow form or not, depending on the Reynolds number. As the no-slip condition on the bodies surface was not learned correctly, two possibilities to integrate a priori knowledge were investigated and proved to be suitable.