Initial studies of predicting flow fields with an ANN hybrid

Principal investigations of the application of a hybrid, consisting of an artificial neural network (ANN) and a conventional numerical method to predict an exemplary flow field are presented. Topics of the work were to show principle possibility of using ANN in fluid mechanics and to show the potential of integrating physical a priori knowledge into the training procedure. The flow fields for training and evaluation were generated by a numerical algorithm. Major result was that prediction of the flow field, including the existence of vortices in the bodies outflow at higher Reynolds numbers can be realized in much shorter times than necessary for numerical calculation. The results were obtained with training data, which represented totally different relations in their dynamical behavior, depending on the geometric location. Furthermore, physical a priori knowledge was included in the learning process with an obvious improvement of the hybrid models performance.