Estimation of flow patterns by applying artificial neural networks

It wasn´t until this decade that attempts were initiated to apply artificial neural networks (ANN) to problems in computational fluid dynamics (CFD). The purpose of the study is to propose a new approach for flow prediction using feedforward neural networks and fluid dynamics knowledge. A representative hydraulic flow problem, the two-dimensional Karman vortex street around a static prism with an elongated rectangular cross section, is examined. Several precalculated flow solutions with different Reynolds numbers and phases of vortex generation are used to train the neural networks. As a result, flow patterns of new Reynolds numbers and phases are obtained. The study reveals the potential of using artificial neural networks for estimating flow patterns without carrying out complicated and time-consuming CFD simulation. The computing time of this ANN approach is greatly reduced compared with CFD simulation. Furthermore, the estimation accuracy is also very encouraging