Development of numerical methods for flow control with the aid of artificial neural networks

In this research, in order to develop a real-time flow control strategy by preventing application of expensive and time-consuming Computational Fluid Dynamics (CFD) simulations, Proper Orthogonal Decomposition (POD) and Artificial Neural Networks (ANN?s) are utilized. Time-dependent two-dimensional laminar (Re=100) and turbulent (Re=20000) fluid flows over a circular cylinder and time-dependent two-dimensional driven cavity flow applications are addressed to develop numerical techniques by observing low-dimensional modeling of the flow fields and to estimate the state of the flow effectively.Proper Orthogonal Decomposition (POD) is a reduced-order modeling technique that enables observation of the chaotic, time-dependent and periodically moving vortices and control of these structures in the flow field. By employing POD, a data ensemble consisting physical flow characteristics within a set of snapshots is separated into its principal components named as modes and relative mode amplitudes according to their energy contents (frequencies).In order to develop a real-time flow control strategy, predictions of mode amplitudes carrying the temporal characteristics of the flow field are essential. The constituted ANN structure predicts mode amplitudes for other test cases where it takes data coming from surface sensors placed on a few discrete points that show the highest activity in terms of static pressure and previously known mode amplitudes of specific test cases.In the application based on time-dependent two-dimensional driven cavity flow, test cases at Reynolds numbers 100, 500, 1000, 5000 and 10000 are investigated numerically by solving two-dimensional Navier-Stokes equations with the vorticity-stream function approach. POD and ANN techniques are also employed for this application to predict temporal characteristics of the flow.