Flow Control Using Neural Networks

Abstract : To improve the computational design procedure, and to prepare for studies on BCU arrays, it is necessary to avoid the long computation times that would be required for tracking the response to real actuator signals by DNS. Since the control units exploit the (linear) wave superposition for wave attenuation, we have developed a highly efficient simulation technique by combining the results of a single DNS run with the Duhamel superposition integral (DSI). We perform the single run for a small ramp motion of a given actuator and record the flow response at sensor locations. From this time series, DSI generates the flow response to arbitrary actuator motion in milliseconds. The flow response to sample signals agrees perfectly with DNS results for these signals. Figure 5 shows a time series for an actuator that performs a ramp motion in the streamwise direction over about 1% of the TS period and remains deflected at the maximum amplitude of 0.02 mm. the flow response is recorded at a hotwire located one TS wavelength downstream of the actuator and 1/4th boundary layer thickness from the wall. The duration of the signal clearly indicates that instantaneous sensor signals contain contributions from actuator signals over some period of time. Accounting for the actuation history is key to successful feedback control.