Application of a Neural Network for Predicting the Control Effect on Sweep Events in a Fully Turbulent Channel Flow

The proposed study explores the application of a neural network to predict the control effect on sweep events within the framework of an opposition control strategy. Experiments were conducted in a fully turbulent channel flow at a friction Reynolds number equal to 365. A dataset was created by employing different control parameters sampled with a Latin hypercube sampling method. A neural network was successfully trained to predict the controlled conditionally averaged sweep event given the actuation parameters and the uncontrolled conditionally averaged sweep event. Results show that the model is able to reliably predict the control signature across different actuation parameters. The predicted model was applied to perform an offline optimization of the actuation parameters of the control strategy to find the best actuation parameters that can reduce the sweep event intensity. Moreover, a sensitivity analysis by using the Jacobian matrix was also carried out to understand which parameters influence the predicted response the most.