Neural Network-Based Prediction of Bursting Events in a Turbulent Channel Flow

This study investigates the use of a hybrid architecture of a Feed-forward Neural Network (FNN) combined with the Proper Orthogonal Decomposition (POD) to estimate the effect of control on sweep events within the context of an opposition control strategy. The experiments were carried out in a fully turbulent channel flow with a friction Reynolds number of 365. The POD was conducted to reduce the database dimensionality, and then a FNN was trained to predict the non-normalized spatial modes matrix, subsequently allowing the reconstruction of the Conditionally Averaged Sweep Event (CASE). The results demonstrate that the hybrid neural network is capable of reproducing the control signature for different actuation parameters with an accuracy comparable to a conventional neural network, while requiring approximately 90 times fewer trainable parameters.