WING BUFFET PRESSURE LOAD PREDICTION BASED ON A HYBRID DEEP LEARNING MODEL

In the present study, a hybrid deep learning reduced order model (ROM) is applied for unsteady transonic wing buffet load prediction. The hybrid model is defined by the combination of a convolutional variational autoencoder (CNN-VAR-AE) and a long short-term memory (LSTM) neural network. In the first step, the CNN-VAR-AE is trained using experimental buffet data. Thereby, the high-dimensional buffet flow field is reduced into a low-dimensional latent space. In the second step, the LSTM is trained and applied in order to predict the temporal evolution of the wing buffet pressure loads. As a test case, the generic XRF-1 configuration developed by Airbus, is applied. The XRF-1 configuration has been investigated at different transonic buffet conditions in the European Transonic Wind Tunnel (ETW). During the test campaign, surface pressure data has been obtained by means of unsteady pressure sensitive paint (iPSP) measurements. As a first step, the trained model is applied in a recurrent multi-step prediction mode in order to reproduce pressure distribution at flow conditions included in the validation data set. In the second step, the trained model is used for the prediction of pressure distributions at an unknown flow condition. A comparison of the experimental data with data predicted by the deep learning model yields an accurate prediction of the buffet flow characteristics.