A NEURAL-NETWORK-BASED SYNTHETIC BOUNDARY CONDITION FOR THE SIMULATION OF CHANNEL TURBULENCE
Résumé
We describe a method combining Proper Orthogonal Decomposition (POD) and an artificial recurrent neural network to implement an inlet boundary condition for a turbulent channel flow. The boundary condition is reconstructed from POD modes, the amplitudes of which are predicted using a neural network (LSTM). We first assess the quality of the LSTM prediction in POD space. The reconstructed velocity is then used as an inlet boundary condition in a spatially developping channel. Statistics are compared with the reference and are found to be in relatively good agreement.
Domaines
Sciences de l'ingénieur [physics]
Origine : Fichiers produits par l'(les) auteur(s)