COMPLETE CHARACTERIZATION OF AXISYMMETRIC TURBULENT JET USING BACKGROUND ORIENTED SCHLIEREN AND PHYSICS-INFORMED NEURAL NETWORK

Axisymmetric turbulent jet of hot air is completely reconstructed from the experimentally measured temperature field using physics-informed neural network (PINN), which takes into account both the experimental data and the governing equations. The proposed data assimilation technique allows determination of the velocity and turbulent viscosity fields without usage of specific turbulence model equations. The input experimental data are obtained using nonintrusive background oriented schlieren (BOS) measurements. The accuracy of the flow reconstruction is assessed for synthetic data and two different experimental setups. The data assimilation results are shown to be in good agreement with the conventional Reynolds-averaged Navier-Stokes (RANS) simulations using the Spalart-Allmaras (SA) and k-ε turbulence models. The following advantages of PINN data assimilation are demonstrated: it does not require regularization of the equations or smoothing of the experimental data and allows omission of the boundary condition for the inlet turbulence level.