4:10 pm Presentation
“A Vortex Sheet as an Interface Retaining Coarsed Model for Multiphase Flow”
Presented by XIANYANG CHEN (Adviser: Prof. Tryggvason)
Multiphase flows are characterized by sharp moving phase boundaries, separating different fluids or phases. In many cases the dynamics of the interfaces determines the behavior of the flow. In a coarse, or reduced order model, either an averaged two-fluid model or a large-eddy-simulation like one, it is therefore critical to retain a sharp interface for the resolved scales. To develop a more general strategy we are experimenting with a particularly simple approach, modeling the interface for a weakly stratified flow using an inviscid vortex sheet with baroclinically generated vorticity, advected using the Biot-Savart law, and adjustable model parameters to match the results from a full resolved simulation. We start by filtering the fully resolved results, using a filter that we call weighted coordinates smoothing (WCS), that simplifies the interface and concentrates the vorticity at the interface. The model parameters (“vortex blob’’ sizes) are found by optimization, using a cost function that compares the velocity of the filtered interface with the model prediction. Once the model parameters have been found, we plan to use machine learning to develop a strategy to evolve them in time. We discuss the overall work plan, the filtering strategy, and show preliminary results.
4:35 pm Presentation
“Optimal Prediction of Wall Shear Stress from Filtered Velocity Datas”
Presented by MENGZE WANG (Adviser: Prof. Zaki)
Resolving near-wall turbulence is difficult in simulations of high-Reynolds-number flows. To mitigate the near-wall resolution requirements and reduce computational cost, wall models are indispensable in practical numerical solvers such as large-eddy simulations (LES). Conventional wall models adopt instantaneous LES velocity as their input and predict a corresponding instantaneous wall shear-stress which is fed back to LES as a boundary condition. Despite the numerous existing wall models, a fundamental question remains unaddressed: given instantaneous filtered off-wall velocity data, what is the highest possible accuracy of predicted wall shear stress? This problem is investigated using an adjoint variational data assimilation approach. By combining a full Navier-Stokes solver with off-wall filtered-velocity data, we reconstruct the initial condition that tracks the full flow state and optimally predicts the spatio-temporal evolution of the wall shear stress. We demonstrate that the optimal prediction is robust to the filter width. As the height of the first available velocity grid point increases, the correlation between the true and optimally predicted wall stresses deteriorates, but remains at least twice as large as that of the true and equilibrium wall-model stresses. Finally, the Reynolds number effect on the optimal prediction is explored.