Dr. Xiang Yang is an Assistant Professor in the Mechanical Engineering Department at the Pennsylvania State University. He received his Ph.D. in Mechanical Engineering from Johns Hopkins University in 2016. After that, Yang joined the Center for Turbulence Research in 2016 as a Postdoctoral Research Fellow. He became a faculty member in the Mechanical Engineering Department at Penn State in 2018 and has been there since then. His group conducts high-fidelity numerical simulations, builds physics- and data-based models, and finds efficient solutions for real-world engineering problems. His group uses tools including direct numerical simulation, large-eddy simulation, Reynolds-averaged Navier Stokes, and machine learning.
报告简介
Abstract This talk will explore the possibility of data-enabled turbulence models for general purposes. We require that the benefits offered by the developed machine learning model in one flow not be at the expense of its performance in other flows. As such, a CFD practitioner can pick up the model, as it is, and use it for predictive modeling without worrying about detrimental effects. We do this through progressive machine learning. The modeling framework builds on two theorems, the extrapolation theorem and the neutral neural network theorem. The former theorem allows one to control how a neural network extrapolates, and the latter enables one to progressively improve an existing model to account for more complex physics. We will apply this modeling framework to re-calibrate the Spalart-Allmaras model. We will show that the re-calibrated model offers improvements in separated flows while preserves known empiricisms.
