Publications Details

Data-driven adaptive physics modeling for turbulence simulations

Ling, Julia L.; Kurzawski, Andrew

For many aerospace applications, there exists significant demand for more accurate tur- bulence models. Data-driven machine learning algorithms have the capability to accurately predict when Reynolds Averaged Navier Stokes (RANS) models will have increased model form uncertainty due to the breakdown of underlying model assumptions. These machine learning models can be used to adaptively trigger relevant model corrections in the regions they are needed. This paper presents a framework for data-driven adaptive physics model- ing that leverages known RANS model corrections and proven machine learning methods. This adaptive physics modeling framework is evaluated for two case studies: fully developed turbulent square duct flow and flow over a wavy wall. It is demonstrated that implement- ing model corrections zonally based on machine learning classification of where underlying RANS model assumptions are violated can achieve the same accuracy as implementing those corrections globally.