Development and Validation of a Machine Learned Turbulence Model

Ian D Dettwiller,Shanti Bhushan,Wesley Brewer,Greg W Burgreen

doi:10.3390/en14051465

Abstract

A stand-alone machine learned turbulence model is developed and applied for the solution of steady and unsteady boundary layer equations, and issues and constraints associated with the model are investigated. The results demonstrate that an accurately trained machine learned model can provide grid convergent, smooth solutions, work in extrapolation mode, and converge to a correct solution from ill-posed flow conditions. The accuracy of the machine learned response surface depends on the choice of flow variables, and training approach to minimize the overlap in the datasets. For the former, grouping flow variables into a problem relevant parameter for input features is desirable. For the latter, incorporation of physics-based constraints during training is helpful. Data clustering is also identified to be a useful tool as it avoids skewness of the model towards a dominant flow feature.

Highlights

Engineering applications encounter complex flow regimes involving turbulent and transition flows, which encompass a wide range of length and time scales that increase with the Reynolds number (Re)
Direct Numerical Simulations (DNS) require grid resolutions small enough to resolve the entire range of turbulent scales and are beyond the current computational capability
Simulations were started from an initial velocity profile obtained from one-equation Reynolds averaged Navier Stokes (RANS) model

Summary

Introduction

Engineering applications encounter complex flow regimes involving turbulent and transition (from laminar to turbulent) flows, which encompass a wide range of length and time scales that increase with the Reynolds number (Re). Direct Numerical Simulations (DNS) require grid resolutions small enough to resolve the entire range of turbulent scales and are beyond the current computational capability. Availability of high-fidelity DNS and experimental datasets is fueling the emergence of machine learning tools to improve accuracy, convergence and speed-up of turbulent flow predictions [1,2]. Machine learning tools depend on neural networks to identify the correlation between input and output features, and have been used in different ways for turbulent flow predictions, such as direct field estimation, estimation of turbulence modeling uncertainty, or advance turbulence modeling. In the direct field estimation approach, the entire flow field is predicted using a ML approach, i.e., the flow field is the desired output feature. Milano and Koumoutsakos [3] estimated mean flow profile in the turbulent buffer-layer region using

Objectives

Results

Conclusion