A machine learning based wall model for LES of turbulent flows
Document typeConference report
PublisherBarcelona Supercomputing Center
Rights accessOpen Access
All rights reserved. This work is protected by the corresponding intellectual and industrial property rights. Without prejudice to any existing legal exemptions, reproduction, distribution, public communication or transformation of this work are prohibited without permission of the copyright holder
The trubulent flow of fluids is still an enigma for mathematicians and engineers alike. The partial differential equation that reperesents the flow of fluids does not have an analytical solution for turbulent regimes. The closest approximation for a solution can be arrived at by using Direct Numerical Solution(DNS). But the application of DNS is quite restricted because of the broad range of scales of turbulent flow. In order to resolve the complete scales that represent the flow, the computational resourses required is beyond the current capacity for industrial flows. The second best option is to employ Large Eddy Simulation(LES). In LES, only large scales of motion, that are dependent on the boundary conditions are resolved. The smaller scales of motion which are universal are modelled. This reduces the computational demand for free flows where there are no boundaries. However, LES is very expensive when it comes to wall-bounded flows. In wallbounded flows. Approximately 50% of the resources are used for resolving the layers close to the wall. If we are able to use an appropriate model that represents the effects of the inner turbulent wall layers, that will save lots of computational resourses. Such models are know as Wall Models in LES. Wall models can be loosely classified into algebraic Equilibrim wall models and Non-Equilibrium wall models. Different types of wall models are described in , , , . In this study Machine Learning(ML) is used to develop a wall model and compared with an algebraic equilibrium wall model(EQBWM)
CitationRadhakrishnan, S.; Lehmkuhl Barba, O. A machine learning based wall model for LES of turbulent flows. A: . Barcelona Supercomputing Center, 2021, p. 63-64.