Wall modeled LES of compressible flows at non-equilibrium conditions

Abstract

METTU, BALACHANDRA REDDY. Wall Modeled LES of Compressible Flows in Non-Equilibrium Conditions. (Under the direction of Pramod Subbareddy.)A turbulent boundary layer is composed of a wide range of length scales and time scales. To resolve all these scales of motion, as with a DNS or traditional wall resolved LES, the grid sizes required scale roughly as the square/cube of the Reynolds number (Re). These requirements make high Re flows computationally very expensive. By modeling the effect of the motions close to the wall, as is proposed in wall modeled LES (WMLES) methods, the computation of high Re flows can be made significantly cheaper. To date, WMLES has been relatively less studied in application with strong compressibility effects.In this study, we implement state-of-the-art wall models for these flows and examine the different challenges which are common to high-speed flow applications: isothermal cold wall conditions, the transition to turbulence, and shock boundary layer interactions (SBLI). It is observed that for cold-wall flows, a more empirical "mixed" scaling for the length scale appearing in the eddy viscosity formulation outperforms the classical semi-local scaling for obtaining predictions of heat-flux and skin-friction. For transitional flows, we develop a modified transition sensor that gave robust predictions for the high Mach number cases we examined. We find that the more computationally expensive non-equilibrium model with an eddy viscosity formulation based on the pressure gradient outperformed the standard equilibrium model for skin-friction predictions in SBLIs. A sensor for detecting regions of non-equilibrium flow was developed based on the resolved Reynolds stress. By using this sensor and switching to an exchange location closer to the wall in the non-equilibrium region for the equilibrium model gave skin-friction and heat flux results close to or even better than the non-equilibrium model at a far cheaper cost. WMLES was performed for SBLI at moderate to high Re with different wall temperature conditions. It was observed that WMLES was able to capture the expansion and shrinking of the separation bubble size when the wall was heated and cooled. Also observed was that using WMLES the low-frequency characteristics of SBLI at high Re was now possible. As a whole, it was seen that WMLES can become a promising tool for exploring the complex physics involved in turbulent high-speed flows at a significantly lower cost than wall resolved LES/DNS and with much higher fidelity compared to pure RANS simulations.

Shock-induced heating and transition to turbulence in a hypersonic boundary layer

“…Recent advances in numerical algorithms, computer hardware and the related computer science have led to successful predictions of complex multi-physics turbulent flows in aerospace applications by using wall-modelled large-eddy simulations (WMLES), but most of these breakthroughs have been limited to systems operating at subsonic and low-supersonic speeds . While notable attempts to employ WMLES have been recently made in supersonic and hypersonic flows (Kawai & Larsson 2012;Bermejo-Moreno et al 2014;Larsson et al 2015;Marco & Komives 2018;Mettu & Subbareddy 2018;Iyer & Malik 2019), this research area is still in its infancy, particularly in relation to aspects connected with hypersonic transitional phenomena (Yang et al 2017b) and thermochemical effects (Di Renzo & Urzay 2019). The present study contributes to this progress by utilizing a relatively simple, yet challenging configuration for benchmarking wall models in hypersonic flows.…”

A new ODE-based turbulence wall model accounting for pressure gradient and Reynolds number effects

“…This estimate of H is then fed into the ODE model for solving U [y] according to Eq. ( 11), (21), and (22). As the first iteration, the resulting inner profile U [y] can be used to update the estimate of H according to Eq.…”

“…The equilibrium wall model has become increasingly popular for engineering applications since only ODEs are solved in the wall-normal direction instead of the more expensive PDEs as in the classical RANS models, e.g. see [17][18][19][20][21][22][23]. To further improve the predictive capability, several variants of non-equilibrium wall models have also been proposed by retaining part of the neglected terms, e.g.…”

Using graph neural networks for wall modeling in compressible anisothermal flows

“…Hybrid methods based on an embedded grid, zonal methods, and seamless methods such as detached-eddy simulation may fall into this category (Cabot, 1995; Balaras et al, 1996; Davidson and Peng, 2003; Temmerman et al, 2005; Piomelli, 2008). This type of approach can handle non-equilibrium effects (Kawai and Larsson, 2013; Park and Moin, 2014) and may take into account the effects of compressibility and temperature variations, provided that relevant RANS models are used (Benarafa et al, 2007; Rani et al, 2009; Zhang et al, 2013; Mettu and Subbareddy, 2018; Iyer and Malik, 2019). However, the computational cost can be large as it requires a mesh that resolves the viscous and conductive sublayers for the RANS computation.…”