Hostname: page-component-7b7567c8c-tnkh4 Total loading time: 0 Render date: 2026-11-16T08:18:11.924Z Has data issue: false hasContentIssue false

Methodology for determining coefficients of turbulent mixing model

Published online by Cambridge University Press:  30 October 2020

You-sheng Zhang
Affiliation:
Institute of Applied Physics and Computational Mathematics, Beijing 100094, PR ChinaCenter for Applied Physics and Technology, HEDPS, and College of Engineering, Peking University, Beijing 100871, PR China
Zhi-wei He
Affiliation:
Institute of Applied Physics and Computational Mathematics, Beijing 100094, PR China
Han-song Xie*
Affiliation:
Institute of Applied Physics and Computational Mathematics, Beijing 100094, PR China
Meng-Juan Xiao*
Affiliation:
Institute of Applied Physics and Computational Mathematics, Beijing 100094, PR China
Bao-lin Tian
Affiliation:
Institute of Applied Physics and Computational Mathematics, Beijing 100094, PR ChinaCenter for Applied Physics and Technology, HEDPS, and College of Engineering, Peking University, Beijing 100871, PR China
*
†Email addresses for correspondence: xiehansong19@gscaep.ac.cn; xiao_mengjuan@163.com
†Email addresses for correspondence: xiehansong19@gscaep.ac.cn; xiao_mengjuan@163.com

Abstract

The accurate prediction of turbulent mixing induced by Rayleigh–Taylor (R–T), Richtmyer–Meshkov (R–M) and Kelvin–Helmholtz (K–H) instabilities is very important in understanding natural phenomena and improving engineering applications. In applications, the prediction of mixing with the Reynolds-averaged Navier–Stokes (RANS) equation remains the most widely used method. The RANS method involves two aspects, i.e. physical modelling and model coefficients. Generally, the latter is determined empirically; thus, there is a lack of universality. In this paper, inspired by the well-known Reynolds decomposition, we propose a methodology to determine the model coefficients with the following three steps: (i) preset a set of analytical RANS solutions by fully using the knowledge of mixing evolutions; (ii) simplify the differential RANS equations to algebraic equations by imposing the preset solutions to RANS equations; (iii) solve the algebraic equations approximately to give the values of the entire model coefficients. The specific application of this methodology in the widely used K–L mixing model shows that, using the same set of model coefficients determined from the current methodology, the K–L model successfully predicts the mixing evolutions in terms of different physical quantities (e.g. temporal scalings and spatial profiles), density ratios and problems (e.g. R–T, R–M, K–H and reshocked R–M mixings). It is possible to extend this methodology to other turbulence models characterised with self-similar evolutions, such as K-$\epsilon$ mixing models.

Information

Type
JFM Papers
Copyright
© The Author(s), 2020. Published by Cambridge University Press
Figure 0

Table 1.The K–L model coefficients used in previous literature and in this paper.

Figure 1

Table 2.Numerical schemes used for problems investigated in this paper.

Figure 2

Figure 1.The schematic demonstration of the traditional (dashed arrows, nodes I-V) and current (solid arrows, nodes 1–7) methods in determining the model coefficients of the turbulent mixing model.

Figure 3

Figure 2.The variations of newly defined integrals $f_{1,2,3,4,6}(r)$, as shown in (3.19)–(3.22), (3.24) with power index $r$. The symbols give the variations by numerically integrating (3.19)–(3.22), (3.24). The lines plot the variations with fitted function, as shown on the right-hand side of (3.19)–(3.22), (3.24). The comparisons show that the fitted functions agree very well with numerical integrals in the range of $r\in (0,1)$.

Figure 4

Figure 3.The variations of newly defined integrals $\varphi _{1,2}(r)$ (as shown in (3.41) and (3.42)) with power index $r$. The symbols give the variations by numerically integrating (3.41) and (3.42). The lines plot the variations with fitted function, as shown on the right-hand side of (3.41) and (3.42). The comparisons show that the fitted functions agree very well with numerical integrals in the range of $r\in (0,1)$.

Figure 5

Figure 4.Comparison of the temporal evolutions (left) of (a) maximum turbulent kinetic energy $\tilde {K}_f^{max }$, (b) mixing width of bubble mixing zone ${h_b}$, and (c) self-similar parameter $\beta$ versus $Agt^2$ and the spatial profiles (right) of (d) the volume fraction of a heavy fluid ${\tilde \phi _1}$, (e) re-normalised turbulent kinetic energy ${\tilde {K}_f}/\tilde {K}_f^{max }$ and (f) re-normalised length scale $\tilde {L}/{\tilde {L}^{max }}$ versus $(x - {h_b})/({h_s} - {h_b})$ between RANS evolutions (solid lines) and preset evolutions (dashed lines). In this study the superscript $max$ denotes the maximum value of quantity over the entire mixing zone at any given time. The RANS results of the R–T test problem with $A = 0.1$ are calculated with a set of model coefficients determined by ${\alpha _b} = 0.05$, ${\alpha _{{{KH}}}} = 0.18$, $\theta = 0.25$, ${\rm \Delta} {E_k}/{\rm \Delta} PE = 0.5$, $s = 4/3$, $\gamma = 1.4$ and ${C_D} = 1/(2\sqrt 2 )$ (Morgan & Greenough 2016). The preset evolutions are given according to the current theory. Specifically, the preset evolutions of ${\tilde {K}_f}/\tilde {K}_f^{max }$, ${h_b}$ and $\beta$ are given theoretically by (3.59), (2.4) and both (3.30) and (3.70), respectively. The spatial profiles of ${\phi _1}$, ${\tilde {K}_f}/\tilde {K}_f^{max }$ and $\tilde {L}/{\tilde {L}^{max }}$ are given by (3.8).

Figure 6

Figure 5.Variation of the measured ${\alpha _b}$ of RANS results with Atwood number $A$ (or equivalent $R$) and model coefficient ${C_D}$. The RANS results are obtained with the model coefficients determined by providing ${\alpha _b} = 0.05$, ${\alpha _{{{KH}}}} = 0.18$, $\theta = 0.25$, ${\rm \Delta} {E_k}/{\rm \Delta} PE = 0.5$, $s = 4/3$, $\gamma = 1.4$ and different values of ${C_D}$. In this figure, ${\alpha _b}$ corresponding to five values of ${C_D}$ = 0.1 (black line), 0.2 (red line), 0.4 (green line), 0.8 (blue line) and 1.6 (cyan line) is plotted. The values of ${\alpha _b}$ vary slightly with the methods used to calculate ${\alpha _b}$. If the mixing width is assumed growing as ${h_b} = {\alpha _b}Ag{(t + {t_s})^2}$, one can derive $h_b^{1/2} = \alpha _b^{1/2}\sqrt {Ag} (t + {t_s})$, where $t_s$ is the time origin of bubbles (Clark & Zhou 2003; Cabot & Cook 2006). In this situation, when plotting $h_b^{1/2}$ versus $\sqrt {Ag} t$, the asymptotic slope of the curve can be regarded as $\alpha _b^{1/2}$, and thus ${\alpha _b}$. In contrast, if the mixing width is assumed growing as ${h_b} = {\alpha _b}Ag{t^2}$, ${\alpha _b}$ is the asymptotic slope when plotting ${h_b}$ versus $Ag{t^2}$. The slight difference between the two methods has been discussed in Cabot & Cook (2006) and Zhou (2017b). In this figure ${\alpha _b}$ is measured using the latter method.

Figure 7

Figure 6.Variation of the profiles of the volume fraction of a heavy fluid ${\tilde {\phi }_1}$, re-normalised turbulent kinetic energy ${\tilde {K}_f}/\tilde {K}_f^{max }$ with power index $s$ (a) and shape factor $\varPi$ (b). The RANS results of the R–T test problem with $A = 0.9\ (R = 20\,{:}\,1)$ are calculated with a set of model coefficients determined by providing ${\alpha _b} = 0.05$, ${\alpha _{{{KH}}}} = 0.18$, $\theta = 0.25$, ${\rm \Delta} {E_k}/{\rm \Delta} PE = 0.5$, $\gamma = 1.4$, ${C_D} = 0.2$ and different $s$ and $\varPi$. In the left figure the shape factor $\varPi$ is fixed as 1, and $s$ increases from 4/3 (black solid lines) to 1.8 (red dashed lines). In the right figure $s$ is fixed as 4/3, and the shape factor $\varPi$ increases from 1 (black solid lines) to 2 (red dashed lines).

Figure 8

Figure 7.The comparison of the mixing evolution predicted by RANS (black lines) and implicit LES (Youngs 2013; Kokkinakis et al.2015) (red dashed lines in the left figure and hollow red rings in the right figure). The left figure plots the evolution of integral mixing width $W$ and maximum turbulent kinetic energy $\tilde {K}_f^{max }$ versus $Ag{t^2}$, and the dashed lines show the corresponding linear growth rate of LES (Kokkinakis et al.2015). The right figure plots the spatial profiles of the volume fraction of a heavy fluid ${\tilde {\phi }_1}$ and re-normalised turbulent kinetic energy ${\tilde {K}_f}/\tilde {K}_f^{max }$ versus re-normalised spatial coordinate $X/W$. The RANS results of the R–T test problem with $A = 0.9\ (R = 20\,{:}\,1)$ are calculated with a set of model coefficients determined by providing ${\alpha _b} = 0.05$, ${\alpha _{{{KH}}}} = 0.18$, $\theta = 0.25$, ${\rm \Delta} {E_k}/{\rm \Delta} PE = 0.5$, $\gamma = 1.4$, ${C_D} = 0.2$, $s = 1.6$ and $\varPi = 1.3$.

Figure 9

Figure 8.Comparison between RANS and LES (${\rm \Delta} {E_k}/{\rm \Delta} PE = 0.55, A=0.9$). For captions, see figure 7.

Figure 10

Figure 9.Comparison between RANS and LES (${\rm \Delta} {E_k}/{\rm \Delta} PE = 0.55, A = 0.5$). For captions, see figure 7.

Figure 11

Figure 10.The temporal evolutions of total mixing width $H(t)$ (a), maximum turbulent eddy scale ${\tilde {L}^{max }}(t)$ (a) and maximum turbulent kinetic energy $\tilde {K}_f^{max }$ (b) with time $t$ for K–H mixing. The solid lines denote the RANS results calculated with the same model coefficients as those of R–T mixing at $A=0.9$. The red dashed lines in the left figure denote the temporal scalings.

Figure 12

Figure 11.The temporal evolutions of the total mixing width $H(t)$ (black lines), bubble mixing width $h_b(t)$ (red lines) and spike mixing width $h_s(t)$ (blue lines) of classical R–M mixing at (a) $A=0.1$, (b) $A=-0.1$, (c) $A=0.5$, (d) $A=-0.5$, (e) $A=0.9$ and (f) $A=-0.9$. The solid lines denote the RANS results calculated with the same model coefficients as those of R–T mixing at $A=0.9$. The dashed lines denote the fitted temporal scalings.

Figure 13

Figure 12.Comparison of $\theta _b$ and $\theta _s$ between the present results at $A=\pm 0.1, \pm 0.5, \pm 0.9$ and LEM experiments (Dimonte & Schneider 2000).

Figure 14

Figure 13.The temporal evolutions of the total mixing width $H(t)$ of reshocked R–M mixing problem with (a) $A=0.67$ and (b) $A=-0.67$. The solid lines denote the RANS results calculated with the same model coefficients as those of R–T mixing at $A=0.9$. The symbols denote the experiment results conducted by Vetter & Sturtevant (1995) (a) and Poggi et al. (1998) (b). See Xiao et al. (2020) for more details.