betaFeedback

WorksPhysics and Astronomy

Eddy thermal diffusivity model and mean temperature profiles in turbulent vertical convection

Abstract

In this paper, we propose a space-dependent eddy thermal diffusivity model for turbulent vertical natural convection in a fluid between two infinite vertical walls at different temperatures. Using this model, we derive analytical results for the mean temperature profile. Our results reveal that mean temperature profiles for different Rayleigh and Prandtl numbers are described by two universal scaling functions in the inner region next to the walls and the outer region near the centerline between the two walls and the characteristic temperature scales in the inner and outer regions are expressed in terms of the two parameters of the model which determine the characteristic velocities for heat transfer in the two regions. We show that these results are in good agreement with direct numerical simulation data.

Introduction

Natural convective flows driven by temperature differences are ubiquitous in nature and engineering applications. To understand buoyancy-driven wall-bounded flows, it is common to study natural convection in a fluid confined between two vertical walls at different temperatures [2, 15, 23, 3, 1, 21, 20, 8] (Kiš & Herwig 2012; Ng, Chung & Ooi 2013) or adjacent to a single heated vertical plate [17, 4, 6, 18, 22] (Kuiken 1968; Ke et al. 2021). The state of these convective flows is determined by two control parameters, Rayleigh number (RaRa) and Prandtl number (PrPr). Laminar vertical convection has been understood by analysis of steady-state boundary-layer equations [17, 20] (Kuiken 1968) but a full understanding of turbulent vertical convection is still lacking. Knowledge of turbulent vertical convection is important for engineering applications such as ventilation in buildings and can shed light on ice-ocean interaction at near-vertical ice surfaces in a polar ocean [24 Howland, Verzicco &] Lohse 2023). Physical quantities of interest include heat flux, wall shear stress, maximum mean vertical velocity and mean temperature and velocity profiles.

A recent theoretical analysis by one of us (Ching [5]) showed that for fluid confined between two infinite vertical walls at different temperatures, the Nusselt number (NuNu) and shear Reynolds number (ReτRe_{\tau}), which describe heat flux and wall shear stress, scale as Ra1/3Ra^{1/3} in the high-RaRa limit. These theoretical results can well describe direct numerical simulation (DNS) data (Howland et al. [8]). The scaling Nu∼Ra1/3Nu \sim Ra^{1/3} is also consistent with experimental results for fluids with different PrPr (Jakob [10]; MacGregor & Emery [15]; Tsuji & Nagano [22]) and the asymptotic law of heat transfer derived by George & Capp [6] for turbulent natural convection next to a semi-infinite heated vertical plate. The analysis by George & Capp [6] also yields the mean temperature and velocity profiles. By proposing scaling functions of temperature and velocity of certain characteristic scales of length, velocity and temperature in an inner layer next to the heated plate and a turbulent outer layer far away from the plate and matching them in an overlap layer, which is assumed to exist in the high-RaRa limit, they obtained an inverse cubic-root dependence on distance for the mean temperature and a cubic-root dependence for the mean velocity in the overlap layer. The result for the mean velocity deviates from both experimental and DNS data (Versteegh & Nieuwstadt [23]; Hölling & Herwig [7]; Shiri & George [19]). Using a different temperature scale in the inner layer, a logarithmic mean temperature profile was obtained by Hölling & Herwig [7]. The inverse cubic-root and the logarithmic mean temperature profiles have been shown to fit experimental (Cheesewright [4]; Tsuji & Nagano [22]) and DNS data (Versteegh & Nieuwstadt [23]; Ng, Chung & Ooi 2013) for air (Pr=0.709Pr = 0.709) over different spatial regions but their validity for general values of PrPr has not been tested. Li et al. (2023) studied the mean velocity and temperature profiles using models for turbulent heat flux and Reynolds stress but their models violate required boundary conditions at the vertical walls.

In this paper, we propose a space-dependent eddy thermal diffusivity model for turbulent vertical convection in a fluid between two infinite vertical walls at different temperatures and use it to derive analytical results for the mean temperature profile. These analytical results reveal that mean temperature profiles for different Rayleigh and Prandtl numbers are described by two universal scaling functions in the inner region near the walls and the outer region near the centerline between the two walls. We validate our results using DNS data for 1≤Pr≤1001 \le Pr \le100 and 106≤Ra≤10810^6 \le Ra \le10^8 (Howland et al. [8]) and show that the two scaling functions can better describe the data than results reported in previous studies.

The problem

We consider a fluid confined between two infinite vertical walls separated by a distance HH. The wall at the wall normal coordinate x=0x = 0 is kept at a temperature ThT_h and the wall at x=Hx = H at a lower temperature Tc=Th−ΔTT_c = T_h - \Delta T. With the Oberbeck-Boussinesq approximation, the equations governing the fluid motion are ∇⋅u=0\nabla\cdot\mathbf{u} = 0 and

∂u∂t+u⋅∇u=−1ρ∇p+ν∇2u+αg(T−Tm)z^(1)\frac{\partial\mathbf{u}}{\partial t} + \mathbf{u} \cdot\nabla\mathbf{u} = -\frac{1}{\rho}\nabla p + \nu\nabla^2 \mathbf{u} + \alpha g(T-T_m)\hat{\mathbf{z}} \tag*{(1)}
∂T∂t+u⋅∇T=κ∇2T(2)\frac{\partial T}{\partial t} + \mathbf{u} \cdot\nabla T = \kappa\nabla^2 T \tag*{(2)}

where u(x,y,z,t)=(u,v,w)\mathbf{u}(x,y,z,t) = (u,v,w) is the velocity, p(x,y,z,t)p(x,y,z,t) is the pressure, T(x,y,z,t)T(x,y,z,t) is the temperature, ρ\rho is the density of the fluid at the temperature at the centerline between the walls, Tm=(Th+Tc)/2T_m = (T_h + T_c)/2, α\alpha, ν\nu and κ\kappa are the volume expansion coefficient, kinematic viscosity and thermal diffusivity of the fluid, respectively, gg is the acceleration due to gravity and z^\hat{z} is a unit vector along the vertical direction. The velocity field satisfies the no-slip boundary condition at the two walls. The Rayleigh number and the Prandtl number are defined by

Ra≡αgΔTH3νκ,Pr≡νκ(3)Ra \equiv\frac{\alpha g \Delta T H^3}{\nu\kappa}, \qquad Pr \equiv\frac{\nu}{\kappa} \tag*{(3)}

The flow quantities are Reynolds decomposed into sums of time averages and fluctuations such as T(x,y,z,t)=T‾(x,y,z)+T′(x,y,z,t)T(x,y,z,t)=\overline{T}(x,y,z)+T'(x,y,z,t), where an overbar denotes an average over time and primed symbols denote fluctuating quantities. As the vertical walls are infinite, all the mean flow quantities depend on xx only. This is also valid when periodic boundary conditions are imposed on the velocity and temperature in the spanwise (yy) and streamwise (zz) directions as in DNS [23, 8] (Ng, Chung & Ooi 2013). Taking the time average of (1) and (2) leads to the mean flow equations [23].

ddxu′w′‾=νd2dx2w‾+αg(T‾−Tm)(4)\frac{d}{dx}\overline{u'w'}=\nu\frac{d^2}{dx^2}\overline{w}+\alpha g(\overline{T}-T_m) \tag*{(4)}
ddxu′T′‾=κd2dx2T‾(5)\frac{d}{dx}\overline{u'T'}=\kappa\frac{d^2}{dx^2}\overline{T} \tag*{(5)}

The boundary conditions are

w‾(0)=0,w‾(H/2)=0,(6)\overline{w}(0)=0,\qquad\overline{w}(H/2)=0, \tag*{(6)}
T‾(0)=Th,T‾(H/2)=Tm(7)\overline{T}(0)=T_h,\qquad\overline{T}(H/2)=T_m \tag*{(7)}

and by symmetry, the mean profiles w‾(x)\overline{w}(x) and T‾(x)\overline{T}(x) are antisymmetric about x=H/2x=H/2. A well-known challenge for solving w‾(x)\overline{w}(x) and T‾(x)\overline{T}(x) is that (4) and (5) are not closed. In this paper, we focus on finding T‾\overline{T} by solving (5) with (7) using a closure model for the eddy thermal diffusivity.

Eddy thermal diffusivity and mean temperature profiles

Eddy thermal diffusivity model

Integrating (5) with respect to xx, one obtains

u′T′‾−κdT‾dx=−κdT‾dx∣x=0=κNuΔTH(8)\overline{u'T'}-\kappa\frac{d\overline{T}}{dx}=-\kappa\left.\frac{d\overline{T}}{dx}\right|_{x=0}=\kappa Nu\frac{\Delta T}{H} \tag*{(8)}

where NuNu, being the ratio of the actual heat flux normal to the walls to the heat flux when there is only thermal conduction, is defined by

Nu=u′T′‾−κ dT‾/dxκΔT/H=qκΔT/H(9)Nu=\frac{\overline{u'T'}-\kappa\,d\overline{T}/dx}{\kappa\Delta T/H}=\frac{q}{\kappa\Delta T/H} \tag*{(9)}

The heat flux is given by the product of qq and ρc\rho c, where cc is the specific heat capacity of the fluid. We introduce a space-dependent function K(x)K(x) for the eddy thermal diffusivity:

u′T′‾≡−K(x)dT‾dx(10)\overline{u'T'}\equiv-K(x)\frac{d\overline{T}}{dx} \tag*{(10)}

Then (8) with (10) can be integrated to give Ruckenstein & Felske [18]

1ΔT[T‾(x)−Tm]=12−Nu∫0x/Hdy1+K(yH)/κ(11)\frac{1}{\Delta T}\left[\overline{T}(x)-T_m\right]=\frac{1}{2}-Nu\int_{0}^{x/H}\frac{dy}{1+K(yH)/\kappa} \tag*{(11)}

and NuNu is obtained by using the boundary condition at x=H/2x=H/2 in (7):

Nu=12[∫01/2dy1+K(yH)/κ]−1(12)Nu=\frac{1}{2}\left[\int_{0}^{1/2}\frac{dy}{1+K(yH)/\kappa}\right]^{-1} \tag*{(12)}

Thus, both T‾\overline{T} and NuNu can be determined if K(x)K(x) is known.

The characteristic temperature scales in the inner region near the wall and the outer region near the centerline between the two walls should have different dependencies on ν\nu and κ\kappa as molecular diffusivities are expected to play a significant role in the inner region but not the outer region. As seen from (11), this suggests that the eddy diffusivity K(x)K(x) is governed by two independent parameters in the inner and outer regions. Thus, we model K(x)K(x) separately in the inner and outer regions and in the intermediate region between them.

Due to the no-slip and isothermal boundary conditions at the two vertical sidewalls, u′u', T′T' and ∂u′/∂x=−(∂v′/∂y+∂w′/∂z)\partial u'/\partial x=-(\partial v'/\partial y+\partial w'/\partial z) vanish at x=0x=0, and this leads to the vanishing of u′T′u'T' and its first- and second-order derivatives with respect to xx at x=0x=0 [18, 5]. Since the temperature gradient at x=0x=0, being proportional to NuNu, is non-zero, K(x)K(x) and its first- and second-order derivatives should also vanish at x=0x=0. Thus, we model K(x)/νK(x)/\nu by a cubic function in the inner region. In the outer region, we make use of the symmetry of the problem. Since u′T′u'T' is symmetric and T‾\overline{T} is antisymmetric about the centerline x=H/2x=H/2, K(x)K(x) is symmetric about x=H/2x=H/2 and attains a maximum value at x=H/2x=H/2. This motivates us to model K(x)/νK(x)/\nu by the simplest symmetric quadratic function with a peak at x=H/2x=H/2 in the outer region. For the intermediate region, we choose a linear function that connects continuously to the other two regions as it is simple and the resulting K(x)K(x) gives rise to a NuNu that scales as Ra1/3Ra^{1/3} in the high-RaRa limit in accord with the theoretical result Ching [5] (c.f. § 4). Putting all these results together, we propose a three-layer model for K(x)/νK(x)/\nu:

K(x)ν={Ay3,0≤y≤y1c1+c2y,y1<y<y2Cm[1−b(1/2−y)2],y2≤y≤1/2(13)\frac{K(x)}{\nu}= \begin{cases} Ay^3, & 0\le y\le y_1 \\ c_1+c_2y, & y_1<y<y_2 \\ C_m[1-b(1/2-y)^2], & y_2\le y\le1/2 \end{cases} \tag*{(13)}

where y=x/Hy=x/H. Near the walls, the turbulent heat flux is expected to be small compared to the conductive heat flux, or equivalently, K(x)K(x) is expected to be small compared to κ\kappa. Thus, a natural characteristic length scale for the inner region is given by

li=H(PrA)1/3(14)l_i=\frac{H}{(PrA)^{1/3}} \tag*{(14)}

which is the position at which our proposed model of eddy thermal diffusivity is equal to κ\kappa, and we set the width of the inner region to be 2li2l_i, i.e., y1H=2liy_1H=2l_i. We note that the equality of the eddy and molecular thermal diffusivities was used to define the thickness of the thermal boundary layer in vertical convection by Howland et al. [8]. We fix the width of the outer region to be 0.2H0.2H with y2=0.3y_2=0.3 and take b=4b=4. These values are guided by the DNS data Howland et al. [8]. The constants c1c_1 and c2c_2 are fixed by the continuity of K(x)/νK(x)/\nu at x=y1Hx=y_1H and x=y2Hx=y_2H:

c1=Ay13y2−Cm[1−b(1/2−y2)2]y1y2−y1,c2=Cm[1−b(1/2−y2)2]−Ay13y2−y1(15)c_1=\frac{Ay_1^3y_2-C_m\left[1-b(1/2-y_2)^2\right]y_1}{y_2-y_1},\qquad c_2=\frac{C_m\left[1-b(1/2-y_2)^2\right]-Ay_1^3}{y_2-y_1} \tag*{(15)}

Hence, the model has two independent parameters: AA and CmC_m. We will show below that they determine the characteristic velocities for heat transfer in the inner and outer regions, respectively.

Mean temperature profiles

Substitute (13) into (11) and evaluate the integral, we obtain T‾(x)\overline{T}(x):

1ΔT[Th−T‾(yH)]=Nu I1(y),0≤y≤y1(16)\frac{1}{\Delta T}\left[T_h-\overline{T}(yH)\right]=Nu\,I_1(y),\qquad0\le y\le y_1 \tag*{(16)}
1ΔT[T‾(yH)−Tm]={Nu[I1(y1)+I2(y)],y1<y<y2,Nu I3(y),y2≤y≤1/2(17)\frac{1}{\Delta T}\left[\overline{T}(yH)-T_m\right]= \begin{cases} Nu\left[I_1(y_1)+I_2(y)\right], & y_1<y<y_2,\\ Nu\,I_3(y), & y_2\le y\le1/2 \end{cases} \tag*{(17)}

where

Nu=12[I1(y1)+I2(y2)+I3(y2)](18)Nu=\frac{1}{2\left[I_1(y_1)+I_2(y_2)+I_3(y_2)\right]} \tag*{(18)}

and

I1(y)≡1Pr1/3A1/3∫0Pr1/3A1/3ydy′1+y′3=13Pr1/3A1/3{12log⁡[(1+Pr1/3A1/3y)31+PrAy3]+3arctan⁡(2Pr1/3A1/3y−13)+3π6}(19)I_1(y)\equiv\frac{1}{Pr^{1/3}A^{1/3}}\int_0^{Pr^{1/3}A^{1/3}y}\frac{dy'}{1+y'^3} =\frac{1}{3Pr^{1/3}A^{1/3}}\left\{\frac{1}{2}\log\left[\frac{\left(1+Pr^{1/3}A^{1/3}y\right)^3}{1+PrAy^3}\right]+\sqrt{3}\arctan\left(\frac{2Pr^{1/3}A^{1/3}y-1}{\sqrt{3}}\right)+\frac{\sqrt{3}\pi}{6}\right\} \tag*{(19)}
I2(y)≡∫y1ydy′1+Pr(c1+c2y′)=1Prc2log⁡[1+Pr(c1+c2y)1+Pr(c1+c2y1)](20)I_2(y)\equiv\int_{y_1}^{y}\frac{dy'}{1+Pr(c_1+c_2y')}=\frac{1}{Pr c_2}\log\left[\frac{1+Pr(c_1+c_2y)}{1+Pr(c_1+c_2y_1)}\right] \tag*{(20)}
I3(y)≡∫y1/2dy′1+PrCm[1−b(1/2−y′)2]=12B(1+CmPr)log⁡∣1+B(1/2−y)1−B(1/2−y)∣,B=bCmPr1+CmPr(21)I_3(y)\equiv\int_y^{1/2}\frac{dy'}{1+PrC_m\left[1-b(1/2-y')^2\right]} =\frac{1}{2B(1+C_mPr)}\log\left|\frac{1+B(1/2-y)}{1-B(1/2-y)}\right|,\qquad B=\sqrt{\frac{bC_mPr}{1+C_mPr}} \tag*{(21)}

We focus on the mean temperature profile in the inner and outer regions. The analytical results (16) and (17) reveal that T‾(x)\overline{T}(x) is given by two universal scaling functions of their respective characteristic temperature and length scales in these two regions, namely

Th−T‾(x)=TiFi(x/li)inner wall region(22)T_h-\overline{T}(x)=T_iF_i(x/l_i)\qquad\text{inner wall region} \tag*{(22)}
T‾(x)−Tm=ToFo(x/H)outer centerline region(23)\overline{T}(x)-T_m=T_oF_o(x/H)\qquad\text{outer centerline region} \tag*{(23)}

where the inner and outer scaling functions are

Fi(z)=16log⁡[(1+z)31+z3]+33arctan⁡(2z−13)+3π18(24)F_i(z)=\frac{1}{6}\log\left[\frac{(1+z)^3}{1+z^3}\right]+\frac{\sqrt{3}}{3}\arctan\left(\frac{2z-1}{\sqrt{3}}\right)+\frac{\sqrt{3}\pi}{18} \tag*{(24)}
Fo(z)=12blog⁡∣1+b(1/2−z)1−b(1/2−z)∣=14log⁡(1−zz)(25)F_o(z)=\frac{1}{2\sqrt{b}}\log\left|\frac{1+\sqrt{b}(1/2-z)}{1-\sqrt{b}(1/2-z)}\right|=\frac{1}{4}\log\left(\frac{1-z}{z}\right) \tag*{(25)}

the inner and outer temperature scales are

Ti=ΔTNu(PrA)1/3(26)T_i = \frac{\Delta T Nu}{(PrA)^{1/3}} \tag*{(26)}
To=ΔTNuPrCm(27)T_o = \frac{\Delta T Nu}{PrC_m} \tag*{(27)}

and the characteristic length scales in the inner and outer regions are lil_i and HH, respectively. To obtain (23) with (25), we have used the approximation PrCm≫1PrC_m \gg1, which follows from the dominance of the turbulent heat flux over the conductive heat flux, ∣u′T′∣≫κ∣dT/dx∣\lvert u'T'\rvert\gg\kappa\lvert dT/dx\rvert, at x=H/2x=H/2 in the outer region for turbulent convective flows. Using (9) and (14), we can rewrite (26) and (27) as

ViTi=VoTo=q(28)V_iT_i=V_oT_o=q \tag*{(28)}

where

Vi≡κli=κH(PrA)1/3(29)V_i \equiv\frac{\kappa}{l_i}=\frac{\kappa}{H}(PrA)^{1/3} \tag*{(29)}
Vo≡K(H/2)H=νCmH(30)V_o \equiv\frac{K(H/2)}{H}=\frac{\nu C_m}{H} \tag*{(30)}

Thus, the two parameters AA and CmC_m determine ViV_i and VoV_o, the characteristic velocities for heat transfer in the inner and outer regions, respectively.

Validation and discussions

Validation

Using DNS data for Pr=1Pr=1, 2, 5, 10 and 100 and RaRa ranging from 10610^6 to 10910^9 [8], we evaluate K(x)/νK(x)/\nu, fit it by A(x/H)3A(x/H)^3 in the region 0≤x/H≤2li/H=2/(PrA)1/30 \le x/H \le2l_i/H=2/(PrA)^{1/3} using least-square method to obtain AA and extract CmC_m directly as Cm=K(H/2)/νC_m=K(H/2)/\nu. The values of AA and CmC_m are shown in Table 1. In determining T‾(x)\overline{T}(x) and NuNu, the relevant quantity is F(x)=1/[1+PrK(x)/ν]F(x)=1/[1+PrK(x)/\nu] [see (11) and (12)]. We compare F(x)F(x) evaluated using the eddy diffusivity model (13) and the DNS data in Figure 1. Good agreement can be seen, which is further supported by the relatively small errors of the values of NuNu estimated using the model with respect to the DNS data (see Table 1). Good agreement between the analytical result for T‾\overline{T} and the DNS data is also shown in Figure 1. If K(x)/νK(x)/\nu is modeled by A(x/H)3A(x/H)^3 throughout the whole region [18], the estimated NuNu would be given by 1/[2I1(1/2)]1/[2I_1(1/2)]. As shown in Table 1, the relative errors of these estimates of NuNu are substantially larger for Pr≤5Pr \le5.

PrPrRaRaAACmC_mNuNu (DNS)NuNu (model)1/[2I1(1/2)]1/[2I_1(1/2)]
11×1061 \times10^610478.1430.776.597.22 (9.51%)9.08 (37.70%)
12×1062 \times10^622731.5841.788.309.06 (9.12%)11.74 (41.36%)
15×1065 \times10^656711.2860.4211.3111.92 (5.38%)15.90 (40.66%)
11×1071 \times10^7109126.2380.6813.8114.63 (5.91%)19.77 (43.18%)
12×1072 \times10^7225663.87109.1917.1018.34 (7.23%)25.19 (47.28%)
15×1075 \times10^7521909.56168.0722.9324.40 (6.39%)33.30 (45.21%)
11×1081 \times10^8972659.79221.8328.1929.91 (6.10%)40.98 (45.34%)
21×1061 \times10^63971.1318.906.507.05 (8.38%)8.28 (27.40%)
22×1062 \times10^68504.9826.178.288.88 (7.30%)10.66 (28.79%)
25×1065 \times10^623296.2240.0011.2512.12 (7.67%)14.90 (32.39%)
21×1071 \times10^743690.9154.0114.1114.85 (5.27%)18.36 (30.14%)
22×1072 \times10^787605.1072.5417.5718.52 (5.39%)23.15 (31.73%)
25×1075 \times10^7204899.56107.2823.4124.49 (4.62%)30.72 (31.25%)
21×1081 \times10^8419021.34141.8929.0730.72 (5.68%)38.99 (34.13%)
51×1061 \times10^61215.978.566.106.74 (10.55%)7.58 (24.38%)
52×1062 \times10^62424.4212.397.828.42 (7.65%)9.53 (21.84%)
55×1065 \times10^65834.4821.2510.7611.35 (5.52%)12.75 (18.54%)
51×1071 \times10^711543.1828.7113.6114.12 (3.78%)16.00 (17.58%)
52×1072 \times10^722446.0539.7517.0817.59 (3.00%)19.96 (16.84%)
55×1075 \times10^755357.5760.4223.0123.71 (3.06%)26.96 (17.17%)
51×1081 \times10^8107172.0579.4928.6429.40 (2.64%)33.59 (17.29%)
101×1061 \times10^6372.053.505.365.82 (8.45%)6.45 (20.28%)
102×1062 \times10^6835.216.087.227.65 (5.97%)8.42 (16.67%)
105×1065 \times10^62062.5011.9810.0910.54 (4.47%)11.36 (12.66%)
101×1071 \times10^73955.0517.9812.7213.18 (3.63%)14.11 (10.92%)
102×1072 \times10^77875.6323.6716.0516.43 (2.37%)17.74 (10.55%)
105×1075 \times10^719137.6439.0921.7122.29 (2.64%)23.84 (9.80%)
101×1081 \times10^837057.3050.2727.1327.59 (1.69%)29.71 (9.52%)
102×1082 \times10^872836.2767.8733.8934.54 (1.91%)37.21 (9.80%)
105×1085 \times10^8167572.3898.0245.2045.61 (0.91%)49.12 (8.66%)
101×1091 \times10^9333601.74129.5755.9157.27 (2.43%)61.79 (10.52%)
1001×1071 \times10^751.940.887.867.02 (10.63%)7.20 (8.33%)
1002×1072 \times10^7151.222.0510.8110.19 (5.73%)10.25 (5.18%)
1005×1075 \times10^7354.764.5715.0913.88 (8.01%)13.61 (9.82%)
1001×1081 \times10^8784.517.0019.1618.12 (5.43%)17.72 (7.55%)
1002×1082 \times10^81550.4611.3624.3522.97 (5.69%)22.23 (8.72%)
1005×1085 \times10^83338.9518.7532.3229.91 (7.47%)28.70 (11.21%)
1001×1091 \times10^96491.3126.6440.1437.44 (6.71%)35.81 (10.78%)

Table 1. Values of the parameters AA and CmC_m of the eddy thermal diffusivity model. NuNu (model) are the NuNu values estimated using the eddy diffusivity model with these values of AA and CmC_m and 1/[I1(1/2)]1/[I_1(1/2)] are the values of NuNu obtained when K(x)/νK(x)/\nu is modeled instead by A(x/H)3A(x/H)^3. The numbers in parentheses are the relative errors of the estimated values of NuNu with respect to the DNS data by Howland et al. [8].

F(x) versus x for different Prandtl numbers

Figure 1. F(x)=1/[1+PrK(x)/ν]F(x)=1/[1+PrK(x)/\nu] vs xx at Ra=108Ra=10^{8} for Pr=1Pr=1 (black), Pr=10Pr=10 (blue) and Pr=100Pr=100 (green). The symbols are DNS data [8] and the solid lines are evaluated using the eddy diffusivity model (13) with the values of AA and CmC_m from Table 1. In the inset, we compare the analytical result of the mean temperature profiles T‾(x)\overline{T}(x) from the model (solid lines) with the DNS data. The dashed lines denote the positions of y1y_1 and y2=0.3y_2=0.3.

In Figure 2, we plot (Th−T‾)/Ti(T_h-\overline{T})/T_i as a function of x/lix/l_i. The DNS data for different PrPr and RaRa collapse onto a single curve that is well described by the theoretical inner scaling function FiF_i in (24) for 0≤x/li<20 \le x/l_i < 2, validating (22). In this region, FiF_i clearly deviates from a linear function showing that the turbulent heat flux u′T′u'T' cannot be neglected even within the inner region. Following George & Capp [6], we express the temperature and length scales in the inner region on heat flux, or equivalently, qq, the buoyancy parameter αg\alpha g and the molecular diffusivities ν\nu and κ\kappa. Dimensional analysis then yields

Plot of $(T_h-\overline{T})/T_i$ versus $x/l_i$ at $Ra_{\min}$ and $Ra_{\max}$ for different $Pr$ values

Figure 2. (Th−T‾)/Ti(T_h-\overline{T})/T_i vs x/lix/l_i at Ramin⁡Ra_{\min} (circles) and Ramax⁡Ra_{\max} (squares) for Pr=1Pr = 1 (black), Pr=2Pr = 2 (red), Pr=5Pr = 5 (orange), Pr=10Pr = 10 (blue) and Pr=100Pr = 100 (green). Here, Ramin⁡Ra_{\min} and Ramax⁡Ra_{\max} are the minimum and maximum values of RaRa for the corresponding PrPr (see Table 1). The solid line is the inner scaling function FiF_i given by (24) and the two dashed lines indicate 0.9Fi0.9F_i and 1.1Fi1.1F_i.

Ti=(q3αgκ)1/4gi(Pr)=Ti,GCgi(Pr)=ΔTNu3/4(RaPr)−1/4gi(Pr)(31)T_i=\left(\frac{q^3}{\alpha g\kappa}\right)^{1/4}g_i(Pr)=T_{i,GC}g_i(Pr)=\Delta T Nu^{3/4}(RaPr)^{-1/4}g_i(Pr) \tag*{(31)}

up to some PrPr-dependent function gi(Pr)g_i(Pr). Using (28) and (29), we have

li=(κ3αgq)14gi(Pr)=li,Cgi(Pr)(32)l_i = \left(\frac{\kappa^3}{\alpha g q}\right)^{\frac{1}{4}} \qquad g_i(Pr) = l_i,Cg_i(Pr) \tag*{(32)}

Our inner temperature and length scales TiT_i and lil_i thus differ from the scales Ti,GC=(q3/αgκ)1/4T_{i,GC}=(q^3/\alpha g\kappa)^{1/4} and li,GC=(κ3/αgq)1/4l_{i,GC}=(\kappa^3/\alpha gq)^{1/4} adopted by George & Capp [6]. The inclusion of ν\nu in our analysis leads to an additional function gi(Pr)g_i(Pr), and this enables us to obtain a universal scaling function FiF_i in the inner region for all PrPr. By comparing (26) with (31), we obtain

(PrA)−1/3=(RaPrNu)−1/4gi(Pr)(33)(PrA)^{-1/3}=(RaPrNu)^{-1/4}g_i(Pr) \tag*{(33)}

which is used to evaluate gi(Pr)g_i(Pr). We find that gi(Pr)g_i(Pr) can be fitted by a power law kiPrβk_iPr^\beta with β=0.425≈3/7\beta=0.425\approx3/7 and ki=2.18k_i=2.18. A commonly used velocity scale in the inner region is the friction velocity uτ=ν dw‾/dx∣x=0u_\tau=\sqrt{\nu\,\mathrm{d}\overline{w}/\mathrm{d}x|_{x=0}}. It is interesting to compare the inner velocity scale ViV_i and uτu_\tau. Using (29) and (33), we obtain

Viuτ=(RaNu)1/4gi(Pr)Pr3/4Reτ(34)\frac{V_i}{u_\tau}=\frac{(RaNu)^{1/4}}{g_i(Pr)Pr^{3/4}Re_\tau} \tag*{(34)}

where Reτ=uτH/νRe_\tau=u_\tau H/\nu is the shear Reynolds number. The theoretical results of Ching [5] that NuNu and ReτRe_\tau scale as Ra1/3Ra^{1/3} in the high-RaRa limit then imply that ViV_i is related to uτu_\tau by a PrPr-dependent function h(Pr)h(Pr) when RaRa is sufficiently large:

Vi≈h(Pr)uτ(35)V_i\approx h(Pr)u_\tau \tag*{(35)}

It can be seen in Figure 3 that (35) is confirmed and h(Pr)h(Pr) is a decreasing function of PrPr. As a result, if the mean temperature profiles are rescaled by lτ≡κ/uτl_\tau\equiv\kappa/u_\tau and Tτ≡q/uτT_\tau\equiv q/u_\tau instead of lil_i and TiT_i, the profiles for different PrPr do not collapse beyond the linear region near the wall [as shown in figure 5 b of [8]].

Plot of $V_i/u_\tau$ as a function of $Ra$ for different $Pr$ values

Figure 3. Vi/uτV_i/u_\tau as a function of RaRa. The dashed lines give the average values of Vi/uτV_i/u_\tau over RaRa for each PrPr.

Next we check (23). As shown in Figure 4, the rescaling of T‾−Tm\overline{T}-T_m by ToT_o results in an approximate collapse of the data in the region 0.3≤x/H≤0.50.3\le x/H\le0.5. The collapsed data are consistent with the outer scaling function FoF_o with b=4b = 4 in (25). These observations support our choice of y2=0.3y_2 = 0.3 and b=4b = 4 and validate our proposed functional form of K(x)K(x) in the outer region. If the outer temperature scale To,GCT_{o,GC} proposed by George & Capp [6] is used instead of ToT_o, the data collapse is worse (see the inset of Figure 5). The scale

Plot of $(\overline{T}-T_m)/T_o$ vs $x/H$

Figure 4. Plot of (T‾−Tm)/To(\overline{T}-T_m)/T_o vs x/Hx/H at Ramin⁡Ra_{\min} and Ramax⁡Ra_{\max} for Pr=1,2,5,10,100Pr = 1, 2, 5, 10, 100. Same symbols as in Figure 2. The solid line is the function FoF_o given by (25). A similar plot is shown in the inset with the temperature rescaled by To,GCT_{o,GC} instead of ToT_o.

Plot of $A/Ra$ and $C_m/Ra^{4/9}$ versus $Ra$

Figure 5. Plot of A/RaA/Ra and Cm/Ra4/9C_m/Ra^{4/9} vs RaRa. The dashed lines for different PrPr are the values of C1(Pr)C_1(Pr) and C2(Pr)C_2(Pr) obtained by using C=0.043C=0.043, f(Pr)≈0.19f(Pr)\approx0.19 [5] and the fitted values of kik_i, β\beta and kok_o [see (40) and (40)].

To,GCT_{o,GC}, given by

To,GC=(q2αgH)1/3,(36)T_{o,GC} = \left(\frac{q^2}{\alpha g H}\right)^{1/3}, \tag*{(36)}

was obtained by assuming that the temperature scale in the outer region depends on qq, αg\alpha g and HH only and not on the molecular diffusivities ν\nu and κ\kappa. We find that

Cm→ko(RaNu)1/3Pr−2/3(37)C_m \to k_o(RaNu)^{1/3}Pr^{-2/3} \tag*{(37)}

with ko=0.162k_o = 0.162 for sufficiently large RaRa. This implies that To→ko−1To,GCT_o \to k_o^{-1}T_{o,GC} when RaRa is greater than certain threshold value Rac(Pr)Ra_c(Pr) and Rac(Pr)Ra_c(Pr) increases with PrPr.

Finally, we show that our model gives Nu∼Ra1/3Nu \sim Ra^{1/3} in the high-RaRa limit. Using (33) and (37), it can be shown that for sufficiently large RaRa, I1(y1)∼(RaNu)−1/4I_1(y_1) \sim(RaNu)^{-1/4}, I2(y2)∼(RaNu)−1/3ln⁡(RaNu)I_2(y_2) \sim(RaNu)^{-1/3}\ln(RaNu) and I3(y2)∼(RaNu)−1/3I_3(y_2) \sim(RaNu)^{-1/3} and hence

Nu≈12I1(y1)⇒Nu∼Ra1/3.(38)Nu \approx\frac{1}{2I_1(y_1)} \Rightarrow Nu \sim Ra^{1/3}. \tag*{(38)}

Substitute the theoretical result of Nu≈[C2f(Pr)]1/3Pr−1/9Ra1/3=γ(Pr)Ra1/3Nu \approx[C^2f(Pr)]^{1/3}Pr^{-1/9}Ra^{1/3} = \gamma(Pr)Ra^{1/3} for Pr≥1Pr \ge1 in the high-RaRa limit [5] into (33) and (37), we obtain

A≈[γ(Pr)]3/4[gi(Pr)]−3Pr−1/4Ra=C1(Pr)Ra(39)\begin{aligned} A &\approx[\gamma(Pr)]^{3/4}[g_i(Pr)]^{-3}Pr^{-1/4}Ra = C_1(Pr)Ra \tag*{(39)} \end{aligned}
Cm≈ko[γ(Pr)]1/3Pr−2/3Ra4/9=C2(Pr)Ra4/9(40)C_m \approx k_o[\gamma(Pr)]^{1/3}Pr^{-2/3}Ra^{4/9} = C_2(Pr)Ra^{4/9} \tag*{(40)}

when RaRa is sufficiently large. Such RaRa-dependencies of AA and CmC_m are confirmed in Figure 5.

Comparison with mean temperature profiles reported in previous studies

One often-cited result for mean temperature profile was obtained by George & Capp [6]. They proposed scaling functions in terms of Ti,GCT_{i,GC} and li,GCl_{i,GC} in the inner region and To,GCT_{o,GC} and HH in the outer region. In an overlap layer in which both scaling functions hold, they obtained

Th−T‾(x)Ti,GC=−K1(xli,GC)−1/3+ϕ1(Pr),(41)\frac{T_h-\overline{T}(x)}{T_{i,GC}}=-K_1\left(\frac{x}{l_{i,GC}}\right)^{-1/3}+\phi_1(Pr), \tag*{(41)}
T‾(x)−TmTo,GC=K1(xH)−1/3+θ1(42)\frac{\overline{T}(x)-T_m}{T_{o,GC}}=K_1\left(\frac{x}{H}\right)^{-1/3}+\theta_1 \tag*{(42)}

with undetermined constants K1K_1, ϕ1(Pr)\phi_1(Pr) and θ1\theta_1; the independence of K1K_1 and θ1\theta_1 on PrPr follows from the independence of the outer scaling function on ν\nu or κ\kappa. Fitting their DNS data for Pr=0.709Pr=0.709 (air), Versteegh & Nieuwstadt [23] and Ng, Chung & Ooi (2013) found K1=4.2K_1=4.2. Using the DNS data of Howland et al. [8], we find that the region that (41) with K1=4.2K_1=4.2 can fit decreases drastically for larger PrPr while our result for the inner region (22) with (24), rewritten as

Th−T‾(x)Ti,GC=gi(Pr)Fi(1gi(Pr)xli,GC),(43)\frac{T_h-\overline{T}(x)}{T_{i,GC}}=g_i(Pr)F_i\left(\frac{1}{g_i(Pr)}\frac{x}{l_{i,GC}}\right), \tag*{(43)}

can give good fits for all the 5 values of PrPr studied. The comparison for Pr=1Pr=1 and Pr=100Pr=100 is shown in Figure 6.

Plots of temperature profiles for Pr = 1 and Pr = 100

Figure 6. Plots of [Th−T‾(x)]/Ti,GC[T_h-\overline{T}(x)]/T_{i,GC} vs x/li,GCx/l_{i,GC} for Pr=1Pr=1 (left panel) and Pr=100Pr=100 (right panel) at Ra=106Ra=10^{6} (plusses), 2×1062\times10^{6} (crosses), 5×1065\times10^{6} (stars), 10710^{7} (circles), 2×1072\times10^{7} (squares), 5×1075\times10^{7} (diamonds), 10810^{8} (triangles), 2×1082\times10^{8} (left triangles), 5×1085\times10^{8} (inverted triangles) and 10910^{9} (right triangles). The red solid lines are (43) and the blue dashed lines are (41) with K1=4.2K_1=4.2, ϕ1(1)=5.10\phi_1(1)=5.10 and ϕ1(100)=5.15\phi_1(100)=5.15.

Using the same approach but with the temperature scale Ti,GCT_{i,GC} for both the inner and outer regions, Hölling & Herwig [7] obtained a logarithmic profile in the overlap layer:

Th−T‾(x)Ti,GC=K2log⁡(xli,GC)+ϕ2(Pr)(44)\frac{T_h-\overline{T}(x)}{T_{i,GC}}=K_2\log\left(\frac{x}{l_{i,GC}}\right)+\phi_2(Pr) \tag*{(44)}
T‾(x)−TmTi,GC=−K2log⁡(xH)+θ2(45)\frac{\overline{T}(x)-T_m}{T_{i,GC}}=-K_2\log\left(\frac{x}{H}\right)+\theta_2 \tag*{(45)}

with undetermined constants K2K_2, ϕ2(Pr)\phi_2(Pr) and θ2\theta_2. Different values of K2K_2 and ϕ2\phi_2, obtained by fitting DNS data for air, were reported ([23, 7] (Kiš & Herwig 2012; Ng, Chung & Ooi 2013)). We thus test instead the relation between NuNu, RaRa and PrPr, obtained by adding (44) and (45) and using (31) and (32):

(Nu−3RaPr)1/4=K22log⁡(NuRaPr)+2[ϕ2(Pr)+θ2](46)\left(Nu^{-3}RaPr\right)^{1/4}=\frac{K_2}{2}\log(NuRaPr)+2\left[\phi_2(Pr)+\theta_2\right] \tag*{(46)}

Balaji, Hölling & Herwig [1] assumed (44) to hold up to x=H/2x=H/2 and obtained a different relation than (46). As shown in Figure 7, the DNS data (Howland et al. [8]) are consistent with (46) with K2=0K_2=0 indicating their incompatibility with (44) and (45). When K2=0K_2=0, (46) reduces to Nu∼Ra1/3Nu\sim Ra^{1/3}.

Plot of $(Nu^{-3}RaPr)^{1/4}$ vs $\log(NuRaPr)$

Figure 7. Plot of (Nu−3RaPr)1/4(Nu^{-3}RaPr)^{1/4} vs log⁡(NuRaPr)\log(NuRaPr) using DNS data of Howland et al. [8]. The dashed lines for different PrPr are the best fits of (46) with K2=0K_2=0.

A recent work (Li et al. 2023) studied the mean velocity and temperature profiles by modeling the Reynolds stress and turbulent heat flux:

u′w′‾=−k1x2(∂w‾∂x)2(47)\overline{u'w'}=-k_1x^2\left(\frac{\partial\overline{w}}{\partial x}\right)^2 \tag*{(47)}
u′T′‾=−k2x2∂w‾∂x∂T‾∂x(48)\overline{u'T'}=-k_2x^2\frac{\partial\overline{w}}{\partial x}\frac{\partial\overline{T}}{\partial x} \tag*{(48)}

where k1k_1 and k2k_2 are dimensionless coefficients. As discussed in § 3.1, u′T′‾\overline{u'T'} and its first- and second-order derivatives with respect to xx vanish at x=0x=0 due to the boundary conditions. Similar arguments require u′w′‾\overline{u'w'} and its first- and second-order derivatives with respect to xx to vanish at x=0x=0. These properties should be satisfied by any closure model but they are violated by (47) and (48). Since both the mean velocity and temperature gradients, ∂w‾/∂x\partial\overline{w}/\partial x and ∂T‾/∂x\partial\overline{T}/\partial x, are non-zero at x=0x=0, the model expressions on the RHS of (47) and (48) have non-vanishing second-order derivatives at x=0x=0. Applying (47) and (48) to the near-centerline outer region, Li et al. obtained an inverse cubic-root dependence for the mean temperature profile [(2.14b) of Li et al. 2023], which we rewrite as:

T‾(x)−TmTτ=cK[(xH)−1/3−21/3]=cKx∗(49)\frac{\overline{T}(x)-T_m}{T_\tau}=cK\left[\left(\frac{x}{H}\right)^{-1/3}-2^{1/3}\right]=cKx^* \tag*{(49)}

where K=(αgHTτ/uτ2)1/3K=(\alpha gHT_\tau/u_\tau^2)^{1/3}, c≈3.8c\approx3.8 and

x∗≡(xH)−1/3−21/3(50)x^*\equiv\left(\frac{x}{H}\right)^{-1/3}-2^{1/3} \tag*{(50)}

We note that (49) implies that the mean temperature profiles in the outer region would collapse onto a straight line in x∗x^* when temperature is rescaled by Tτ/K=To,GCT_\tau/K = T_{o,GC}. Using (50), we rewrite our result for the outer region, (23) with (25), in terms of x∗x^*:

Tˉ(x)−TmTo=14log⁡[(x∗+21/3)3−1](51)\frac{\bar{T}(x)-T_m}{T_o} = \frac{1}{4}\log\left[(x^*+2^{1/3})^3-1\right] \tag*{(51)}

In Figure 8, it can be clearly seen that the mean temperature profiles in the outer region do not collapse onto the straight line cx∗cx^* when rescaled by To,GCT_{o,GC}, contrary to the prediction by (49), but they are well represented by (51).

Plot of $(\bar{T}-T_m)/(T_{o,GC})$ vs $x^*$ at $Ra_{min}$ and $Ra_{max}$

Figure 8. Plot of (Tˉ−Tm)/(To,GC)(\bar{T}-T_m)/(T_{o,GC}) vs x∗x^* at RaminRa_{min} and RamaxRa_{max} for Pr=1,2,5,10,100Pr = 1, 2, 5, 10, 100. Same symbols as in Figure 2 and the solid straight line is cx∗cx^*. In the inset, a similar plot is shown with the temperature rescaled by ToT_o instead of To,GCT_{o,GC} and the solid line is (51).

Conclusions

In this study, we focus to determine the mean temperature profiles in turbulent natural convection between two infinite vertical walls at different temperatures. A three-layer model for the eddy thermal diffusivity with two parameters has been proposed. Using this model, we have derived analytical results for the mean temperature profiles in terms of the two parameters of the model. These analytical results reveal that the mean temperature profiles for different RaRa and PrPr are described by two universal scaling functions in the inner region next to the walls and the outer region near the centerline between the walls. The characteristic temperature scales in these two regions are expressed in terms of the two parameters, which determine the characteristic velocities for heat transfer in the two regions. The dependencies of the two parameters on RaRa and PrPr are obtained. Analytical expressions of the two scaling functions are fully determined and there is no overlap region in which both functions hold. Our theoretical results for the mean temperature profiles in the inner and outer regions are in good agreement with DNS data for 1≤Pr≤1001 \leq Pr \leq100 and 106≤Ra≤10910^{6} \le Ra \le10^{9} [8] and can give a better description of the data than results reported in the literature [6, 7] (Li et al. 2023).

Funding. This work was funded by the Hong Kong Research Grants Council (Grant No. CUHK 14303623).

Declaration of interests. The authors report no conflict of interest.

Author ORCIDs. Ho Yin Ng, https://orcid.org/0009-0003-8811-2743; Emily S.C. Ching, https://orcid.org/0000-0001-5114-5072

References

  1. [1]Balaji, C., Hölling, M. and Herwig, H. 2007 Nusselt number correlations for turbulent natural convection flows using asymptotic analysis of the near-wall region, ASME. J. Heat Transfer 129, 1100–1105.DOI
  2. [2]Batchelor, G.K. 1954 Heat transfer by free convection across a closed cavity between vertical boundaries at different temperatures, Q. Appl. Maths 12, 209–233.DOI
  3. [3]Betts, P., and Bokhari, I. 2000 Experiments on turbulent natural convection in an enclosed tall cavity, Intl. J. of Heat and Fluid Flow 21, 675–683.DOI
  4. [4]Cheesewright, R. 1968 Turbulent natural convection from a vertical plane surface, J. Heat Transfer 90, 1–8.
  5. [5]Ching, E.S.C. 2023 Heat flux and wall shear stress in large-aspect-ratio turbulent vertical convection, Phys. Rev. Fluids, 8, L026201.DOI
  6. [6]George, W.K.J. and Capp, S.P. 1979 A theory for natural convection turbulent boundary layers next to heated vertical surfaces, Intl. J. Heat Mass Transfer 22, 813–826.DOI
  7. [7]Hölling, M. and Herwig, H. 2005 Asymptotic analysis of the near-wall region of turbulent natural convection flows, J. Fluid Mech. 541, 383–397.DOI
  8. [8]Howland, C.J., Ng, C.S., Verzicco, R. and Lohse, D. 2022 Boundary layers in turbulent vertical convection at high Prandtl number, J. Fluid Mech. 930, A32.DOI
  9. [10]Jakob, M. 1949 Heat Transfer (Wiley & Sons).
  10. [15]MacGregor, R.K. and Emery, A.F. 1969 Free convection through vertical plane layers — moderate and high Prandtl number fluids, Trans. ASME, J. Heat Transfer 93, 253.DOI
  11. [17]Ostrach, S. 1953 An analysis of laminar free-convection flow and heat transfer about a flat plate parallel to the direction of the generating body force, NACA Report 1111, 63.
  12. [18]Ruckenstein, E. and Felske, J.D. 1980 Turbulent natural convection at high Prandtl numbers, ASME. J. Heat Transfer 102, 773–775.DOI
  13. [19]Shiri, A. and George, W.K. 2008 Turbulent natural convection in a differentially heated vertical channel, Proceedings of 2008 ASME Summer Heat Transfer Conference, 285–291.DOI
  14. [20]Shishkina, O. 2016 Momentum and heat transport scalings in laminar vertical convection, Phys. Rev. E 93, 051102(R).DOI
  15. [21]Trias, F.X., Soria, M., Oliva, A. and Pérez-Segarra, C.D. 2007 Direct numerical simulations of two- and three-dimensional turbulent natural convection flows in a differentially heated cavity of aspect ratio 4, J. Fluid Mech. 586, 259–293.DOI
  16. [22]TSUJII, T. AND NAGANO, Y. 1988 CHARACTERISTICS OF A TURBULENT NATURAL CONVECTION BOUNDARY LAYER ALONG A VERTICAL FLAT PLATE, Intl. J. Heat Mass Transfer 31, 1723–1734.
  17. [23]VERSTEEGH, T.A.M. AND NIEUWSTADT, F.T.M. 1999 A DIRECT NUMERICAL SIMULATION OF NATURAL CONVECTION BETWEEN TWO INFINITE VERTICAL DIFFERENTIALLY HEATED WALLS SCALING LAWS AND WALL FUNCTIONS, Intl. J. Heat mass Transfer 42, 3673-3693.DOI
  18. [24]WELLS, A.J. AND WORSTER, M.G. 2008 A GEOPHYSICAL-SCALE MODEL OF VERTICAL NATURAL CONVECTION BOUNDARY LAYERS. J. Fluid Mech. 609, 111-137.DOI

Paper details

vlnc:W_rm4c3mpf
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Physics and Astronomy

Not recorded for this paper

View statement
Looking for related papers…
Contents