Natural Convection in Shear-Thinning Yield Stress

Fluids in a Square Enclosure

Chong Li and Albert Magnin
Universite Grenoble Alpes, LRP, F-38000 Grenoble, France
CNRS, LRP, F-38000 Grenoble, France
Christel Metivier
Laboratoire dEnergetique et de Mecanique Theorique et Appliquee, UMR 7563,
Universite de Lorraine, CNRS, France

DOI 10.1002/aic.15112
Published online December 8, 2015 in Wiley Online Library (wileyonlinelibrary.com)

The influence of viscoplastic rheological features on the Rayleigh-B enard convection is investigated by numerical means
in order to compare with first experimental results given by Darbouli et al. The fluid is modeled by a regularized
Herschel-Bulkley law which is often used to fit numerous pasty fluids. Natural convection in a two-dimensional square
cavity heated from below is considered. Critical values of Oldroyd number Od and yield number Y are provided.
Numerical results highlight a stabilizing effect of the yield stress as well as a destabilizing effect of increasing shear-
thinning coefficient n as the increase in n enhances the heat transfer in the range of our calculations. Unyielded regions
are located in the square corners of the cavity and in the cavity where convection occurs. The unyielded zones size
increases with the increase in Od and can invade all the cavity for sufficiently large values of Od. V C 2015 American

Institute of Chemical Engineers AIChE J, 62: 13471355, 2016

Keywords: yield stress fluids, shear-thinning fluids, Rayleigh-B
enard convection, Herschel-Bulkley fluid

Introduction
The Rayleigh-Benard Convection (RBC) in Newtonian flu-
ids has been intensively studied due to a large interest in ther-
mal transfer in fluids. Reviews on the RBC in Newtonian
fluids can be found in Refs. 13. The Rayleigh-Benard config-
uration consists on heating a fluid layer from below which
involves a buoyancy force. The fluid remains motionless as far
as the buoyancy effect does not overcome dissipative effects
due to viscosity and thermal diffusivity. This competition is
governed by the Rayleigh number Ra and above a critical
value Rac, convection occurs in the fluid. In the Newtonian
case, for an infinite cavity extent with no-slip conditions and
perfect conducting walls, the criticality is Rac 5 1708.14
Despite a wide range of industrial and natural applications
such as nuclear safety, bioreactor, food processes, transportation
of oil, earth sciences, and so forth, the RBC in soft materials
such as viscoplastic fluids has been little studied. This is prob-
ably due to the complexity to treat the solid-liquid transition in
yield stress fluids. The first theoretical study is due to Zhang
et al.5 which considered an inelastic Bingham fluid. The authors
shown that the Rayleigh-Benard for a Bingham fluid, when sub-
mitted to small perturbations, is linearly stable. By theoretical
considerations and numerical means (Augmented Lagrangian
method), Zhang et al. shown that the convection could occur if
Correspondence concerning this article should be addressed to C. Metivier at
Christel.Metivier@univ-lorraine.fr.
C 2015 American Institute of Chemical Engineers
V any external perturbation. For no-slip boundary conditions and
the perturbation amplitude is finite, that is, large enough. These
results are confirmed by Vikhansky6 who considered the RBC
for inelastic Bingham fluid in a two-dimensional (2-D) square
cavity using a Lattice-Boltzmann numerical method. The 2-D
square cavity has also been considered by Turan et al.7 for a
regularized Bingham model. In Turan et al.,7 the authors high-
light the stabilizing effect of the yield stress on the perturbed
flow for different Ra number as well as on the heat transfer.
Furthermore, they also display the variations of the unyielded
regions with the increase in the yield stress value. These regions
are located in the corners of the cavity and could develop in
some cases inside the convective roll.
From an experimental viewpoint, the first controlled study
to consider the RBC in viscoplastic fluids is proposed by
Darbouli et al.8 The authors investigate the onset of convec-
tion in Carbopol gels which are known to behave as an elasto-
viscoplastic material. Carbopol gels are widely used in numer-
ous experimental studies. They are considered as model fluids
as they are not thixotropic.9 In Darbouli et al.,8 several values
of aspect ratio C (cavity diameter over fluid depth d) between
18 and 6 have been investigated. In their experiments, the
authors show that the control parameter is the yield number
^s y
Y5 q^ g^bD
^ T^L^, which represents the ratio between the yield stress
^s y effect and the buoyancy effect, with q ^ the fluid density, g^
the gravitational acceleration, b ^ the thermal volume expan-
sion, and DT^ the temperature difference between the horizon-
tal walls. It is also highlighted that convection occurs without

AIChE Journal April 2016 Vol. 62, No. 4 1347

 well as the code validation are developed. Results are then dis-
played and they are discussed in the final section.

Mathematical Formulation
Rayleigh-B
enard configuration
Consider a Herschel-Bulkley fluid heated from below and
cooled from above in a square cavity of side L^ (Figure 1). The
wall 1 has a fixed temperature T^h larger than the temperature
T^c as that of the upper wall 2 while walls 3 and 4 are insulated.
Constitutive equation
In the dimensional form, denoted by the hat notation, the
Herschel-Bulkley model is given by
Figure 1. Square cavity scheme.  
^ ^n21 ^sy ^
^s 5 K c_ 1 c_ if f ^s > ^s y (1)
^c_

conducting horizontal walls, Darbouli et al. show that critical- ^c_ 50 if f ^s  ^s y (2)
ity is such as 1=Yc  80. This value is close to that one
obtained in Davaille et al.10 which considers natural convec- where K^ corresponds to the consistency, n the shear-thinning coef-
tion involved by a local heat source in Carbopol gels. In Dar- ficient, ^s y the yield stress and ^c_ 51=2^c_ ij ^c_ ij 1=2 ; ^s 51=2^s ij^s ij 1=2
bouli et al.,8 it is also shown that slip at walls can modify are the second invariants of, respectively, the shear rate ^c_ and shear
criticality by decreasing its value which is in agreement with stress ^s tensors. The shear rate tensor is defined by
theoretical results obtained in the Newtonian case11 and in the ^c5$
_ ^
U1$ ^ T
U (3)
Bingham case.12 A second experimental study has been pro-
posed in Kebiche et al.13 which considers the RBC in Carbo- with T the transposition.
pol gels for a rectangular cavity characterized by an aspect Based on the Von Mises criterion, the inelastic Herschel-
ratio of 9.3. The results provided in Ref. 13 in terms of critical Bulkley model (Eqs. 1 and 2) stipulates that the material
temperature difference DTc and 1=Yc show smaller values than behaves as a rigid solid below the yield stress and as a viscous
that of Darbouli et al.8 We think that this discrepancy is due to fluid above. One can also notice that the viscosity tends to
the difference in boundary conditions. In Ref. 13, thermal infinity when the shear rate is weak and that the stress is
conditions (walls conductivity) and slip conditions are not unknown when ^s  ^s y . To avoid these limitations, one can use
quantitatively provided but the authors highlight a small con- regularized models which consider a viscous behavior prior
ductivity as well as smooth and untreated horizontal walls the yield stress and afterward. In the present study, we use the
(polycarbonate plates). It is well known that smooth walls regularization proposed by Papanastasiou21
involve slippery of Carbopol gels.9,1417 As studied by Ref. 8,  n21 
^
slip boundary conditions have a destabilizing effect on the ^s 5 K^ ^c_ 1^s y ^c_ 12e2m^ c_ ^c_ (4)
RBC in Carbopol gels. Similarly, insulated horizontal walls
can have a destabilizing effect as shown in the Newtonian where m ^ is the regularized coefficient. Note that the Herschel-
case1820 for which the critical Ra number decreases from Bulkley model is recovered when m ^ tends to infinity. Using the
1708 (perfect conducting walls) to the value 734 in Cerisier q
buoyancy time L=^ ^ gb ^ dT ^ as the time scale, L^ as the length
et al.18 or 720 in Gershuni et al.19,20 for insulated horizontal
walls.  n21=2
^ dT
scale and K^ g^b ^ =L^ as the characteristic viscosity, the
The aim of the present study is to investigate by numerical
means, the influence of viscoplastic rheological parameters on dimensionless regularized model can thus be written as
the RBC in order to compare computational results with first  
Od
experimental results provided in Refs. 8 and 13. As a first s5 c_ n21 1 12e2q_c c_ (5)
c_
approach, we propose to extend the study realized by Turan
et al.7 to a Herschel-Bulkley fluid which is often used to model- q
where q5 g^=b ^ dT ^ L^m, ^ Od represents the ratio between the
ize shear-thinning yield stress fluids such as Carbopol gels. We
consider here the bench case corresponding to a 2-D square cav- yield stress and the viscous stress termed as the Oldroyd num-
ity heated from below and cooled from above. The small aspect ber and defined by
ratio (C 5 1) has the advantage to be less time consuming in !n2
calculations than a larger one and it can give first quantitative ^ L^
Od5^s y K (6)
tendencies. The boundary conditions considered in our study ^ dT
g^b ^
are classical, meaning no-slip conditions at walls, perfect con-
ducting horizontal walls and insulated lateral walls. In this The effective viscosity of the Papanastasiou model writes
sense, the conditions are close to the experimental conditions of Od
Ref. 8. Values of the control parameters such as Pr; Ra; Od; n, l5 12e2q_c 1_c n21 (7)
c_
defined in the next section, correspond to that of the experiences
realized by Ref. 8, meaning that we consider weakly shear- Let define l1 5Odc_ 12e2q_c and l2 5_c n21 . l2 depends only
thinning yield stress fluids. The mathematical formulation of on n and display a larger shear-thinning effect when n< 1
the problem is first detailed. Then, the numerical method as decreases as shown in Figure 2b. For n 5 1, we recover the

1348 DOI 10.1002/aic Published on behalf of the AIChE April 2016 Vol. 62, No. 4 AIChE Journal
 that l1  l2 when c_ > O1029  0 as displayed in Figure
3b. Then, in our framework we consider that the effective vis-
cosity given by the Papanastasiou regularization (Eq. 7) tends
to a large (but finite) value equal to Od q at small c_ values. By
increasing c_ values, one notices that at fixed value of
Od51022 , the decrease in n implies the increase in l for
c_ < 1. When c_ 51, all curves collapse to the value l  Od11
and for c_ > 1, the decrease in n implies a decrease in l. In
summary, considering c_ < 1, for weak values of c_ < 1=q we
obtain a constant viscosity (Newtonian plateau) whose value
depends on the yield stress, then by increasing c_ , the viscosity
decreases all the more rapidly as n increases (n  1).
Conservation equations
^ dT
^ g^b
Characteristic scales are completed by considering q ^ L^
as the modified pressure scale, and dT ^ as the temperature
scale. The dimensionless governing equations under the Bous-
sinesq approximation read
$  U50 (9)
@U Y
1U  $U52$P1Te2 1 $s (10)
@t Od

Figure 2. l1 and l2 for q5103 and different values

of Od and n.

Bingham model for which l is characterized by a constant plas-

tic viscosity l2. By definition, l1 does not depend on n as dis-
played in Figure 2a. One can notice that when c_ tends to zero
and under the condition c_  1=q, one obtains
l1  Od q (8)
It means that when c_ tends to zero, l1 is finite and depends
on Od and q values. Furthermore, for larger values of c_ , such
as c_  1=q, then e2q_c ! 0 and
Od
l1 
c_
as displayed in the Figure 2a (log-log scale).
Finally, the effective viscosity l5l1 1l2 is represented in
Figure 3. The shear-thinning index n plays a major role for large
values of c_ . For shear-thinning fluids (n < 1), the effective vis-
cosity is dominated by l1 as far as Odq  c_ n21 . For values of c_
such as c_ n21  Od q, l is dominated by l2 and tends to infinity.
This is the case for instance for the case Od 5 1, n 5 0.5 for
which one observes that l2 > l1 when c_ < 1025 .
In the frame of our computations, we set q5O105 ; Od5O Figure 3. Effective viscosity l given by Eq. 7 for (a) q5103 ,
1022 and 0:6 < n  1. Under these conditions, one finds (b) q5105 and different values of Od and n.

AIChE Journal April 2016 Vol. 62, No. 4 Published on behalf of the AIChE DOI 10.1002/aic 1349
 Table 1. Nusselt and Maximal Velocity Results for Newtonian, Bingham, and Herschel-Bulkley Fluids at Ra515; 000 and
Pr 5 1200 for Different Uniform Meshes. The error (%) is evaluated from the M4 results.
Nusselt Maximal Velocity (1024 m/s)
M1 M2 M3 M4 M1 M2 M3 M4
Newtonian fluid Value 3.336 3.312 3.304 3.300 2.792 2.797 2.801 2.80
n51; Od50 Error (%) 1.09 0.35 0.13 / 0.42 0.25 0.08 /
Bingham fluid Value 2.962 2.907 2.894 2.882 1.520 1.496 1.490 1.483
n51; Od50:03 Error (%) 2.77 0.86 0.40 / 2.49 0.91 0.49 /
Herschel-Bulkley fluid Value 2.437 2.285 2.275 2.263 1.001 0.907 0.904 0.897
n50:9; Od50:03 Error (%) 7.83 1.10 0.69 / 11.67 1.19 0.85 /

 
p @T the convective terms, the pressure and velocity are coupled in
Ra Pr 1U  $T 5DT (11)
@t the moment equation using the Semi-Implicit Method for
Pressure-Linked Equations algorithm. The convergence crite-
where U, P, and T denote, respectively, the dimensionless ria were set to 1028 for the temperature and velocity. In the
velocity, modified pressure fields, and the modified tempera- Newtonian case, in the range of Rayleigh values investigated,
^ T^
ture defined by T5 T hdT2
^ . In addition to the Oldroyd number the numerical round-off errors are sufficient to initiate convec-
(Eq. 6), the dimensionless numbers are defined by tion from the conductive regime as initial condition. This is
32n n15 not the case for the Herschel-Bulkley fluids. For these latter
^ ^ dT
q g^b ^ 2 L^ 2
Ra5 (12) fluids, we implement the steady convective Newtonian solu-
^ K^
j tion at the wanted Ra, Pr, and n values and Od 5 0 as initial
^ condition. The converged result is then used as a new initial
for the Rayleigh number, with j^ the thermal diffusivity and b
condition for similar values of Ra, Pr, and n but for an
the thermal expansion coefficient
increased value of Od. This procedure is repeated for increas-
!n21 ing Od values.
K^ g^b ^ dT^ 2
Pr5 (13)
q
^j^ L^ Grid independency and code validation
The grid independency of our results has been examined for
for the Prandtl number and different uniform meshes M1 (30 3 30), M2 (50 3 50), M3 (70
^s y 3 70), and M4 (100 3 100). Simulations are done for a Newto-
Y5 (14) nian fluid (n51; Od50), a Bingham fluid (n51; Od50:03),
^
^ g^b dT
q ^ L^
and a Herschel-Bulkley fluid (n50:9; Od50:03) at Ra 5
for the yield number. One considers the set Ra; Pr; Od as 15,000 and Pr 5 1200. Results in terms of Nusselt number and
the control parameters as maximal velocity are displayed in Table 1. One can notice that
r the introduction of both n < 1 and Od 6 0 (yield stress)
Pr increases the relative error compared with the Newtonian case.
Y5 Od (15)
Ra For all fluids, one notices that convergence of results below 1%
of relative error is obtained with M3 and M4. In the following,
However, results for Y will be displayed and discussed in
all results are obtained with the mesh M3 characterized by a
the following as this parameter plays a fundamental role in the
spatial step equal to 0.017/70.
onset of convection as shown in Refs. 8 and 10 for instance.
The code has been validated by testing several cases. First,
In addition to these control parameters, we also introduce
we have tested the Newtonian case by fixing Pr 5 0.71,
the dimensionless heat flux averaged over the hot wall, that is,
Od 5 0 (^s y 50, n 5 1). Results obtained with the mesh M3
the Nusselt number Nu
have been compared with Ouertatani et al.,22 they are dis-
1  
@T played in Table 2 for different Ra values. As expected, numer-
Nu52 dx1 (16) ical errors increase with Ra. We obtain relative errors of less
0 @x2 x2 50
than 1% for Ra < 106 . In our study, we focus on results such
The systems of Eqs. 911 are completed by no-slip condi- as Ra  O104 , that is, the range for which the relative error
tions at walls is considered negligible (<1%). Furthermore, we have also
U50 tested a shear-thinning fluid case which corresponds to the
Herschel-Bulkley model when Od 5 0, that is, the power law
at x1 50 or 1 and x2 50 or 1.
model case for different values of Prandtl numbers. In this
The isothermal horizontal walls read
case, one knows that the decrease in n increases variations in
Tx2 5050 and Tx2 5151 viscosity and could involve a loss in the results accuracy. In
Finally, vertical walls are considered adiabatic
@T Table 2. Comparison of the Nusselt Values Obtained with
50
@x1 our Simulations with the Results by Ouertatani et al.22 for a
at x1 50 and x1 51. Newtonian Fluid (Pr 5 0.71)
Ra Present Study Ouertatani22 Relative Error (%)
Numerical Procedure 10,000 2.168 2.158 0.4
A finite-volume code is used to solve the RBC in Herschel- 100,000 3.93 3.91 0.5
Bulkley fluids. An upwind second-ordered scheme is used for 1,000,000 6.225 6.309 1.3

1350 DOI 10.1002/aic Published on behalf of the AIChE April 2016 Vol. 62, No. 4 AIChE Journal
 amplitude perturbation A, one can have an idea on a needed
minimal amplitude perturbation Amin variations. In this
respect, one notices that Amin increases with increasing n val-
ues leading to a more intense convection at the transition from
conduction to convection. Furthermore, the numerical results
show that the convection occurs (or stops) at a finite circula-
tion velocity, that is, at a Nusselt value such as Nu > 1. When
the perturbation amplitude is minimal, the onset occurs at a
minimal Nusselt value Numin > 1. This is consistent with the
results obtained by Vikhansky29 in the Bingham case. In this
article, it was shown that the stopping of convection is a limit
point in the parameter space and that convection occurs with a
velocity value larger than U which is proportional to the yield
stress over the Rayleigh number. In the Herschel-Bulkley flu-
ids, we recover similar tendencies as Numin increases with Od
as observed in Figure 5. The influence of n on Numin is shown
Figure 4. Nusselt values Nun =Nun51 as a function of n for in Figure 6 for different values of Ra number. One observes
Ra5105 and different values of Pr (3: Pr 5 100, that the increase in n, and also the increase of Ra as expected,
: Pr 5 100 results from Ref. 23, : Pr 5 10,000, enhance the convective motion at the subcritical threshold,
1: Pr 5 10,000 results from Ref. 23). that is, increase the Numin values.
Similarly, the influence of n on values of Odc (or Yc) is dis-
played in Figures 5 and 7. In these figures, the increase in Odc
order to test large Prandtl values, we have chosen to compare
(or Yc) highlights a destabilizing effect of increasing n. To
our numerical code to results given by Kim et al.23 which con-
sider natural convection in a square enclosure with differen-
tially heated side walls for Pr5100; 1000; 10; 000 and
Ra5105 . Results obtained by Ref. 23 are compared with our
numerical results in Figure 4. One can underline a very close
agreement between results when n varies. The shear-thinning
coefficient n can involve a decrease or an increase in heat
transfer according to the values of Pr and Ra numbers.
Finally, considering the Herschel-Bulkley model, we have
also checked our results convergence by varying the regulari-
zation parameter m ^ from 104 to
^ defined in Eq. 4. By varying m
6
10 , one finds that the Nu value converge within 0.1%. In the
5
following, all results are obtained for m510 ^ involving
q5O105 .

Results
Dimensionless heat transfer is displayed in Figure 5 as func-
tion of Od number for different values of n. Maximal Nusselt
values are obtained when Od 5 0, that is, when the yield stress
is equal to zero. By increasing the Oldroyd values at fixed
value of n, the heat transfer decreases until the value Nu 5 1
which means that the convection stops for Od  Odc . In this
sense, the yield stress has a stabilizing effect on the RBC. This
effect is not surprising as the effective viscosity dominated by
l1 at low values of c_ is constant and depends on the yield
stress (Oldroyd number). Furthermore, this result is consistent
with those given by Turan et al.7 as the authors highlight a sta-
bilizing effect of the yield stress via the Bingham number.
One observes that the transition from convection to conduc-
tion is abrupt in terms of the Nusselt values close to OdOdc .
This phenomenon highlights a subcritical bifurcation. It is not
surprising as one knows that shear-thinning fluids can involve
subcritical bifurcation as shown in Refs. 2427. A theoretical
study28 has also shown the possibility of subcritical bifurcation
in yield stress fluids in the case of the Rayleigh-Benard Pois-
euille flow for low values of Reynolds number. This is also
consistent with results obtained in Ref. 5 which highlight a Figure 5. Nusselt values as a function of (a) Od, (b) 1=Y;
global stability limit RaE meaning that for Ra > RaE , the for Ra515; 000, Pr 5 600, q51:5 3 105 and dif-
occurrence of convection depends on a sufficiently large initial ferent values of n (3: n 5 1, : n 5 0.9, :
perturbation. Because the Nusselt value is correlated to the n 5 0.8, D: n 5 0.7).

AIChE Journal April 2016 Vol. 62, No. 4 Published on behalf of the AIChE DOI 10.1002/aic 1351
Figure 6. Minimal Nusselt values as a function of n for Figure 8. Nusselt values as a function of Od=Odc for
Pr 5 600 and different values of Ra (: Ra515; 000, Pr 5 600, q51:5 3105 and differ-
Ra522; 000, : Ra515; 000, D: Ra 5 9500). ent values of n (3: n 5 1, : n 5 0.9, :
n 5 0.8, D: n 5 0.7).
study the influence of n in the convective regime, one can dis-
play heat transfer as a function of Od=Odc as done in Figure 8.
One can notice that heat transfer is increased for increasing n.
This is consistent with the fact that the effective viscosity
decreases all the more that n increases as shown previously.

Figure 9. Critical values of (a) Oldroyd number Odc, (b)

Figure 7. Critical values of (a) Oldroyd number Odc,
(b) Yield number 1=Yc ; as a function of n for
Ra515; 000, Pr 5 600, q51:5 3 105 .
yield stress number 1=Yc as a function of n
for Pr 5 600 and different values of Ra
(: Ra55 000; q52:67 3105 ; : Ra59 500; q5
4 3 105 ; 3: Ra515; 000; q51:5 3105 ; D: Ra5
22; 000; q5 1:9 3 105 ).

1352 DOI 10.1002/aic Published on behalf of the AIChE April 2016 Vol. 62, No. 4 AIChE Journal
Figure 10. Isovalues of dimensionless temperature (left) defined by h5T^ 2T^ h =dT ^ and dimensionless stream func-
tion (right) W for Ra 5 15,000, Pr 5 600, for different n values and (a) Od 5 0.00375, (b) Od 5 0.0225.
Gray zones correspond to unyielded regions.

Then, a lower viscosity leads to a more intense convection. It
is worth noting that in the Carreau fluid case, the heat transfer
is increased with decreasing n as shown by Refs. 2427 and
30. Contrarily to the Herschel-Bulkley model, in the Carreau
case, the decrease in effective viscosity is larger when n
decreases. However, phenomena are similar in the sense that
heat transfer is increased for decreasing viscosity values.
In the Herschel-Bulkley case, the stabilization of the system
when n is decreased, is also shown in Figure 9, for several Ra
values, that is, Ra 2 5000; 22; 000
. In addition, one can also case by Refs. 6,7, and 31 which consider natural convection of
notice that fixing n and increasing Ra values leads to the
increase in Odc, meaning a destabilization, as the increase in
Ra values enhances the buoyancy effect which is the driving
source of convection.
Isovalues of temperature and stream function are displayed
in Figure 10 for two values of Od and several values of n. One
can observe the distortion of temperature which increases with
the convection strength, that is, with increasing Nu values.
Concerning isovalues of the stream function W, it is worth not-
ing that for Od 5 0.00375, the isovalues gradient of W
decreases with n while it remains quite constant for
Od 5 0.022 (except for n < 0.7). For both values of Od and
n > 0.7, one can notice that rigid regions are located at the
square corners but also in the center of the convective cell
which is in agreement with results obtained in the Bingham
a Bingham fluid in a square enclosure with differentially
heated side walls. Similarly to Turan et al.7 and Huilgol and
Kefayati,31 the size of these regions increases with increasing
Od values. However, one can highlight that the size of these
regions is not clearly correlated with the Nu values at fixed
Od, as the unyielded zones can be larger, for example, for

AIChE Journal April 2016 Vol. 62, No. 4 Published on behalf of the AIChE DOI 10.1002/aic 1353
Nu 5 2.253 (n 5 1, Od 5 0.0225 case) than for Nu 5 1.722
(n 5 0.8, Od 5 0.0225 case) for which the convective strength
is smaller. Finally, for Od 5 0.0225 and n  0:7, unyielded
regions invade all the cavity meaning that we do not observe
any convection as previously highlighted.
Discussion and Conclusion
In this section, a qualitative comparison between our numeri-
cal results and experimental results8 is developed as well as per-
spectives are provided. The fluid has been modeled by a
regularized Herschel-Bulkley law which is often used to fit
the Carbopol gels as the ones used in Darbouli et al.8 In this
last article, parameters are such as 0:4 < n < 0:8; Pr5O102
2O103 ; Ra5O103 2O104 at the onset of convection. since gels such as Carbopol are known to also present an elastic
Experimental results lead to consider 1=Y as the control parame-
ter and it is found that 1=Yc  80. In our computations, smallest
values of 1=Yc are obtained for largest tested Ra values, that is,
for Ra 5 15,000; Ra 5 22,000 and for 0:7 < n < 1, one obtains
1=Yc  90 as displayed in Figure 9b. These results are qualita-
tively in good agreement with experimental results.8 For decreas-
ing values of n and fixed values of Ra, our computations indicate
increasing values of 1=Yc . Recent experiments led by Li32 show
that the critical value 1=Yc  80 remains constant, within the
experimental accuracy, for different Carbopol gels concentra-
tions and thus different values of n. In Li,32 values of Ra are eval-
depends on the material. From these
uated at criticality, RaYc , it is shown that they vary with
rheological features: RaYc decrease with decreasing n. In this
sense, tendencies between our numerical results and experimen-
tal ones8,32 are similar. From a quantitative viewpoint, numerical
results lead to larger values of 1=Yc as that ones obtained in these
later experimental works. That could be explained by the differ-
ent geometries considered in these studies. One expects that
results obtained for a small aspect ratio C 5 1 lead to a stabiliza-
tion of the system,33,34 meaning in our case larger values of 1=Yc
than that of the cases 6  C  18 considered in experiments.
Our numerical results highlight the existence of a subcritical
bifurcation which is enhanced by the increase of n. From an
experimental viewpoint, Darbouli et al.8 have shown that a
small hysteresis (few degrees) is observed when DT is increased
and then decreased meaning that a subcritical bifurcation is
observed but remains small. Concerning the unyielded regions,
our computations provide information (unyielded zones at the
corner and in the center of convective cells) that cannot be com-
pared with observations as any experimental data do not exist.
Further investigations are needed to refine the comparison
between experiments and computations. For instance, one per-
spective should be to consider larger values of aspect ratio in a
future numerical work. From experimental and numerical
viewpoints, one could investigate RBC in viscoplastic fluids
characterized by a larger range of n values, in particular to
investigate the limit of n going to zero. The knowledge of
velocity fields (by Particle Image Velocimetry or Magnetic
Resonance Imaging technics) in the cavity is also of interest in
order to observe (un)yielded regions.
In our numerical results, variations in 1=Yc are clearly corre-
lated with viscosity at low values of c_ as already discussed. In
the viscous fluid case, it is shown that criticality, that is Rac, is
governed by the flow average viscosity.28,35 In our study, c_
values are weak close to the onset of convection and viscosity
in the cavity is dominated by l1  Od q. This expression
highlights a complex dependence on the critical parameters Odc
or 1=Yc with rheological parameters as well as the regularized
parameter q. In this sense, the choice of the model is debatable.
1354 DOI 10.1002/aic Published on behalf of the AIChE
The regularization of the Herschel-Bulkley model is useful
since the shear stress is known for all c_ values while s is inde-
terminate in the inelastic Herschel-Bulkley (or Bingham)
model. The regularization provides good results for several sets
of sheared flows. However, it introduces mathematical parame-
ters which could have a consequence on rheological (physical)
parameters particularly for low c_ values. In some cases, the use
of regularized models are not indicated, see for instance theoret-
ical stability analyses.36,37 From a computational viewpoint, the
use of inelastic models necessitate to use more complex numeri-
cal methods such as, for example, Augmented Lagrangian,5
Lattice-Boltzmann,6 or splitting method.31 However, inelastic
models are known to be the simplest ones but not realistic ones
behavior. These last decades, several studies, for example, Refs.
3840, have proposed to take into account the elastic behavior
leading to elasto-viscoplastic models. More recently, several
studies4145 have focused on the gels behavior below yielding.
In these studies, the authors have observed viscoplastic creep
below the yield stress which is the signature of strain hardening.
Furthermore, these studies have shown that an apparent viscos-
ity la can be measured below the yield stress when measure-
ment times are long enough, that is, O103 2O104 s. This
viscosity is time dependent with a relation la / ta , where the
exponent a 2 0:6; 1
experiments, a model proposed by Dimitriou et al.44 for the 2-D
version and Dimitriou et al.45 for the 3-D version, has emerged
to characterize gels behavior in a more realistic way.
Finally, to consider these more realistic models represents
one perspective to our work.
An additional perspective to our work is to consider in a
closer look the influence of the boundary conditions on the
onset of convection. For instance, it would be interesting to con-
sider slip at walls or to study the influence of thermal conditions
at walls as it is known, in the Newtonian case, that variations in
the wall conductivity can modify significantly the critical-
ity.1820 That could help to understand large discrepancies in
results observed in Darbouli et al.8 and Kebiche et al.13
Acknowledgments
The work of the authors has been supported by the French
National Agency of Research (ANR), grant called ThiM
(ANR-10-BLAN-0925-01). The Laboratoire de Rheologie et
Procedes is part of the LabEx Tec21 (Investissements dAve-
nirGrant ANR-11-LABX-0030) and of the Polynat Carnot
Institute (Investissements dAvenirGrant ANR-11-CARN-
030-01).
Literature Cited
1. Chandrasekhar S. Hydrodynamic and Hydromagnetic Stability. New
York: Dover Publications, 1981.
2. Koschmieder EL. B enard Cells and Taylor Vortices. Cambridge:
Cambridge University Press, 1993.
3. Bodenschatz E, Pesch W, Ahlers G. Recent developments in Rayleigh-
Benard convection. Annu Rev Fluid Mech. 2000;32:709778.
4. Lord Rayleigh. On the convective currents in a horizontal layer of
fluid when the higher temperature is on the underside. Philos Mag.
1916;32:529546.
5. Zhang J, Vola D, Frigaard IA. Yield stress effects on Rayleigh-
Benard convection. J Fluid Mech. 2006;566:389419.
6. Vikhansky A. Thermal convection in a viscoplastic liquid with high
Rayleigh and Bingham numbers. Phys Fluids. 2009;21:103103.
7. Turan O, Chakraborty N, Poole RJ. Laminar Rayleigh-Benard con-
vection of Bingham fluids of yield stress fluids in a square enclosure.
J Non-Newtonian Fluid Mech. 2012;171172:8396.
April 2016 Vol. 62, No. 4 AIChE Journal
 8. Darbouli M, Metivier C, Piau JM, Magnin A, Adbelali A. Rayleigh-
Benard convection for viscoplastic fluids. Phys Fluids. 2013;25:023101.
9. Piau JM. Carbopol gels: elastoviscoplastic and slippery glasses made
of individual swollen sponges. Meso- and macroscopic properties,
constitutive equations and scaling laws. J Non-Newtonian Fluid
Mech. 2007;144:129.
10. Davaille A, Gueslin B, Massmeyer A, Di Giuseppe E. Thermal insta-
bilities in a yield stress fluid: existence and morphology. J Non-
Newtonian Fluid Mech. 2013;193:144153.
11. Webber M. The destabilizing effect of boundary slip on Benard con-
vection. Math Methods Appl Sci. 2006;29:819838.
12. Metivier C, Magnin A. The effect of wall slip on the stability of the
Rayleigh-Benard Poiseuille flow of viscoplastic fluids. J Non-
Newtonian Fluid Mech. 2011;166:839846.
13. Kebiche Z, Castelain C, Burghelea T. Experimental investigation of
the Rayleigh-Benard convection in a yield stress fluid. J Non-
Newtonian Fluid Mech. 2014;203:923.
14. Magnin A, Piau JM. Cone-and-plate rheometry of yield stress fluids.
Study of an aqueous gel. J Non-Newtonian Fluid Mech. 1990;56:
221251.
15. Barnes HA. A review of the slip (wall depletion) of polymers solu-
tions, emulsions and particle suspensions in viscometers: it causes,
character and cure. J Non-Newtonian Fluid Mech. 1995;36:85108.
16. Meeker SP, Bonnecaze RT, Cloitre M. Slip and flow in soft particles
pastes. Phys Rev Lett. 2004;92:198302.
17. Meeker SP, Bonnecaze RT, Cloitre M. Slip and flow in pastes of
soft particles: direct observation and rheology. J Rheol. 2004;48(6):
12951320.
18. Cerisier P, Rahal S, Cordonnier J, Lebon G. Thermal influence of
boundaries on the onset of Rayleigh-Benard convection. Int J Heat
Transf. 1998;41:33093320.
19. Gershuni GZ, Zhukhovitskii EM, Semakin IG. On convective insta-
bility in a horizontal fluid layer separating walls of different conduc-
tivity. Uchen Zap Perm Univ. 1967;248:1829.
20. Gershuni GZ, Zhukhovitskii EM. Convective Stability of Incompres-
sible Fluids. Jerusalem: Israel Programm for Scientific Translations,
1976:32.
21. Papanastasiou TC. Flow of materials with yield. J Rheol. 1987;31:
385404.
22. Ouertatani N, Cheikh NB, Beya BB, Lili T. Numerical simulation of
two-dimensional Rayleigh-Benard convection in an enclosure. C R
M ecanique. 2010;336:464470.
23. Kim GB, Huyn JM, Kwak HS. Transient buoyant convection of a
power-law non-Newtonian fluid in an enclosure. Int J Heat Mass
Transf. 2003;46:36053617.
24. Balmforth NJ, Rust AC. Weakly nonlinear viscoplastic convection.
J Non-Newtonian Fluid Mech. 2009;158:3645.
25. Albaalbaki B, Khayat RE. Pattern selection in the thermal convec-
tion of non-Newtonian fluids. J Fluid Mech. 2011;668:500550.
26. Jenny M, Plaut E, Briard A. Numerical study of subcritical
Rayleigh-Benard convection rolls in strongly shear-thinning Carreau
fluids. J Non-Newtonian Fluid Mech. 2015;219:1934.
AIChE Journal April 2016 Vol. 62, No. 4 Published on behalf of the AIChE
27. Bouteraa M, Nouar C, Plaut E, Metivier C, Kalck A. Weakly nonlin-
ear analysis of Rayleigh-Benard convection in shear-thinning fluids:
nature of the bifurcation and pattern selection. J Fluid Mech. 2015;
767:696734.
28. Metivier C, Nouar C, Brancher JP. Weakly nonlinear dynamics of
thermoconvective instability involving viscoplastic fluids. J Fluid
Mech. 2010;660:316353.
29. Vikhansky A. On the stopping of thermal convection in a viscoplas-
tic liquid. Rheol Acta. 2011;50:423428.
30. Liang SF, Acrivos A. Experiments on buoyancy driven convection
in non-Newtonian fluid. Rheol Acta. 1970;9(3):447455.
31. Huilgol RR, Kefayati GHR. Natural convection problem in a Bing-
ham fluid using the operator-splitting method. J Non-Newtonian
Fluid Mech. 2015;220:2232.
32. Li C. Instabilite de Rayleigh-Benard dans les fluides seuil: critere de
demarrage, experiences et modelisation. Ph.D. Thesis. Universite de
Grenoble, 2015.
33. Buell JC, Catton I. The effect of wall conduction on the stability of
a fluid in a right circular cylinder heated from below. ASME J Heat
Transf. 1983;105:255260.
34. Charlson GS, Sani RL. Thermoconvective instability in a bounded
cylindrical fluid layer. Int J Heat Mass Transf. 1970;13(9):14791496.
35. Parmentier E. Study of thermal convection in non-Newtonian fluids.
J Fluid Mech. 1978;84(1):111.
36. Frigaard IA, Nouar C. On the usage of viscosity regularisation
method for viscoplastic fluid flow computation. J Non-Newtonian
Fluid Mech. 2005;127:126.
37. Metivier C, Nouar C. Stability of a Rayleigh-Benard Poiseuille flow
for yield stress fluids - comparison between Bingham and regular-
ized models. Int J Non-Linear Mech. 2011;46:12051212.
38. Fusin L, Farina A. Modelling of Bingham-like fluids with deforma-
ble core. Comput Math Appl. 2007;53(34):583594.
39. Saramito P. A new constitutive equation for elastoviscoplastic fluid
flows. J Non-Newtonian Fluid Mech. 2007;145(1):114.
40. Saramito P. A new elastoviscoplastic model based on the Herschel-
Bulkley viscoplastic model. J Non-Newtonian Fluid Mech. 2009;
158:154161.
41. Divoux T, Tamarii D, Barentin C, Manneville S. Transient shear band-
ing in a simple yield stress fluid. Phys Rev Lett. 2010;104:208301.
42. Moller P, Fall A, Chikkadi V, Derks D, Bonn D. An attempt to cate-
gorize yield stress fluid behaviour. Philos Trans R Soc London Ser
A. 2009;367:51395155.
43. Moller P, Fall A, Bonn D. Origin of apparent viscosity in yield
stress fluids below yielding. Europhys Lett. 2009;87:38004.
44. Dimitriou C, Ewoldt RH, McKinley GH. Describing and prescribing
the constitutive response of yield stress fluids using large amplitude
oscillatory shear stress (LAOStress). J Rheol. 2013;57(27):144.
45. Dimitriou C, McKinley GH. A comprehensive constitutive law for
waxy crude oil: a thixotropic yield stress fluid. Soft Matter. 2014;10:
66196644.
Manuscript received July 31, 2015, and revision received Oct. 14, 2015.
DOI 10.1002/aic 1355