Escolar Documentos
Profissional Documentos
Cultura Documentos
Abstract. Simulating flow of a Bingham fluid in porous media still remains a challenging task as the yield
stress may significantly alter the numerical stability and precision. We present a Lattice-Boltzmann TRT
scheme that allows the resolution of this type of flow in stochastically reconstructed porous media. LB
methods have an intrinsic error associated to the boundary conditions. Depending on the schemes this
error might be directly linked to the effective viscosity. As for non-Newtonian fluids viscosity varies in
space the error becomes inhomogeneous and very important. In contrast to that, the TRT scheme does not
present this deficiency and is therefore adequate to be used for simulations of non-Newtonian fluid flow. We
simulated Bingham fluid flow in porous media and determined a generalized Darcy equation depending on
the yield stress, the effective viscosity, the pressure drop and a characteristic length of the porous medium.
By evaluating the flow in the porous structure, we distinguished three different scaling regimes. Regime I
corresponds to the situation where fluid is flowing in only one channel. Here, the relation between flow rate
and pressure drop is given by the non-Newtonian Poiseuille law. During Regime II an increase in pressure
triggers the opening of new paths and the relation between flow rate and the difference in pressure to the
critical yield pressure becomes quadratic: q ∝ (dp ˜ − dp
˜ )2 . Finally, Regime III corresponds to the situation
c
where all the fluid is flowing. In this case, q ∝ (dp ˜ − dp
˜ ).
c
methods have been successfully applied to solve flow of For very low shear rates, the effective viscosity diverges
non-Newtonian fluids without yield stress (see [20–22]). to infinity. However, in order to overcome this restriction
Yet, the presence of a threshold in the stress induces nu- numerically, we use the classical regularized viscosity func-
merical errors and instabilities. This is due to the fact that tion as in [26,24]
it is necessary to define an effective viscosity that diverges τ0
for small shear. One possibility to cope with this problem νeff (γ̇) = ν0 + (1 − e−mγ̇ ), (4)
ργ̇
has been introduced by Vikhansky [23] and it is based
on an implicit Lattice-Boltzmann scheme. In the present where m is a regularization coefficient.
article, we will follow the work of [24] who suggested a It is important to note that the main consequence of
multiple-relaxation-time scheme (MRT) to simulate Bing- such regularized function is the fact that now the viscosity
ham fluids. tends to a finite (large) value for low shear. Indeed, we
From a physical point of view, as demonstrated us- have τ0 m
ing the “pore-network model” approach by [3], the flow lim νeff (γ̇) = ν0 + . (5)
γ̇→0 ρ
rate curve might display different scaling regimes with the Thus, in the numerical simulations, the fluid is not strictly
pressure depending on the rheological parameters and the “solid” but only extremely viscous. In that case, its vis-
structure of the medium. These regimes are characterized cosity is mostly characterized by m.
by a power law whose exponent depends on the amount of
fluid that is flowing. The exponent is one when the fluid
is flowing in either only one chanel or in the entire do- 2.2 Poiseuille flow
main. An intermediate regime, with an exponent of two,
characterizes the regime where the amount of flowing fluid The velocity profile of a Bingham fluid driven by a (nega-
increases with the applied pressure. One should also men- tive) pressure gradient in a two-dimensional configuration
tion that in a recent paper, Sinha and Hansen [25] have can be easily computed. Naming x and y the gap-wise and
predicted this quadratic dependence with a mean field ap- stream-wise direction respectively, eq. (1) becomes
proach. ∂P ∂
The objective of this article is then mainly twofold. = Πxy . (6)
∂x ∂y
First, we will present a numerical scheme able to solve
Bingham fluid displacement in a stochastically recon- By symmetry, the shear stress is nil in the centre. Thus,
structed porous media. Secondly, we will determine a gen- after integration, eq. (6) leads to
eralized Darcy equation and investigate the different flow Πxy = y∂x P. (7)
regimes induced by the yield stress and the heterogeneities
Equation (2) can be written as
of the media.
∂ux
= 0, if |Πxy | < τ0 , (8)
∂y
2 Equation of flow — Numerical ∂ux
=
1
(Πxy − sign(Πxy )τ0 ), if |Πxy | > τ0 .
Implementation ∂y ρν0
Combining, eqs. (7) and (8) and assuming a negative pres-
2.1 Bingham fluid sure gradient (∂x P < 0), the velocity field becomes
τ0
The momentum equation defining Stokes flow is given by ux (y) = U0 , for |y| < b , (9)
τw
0 = −∇P + ∇ · Π, 2
(1) 1 τw τ0 τ0
ux (y) = U0 − |y| − b
, for |y| > b ,
where Π is the shear stress tensor and P the pressure. 2ρν0 b τw τw
For a Bingham fluid, the shear stress follows where 2b is the width of the channel. The shear at the wall
is given by τw = −b ∂x P > 0 and
τ0 2
Π ij = 2ρν0 1 + Δij , (2) 1 τ0
2γ̇ρν0 U0 = bτw 1 − . (10)
2ρν0 τw
where By integration, the flow rate is determined as a function
1 ∂ui ∂uj of the pressure gradient
Δij = +
2 ∂xj ∂xi 2
2b3 1 τ0 2 τ0
is the deformation rate tensor and γ̇ = 2Δij Δij is the Qth = ∂x P − ∂x P + . (11)
3ρν0 ∂x P b 2b
shear rate. u is the velocity field, τ0 is the yield stress, ν0
the kinematic Bingham viscosity and ρ the density. It is important to note that the flow condition is given
An effective kinematic viscosity can thus be deter- by the non-dimensional number τ0 /τw = τ0 /b∂x P which
mined by compares the stress at the wall to the yield stress. This
τ0 quantity gives also the size ratio of the solid region to the
νeff (γ̇) = ν0 + . (3) flowing one (see eq. (9)).
ργ̇
Eur. Phys. J. E (2013) 36: 139 Page 3 of 10
tion of two quantities: the local mass = −s+ n cqi cqj . (19)
ρ0 1 q
9
4
ρ= fq = f0 + 2 fq+ , (15) The algorithm is implemented as follows. Initially, the
q=0 q=1 fluid has a homogeneous effective viscosity νeff . We apply
a pressure drop ΔP between the inlet and outlet. At each
and local momentum time step, the local shear rate is computed using eq. (19)
9
4 to update the local effective viscosity using eq. (4). The
j= fq cq = 2 fq− cq . (16) iteration is pursued until convergence of the flow field is
q=1 q=1 reached (which typically takes around 105 -109 time steps).
Page 4 of 10 Eur. Phys. J. E (2013) 36: 139
1 is relative and due to the fact that the theoretical flow rate
tends to zero (see eq. (10)) wheras the numerical one can
0.5 not as the numerical viscosity tends not to infinity but to
a large finite value. It should be noted that this discrep-
ancy can be reduced by increasing the numerical value m.
y/b
0
However, for too high values of m, numerical instabilities
have been observed. Moreover, on this figure, we show that
−0.5 ν0 Q is independent of the numerical parameter ν0 .
Figure 3 displays the relative error of the flow rate vis
−1 à vis of the discretization b. As expected, the numerical
0 0.2 0.4 0.6 0.8 1 1.2
results become more accurate when the number of grid
Ux /U0 points is increased.
Fig. 1. Numerical velocity field (normalized by the theoreti-
Finally, we evaluated the influence of Λ on the velocity
cal velocity at the centre U0 ) inside a two-dimensional chanel profile. To this goal, Λ was varied in the intervall Λ ∈
compared to the theoretical prediction (eq. (9)). The numer- [0.1, 0.4] with the resolution b = 4. We found a maximal
ical parameters are b = 8 δx, m = 109 , Λ = 0.2, ν0 = 0.001, error of Q/Qth of 2%. Thus, the influence of the value of
τ0 = 10−5 , τ0 /τw = 0.5. Λ on the results of the present simulations is negligible.
For the lattice-Boltzmann scheme, typically used val- 2.5 Porous media generation
ues are Λ = 0.2, ν0 = 10−3 , τ0 /ρ 10−5 and m = 109 ,
which allow to have several orders of magnitude between We used the spectral method proposed in [19,32] to gen-
the viscosity of the “solid” and the fluid. In the following, erate the porous medium. To this goal, we first generate
units of the presented quantities are expressed in terms of a matrix W (r) of white random noise. After multiplying
the lattice grid unit and iteration time step. its Fourier transform (Z(r)) with a Gaussian function we
The correspondance between the numerical and exper- obtain: 2 |k|
imental data requires −
√ the determination of a characteristic Z (k) = αZ(k)e
2
k0
. (23)
length (l∗ = λ, b or K), a characteristic pressure p∗ = τ0
−1
and a characteristic velocity v ∗ = l∗ p∗ /ρν0 (based on the A fast inverse Fourier transform f (r) = F T (Z (k))
dynamic viscosity). The correspondance for any distance, leads then to a Gaussian distributed noise correlated with
pressure or velocity are then obtained by a Gaussian correlation function
∗ −2
|k|2
lexp
F T (f · f ∗ ) = α2 e
2
k0
xexp = ∗ xnum , (20) .
lnum
This leads to the autocorrelation function
p∗exp
Pexp = ∗ Pnum , (21) 2
k0 π|r| 2
f ∗ f (x) ∝ e− |r|2
= e− 2 ( ) ,
1
pnum 8 2λ (24)
∗
vexp where λs = π/k0 . The prefactor α is set, without loss in
qexp = ∗
qnum . (22)
vnum generality, to have a standard deviation equal to one. The
solid lattices are then deduced by a level-set with a given
value, f0 : S = {r|f (r) < f0 }. The porosity φ is related
2.4 Validation to the cumulative distribution function, P (f ), which is by
construction an error function centered around 0 with a
In order to validate the scheme, we will present the re- standard deviation of 1
sults of the flow simulation performed in a Poiseuille con-
figuration as described in sect. 2.2. Figure 1 displays the f0 = P −1 (φ).
theoretical velocity profile as well as the numerical one.
The following parameters were used: b = 8, ν0 = 10−3 , We note that this method is not restricted to 2D and
m = 109 , Λ = 0.2 and τ0 /τw = 0.5. We remark here that has also the advantage of generating periodic media which
despite the coarse grid used (given by the points in the allow to apply periodic lateral boundary conditions.
figure), the numerical profile is in good accordance with The porous media is then characterized by its size
the theoretical prediction. Moreover, it is also important (Lx × Lx ), correlation length (λ) and porosity (φ). A pres-
to compare the flow for different size ratio of the sheared sure drop ΔP is then applied to the boundaries in order
and unsheared regions. Figure 2 (left) displays the flow to drive the fluid. After a transient time (between 106 and
rate Q normalized by the theoretical one (eq. (11)) as a 109 δt), a steady velocity field u(r) is reached, from which
function of the unsheared width bτ0 /τw . We can see a good we compute the volume average flow rate
agreement for bτ0 /τw < 6. However, a significant discrep-
1.1 −4
10
0.9 −6
10
0.8
ν0 Q, ν0 Qth
Q/Qth
−8
0.7 10
0.6
−10
0.5 10
= 0.01 0 = 0.01
0
0.4
0 = 0.001 = 0.001
0
−12
10
0 1 2 3 4 5 6 7 8 0 1 2 3 4 5 6 7 8
bτ0 /τw [δx] bτ0 /τw [δx]
Fig. 2. Left: numerical flow rate Q normalized by the theoretical flowrate Qth as a function of the unsheared width for two
different viscosities. Right: numerical flow rate (symbols) and theoretical (line) multiplied by ν0 as a function of the unsheared
width bτ0 /τw . The parameters are the same as in fig. 1.
1.2 x 10
−3
5
= 0.65
1
= 0.7
4
= 0.75
0.8
= 0.8
3
Q/Qth
= 0.85
0.6
q
0/w = 0.1 2
0.4
0/w = 0.2
0.2 / = 0.5 1
0 w
0/w = 0.7
0 0
0 10 20 30 40 50 60 70 80 0 1 2 3 4 5 6
−4
2b[δx] ΔP x 10
Fig. 3. Relative error for the flow rate Q/Qth as function of Fig. 4. Average flow rate q as a function of the applied pressure
the grid width 2b for different unsheared width ratio τ0 /τw . drop ΔP for different porosities. For each random generation,
The parameters are the same as in fig. 1. the same seed has been used. The statistical properties are then
similar. The numerical parameters are Lx = 512 δx, λ = 6 δx,
m = 109 , Λ = 0.2 and ν0 = 10−3 .
3 Results and discussion
3.1 Flow rate curve where n is the Herschel-Bulkley exponent and k the New-
tonian permeability. In particular, for Bingham fluids
In this section we investigate the dependence of the av- (n = 1), this law predicts that at high pressure drop,
erage flow rate q on the characteristics of the porous one should obtain Darcy’s law, explaining the fact that
medium, the yield stress and the pressure drop in order the slope increases with the porosity. We also note that
to determine a generalized Darcy equation. porosity is affecting the pressure threshold.
Figure 4 displays the average flow rate q as a func- On fig. 5, different velocity fields are represented as a
tion of the applied pressure drop ΔP for a given value of function of the pressure drop ΔP . Close to the thresh-
the yield τ0 = 10−6 and different porosities. The trend is old ΔPc fluid is flowing in only few paths. For a pressure
quite similar for all porosities. It can be seen that a min- difference higher by several orders of magnitude, one can
imal pressure drop ΔPc is required for the fluid to flow. clearly observe that all the fluid in the medium is now
Beyond this threshold, the flow rate seems to increase lin- flowing, corresponding to Darcy’s regime.
early with the applied pressure. This trend is qualitatively Figure 6 (left) displays the flow curves as a function
in agreement with the experimental and numerical works of the pressure drop for different values of the yield stress
of [10,11,13], where it has been proposed that the flow τ0 (including the Newtonian case τ0 = 0) and a given
rate behaves like porosity φ = 0.75. As expected, for large enough pres-
sure drop, all curves collapse on the Newtonian one, given
q ∝ k/μeff (∇P − ∇Pc )n , (26) by the standard Darcy’s law. The threshold pressure is
Page 6 of 10 Eur. Phys. J. E (2013) 36: 139
Fig. 5. Flow field example inside the porous media (solid sites are darker) for different applied pressure drop ΔP . The numerical
parameters are φ = 0.75, Lx = 1024 δx, λ = 6 δx, m = 109 , Λ = 0.2, ν0 = 10−3 and τ0 = 10−5 .
0 2
10 10
0=0 =0
0
0=1e−08 0=1e−08
−2 0
10 0=1e−07 10 =1e−07
0
0=1e−06 0=1e−06
−4 0=1e−05 −2 =1e−05
10 10
ρν0 q/λτ0
0
q
−6 −4
10 10
−8 −6
10 10
−10 −8
10 −6 −4 −2
10 −1 0 1 2 3
10 10 10 10 10 10 10 10
ΔP λΔP
τ 0 Lx
Fig. 6. Left: average flow rate q as a function of the applied pressure drop ΔP for different yield stress τ0 plotted on a log-log
scale as a function of the pressure drop. Right: dimensionless flow rate ρν0 ∗ q/λτ0 as a function of the dimensionless parameter
˜ = λΔP/(τ0 Lx ).
dp
Taking into account the pressure threshold, one can In order to determine the critical length λc from dp ˜
c
then write the flow rate curve in the following form: we follow the argument of [3] and [6] for a pore network
model, suggesting that the critical pressure has a simple
KDarcy τ0 λΔP ˜ geometrical meaning. Indeed, the critical pressure can be
q= H − dpc , (29)
λρν0 τ0 Lx determined by finding the path which
minimizes the sum
of the pressure threshold ΔP = min( δpc ), where δpc de-
with lim H(x) = x + O(x). notes the critical pressure of the links. In our context, the
x→∞
argument can be understood as follows. As we have seen
in fig. 5, close to the pressure threshold, only one single
3.2 Pressure threshold-critical length chanel path remains flowing. This path can be approxi-
mated by a channel with variable opening b(s). Assuming
In order to finalize the generalized Darcy equation the lubrication approximation, one can then compute the
(eq. (29)), we determine in the next step the critical pres- flow rate along this channel using eq. (11)
˜ and relate it to a critical length λc , a ge-
sure threshold dp c
ometrical characteristic of the generated porous medium. 2 2
2b(s)3 1 τ0 τ0
Different methods to determine the critical pressure Q= ∇P − ∇P + .
3ρν0 ∇P b(s) 2b(s)
threshold exist. Unfortunately, none of them is exact and (30)
they imply some errors. The easiest method one could When Q tends to zero, because of mass conservation, it
imagine is to decrease the flow rate or the applied pressure implies that the quantity ∇P − b(s)
τ0
tends to zero all along
until the flow is stopped. This method is however difficult
to achieve experimentally and numerically. Indeed, exper- the channel. The pressure gradient is thus known every-
imental set-ups are usually bond to a minimal flow rate where. Since the pressure drop is the integration of the
or pressure that limits the range in which the flow rate- gradient, we have then ΔPc = τ0 /b(s)ds. Consequently,
pressure curve can be determined (see [33] for instance). the critical pressure drop can be computed by finding the
In the present numerical case, the difficulty results from path that minimizes the quantity:
the fact that the flow rate never drops to zero, due to
1.6
is clearly not the case in the present work. Additionally,
1.4 they noted that the exponent α = 2 is independent of the
threshold distribution. However, Roux and Hermann have
1.2 also reported that this scaling regime changes by modify-
1
ing the current-voltage relationship (linear to quadratic).
λc /λ
1 1
0.8 0.8
0.6 0.6
O
O
0.4 0.4
0.2 0.2
0 −3 −2 −1 0 1 2 3
0 −3 −2 −1 0 1 2 3
10 10 10 10 10 10 10 10 10 10 10 10 10 10
˜
dp ˜ − dp
dp ˜
c
4 Conclusion References
In this work, we have used a Lattice-Boltzmann scheme 1. G. Barenblatt, V. Entov, V. Ryzhik, Theory of fluid flows
able to solve yield-stress fluid flow in porous media. We through natural rocks (Norwell, MA (USA); Kluwer Aca-
have analysed the different flow structures inside the me- demic Publishers, 1989).
dia. Three different scaling regimes could be distinguished. 2. W.R. Rossen, J. Colloid Interface Sci. 136, 1 (1990).
Regime I that corresponds to the situation where fluid is 3. S. Roux, H.J. Herrmann, Europhys. Lett. 4, 1227 (1987).
flowing in only one channel. Here, the relation between 4. C.B. Shah, Y.C. Yortsos, AIChE J. 41, 1099 (1995).
flow rate and pressure drop is given by the non-Newtonian 5. M.T. Balhoff, K.E. Thompson, AIChE J. 50, 3034 (2004).
Poiseuille law. During Regime II an increase in pressure 6. M. Chen, W. Rossen, Y.C. Yortsos, Chem. Eng. Sci. 60,
triggers the opening of new paths and the relation be- 4183 (2005).
tween flow rate and the difference in pressure to the criti- 7. T. Sochi, M.J. Blunt, J. Petrol. Sci. Engin. 60, 105 (2008).
cal yield pressure becomes quadratic: q ∝ (dp ˜ − dp˜ )2 . Fi- 8. T. Sochi, Polymer 51, 5007 (2010).
c
nally, Regime III corresponds to the situation where all the 9. M. Balhoff, D. Sanchez-Rivera, A. Kwok, Y. Mehmani, M.
˜ − dp˜ ). Prodanović, Transport Porous Media 93, 363 (2012).
fluid is sheared. In this case, we obtained q ∝ (dp c 10. H. Park, M. Hawley, R. Blanks, SPE (11), 4722 (1973).
It is interesting to note that most of the generalizations of
11. T. Al-Fariss, K.L. Pinder, Cana. J. Chem. Engin. 65, 391
Darcy’s law that have been proposed (see eq. (26)) from (1987).
experimental or numerical studies correspond only to the 12. G.G. Chase, P. Dachavijit, Sep. Sci. Technol. 38, 745
last regime. (2003).
In a recent article, Sinha and Hansen [25] proposed an 13. X. Clain, Ph.D. thesis, Université Paris-Est (2010).
analogy between two phase flow with the Bingham flow 14. D. Rothman, Geophysics 53, 509 (1988).
problem in porous media. The basic idea is that immis- 15. S. Succi, E. Foti, F. Higuera, Europhys. Lett. 10, 433
cible bubbles require a minimal pressure drop in order (1989).
to go through pore throats. This threshold depends on 16. Y. Qian, D. D’Humières, P. Lallemand, Europhys. Lett.
the throat radius and the local saturation. Using a mean 17, 479 (1992).
field approach, they derived analytically a quadratic de- 17. L. Talon, J. Martin, N. Rakotomalala, D. Salin, Y. Yortsos,
pendence of the flow rate on the pressure. In recent sim- Water Resour. Res. 39, 1135 (2003).
ulation of blob dynamics in the same stochastic porous 18. L. Talon, J. Martin, N. Rakotomalala, D. Salin, Y. Yortsos,
media, Yiotis et al. [32] demonstrated numerically such Phys. Rev. E 69, 066318 (2004).
a scaling regime, which confirms the pertinence of this 19. L. Talon, D. Bauer, N. Gland, S. Youssef, H. Auradou, I.
analogy. Further work will be dedicated to the study of Ginzburg, Water Resour. Res. 48, W04526 (2012).
two-phase flow or other rheological laws and boundary 20. E. Aharonov, D.H. Rothman, Geophys. Res. Lett. 20, 679
conditions (the effect of wall slippage particularly) in or- (1993).
der to test the robustness of those regimes. 21. S. Gabbanelli, G. Drazer, J. Koplik, Phys. Rev. E 72,
046312 (2005).
22. J. Psihogios, M. Kainourgiakis, A. Yiotis, A. Papaioannou,
The authors would like to thank Alex Hansen and Irina A. Stubos, Transport Porous Media 70, 279 (2007).
Ginzburg for usefull discussions, the “Agence National de la 23. A. Vikhansky, J. Non-Newtonian Fluid Mech. 155, 95
Recherche” for financial support of the project LaboCothep (2008).
ANR-12-MONU-0011 and the CCR of UPMC for computa- 24. I. Ginzburg, K. Steiner, Philos. Trans. R. Soc. London,
tional resources. Ser. A 360, 453 (2002).
Page 10 of 10 Eur. Phys. J. E (2013) 36: 139
25. S. Sinha, A. Hansen, EPL 99, 44004 (2012). 31. C. Pan, L.S. Luo, C.T. Miller, Comput. Fluids 35, 898
26. T.C. Papanastasiou, J. Rheol. 31, 385 (1987). (2006).
27. I. Ginzburg, J. Stat. Phys. 126, 157 (2007). 32. A.G. Yiotis, L. Talon, D. Salin, Phys. Rev. E 87, 033001
28. I. Ginzburg, F. Verhaeghe, D. d’Humières, Commun. Com- (2013).
put. Phys. 3, 427 (2008). 33. T. Chevalier, C. Chevalier, X. Clain, J. Dupla, J. Canou,
29. I. Ginzburg, Phys. Rev. E 77, 066704 (2008). S. Rodts, P. Coussot, J. Non-Newtonian Fluid Mech. 195,
30. D. d’Humières, I. Ginzburg, Comput. Math. Appl. 58, 823 57 (2013).
(2009). 34. M. Kardar, Y.C. Zhang, Phys. Rev. Lett. 58, 2087 (1987).
35. L. Talon, H. Auradou, M. Pessel, A. Hansen, EPL 103,
30003 (2013).