Professional Documents
Culture Documents
Abstract
Innovation in biotechnology research has resulted in a number of fungi being identified for diverse industrial applications. Much research
has been done in developing optimised membrane bioreactor (MBR) systems for the cultivation of these fungi as a consequence of their potent
industrial applications. This research has been hampered by the lack of a thorough understanding of the fluid mechanics through these devices.
In this article, analytical and numerical solutions to the Navier–Stokes equations were developed to describe the hydrostatic pressure and velocity
profiles in a single fiber membrane gradostat reactor (SFMGR). A generic equation for low wall Reynolds number (Rew = ρvw rH /μ) flows was
developed and solved for the case of negligible angular variations of the flow profiles. The mathematical expressions were proposed as solutions
to transient state, laminar, incompressible, viscous and isothermal flow through a membrane with a variable hydraulic permeability. These profiles
were developed for the lumen and shell sides, taking into account the osmotic pressure and gel formation that occurs when solute particles are
rejected on the membrane. The models developed are applicable to different orientations and configurations. A numerical scheme, with a complete
stability analysis, was developed to complement the analytical models. The models were tested on a vertically orientated MGR, operated in the
dead-end mode. The model solutions gave profiles that are in agreement with the experimental results.
© 2007 Elsevier B.V. All rights reserved.
Keywords: Capillary membrane bioreactor; Momentum transfer; Finite difference; Numerical technique
1. Introduction are inherent of hollow fiber (HF) and capillary MBR’s, for
the continuous production of enzymes from Phanerochaete
Innovation in biotechnology research has resulted in a num- chrysosporium. The MGR showed higher enzyme activities
ber of fungi being identified for diverse industrial applications. than previous and subsequent conventional bioreactor systems
Much research has been done in developing optimised mem- [1,2,4,5]. Following from the work done by Leukes [3], a number
brane bioreactor (MBR) systems for the cultivation of these of other investigations [6–9] have demonstrated the suitability
fungi as a consequence of their potent industrial applications and viability of a capillary polysulphone (PSu) MGR for lignin
[1–5]. Leukes [3] developed a membrane gradostat reactor peroxidase (LiP) and manganese peroxidase (MnP) production.
(MGR), which uses to its advantage the nutrient gradients that The performance of HF and capillary membranes is determined
in large by the transport rate of the key nutrients and/or wastes
[10,11]. It is therefore crucial to have a complete description
Abbreviations: BC, boundary condition; HF, hollow fiber; IC, initial con- and understanding of the momentum as well as the mass transfer
dition; LiP, lignin preoxidase; MnP, manganese preoxidase; P. chrysosporium,
Phanerochaete chrysosporium; SFMGR, single fiber membrane gradostat reac-
through these devices for an optimum MBR design.
tor; SSC, steady-state condition; WRF, white rot fungi A number of experimental and theoretical investigations have
∗ Corresponding author. Tel.: +27 21 460 3160; fax: +27 21 460 3282. been conducted with the aim of modelling the momentum trans-
E-mail addresses: sheldonm@cput.ac.za (M.S. Sheldon), fer inside HF and capillary membranes [10–18]. Most of these
Deon.Solomons@uct.ac.za (D.M. Solomons). investigations however, were unsuccessful in providing adequate
1 Tel.: +27 21 460 3160; fax: +27 21 460 3282.
2 Address: Department of Mathematics and Applied Mathematics, University mathematical models for pressure and velocity profiles, which
of Cape Town, Private Bag, Rondebosch 7700, South Africa. also account for the different modes of operations and/or orien-
Tel.: +27 21 460 3160; fax: +27 21 460 3282. tations of MBR’s. Out of the theoretical investigations [10–18],
0376-7388/$ – see front matter © 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.memsci.2007.06.070
B. Godongwana et al. / Journal of Membrane Science 303 (2007) 86–99 87
Table 2
Characteristics of the internally skinned and externally unskinned polysulphone
capillary membrane
Outer diameter (m) 0.0018
Inner diameter (m) 0.0013
Wall thickness (m) ±200 × 10−6
Average pore size (m) 11
Burst pressure (kPa) 1400
Operating pH 4–11
Maximum temperature (◦ C) 50
3. Analysis
goes to zero, B2 in Eq. (31) has to be zero. Imposing the ‘no slip The dimensionless radial velocity is obtained by integrating Eq.
condition’ in Eq. (22c), the axial lumenal velocity, UL , should (39) and making use of BC3 to obtain:
be equal to zero at R = 1, Ξ must therefore also vanish at R = 1. ∞
Since B1 cannot be set equal to zero without obtaining the trivial 1 R R2 J1 (αn R) −α2n τ
VL (Z, R, τ) = 1− −8 e
solution
τ = 0, the Bessel function J0 (αR) must be set equal to 4 2 2 α4n J1 (αn )
n=1
0 in Eq. (31):
d2 PL
∴ J0 (αR) = 0 (32) × (40)
dZ2
But Eq. (32) has an infinite number of roots α (often called The velocity through the membrane matrix, VM , is derived from
eigenvalues) that will satisfy the boundary conditions of Eq. Darcy’s law, taking into account osmotic pressure, gel formation
(31). These roots are infinite and follow the sequence α1 = 2.405, and gravity, and is given by
α2 = 5.520, α3 = 8.654, . . ., αn . Hence, there are many solutions,
Ξ = B1n J0 (αn R) with n = 1, 2, 3, . . ., ∞, which will satisfy Eq. VM = −κ[PSbπ − PL ] (41)
(30) and the corresponding boundary conditions. Substituting
Eqs. (18) and (31) into Eq. (16), and considering the above where
arguments results in the following expression for Uτ : dw
PSbπ = PS + Π − b (42)
∞
L
−α2n τ
Uτ (R, τ) = Bn e J0 (αn R) (33)
where Π is the dimensionless osmotic pressure on the lumen
n=1
side of the membrane, and the term PSbπ is a function of PS , Π,
where and b. For most work, the van’t Hoff approximation for osmotic
pressure gives an adequate estimate [23]:
Bn = B0 B1n (34)
Φ = cw R∗ T (43)
The initial condition, Eq. (22a) stipulates that at τ = 0, Uτ = U∞ ,
therefore where cw is the total concentration of ions at the wall–membrane
∞ interface on the feed side in kmol/m3 , T the absolute tempera-
1 dPL
− (1 − R2 ) −b = Bn J0 (αn R) (35) ture of the solution in K, and R* is the universal gas constant
4 dZ 8.31451 J/(g mol) K. The determination of the wall concentra-
n=1
tion requires an estimation of the local mass transfer coefficient
Eq. (35), the Fourier–Bessel series, may be solved by making use
between the bulk feed stream and the membrane surface, which
of some standard properties of Bessel functions (e.g. Lommel
in turn requires an account of the mass transfer characteristics
integrals) to obtain the function Bn :
(e.g. the solute diffusion coefficient) of the membrane bioreac-
2 dPL tor. The dimensionless osmotic pressure, Π, inside the lumen
Bn = − 3 −b (36) is therefore obtained by substituting Eq. (43) into Eq. (9). The
αn J1 (αn ) dZ
luminal pressure profile, PL , can be obtained by combining Eqs.
Substitution of the new expression for Bn in Eq. (33) results in (40) and (41) and making use of BC4, and this substitution results
∞
in
J0 (αn R) dPL
−α2n τ
Uτ = −2 e − b (37) d2 PL
α3n J1 (αn ) dZ − 16κPL = −16κPSbπ (44)
n=1
dZ2
where an is the eigenvalues of Eq. (30) as previously explained. Eq. (44) is a simple ordinary differential equation (ODE) that
The transient axial velocity distribution in the membrane lumen can be resolved using Laplace transforms with BC5, and has a
is therefore given by substituting Eqs. (26) and (37) into Eq. solution of the form:
(23):
√
∞
PL (Z) = (P0 − PSbπ ) cosh(4 κ)Z
1 J0 (αn R) −α2n τ
UL (Z, R, τ) = − (1 − R ) − 82
e a √
4 α3 J (α ) + √ sinh(4 κ)Z + PSbπ (45)
n=1 n 1 n 4 κ
dPL where
× −b (38)
dZ √ √
4 κ(P0 − PSbπ ) sinh(4 κ) − b(1 + f )
The corresponding luminal radial velocity profile is obtained by a= √ (46)
f − cosh(4 κ)
making use of the continuity equation, which in its dimension-
less form for an incompressible fluid is given by The parameter f was first defined by Bruining [14] as
For a constant shell pressure, as was the case in the MGR, there Table 5
is no radial shell flow (i.e., VS = 0), and the axial shell velocity Discretization of the dimensionless Navier–Stokes equations
profile, US , is given by Differential Discretization Type
∞
n+1
Ui,j −Ui,j
n
b R 4 −α2n τ
∂U
∂τ n τ−U n
Forward difference
US = 2
2R2 ln + R3 − R −
2 2
Bn e J0 (αn R) ∂U
Ui,j+1 i,j−1
4 R3 b Central difference
n=1
∂R n 2R
Ui+1,j −Ui−1,j
n
∂U
(48) ∂Z n 2Z
Central difference
Ui+1,j −2Ui,j
n +U n
∂2 U i−1,j
∂Z2 ΔZn2 n
Central difference
where ∂2 V
n
Vi,j+1 −2Vi,j +Vi,j−1
Central difference
∂R2 n
Pi+1,j
R
−Pi,j
n
2
2 1bR22 1 ∂P
∂Z n Z
Forward difference
Bn = − ln R3 + J0 (αn ) ∂P
Pi,j+1 −Pi,j
n
In Eqs. (48) and (49), R1 , R2 and R3 are the membrane exit 3.4. Numerical scheme development
radius, the extra capillary space radius and the glass manifold
inner radius, respectively. The volumetric flow rate on the lumen The numerical solutions of the two-dimensional
of the membrane is obtained from the equation: Navier–Stokes equations were obtained by finite differ-
1 encing the dimensionless partial differential equations as shown
ΩL = 4 UL R dR (50) in Table 5. These solutions are restricted to luminal flows only.
0 The assumptions used in the development of the analytical
Substituting Eq. (38) in Eq. (50) and integrating between the solutions are still maintained, however in the numerical scheme
limits 0 and 1 results in the following expression for the luminal the assumptions of negligible inertial and normal stress effects
volumetric flowrate: have been lifted. The extension of the numerical scheme to
∞ −α2 τ
account for these effects makes it a more accurate approach
1 e n dPL vis-à-vis the analytical, however, the convergence of this
ΩL = − −8 − b (51) scheme is also influenced by a number of factor as will be
4 α4n dZ
n=1
discussed. The dimensionless form of Eq. (1) is
A stream function is defined such that
∂U ∂U ∂U
1 ∂ψ + Reb V + βU
U=− (52) ∂τ ∂R ∂Z
R ∂R
1 ∂ ∂U ∂2 U ∂P
and, = R + β2 2 − β +b (56)
R ∂R ∂R ∂Z ∂Z
1 ∂ψ
V = (53) and the dimensionless form of Eq. (2) is
R ∂Z
∂V ∂V ∂V
From Eqs. (52) and (53) the stream functions in the lumen and + Reb V + βU
shell of the membrane are respectively given by ∂τ ∂R ∂Z
1 ∂ ∂V V ∂2 V ∂P
ψL =
1 1 dPL
−b = R − 2 + β2 2 − (57)
4 2 dZ R ∂R ∂R R ∂Z ∂R
2
R R2 dPL(1) dPL(0) where Reb is the bulk fluid Reynolds number, which is different
+ 1− − (54) from the wall Reynolds number (Rew ) that is calculated using
2 2 dZ dZ
the wall velocity. Reb is defined as
and
ρvz0 RL
Reb = (58)
R22 μ
ψS = −b [ln(R3 − R2 − ln R3 )](R23 − R22 )
4
and β is the aspect ratio defined as
R22 R23 R43 − R42
− (R23 − R22 ) + (R23 − R22 ) − (55) RL
8 8 16 β= (59)
L
B. Godongwana et al. / Journal of Membrane Science 303 (2007) 86–99 93
4. Results
Fig. 8. Dimensionless luminal hydrostatic pressure profile as a function of the Fig. 9. A comparison of the developed model with literature models and exper-
dimensionless membrane hydraulic permeability (κ). imental data.
Table 6
A comparison of the developed analytical pressure model against experimental pressure data for different nutrient flowrates and membrane lengths
Membrane length (m) Flowrate (ml/min) Experimental inlet pressure (Pa) Experimental outlet pressure (Pa) Developed model outlet pressure (Pa)
gravitational force, whereas the others do not. Eq. (45) indicates points then the round off error ε may be expressed as the series
a linear decline in the hydrostatic pressure along the length of of the form:
the membrane. This result is in agreement with Catapano’s [11]
contention that the axial pressure profile is linear when there
N/2
is ‘no-slip’ velocity. A summary of other experimental condi- ε(Z, R, τ) = exp(am τ + ikm R + ilm Z) (65)
tions of operation including variations in membrane length and m=1
The difference equations developed and solved in this study, U(Z, R, τ) = μni,j + εni,j (66)
Eqs. (60) and (61), each contain only one unknown and can be
solved explicitly for this unknown in a straightforward man- results in a difference equation for εni,j
n+1 n+1
ner. The unknowns are Ui,j and Vi,j , since we assume that
U and V are known at all grid points at time level n. By defi-
i,j − εi,j
εn+1 n εni,j+1 − 2εni,j + εni,j−1 εni,j+1 − εni,j−1
nition therefore, the solution of Eqs. (60) and (61) follows an = +
explicit approach. Besides its simplicity compared to the implicit τ (R)2 2RR
approach, a major disadvantage of an explicit approach is that εn
2 i+1,j
− 2εni,j + εni−1,j
there are restrictions on the values of the independent variables. +β 2
(67)
(Z)
An exact stability analysis of the difference representation of the
nonlinear Navier–Stokes equations does not exist [22]. Since a time step has size τ, an axial space step Z and a
radial space step size R, there is an amplification factor for the
4.2.1. Possible sources of error rounding off error ε, viz.
Sources of error in the numerical scheme could possibly arise
n+1
ε
i,j am τ
in the accuracy of the experimental determination of the input 2[1−cos(km R)] sin(km R)
variables (e.g. κ, p0 , p1 , π, ρ), and the fit of the criteria for n =e = 1 − τ −
εi,j (R)2 RR
convergence of the scheme (i.e. the continuity equation). The
most sensitive of the input variables are the inlet and outlet pres- 2β2 [1 − cos(lm Z)]
sures, in that, an increase of 1 kPa in the inlet pressure will + ≤1 (68)
(Z)2
result in a 17% error in the prediction of the axial velocity
n+1
for the next time step Ui,j . The convergence of the scheme This inequality has to be satisfied in order for the linearized form
is also very sensitive to changes in the hydraulic permeability, of the difference Eq. (56) to have a stable solution, and implies
κ. Slight changes/errors in the other input variables will also that
affect the accuracy of the scheme but to a very small degree in
comparison to the input and output pressures and the hydraulic 2[1 − cos(km R)] sin(km R)
τ −
permeability. (R)2 RR
2β [1 − cos(lm Z)]
2
4.2.2. Von Neumann stability analysis + ≤2 (69)
If one looks for a stability criteria for the Navier–Stokes dif- (Z)2
ferential Eqs. (56) and (57), a fairly accurate treatment entails
besides the fact that the curly brackets of (68) is non-negative, i.e.
ignoring the inertial and normal stresses for negligible bulk
2[1 − cos(km R)]/(R)2 + 2β2 [1 − cos(lm Z)]/(Z)2 ≥ sin(km
Reynolds number Reb 1, so that Eqs. (56) and (57) become
R)/RR. [This latter stability condition simplifies to
linear in the functions U and V. Despite an exact treatment not
(1 + β2 )Rkm = 2πR(1 + β2 )m ≥ 1 in the limit that R and Z
being presented here, one can still get an idea of the stability
tend to 0.] The stability requirement for the solution of the
requirement upon the difference Eqs. (60) and (61) by studying
difference Eq. (67) [having set sin(km R) = 1, cos(km R) = 0
the linearized Navier–Stoke equations:
and cos(km Z) = 0, at the extremity of the curly brackets of
∂U 1 ∂ ∂U ∂2 U ∂P (69)] is the constraint:
= R + β2 2 − β +b (63)
∂τ R ∂R ∂R ∂Z ∂Z
(ΔR)2
and τ ≤ (70)
1 − 1/2(R/R) + β2 (R/Z)2
2
∂V 1 ∂ ∂V V 2∂ V ∂P
= R − 2
+ β 2
− (64) Here (R/R) is the fractional radial increment. Also of inter-
∂τ R ∂R ∂R R ∂Z ∂R
est is the differential equation for V, Eq. (64). The round
for negligible values of the bulk Reynolds number or by ignoring off error δ(Z, R, τ) of its linear difference equation satisfies
some of the inertial and normal stress effects. If there are N grid another difference equation similar to that of the previous case
96 B. Godongwana et al. / Journal of Membrane Science 303 (2007) 86–99
for ε: Table 7
Grid independence of the flow field
i,j − δi,j
δn+1 n δni,j+1 − 2δni,j + δni,j−1 δni,j+1 − δni,j−1
= + τ U V (Z; R)
τ (R)2 2RR
0.1 2.33 × 10−6 9.74 × 10−8 (0.38; 0.09)
δn
2 i+1,j
− 2δni,j + δni−1,j 0.05 2.29 × 10−6 1.38 × 10−7 (0.38; 0.09)
+β (71) 0.01 2.26 × 10−6 1.70 × 10−7 (0.38; 0.09)
(Z)2 0.005 2.26 × 10−6 1.75 × 10−7 (0.38; 0.09)
0.001 2.25 × 10−6 1.78 × 10−7 (0.38; 0.09)
the main difference being the presence of an additional curvature
Analytical solution 2.29 × 10−6 1.79 × 10−7 (0.38; 0.09)
term −δnij /R2 . Generalise the round off error to
is given by
1 − τ 2[1 − cos(km R)] − sin(km R) + 1
(R)2 RR R2
2β2 [1 − cos(lm Z)]
+ ≤1 (73)
(Z)2
In the limit that R and Z tends to zero, one of the two
possibilities is that
2πm 1
4(1 + β2 )π2 m2 − + 2 ≥0 (74) Fig. 10. Convergence of the numerical solution as a function of grid spacing.
R R
and the other [having put sin(km R) = 1, cos(km R) = 0, and The rate of convergence of the numerical solution is influ-
cos(lm Z) = 0] is that enced by a number of factors including the aspect ratio, the
Reynolds number and the grid spacing. In Fig. 9 the convergence
(ΔR)2
τ ≤ is defined as Un + 1 /Un , and the solution is said to be converged
1 − (1/2)R/R) + (1/2)(R/R)2 + β2 (R/Z)2 when the value of the convergence is unity. The influence of
(75) the grid spacing is that the rate of convergence increases with
This is the stability condition for the linearized difference Eq. increasing grid spacing, and this is demonstrated by a compar-
(61), and is in fact a stronger form of stability condition (70). ison of a grid spacing of 0.02 and 0.0007 in Fig. 10. On the
Unfortunately a treatment of non-linear Eqs. (60) and (61) will other hand, the accuracy of the solution decreases with increas-
not be done here, but constraint (75) at least gives us a good idea ing grid spacing. The grid spacing required therefore and the
of the stability condition for linearized Navier–Stokes equations. corresponding accuracy will be dictated by the application of
The fractional radial increment is always less than unity, except the numerical solver, and a trade off will have to be reached
possibly at the start j = 0. The denominator of (75) is positive between a comprehensive solution that takes up a lot of comput-
definite and greater than one, which guarantees that the numer- ing time and a little less precise solution that is fairly quick to
ical solution is stable in almost all cases where the increments solve [22].
R and Z are chosen to be much less than one.
A tenable solution of a differential equation, when using a 5. Conclusion
finite difference scheme (or finite volume), should be indepen-
dent of the number of grid points (or volume cells) used. If this A generic equation for small Rew flows through SFMGR’s
condition is not met then the solution is not stable. This means was developed. The model was solved for the specific case of
that the steady-state values of U and V should be independent of negligible angular variations in the flow profiles (i.e., m = 0).
the time increment chosen. Generally to remedy this situation From this generic equation, detailed models of momentum trans-
the number of grid points used must be increased, that is, the fer accounting for gel formation, osmotic pressure, the mode
time increment needs to be decreased. In Table 7, decreasing of operation, and orientation of a HF and/or capillary MGR
τ from 0.01 to 0.005 improves the numerical solution only were developed. The accuracy of these models was tested on a
marginally; therefore the solution that uses τ equal to 0.01 is vertically orientated SFMGR operated in the dead-end mode.
essentially grid-independent. The model parameter values cor- These models were based on a clean membrane without any
responding to the velocity values in Table 7 are listed in Table 1, bio-organism; however, since the application of interest of the
and the spatial increments R and Z were chosen to be 0.02. developed models is the optimisation of enzyme production,
B. Godongwana et al. / Journal of Membrane Science 303 (2007) 86–99 97
Appendix A. Vertical orientation (variable shell side ωRx f (Rx )(PL(0) − PS(0) ) sinh ω
pressure) H =b+
[f (Rx ) − Rx ][f (Rx )(f − cosh ω) + Rx (f − 1)]
(A11)
The solutions developed in this section are applicable to a
system with a shell side hydrostatic pressure that is a function the dimensionless shell volumetric flowrate, ΩS , is given by
of the axial position. The treatment is the same as for the con-
stant shell side pressure bioreactor system. The axial and radial 1 R23 1
ΩS = R2 ln R3 +
2
− +
velocity profiles, as well as the luminal volumetric flowrate UL , 2 2 4
VL , ΩL are still governed by the same expressions; i.e., Eqs. ∞
(38), (40) and (51), respectively. The new expressions for the 4 dPS
Bn e−αn τ J1 (αn )
2
+ −b (A12)
shell profiles are obtained, as before, from Eq. (1) neglecting αn b dZ
n=1
the inertial terms:
the stream function in the shell side of the membrane bioreactor
∂vzS 1 ∂ ∂vzS ∂pS
ρ =μ r − + ρgz (A1) is given by
∂t r ∂r ∂r ∂z
dPS 1 4
From Eq. (A1) making use BC7–11, the new expressions for the ψS = −b + R2 − R4 − 2R23 (R22 − R2 )
shell side profiles are given by dZ 16
∞
R22 R2 R22 R2 R R2
1 R 4 −α2n τ − 8R22
ln − − ln −
US = − 2
2R2 ln +R3 −R −
2 2
Bn e J0 (αn R) 2 R3 4 2 R3 4
4 R3 b
n=1
dPS(1) dPS(0)
dPS × − (A13)
× −b (A2) dZ dZ
dZ
where
1 d 2 PS R22 R2
VS = f (R) (A3) = [ln(R3 − R2 ) − ln R3 ](R23 − R22 ) − 2 (R23 − R22 )
16R dZ2 4 8
ARx cosh ωZ CRx sinh ωZ R23 2 R4 − R42
PS = + + HZ + D (A4) + (R3 − R22 ) − 3 (A14)
ω2 f (Rx ) ω3 f (Rx ) 8 16
where A, C, D and H are constants. The corresponding luminal
pressure profile is Nomenclature
A cosh ωZ C sinh ωZ
PL = + + HZ + D (A5) a dimensionless entrance pressure drop
ω2 ω3 A constant in pressure equation (bioreactor of vari-
where able shell side pressure)
b dimensionless function of gravitational accelera-
f (Rx ) − Rx
ω = 16κ
2
(A6) tion
f (Rx )
98 B. Godongwana et al. / Journal of Membrane Science 303 (2007) 86–99
transmembrane pressure and substrate feed concentration on reactor per- [19] S.K.O. Ntwampe, M.S. Sheldon, Quantifying growth kinetics of
formance, J. Membr. Sci. 35 (1988) 325–338. Phanerochaete chrysosporium immobilised on a vertically orientated
[14] W.J. Bruining, A general description of flows and pressures in hollow fiber polysulphone capillary membrane: biofilm development and substrate con-
membrane modules, Chem. Eng. Sci. 44 (1989) 1441–1447. sumption, Biochem. Eng. J. 30 (2006) 147–151.
[15] Y. Moussy, Bioartificial kidney. I. Theoretical analysis of convective flow in [20] E.P. Jacobs, W.D. Leukes, Formation of an externally unskinned polysufone
hollow fiber modules: application to a bioartificial hemofilter, Biotechnol. capillary membrane, J. Membr. Sci. 121 (1996) 149–157.
Bioeng. 68 (1999) 142–152. [21] M. Tien, T.K. Kirk, Lignin peroxidase of Phanerochaete chrysosporium,
[16] K. Damak, A. Ayadi, B. Zeghmati, P. Schmitz, A new Navier–Stokes and Methods Enzymol. 16 (1988) 238–249.
Darcy’s law combined model of fluid flow in crossflow filtration tubular [22] J.D. Anderson, Computational Fluid Dynamics: the Basics with Applica-
membranes, Desalination 161 (2004) 67–77. tions, McGraw Hill, New York, 1995.
[17] M. Elshahed, Blood flow in capillary under starling hypothesis, Appl. Math. [23] R.H. Perry, D.W. Green, J.O. Maloney (Eds.), Perry’s Chemical Engineer’s
Comput. 149 (2004) 431–439. Handbook, 7th ed., McGraw Hill, New York, 1998.
[18] A.S. Berman, Laminar flow in channels with porous walls, J. Appl. Phys. [24] X.-Y. Li, B.E. Logan, Permeability of fractal aggregates, Water Res. 35
24 (1953) 1232–1235. (2001) 3373–3380.