Sei sulla pagina 1di 10

Composites: Part A 40 (2009) 860869

Contents lists available at ScienceDirect

Composites: Part A
journal homepage: www.elsevier.com/locate/compositesa

A general model for the permeability of brous porous media based on uid ow simulations using the lattice Boltzmann method
Aydin Nabovati a,*, Edward W. Llewellin b, Antonio C.M. Sousa a,c
a

Department of Mechanical Engineering, University of New Brunswick, Fredericton, NB, Canada Department of Earth Sciences, Durham University, Durham, UK c Department of Mechanical Engineering, University of Aveiro, Aveiro, Portugal
b

a r t i c l e

i n f o

a b s t r a c t
Fluid ow analyses for porous media are of great importance in a wide range of industrial applications including, but not limited to, resin transfer moulding, lter analysis, transport of underground water and pollutants, and hydrocarbon recovery. Permeability is perhaps the most important property that characterizes porous media; however, its determination for different types of porous media is challenging due its complex dependence on the pore-level structure of the media. In the present work, uid ow in three-dimensional random brous media is simulated using the lattice Boltzmann method. We determine the permeability of the medium using the Darcy law across a wide range of void fractions (0.08 6 / 6 0.99) and nd that the values for the permeability that we obtain are consistent with available experimental data. We use our numerical data to develop a semi-empirical constitutive model for the permeability of brous media as a function of their porosity and of the bre diameter. The model, which is underpinned by the theoretical analysis of ow through cylinder arrays presented by [Gebart BR. Permeability of unidirectional reinforcements for RTM. J Compos Mater 1992; 26(8): 110033], gives an excellent t to these data across the range of /. We perform further simulations to determine the impact of the curvature and aspect ratio of the bres on the permeability. We nd that curvature has a negligible effect, and that aspect ratio is only important for bres with aspect ratio smaller than 6:1, in which case the permeability increases with increasing aspect ratio. Finally, we calculate the permeability tensor for the brous media studied and conrm numerically that, for an isotropic medium, the permeability tensor reduces to a scalar value. 2009 Elsevier Ltd. All rights reserved.

Article history: Received 22 September 2008 Received in revised form 24 February 2009 Accepted 12 April 2009

Keywords: A. Fibres B. Mechanical properties C. Computational modelling E. Resin transfer moulding (RTM)

1. Introduction Permeability prediction, and more generally, the investigation of the effect of pore structure on the bulk properties of porous media, has posed a major challenge to researchers and engineers in a wide range of industrial and academic disciplines. These include, but are not limited to, resin transfer moulding [1,2], biomedical engineering [35], subsurface ow of oil and groundwater [6,7], lter simulation [8,9] and fuel cell simulations [1012]. Macroscopic approaches for uid ow simulation in porous media, either using the Darcy law [13] or more complicated models [14], require as an input the permeability; however, the analysis of the effect of pore-scale parameters on the macroscopic bulk properties is a cumbersome task. The pore structure in porous media is often complex, and complicated ow patterns exist within the pores and between the grains. Consequently, permeability is found to be highly medium-specic, hence there is no general

model for permeability as a function of the bulk properties of a medium. The determination of permeability for a specic material typically requires time-consuming experimental work. Most experimental methods of permeability prediction apply a constant pressure gradient to the porous medium and determine the average ow velocity from the measured uid ow-rate. The mediums permeability is subsequently determined using the Darcy law [13], as follows:

hui

:rp

* Corresponding author. Tel.: +1 506 452 6128; fax: +1 506 453 5025. E-mail address: a.nabovati@unb.ca (A. Nabovati). 1359-835X/$ - see front matter 2009 Elsevier Ltd. All rights reserved. doi:10.1016/j.compositesa.2009.04.009

where hui, K, rp and l are the volume-averaged ow velocity, permeability tensor, pressure gradient vector and the dynamic viscosity of the uid, respectively. This relationship is valid in the creeping-ow regime (Reynolds number ( 1). Earlier studies of porous media ow were conducted experimentally and some of the best-known models for the permeability of porous media are based on experimental data [1315]. In these studies, primarily due to the macroscopic nature of the experimental approach, the details of the pore-scale ow-pattern in the

A. Nabovati et al. / Composites: Part A 40 (2009) 860869

861

porous medium cannot be captured. In general, studying the effect of pore-scale parameters on bulk properties requires a large experimental dataset, which is time-consuming and expensive to generate. As alternative approaches, analytical and numerical methods aim to predict the permeability by solving the uid ow inside the pores of the porous medium. The numerical procedure echoes the experimental approach: a pressure gradient is applied; the uid ow inside the pores of the medium is solved; the permeability of the medium is then determined using the Darcy law (Eq. (1)). This approach has two main advantages over the experimental approach. First, the geometry of the digitally constructed medium can be varied rapidly and arbitrarily. Second, simulated experiments are typically much quicker to run than their laboratory counterparts. Together, these benets allow a more-rapid and more-thorough exploration of parameter space than can be achieved in the laboratory. The very rst challenge in the numerical simulation of permeating ow is that the pore-level structure of the medium is required as input. Imaging techniques, such as computed tomography [16,17], have been developed to capture the complex structure of real porous media; however, these methods may be costly and time consuming and impose limits on the resolution that can be achieved. The alternative approach, which has been widely used in the literature, is to reconstruct the pore-level structure virtually. The level of the reconstructed structures complexity depends on the computational resources available and the nature of the problem under study. The reconstructed medium can be in the form of ordered or random arrangements of two and three-dimensional obstacles. Analytical studies of the pore-level ow, in general, employ the Stokes equation (a simplied form of the NavierStokes equation, which is valid for creeping ow) for a specied domain with periodic boundary conditions. Due to the limitations of the methods applied, the computational domain, which is the building block of the pore structure, is in the form of a simplied, well-dened structure in which the grains of the porous medium are represented in the form of two dimensional obstacles [18,19], ordered sphere packing [20], or ordered packing of cylinders [2124]. Rapid increase in available computing power and the development of advanced numerical algorithms mean that detailed numerical simulations of ow in porous media are now feasible. Removing the constraints of the analytical approaches, more complex pore geometries, which resemble the real porous-media structures more closely, can be used in uid ow simulations. Ordered or random packing of different geometric congurations, such as square blocks, spheres, cylinders, and parallelepipeds [25] have been used in the literature to reconstruct the pore structure. The choice of the constructing elements depends on the nature and application of the porous medium to be modeled. Random arrangements of spheres with mono-dispersed, bi-dispersed, or distributed diameter are often used in simulations of ow in geological materials, including studies of groundwater ow and hydrocarbon recovery, and ow in packed beds, and rocks [6,7,2628]. Simulations of ow in the preform matrix in resin transfer moulding (RTM) [24,29,30], and through paper bres [31,32] and woven materials [9,3336] often use random arrangements of bres. 1.1. Porositypermeability relationships for brous materials The key parameter controlling the permeability of brous materials and, indeed, all porous materials is the porosity / = Vpore/ Vsample, where Vsample is the total volume of the sample and Vpore is the volume not occupied by solid bres. Several workers have published relationships for the permeability of brous materials as a function of their porosity.

Gebart [24] presents a combined theoretical, numerical and experimental investigation of the permeability of ordered arrays of bres. The analytical treatment of creeping ow perpendicular to the long axis of the bres is predicated upon the assumptions that permeability is controlled by the narrow slots formed between the bres at their closest approach and that the width of these slots varies only slowly. These assumptions are most valid in the limit of close-packed bres. Gebart derives the following functional form for K(/):

K C a2

s !5=2 1 /c 1 1/

where a is the bre radius, /c is the critical value of porosity below which there is no permeating ow (the percolation threshold) and C p is a geometric factor (Gebart calculates C 16=9 2p and p p /c = 1 p/4 for a square array, C 16=9 6p and /c 1 p=2 3 for a hexagonal array). Gebart presents numerical results, obtained using a nite difference solution of the NavierStokes equations, that show excellent agreement with the relationship up to at least / = 0.65. Koponen et al. [31] used the LBM on a D3Q19 lattice to study creeping ow through three-dimensional random-bre sheets, analogous to paper and non-woven fabrics. They report that the permeability of such materials is exponentially dependent on the porosity and independent of whether the bres were placed randomly or not. They present an empirical relationship for the permeability as a function of the porosity, based on a t to their data, for porosities in the range 0.4 < / < 0.95:

K 5:55 a2 e10:11/ 1

Clague et al. [37] also studied the permeability of three-dimensional ordered and disordered brous media. They used the lattice Boltzmann method (LBM) on a D3Q15 and D3Q19 lattice to simulate creeping ow through fully three-dimensional random bre networks, in which free overlapping of the bres was allowed. Both wall-bounded and unbounded media were considered and the effect of the wall on the overall permeability of the brous media was investigated. They use a scaling analysis to develop a phenomenological relationship between permeability and porosity for both the bounded and unbounded brous media. For the case of an unbounded medium, they nd:

K b1 a2

s !2 1 /c 1 eb2 1/ 1/

where b1 and b2 are curve tting parameters. For a disordered (random) arrangement of bres, Clague et al. calculate a typical value of /c = 1p/4 % 0.21 but with a 1r variation that gives minimum and p maximum values of /c = 0 and /c 2 p p % 0:4, respectively. They show an excellent t to their data, which span the range 0.33 < / < 0.95 (i.e. they do not cover the region near the percolation threshold). In the present work, uid ow is simulated in 3D random brous media at the pore level. We employ the lattice Boltzmann method for the uid ow simulation and calculate the permeability of the medium using the Darcy law. We cover a wide range of porosity, from near the percolation threshold to very dilute systems (0:08 / 0:99). Based on curve tting of our numerical experimental results, we propose a semi-empirical constitutive relationship for the permeability as a function of porosity. We also investigate the effect of various other pore-level parameters, including the curvature, diameter and aspect ratio of the bres, on the predicted permeability.

862

A. Nabovati et al. / Composites: Part A 40 (2009) 860869

2. Methodology The complex structure of the pore-level geometry, especially in media of low porosity, yields small pores and narrow ow passages. As a rule of thumb, the local ow velocity in these narrow pores is proportional to the volume-averaged ow velocity divided by the porosity. The narrow ow passages and locally high velocities limit the applicability of the conventional computational uid dynamics approaches. Mesoscopic methods such as the smoothed particle hydrodynamics (SPH) [38,39], lattice gas automata (LGA) [40,41] and lattice Boltzmann method (LBM) [4244] have been successfully used for macroscopic uid ow simulations, which require the mesoscopic details of the ow to be considered. The LGA and SPH methods, in their current state-of-the-art, tend to be computationally costly to perform three-dimensional simulations of ow in porous media that are of a size adequate to yield physically meaningful results. Although the early versions of LBM [45] suffered from similar difculties, later developments of the LBM have seen dramatic improvements in the computational efciency, making it a suitable tool for mesoscopic, three-dimensional simulations [42,46,47]. Other attractive features of the LBM are that numerical operations are spatially local, easing implementation, while solid boundaries of arbitrary complexity can be included without performance penalty. Furthermore, the LBM is well suited to ow simulation at the mesoscopic scale, and is amenable to parallelization. These characteristics have made the LBM the most popular method for numerical pore-level analysis; indeed this was one of the rst applications of LBM [48]. The LBM has been shown to be a more efcient tool for ow simulation in such complex geometries than conventional uid dynamic approaches [49,50]. In the present work, three-dimensional uid ow was simulated in brous porous media using LBow,1 an implementation of the single-relaxation-time (SRT) LBM on a D3Q15 lattice [51,52]. As in all such implementations, the ow is represented by the propagation of uid mass through a lattice. The lattice is a discrete representation of physical space; in the present case a three-dimensional cubic lattice. At any time t, uid mass can propagate from a node with position r to any of its six orthogonal neighbours or eight long-diagonal neighbours, or it can remain at the present node. Since time is also discrete, the propagating uid arrives at its new location at time t + 1, hence, there are i = 15 possible uid velocities at each node, represented by the vector ei. The spatial discretization Dx and the time step Dt dene the units of the simulation. The total density at each node is given by:

and average velocity u as the incoming uid masses in the simulation. The propagation and collision steps are encapsulated in the lattice Boltzmann equation:

fi r; t fieq r; t fi r ei Dt; t Dt fi r; t s |{z} |{z}


propagation collision

where s is a relaxation parameter, related to the uid viscosity. In the implementation of the LBM adopted in the present work, ow is driven by imposing a constant, uniform body force G on the uid at every point, which is physically analogous to a gravitational force acting on the uid. This is achieved by adding an extra term gi to each of the mass components fi prior to propagation. The term gi is formulated in such a way that the total mass is conserved but the momentum is adjusted to account for the force acting on the mass at the node for the duration of the time step:

g i r; t

x i Dt
c2 s

G ei

qr; t

X
i

fi r; t

This term is added to the right hand side of Eq. (8). The halfway bounceback method [44] was used to implement the solid wall boundary condition. LBow uses a scripting language to set up the ow simulation; via an interpreted text le, the user can specify the geometry and parameters of the simulation. In this study, the geometries of the porous media of interest are specied as a three-dimensional binary mask of the simulation lattice. Dimensional quantities for the simulation are specied in SI units, rather than simulation units. In this study, the working uid has the properties of water at 20 C: kinematic viscosity m = 1.004 106 m2 s1, density q = 998.29 kg m3. The driving pressure gradient rp is specied in units of Pa m1. Once the ow has settled to steady state, the  average velocity of the uid nodes u is output in SI units. The vol ume-averaged uid velocity hui /u is determined allowing the permeability of the medium to be calculated using the Darcy law, Eq. (1). Note that the choice of the working uid properties is somewhat arbitrary, since the permeability is independent of viscosity, density and driving pressure gradient for creeping ow (i.e. Reynolds number Re ( 1). This was conrmed numerically by repeating a typical simulation multiple times for a range of rp spanning several orders of magnitude, all at low Reynolds number; the calculated permeability was indeed independent of pressure gradient. The uid viscosity is related to the relaxation parameter s of the SRT LBM:

where q is also in simulation units (typically initialized to q = 1 throughout the lattice). Similarly, the average uid velocity at each node is given by:

m s

 1 2 c Dt 2 s

10

P ur; t

i fi ei r; t

qr; t

In addition to the propagation step, at each node r, at each time step t, the incoming uid masses fi undergo collision, in which they relax towards the equilibrium distribution:

fieq

ei u u u ei u2 qxi 1 2 cs 2c2 2c4 s s

"

# 7

where the weights x0 = 2/9, x1. . .6 = p and x7. . .14 = 1/72, and the 2/9 lattice pseudo-sound speed cs Dx= 3Dt. This equilibrium distribution is a discrete analogue of the Maxwell-Boltzmann distribution for a population of uid particles having the same density q
1

Available from http://www.lbow.co.uk.

The relaxation parameter represents the degree to which the uid populations are relaxed towards the equilibrium value during the collision step (Eq. (8)). Consequently, the larger the value of the relaxation parameter, the more rapid is the ow settlement. However, we should note the use of the halfway bounceback method for the solid wall boundary condition with the SRT LBM, yields predictions of the ow eld in porous media, and therefore of their permeability, that are dependent on the choice of the relaxation parameter. Consequently, the accuracy of the simulation is dependent on the relaxation parameter. Pan et al. [53] evaluated different LB methods and solid wall boundary condition treatment methods for ow simulations in porous media. They showed that results obtained using the SRT LBM with the halfway bounceback method for the solid boundaries are in good agreement with the results of the MRT LBM with interpolated bounceback boundaries, and that the effect of the relaxation parameter dependency is negligible for s = 1. We set the relaxation parameter equal to 1 in all the

A. Nabovati et al. / Composites: Part A 40 (2009) 860869

863

Fig. 1. Schematic of two dimensional section through the computational unit cell used for the permeability prediction in hexagonal arrangement of cylinders; ow domain with periodic boundary conditions on both sides is shown in gray colour.

simulations performed. This value represents the best compromise between simulation accuracy and the rate of ow settlement. 3. Results and discussion 3.1. Validation To validate our methodology, we simulate creeping uid ow in a hexagonal array of innite, parallel cylinders. The radius of the cylinders was kept constant and equal to 40 lattice units in a dop main of l 3l 1, where l was varied between 82 and 230 lattice units; this yields a medium of porosity between 0.15 and 0.89. Fig. 1 depicts the schematic of the computational domain unit cell with periodic boundary conditions on both sides; the gray colour identies the simulation domain. Due to the invariance of the domain geometry under translation in the z-direction, a fully 3D simulation can be achieved with a domain 1 lattice unit thick in that direction by using periodic boundary conditions. The results of these simulations are presented in Fig. 2, which compares the permeability, calculated according to the methodology presented in Section 2, with the value determined analytically using the relationship for this geometry presented in [24] (Eq. (2)). This model, described in Section 1.1, has been widely used for ow in regular arrays of bres [54,55]. The agreement over the range of porosities for which that relationship is valid, i.e. 0.10 < / < 0.65, is excellent. It is noteworthy that the t between the data and the analytical solution is good even beyond the upper limit of validity claimed by Gebart (dotted line in Fig. 2). 3.2. Flow in brous media To create the brous medium structure, randomly oriented, straight, cylindrical bres of constant diameter are randomly
10
2

placed in a cubic domain with free overlapping. By allowing the bres to overlap freely, we are able to investigate ow in brous media with porosities across the full range, right down to the percolation threshold. Media in which the bres are not able to overlap have a minimum porosity, which is higher than the percolation threshold. Koponen et al. [31] studied the uid ow in three dimensional bre webs, where exible bres were placed randomly in the computational domain without overlapping, the minimum reported porosity was higher than 0.4. Nabovati and Sousa [56] investigated the permeability of sphere packs with and without free overlapping. They found that overlapping has a negligible impact on permeability for media with porosities higher than 0.85 and leads to a decrease of less than 35% in permeability for low porosity media. The minimum reported porosity for random packing of spheres without overlapping was 0.55. The computational domain comprises a cube of 128 lattice sites on each side and the periodic boundary condition was applied to all faces of the cube. The bres extend to the boundaries of the domain. The bres are placed in the computational domain by using the following algorithm: (1) a random position vector is chosen within the computational domain, or one bre radius of it, as the origin of the bres core, (2) a random vector representing the spatial orientation of the bre core is generated, (3) the bre core line is extended from the origin point along the randomly determined orientation in both directions until it intersects the domain boundaries, and (4) lattice sites that are closer to the core line than the radius of the bre are designated as solid. The radius of the bres is constant along their length and equal to 2, 3, 4, 5 or 6 lattice units, Dx, depending on the experimental run. Fig. 3 shows a high porosity sample of the reconstructed brous medium; the porosity of the sample is 0.80.

Fig. 3. Reconstructed medium with straight bres and porosity / = 0.80; the radius of the bres is four lattice units and their length is such that they span the computational domain.

10
K/a 2

10

-2

10

-4

Present Work Gebart [24]

10

-6

0.1

0.2

0.3

0.4

0.5 0.6 Porosity ( )

0.7

0.8

0.9

Fig. 2. Normalized permeability, calculated using the methodology presented in Section 2, is plotted with the analytically-determined relationship (Eq. (2)) presented by Gebart in [24] for a hexagonal arrangement of solid cylinders. Gebarts relationship is shown dotted outside the range of values of / for which Gebart claims validity.

864

A. Nabovati et al. / Composites: Part A 40 (2009) 860869

Fig. 4. Velocity vectors in a slice of the three-dimensional brous medium; the porosity of the sample is 0.2 and the pressure gradient drives ow in the positive x direction.

A pressure gradient is applied in the x-, y- or z-direction and the velocity eld is calculated in the pores of the reconstructed medium. Fig. 4 shows the velocity vectors in a slice of a three-dimensional brous medium which has a porosity equal to 0.2; the pressure gradient, and the mean ow direction, are in the positive x-direction. We should note that, whilst some pores appear to be dead-end or isolated in the two-dimensional slice, in the third dimension, these pores are effectively connected to each other and they contribute to the uid ow pattern. As a result, the percolation threshold reported to be equal to 0.33 for two-dimensional media made up of random arrangements of square obstacles [57], in fact, takes a lower value for three-dimensional media. In the present study, permeating pathways for uid ow exist for porosity as low as 0.08. 3.2.1. Effect of bre radius Permeability has dimensions of length squared, and it is usual to normalize permeability by the square of a length scale that is characteristic of the system; for brous media, this is typically the bre radius. To validate this approach, we determine the permeability of random networks of bres with radii of 2, 3, 4, 5 and 6 lattice units. Results are shown in Fig. 5. Fig 5 demonstrates: (1) permeability increases as porosity increases for constant bre radius; (2) permeability increases as bre radius increases for constant porosity. Fig. 5b demonstrates that normalizing the permeability by the square of the bre radius causes the curves for different bre radius to collapse onto a single curve, hence, that this non-dimensionalization is appropriate. Note that the normalized permeability is not completely independent of bre radius; the two show an inverse relationship with the normalized permeability for the thinnest bre typically about 20% higher than for the fattest on average. We postulate that this dependence is due to the nite length of the bres in our simulations giving rise to a variable aspect ratio as the bre radius is increased. This is supported by the results presented later in Section 3.2.4. 3.2.2. Constitutive permeabilityporosity relationship for random bre networks We determine the permeability of around 50 random bre networks in the x-, y- and z-direction (giving $150 data points in to-

tal). For each network, the bres all have the same radius: either 2 or 4 lattice nodes. The permeability of the networks is in the range 0.08 < / < 0.99. Results are plotted in Fig. 6 (note the logarithmic scale on the permeability axis). The normalized permeability is approximately exponentially dependent on porosity in the range 0.2 < / < 0.9. At lower and higher porosities, the dependence is stronger. As expected, the permeability tends towards innity in the limit / ? 1 and drops towards zero at low (but nite) porosity. We note that the variation in the data is greatest at very low porosities, where random differences in the placement of the bres between different domains lead to large relative changes in the predicted permeability. We nd that a modied version of the Gebart [24] relationship provides an excellent t to data across the full range of porosity (Fig. 6). We adapt Gebarts original relationship (Eq. (2)) by allowing the three constants it contains to vary:

K C1 a2

s !C 2 1 /c 1 1/

11

where /c is the critical value of porosity above which permeating ow can occur (the percolation threshold). C1 and C2 are related to the geometry of the network (compare with the values determined by Gebart for a regular array of bres, presented in Section 1.1). We use the freely-available statistical analysis package SimFit2 to t Eq. (11) to the data presented in Fig. 6. The t is carried out in logarithmic space to avoid biasing the t towards large values of permeability at high porosity. Table 1 shows the best t values obtained (R2 = 0.999). The values are consistent with those determined analytically by Gebart for a hexagonal array of aligned bres (Section 1.1); differences are due to the different geometries of the networks. The value of /c falls within the range calculated, on statistical grounds, by Clague et al. [37], for a random network of bres (Section 1.1). The published relationships of Koponen et al. [31] and Clague et al. [37] Eqs. (3) and (4), respectively in Section 1.1 were similarly adapted and tted to the data in Fig. 6. The Koponen relationship provided a very poor best t to data, and was abandoned. The
2

Available from: http://www.simt.man.ac.uk.

A. Nabovati et al. / Composites: Part A 40 (2009) 860869


-5 1

865

10 10 10 10 10

10 a = 2 lu 3 4 5 6

-6

10

K (m 2 )

-7

K/a 2

10

-1

-8

-9

10

-2

10

-10

0.2

0.4

0.6 0.8 Porosity ( )

10

-3

0.2

0.4

0.6 0.8 Porosity ( )

(a)

(b)

Fig. 5. (a) Numerically-determined permeability as a function of the porosity for bres with a range of radii. Within each simulation suite, all bres have the same radius; (b) numerically-determined permeability normalized by bre radius as a function of the porosity.

10

10

K/a

10

-2

10

-4

Numerically Predicted Values Proposed Relation, Eq. 11 10


-6

0.1

0.2

0.3

0.4

0.5 0.6 Porosity ( )

0.7

0.8

0.9

Fig. 6. Dimensionless permeability as a function of porosity for random networks of straight bres. Fibre radius is 2 or 4 lattice units in all cases. The solid line shows the t of our semi-empirical relationship (Eq. (11)) which is adapted from Gebart [24]. Best t parameters are given in Table 1. The t (R2 = 0.999) indicates a percolation threshold of /c % 7.4%.

Table 1 Best t parameters of Eq. (11) to data presented in Fig. 6; regression coefcient R2 = 0.999. See main text for details of tting procedure. Parameter /c C1 C2 Best t value 0.0743 0.491 2.31

best t of the Clague relationship was as good as the Gebart t. We favour the Gebart relationship because it has a sound theoretical basis, whereas the Clague relationship is purely phenomenological. To assess the variation of the predicted permeability values due to the random nature of the reconstructed brous porous media, we performed multiple repeated simulations for media at three different porosities: $0.1, $0.5 and $0.95, respectively. Due to the discrete nature of the bre placement procedure, it is not possible to create different random media with exactly the same porosity; hence, there is a variation of around 0.5% for each prescribed porosity. For each porosity, we created 21 media with different random bre placement and simulated ow in x, y, and z directions, resulting in 63 permeability determinations for each porosity. Each dataset is separately plotted against porosity in Fig. 7; the permeability predicted using Eq. (11), with the parame-

ter values presented in Table 1, is shown on each plot as a solid line. Table 2 shows the mean permeability for each dataset and the permeability normalized by the value predicted using Eq. (11). The standard deviation is also presented, expressed as a percentage of the mean permeability. From Fig. 7 and Table 2, it can be seen that both the extent of the variation in permeability and the quality of t of Eq. (11) depend on the porosity. The standard deviation is small (around 10% of the mean) for mid and high-range porosity; at very low porosity, the standard deviation is larger (around 50% for / % 0.1), reecting the large changes in permeability that arise from small structural differences near the percolation threshold. Eq. (11) has a tendency to slightly under-predict permeability at low porosity and to over-predict at high porosity. Given the $6 order of magnitude difference in permeability between / % 0.1 and / % 0.95, however, we consider Eq. (11) to provide an accurate and exible tool for permeability prediction across the porosity range and note that the model value is within one standard deviation of the numerical data across the porosity range. The literature contains permeabilityporosity data from a number of laboratory investigations into brous media; Jackson and James [23] provide a summary. In Fig. 8, we compare our permeabilityporosity relationship (Eq. (11)) with experimental data for high porosity brous media with randomly oriented straight

866

A. Nabovati et al. / Composites: Part A 40 (2009) 860869

9.0 8.0 7.0


2

0.060 0.055 0.050 K/a 0.947 0.948 0.949 Porosity ( ) 0.950 0.045 0.040
2

K/a

6.0 5.0 4.0 0.946

0.035 0.030 0.497 0.498 0.499 Porosity ( ) 0.500

(a)
1.4E-04 1.2E-04 1.0E-04 K/a
2

(b)

8.0E-05 6.0E-05 4.0E-05 2.0E-05 0.0E+00 0.0995 0.0996 0.0997 0.0998 0.0999 Porosity ( ) 0.1

(c)
Fig. 7. Dimensionless permeability for three different porosities: (a) $0.95, (b) $0.5, and (c) $0.1. At each porosity, the permeability was determined for 21 different random networks of straight bres to determine the random variation in permeability. The solid line represents Eq. (11) with the parameters given in Table 1. Note the dramatically expanded scales compared with Fig. 6. See Section 3.2.2 for discussion.

Table 2 Mean permeability and standard deviation for three different porosities: $0.1, $0.5, and $0.95. See Fig. 7. Porosity $0.1 $0.5 $0.95 Mean permeability (K/a2) 2.58 105 4.62 102 7.50 Mean permeability (normalized to Eq. (11)) 2.09 1.01 0.877 Standard deviation (%) 52 10 11

10 10 10 10

K/a

10 10

-1

-2

cylindrical bres of constant diameter presented in that work. The data were obtained using a range of experimental methods, a broad spectrum of brous materials including lter pads, nylon bres, Kapron bres, collagen, metal bres and polymer bres, and various working uids including water, glycerol and air. Despite the diversity of these investigations, there is a broad agreement (within an order of magnitude) among the resulting datasets. Overall the agreement between Eq. (11) and the experimental data is good, especially for porosities higher than approximately 0.75. 3.2.3. Effect of bre curvature We investigate the effect of bre curvature on the permeability of brous media, by replacing straight bres with randomly curved bres as the constituting elements of the medium. We generate the bres by constructing a cylinder of constant diameter around a randomly curved bre core. The origin of the bres is chosen randomly. Three different third-order polynomials represent the three coordinates dening the orientation of the bre core and they are a function of a single variable, t:

10 10

-3

-4

Chen (1995) Ingmanson et al. (1959) Kirsch & Fuchs(1967) Stenzel et al. (1971) Kostornov & Shevchuck (1977) Jackson and James (1982) Proposed relation, Eq. 11
-3 -2 -1 0

10

10

10 Solid Fraction (1- )

10

Fig. 8. Comparison of dimensionless permeability of brous media, calculated using Eq. (11) with the parameters given in Table 1, with the experimental data reported by Jackson and James [23]. Readers are directed to that work for full references to the original experimental studies.

bre reaches the domain boundaries, producing a smoothly curved bre, which crosses the domain. Fig. 9, shows the permeability we determine from simulations of ow in media composed of curved bres. The results are almost indistinguishable from the results for straight bres; and the effect of the bres curvature on the overall permeability of the medium can be considered negligible. 3.2.4. Effect of bre aspect ratio We investigate the effect of bre aspect ratio a (length to diameter ratio of straight cylindrical bres of nite length) on permeability for a range of aspect ratios between 1 and 20, for two different values of porosity, namely 0.51 and 0.73. The bres diam-

xi ai;3 t 3 ai;2 t 2 ai;1 t ai;0

i 1; 2; 3

12

where the coefcients of these polynomials are chosen randomly in the range of [1,1], this ensures that bres are smoothly curved and avoids tight spirals. The variable t is incremented and decremented in appropriate steps to extend the bre in both directions until the

A. Nabovati et al. / Composites: Part A 40 (2009) 860869


2

867

10

Straight Fibres - Equation 11 Curved Fibres 10


0

K/a 2
10
-2

10

-4

0.2

0.3

0.4

0.5 0.6 Porosity ( )

0.7

0.8

0.9

Fig. 9. Numerically-determined permeability of brous media as a function of the porosity for media formed of straight (solid line, Eq. (11) with parameters given in Table 1) and randomly curved bres (circles, simulation results).

eter is constant and equal to 6 lattice units and the aspect ratio is varied by changing the bres length. The numerically-determined values of the permeability are shown in Fig. 10 as a function of the bre aspect ratio; the permeability values are normalized by the permeability calculated using Eq. (11). It can be seen that permeability increases with aspect ratio for a < 6; for a > 6, permeability is largely independent of aspect ratio. This nding explains the imperfect collapse of the data for bres of varying radius, presented in Fig. 2, when normalized by the bre radius. Since the length of the bres is nite limited by intersection with the domain boundaries the aspect ratio of the bres in that suite of simulations varies somewhat with bre radius. For a cylinder of diameter d, the ratio of the surface area to the 1 volume of the bre (specic surface area) is equal to 1 4 a. An d increase in aspect ratio therefore yields a decrease of the specic surface area of the bres, leading to a reduction in the frictional drag force and higher values for the permeability. 3.2.5. Permeability tensor Any experimental or numerical attempt to determine the permeability tensor and the principle directions of a porous medium should be performed in three dimensions. This is an expensive and time consuming task. Neuman [58] proved analytically that the permeability tensor is a symmetric second order tensor, which has six distinct elements in general. For isotropic materials, the three diagonal elements are equal and non-zero and all off-diagonal elements are zero. In this case, the permeability can be represented as a single, scalar value. Ahn et al. [59], Weitzenbck et al. [60,61] and Parnas et al. [62] discuss experimental methods and approaches for permeability tensor measurements. Kolodziej et al. [63] analytically investigated the permeability tensor of high porosity brous porous media. In their study, bres had a unidirectional arrangement with non-uni-

form spacing. Nedanov and Advani [64] studied uid ow in twoscale brous porous media numerically; governing equations of the uid ow around and inside the solid and permeable bres were developed using the homogenization method and were solved numerically. They report the diagonal elements of the permeability tensor for single and three-ply fabrics. Song et al. [35] calculated the permeability tensor of a three-dimensional, woven brous medium using the nite volume method. In this study, to evaluate the viability of the LBM for permeability tensor determination, the permeability tensor elements of three samples of different porosities of 0.90, 0.70 and 0.50, respectively, are determined based on the simulated ow eld. For each porosity, a pressure gradient is applied in one of the three principle directions. The mean ow is in the direction of the applied pressure gradient, which builds the diagonal element of the permeability tensor in the specied direction. We calculated the off-diagonal elements of the permeability tensor using the net ow in each of the other two directions, and repeated this procedure for each of the three directions to obtain the nine elements of the permeability tensor. The predicted permeability tensors for these media are as follows:

6 7 K0:90 4 0:0073 1:0526 0:0138 5 107 m2 2 0:0206 0:0138 1:0820 0:0021 1:1643 0:0262 0:0021 0:0006 3

1:0866 0:0073

0:0203

6 K0:70 4 0:0262 1:1726 1:8851 0:0478 6 4 0:0478 1:9017 0:0395 0:0029 2

7 0:0006 5 108 m2

1:1632

K0:50

3 0:0395 7 0:0029 5 109 m2

1:8484

1
K / K model (Eq. 11)

0.9 0.8 0.7 0.6 0.5 0 5 10 Fiber Aspect Ratio 15


Phi = 0.505 Phi = 0.730

As the media are created with randomly-oriented bres, there is no preferential ow direction and the media are expected to be isotropic. In each case, we nd that the diagonal elements of the permeability tensor are the same to within $3%; this small difference can easily be accounted for by the random nature of the network. Furthermore, the off-diagonal elements are approximately symmetrical, and are smaller than the diagonal elements by around two orders of magnitude. These results support the analysis of Neuman [58] and demonstrate that it is valid to report a scalar value for permeability of these random porous media, and a determination of the full second order tensor is unnecessary. 4. Conclusion Three-dimensional uid ow simulations in brous media are conducted using the SRT LBM; the brous media are reconstructed by random placement of cylindrical bres, with random

20

Fig. 10. Effect of the bre aspect ratio on the numerically-determined permeability for two different porosities, 0.51 and 0.73, respectively; permeability values are normalized by the permeability calculated using Eq. (11) with the parameters given in Table 1.

868

A. Nabovati et al. / Composites: Part A 40 (2009) 860869 [13] Vafai K. Handbook of porous media. Taylor & Francis; 2005. [14] Vafai K, Tien CL. Boundary and inertia effects on ow and heat transfer in porous media. Int J Heat Mass Transfer 1981;24(2):195203. [15] Bear J, Bachmat Y. Introduction to modeling of transport phenomena in porous media. Dordrecht: Kluwer Academic Publishers; 1990. [16] Hazlett RD. Simulation of capillary-dominated displacements in microtomographic images of reservoir rocks. Transport Porous Med 1995;20(12):2135. [17] Kerckhofs G, Schrooten J, Van Cleynenbreugel T. Validation of X-ray microfocus computed tomography as an imaging tool for porous structures. Rev Sci Instrum 2008;79(1):013711. [18] Larson RE, Higdon JJL. Microscopic ow near the surface of two-dimensional porous media. Part 1: axial Flow. J Fluid Mech 1986;166:44972. [19] Larson RE, Higdon JJL. Microscopic ow near the surface of two-dimensional porous media. Part 2: transverse ow. J Fluid Mech 1987;178:11936. [20] Sangani AS, Acrivos A. Slow ow through a periodic array of spheres. Int J Multiphase Flow 1982;8(4):34360. [21] Sangani AS, Acrivos A. Slow ow past periodic arrays of cylinders with application to heat transfer. Int J Multiphase Flow 1982;8(3):193206. [22] Drummond JE, Tahir MI. Laminar viscous ow through regular arrays of parallel solid cylinders. Int J Multiphase Flow 1984;10(5):51540. [23] Jackson GW, James DF. Permeability of brous porous media. Can J Chem Eng 1986;64(3):36474. [24] Gebart BR. Permeability of unidirectional reinforcements for RTM. J Compos Mater 1992;26(8):110033. [25] Nabovati A, Sousa ACM. Fluid ow simulation at open-porous medium interface using the lattice Boltzmann method. Int J Numer Method Fluids 2008;56(8):144956. [26] Van Der Hoef MA, Beetstra R, Kuipers JAM. Lattice-Boltzmann simulations of low-Reynolds-number ow past mono- and bidisperse arrays of spheres: results for the permeability and drag force. J Fluid Mech 2005;528:23354. [27] Yang A, Miller CT, Turcoliver LD. Simulation of correlated and uncorrelated packing of random size spheres. Phys Rev E 1996;53(2):1516. [28] Martys NS, Torquato S, Bentz DP. Universal scaling of uid permeability for sphere packings. Phys Rev E 1994;50(1):4038. [29] Yang J, Jia Y, Sun S. Mesoscopic simulation of the impregnating process of unidirectional brous preform in resin transfer molding. Mater Sci Eng A 2006:51520. [30] Shojaei A, Ghaffarian SR, Karimian SMH. Modeling and simulation approaches in the resin transfer molding process: a review. Polym Compos 2003;24(4): 52544. [31] Koponen A, Kandhai D, Hellen E. Permeability of three-dimensional random ber webs. Phys Rev Lett 1998;80(4):716. [32] Koponen A. Simulations of uid ow in porous media by lattice gas and lattice Boltzmann methods. PhD dissertation, Jyvaskyla, Department of Physics: University of Jyvaskyla; 1998. [33] Wang Q, Maze B, Tafreshi HV. A note on permeability simulation of multilament woven fabrics. Chem Eng Sci 2006;61(24):80858. [34] Wang Q, Maze B, Tafreshi HV. On the pressure drop modeling of monolament-woven fabrics. Chem Eng Sci 2007;62(17):481721. [35] Song YS, Chung K, Kang TJ. Prediction of permeability tensor for three dimensional circular braided preform by applying a nite volume method to a unit cell. Compos Sci Technol 2004;64(1011):6291636. [36] Song YS, Youn JR. Asymptotic expansion homogenization of permeability tensor for plain woven fabrics. Compos A: Appl Sci Manuf 2006;37(11):20807. [37] Clague DS, Kandhai BD, Zhang R. Hydraulic permeability of (un)bounded brous media using the lattice Boltzmann method. Phys Rev E 2000;61(1): 61625. [38] Jiang F, Sousa ACM. Smoothed particle hydrodynamics modeling of transverse ow in randomly aligned brous porous media. Transport Porous Med 2008;75(1):1733. [39] Jiang FM, Oliveira MSA, Sousa ACM. Mesoscale SPH modeling of uid ow in isotropic porous media. Comput Phys Commun 2007;176(7):47180. [40] Frisch U, Hasslacher B, Pomeau Y. Lattice-gas automata for the NavierStokes equation. Phys Rev Lett 1986;56(14):15058. [41] Krafczyk M, Schulz M, Rank E. Lattice-gas simulations of two-phase ow in porous media. Commun Numer Methods Eng 1998;14(8):70917. [42] Qian YH, DHumieres D, Lallemand P. Lattice BGK models for NavierStokes equation. Europhys Lett 1992;17(6):47984. [43] Chen S, Doolen GD. Lattice Boltzmann method for uid ows. Ann Rev Fluid Mech 1998;30:32964. [44] Succi S. The lattice Boltzmann equation for uid dynamics and beyond. Oxford: Oxford University Press; 2001. [45] McNamara GR, Zanetti G. Use of the Boltzmann equation to simulate latticegas automata. Phys Rev Lett 1988;61(20):23325. [46] Higuera FJ, Succi S, Benzi R. Lattice gas dynamics with enhanced collisions. Europhys Lett 1989;9(4):3459. [47] Higuera FJ, Jimenez J. Boltzmann approach to lattice gas simulations. Europhys Lett 1989;9(7):6638. [48] Succi S, Foti E, Higuera F. Three-dimensional ows in complex geometries with the lattice Boltzmann method. Europhys Lett 1989;10(5):4338. [49] Bernsdorf J, Durst F, Schfer M. Comparison of cellular automata and nite volume techniques for simulation of incompressible ows in complex geometries. Int J Numerical Method Fluids 1999;29(3):25164.

orientations, within the computational domain. The radius, curvature and length of the bres are varied systematically. We nd that dividing the permeability by the square of the bre radius yields an appropriate non-dimensionalization. We also nd that bre curvature has a negligible impact on the permeability of the medium. For bres of nite length with aspect ratios smaller than $6, permeability increases with increasing aspect ratio; this effect is negligible for values of the aspect ratio greater than 6. The permeability values that we obtain are compatible with the available experimental data and, based on the determined values, we develop a relationship (Eq. (11)) for the permeability as a function of the materials porosity and the bre diameter. The relationship that we present results from a semi-empirical parameterization of a published analytical relationship for ordered arrays of bres [24]. The t of the relationship to the data is excellent across the range of porosities investigated (0.08 < / < 0.99). Our analysis shows that the percolation threshold for a threedimensional network of randomly oriented bres is /c = 0.0743 . The media in this study were created randomly, hence there was no preferential direction for the ow through the medium; therefore, the media can be assumed isotropic. Prediction of the permeability tensor for media of three different porosities supported this assumption, as the diagonal elements of the numerically-determined permeability tensors differ by less than $3%, and the off-diagonal elements were two orders of magnitude smaller than the average value of the diagonal elements. This nding indicates that it is valid to report a scalar value, instead of a second order tensor, for the permeability of the brous media studied. The results obtained in this study and the general relationship proposed for the permeability, can be fed to the macroscopic ow modelling approaches for the industrial applications, e.g. resin transfer moulding process, where the pore level approach is not applicable due to the large computational resources requirement. Acknowledgment The authors acknowledge the support received from NSERC (Natural Sciences and Engineering Research Council of Canada) Discovery Grant 12875 (ACMS) and from the Foundation for Science and Technology (FCT, Portugal) through the research grant POCTI/EME/59728/2004 (ACMS). EWL is supported by NERC (UK) Research Fellowship NE/D009758/2. We thank two anonymous reviewers for their helpful comments. References
[1] Phelan Jr FR. Simulation of the injection process in resin transfer molding. Polym Compos 1997;18(4):46076. [2] Abrate S. Resin ow in ber preforms. Appl Mech Rev 2002;55(6):57999. [3] Khanafer K, Vafai K. The role of porous media in biomedical engineering as related to magnetic resonance imaging and drug delivery. Heat Mass Transfer 2006;42(10):93953. [4] Vafai K, Ai L. A coupling model for macromolecule transport in a stenosed arterial wall. Int J Heat Mass Transfer 2006;49(910):156891. [5] Yang N, Vafai K. Modeling of low-density lipoprotein (LDL) transport in the artery-effects of hypertension. Int J Heat Mass Transfer 2006;49(56):85067. [6] Pan C, Hilpert M, Miller CT. Pore-scale modeling of saturated permeabilities in random sphere packings. Phys Rev E 2001;64(6):066702-1. [7] Pan C. Use of pore-scale modeling to understand transport phenomena in porous media. PhD dissertation in environmental science and engineering. University of North Carolina at Chapel Hill; 2003. [8] Lu WM, Tung KL, Hwang KJ. Fluid ow through basic weaves of monolament lter cloth. Textile Res J 1996;66(5):31123. [9] Tung KL, Shiau J, Chuang C. CFD analysis on uid ow through multilament woven lter cloths. Separat Sci Technol 2002;37(4):799821. [10] Koido T, Furusawa T, Moriyama K. An approach to modeling two-phase transport in the gas diffusion layer of a proton exchange membrane fuel cell. J Power Sources 2008;175(1):12736. [11] Djilali N. Computational modelling of polymer electrolyte membrane (PEM) fuel cells: challenges and opportunities. Energy 2007;32(4):26980. [12] Wang L, Afsharpoya B. Modeling uid ow in fuel cells using the latticeBoltzmann approach. Math Comput Simul 2006;72(2-6):2428.

A. Nabovati et al. / Composites: Part A 40 (2009) 860869 [50] Breuer M, Bernsdorf J, Zeiser T. Accurate computations of the laminar ow past a square cylinder based on two different methods: lattice-Boltzmann and nite-volume. Int J Heat Fluid Flow 2000;21(2):18696. [51] Llewellin EW. LB ow: an extensible lattice Boltzmann framework for the simulation of geophysical ows. Part I: theory and implementation. Comput Geosci, accepted for publication. [52] Llewellin EW. LBow: an extensible lattice-Boltzmann framework for the simulation of geophysical ows. Part II: usage and validation. Comput Geosci, accepted for publication. [53] Pan C, Luo L-S, Miller CT. An evaluation of lattice Boltzmann schemes for porous medium ow simulation. Comput Fluids 2006;35(89): 898909. [54] Wang JF, Hwang WR. Permeability prediction of brous porous media in a biperiodic domain. J Compos Mater 2008;42(9):90929. [55] Grujicic M, Chittajallu KM, Walsh S. Lattice Boltzmann method based computation of the permeability of the orthogonal plain-weave fabric preforms. J Mater Sci 2006;41(23):79898000. [56] Nabovati A, Sousa ACM. LBM mesoscale modeling of porous media. Int Rev Mech Eng 2008;2(4).

869

[57] Koponen A, Kataja M, Timonen J. Permeability and effective porosity of porous media. Phys Rev E 1997;56(3):331925. [58] Neuman SP. Theoretical derivation of Darcys law. Acta Mech 1977;25(34): 15370. [59] Ahn SH, Lee WI, Springer GS. Measurement of the three-dimensional permeability of ber preforms using embedded ber optic sensors. J Compos Mater 1995;29(6):71433. [60] Weitzenbock JR, Shenoi RA, Wilson PA. Measurement of three-dimensional permeability. Compos Part A: Appl Sci Manuf 1998;29(12):15969. [61] Weitzenbock JR, Shenoi RA, Wilson PA. A unied approach to determine principal permeability of brous porous media. Polym Compos 2002;23(6):113250. [62] Parnas RS, Liu Q, Giffard HS. New Set-Up for in-plane permeability measurement. Compos Part A (Appl Sci Manuf) 2007;38(3):95462. [63] Kolodziej JA, Dziecielak R, Konczak Z. Permeability tensor for heterogeneous porous medium of bre type. Transport Porous Med 1998;32(1):119. [64] Nedanov PB, Advani SG. Numerical computation of the ber preform permeability tensor by the homogenization method. Polym Compos 2002;23(5):75870.

Potrebbero piacerti anche