Composites: Part A 37 (2006) 20572068 www.elsevier.

com/locate/compositesa

Permeability characterization of dual scale brous porous media

Nina Kuentzer a, Pavel Simacek a, Suresh G. Advani
a

a,*

, Shawn Walsh

Department of Mechanical Engineering, Center for Composite Materials, University of Delaware, Newark, DE 19716, USA b Army Research Laboratory, Aberdeen Proving Ground, MD 21005-5066, USA Received 17 June 2005; received in revised form 7 December 2005; accepted 7 December 2005

Abstract The proper characterization of fabrics used in liquid composite molding (LCM) is integral to accurately model the ow through these porous preforms. The dual-scale nature of many fabrics has brought about a need for a methodology, which characterizes not only the bulk permeability of the preform, but the micro-scale permeability of the ber tows. These two permeability values can then be used in LCM simulations that can separately track the bulk ow front progression and the saturation of the ber tows in preforms that exhibit dual-scale porosity. A three dimensional simulation called liquid injection molding simulation (LIMS) has been developed at the University of Delaware that can predict the impregnation of the ber preform with resin in closed molding processes such as resin transfer molding (RTM) and vacuum assisted resin transfer molding (VARTM). To address the dual-scale porosity, standard 2D or 3D mesh elements are combined with 1D elements, which are attached at each node and represent the ber tows. This implementation allows for the interactions between the bulk and micro ow, and can predict the saturation of the ber tows, along with the movement of the bulk resin ow front. However, it does require two permeability inputs: one for the elements representing the bulk preform and another for the 1D elements representing the ber tows. A methodology is proposed to determine the bulk permeability and a parameter that is closely associated with the micro permeability of the tows for dual-scale fabrics. This is accomplished by comparing the inlet pressure proles of one-dimensional constant ow rate injection RTM experiments with a simulation of ow in a dual-scale fabric. The methodology is validated and characterizations for four dierent fabrics are performed to demonstrate the versatility and limitation of the methodology. 2006 Elsevier Ltd. All rights reserved.
Keywords: A. Fabrics/textiles; E. Cutting

1. Introduction It has been known for some time that woven and stitched fabrics inherently induce a dual-scale ow behavior [1,2]. These porous preforms are comprised of distinct bulk- and micro-ow regions, which exhibit distinctly different impregnation rates and, therefore, each correspond to a dierent permeability value. The pores of the bulk region will ll more rapidly, while the micro space between bers within the tows, which are bundles of individual bers, will saturate more slowly, due to the smaller size
Corresponding author. Tel.: +1 302 831 8975; fax: +1 302 831 8149. E-mail addresses: advani@me.udel.edu, advani@udel.edu (S.G. Advani). 1359-835X/$ - see front matter 2006 Elsevier Ltd. All rights reserved. doi:10.1016/j.compositesa.2005.12.005
*

of the pores. The size of the pores between the tows, the diameter of the bers within a tow, and the architecture of the preform all play a role in the determination of the preform permeability. The distinction between these separate bulk- and microow regions is important, due to the delay that occurs in the complete saturation of the preform. The bulk or the macro ow front rst lls the more permeable pores between the tows; then, the uid from the bulk region lls the less permeable pores within the tows, at which point the more slowly advancing micro ow front dictates the lling progression. For this reason, the location of the bulk ow front seldom coincides with the boundary of the preform that is fully saturated with uid. For such preforms, a partially-saturated zone exists and can be dened as the region

2058

N. Kuentzer et al. / Composites: Part A 37 (2006) 20572068

Saturated

Ls

Unsaturated

saturated bulk permeability as the tows are completely lled while in the later case it is the unsaturated bulk permeability as the tows are completely dry. For the linear cases outlined above, the conventional Darcys law is valid and can be expressed as h md i K rp . l 1

Fiber tows Fully-saturated front Flow front

Fig. 1. A schematic that depicts the saturated, partially saturated (Ls), and unsaturated ow regions within dual-scale preforms, and highlights the delayed impregnation of the ber tows.

This relation expresses the average Darcy ow velocity vector, h md i, as a function of the permeability tensor of the preform, K , the viscosity of the test uid, l, and the pressure gradient, $p. When applying (1) to a planar onedimensional ow experiment, Darcys law can be simplied to 1D and re-expressed as hmd i K dp ; l dx 2

in which the pores of the bulk region may be lled, but the ber tows are only partially lled. In one-dimensional ow, one can dene a partially-saturated length, Ls, which is the distance between the bulk ow front and the fullysaturated front, as depicted in Fig. 1. The tows ahead of the bulk ow front are completely unsaturated and the tows behind the fully-saturated front are completely saturated. The partially-saturated length decreases as the micro permeability of the ber tows gets closer to the bulk permeability of the preform. Single-scale fabrics, such as most random mats, only have one length scale and, therefore, only one permeability value. They are the most extreme case where the micro permeability is inherently equal to that of the bulk permeability, since there are no distinct ber tows; accordingly, these fabrics have no partially-saturated length. In order to characterize the permeability of a dual-scale fabric preform, the pressure history at the injection gate of a one-dimensional, linear constant ow rate experiment is used to determine the bulk permeability and a parameter that is associated with the tow permeability. Darcys law, which is used to describe ow through porous media, must be correctly applied to the bulk and the tow regions due to the presence of a delayed impregnation in the tows for such fabrics. For single-scale fabrics, the inlet pressure for onedimensional constant ow rate experiment exhibits a linear behavior with time. When experimentally evaluating dualscale fabrics, this linear trend only holds for two limiting cases. On one extreme the permeability of the tows may be of the same order as the bulk permeability; therefore, both regions ll simultaneously and the pressure as a function of time grows linearly. At the other extreme, the permeability of the tows may be many orders of magnitude smaller than the bulk permeability, in which case the tows remain completely empty (act as solid rods) in the time it takes to ll the bulk regions. For this case too, the pressure as a function of time grows linearly, but more quickly than the previous case. The permeability that is measured from the linear pressure prole for the former case represents the

where the bulk permeability in the ow direction, K, is a scalar value. Since the test uid is injected into the preform through a mold cross-section, A, at a constant ow rate, Q, the denition of velocity, hmd i Q , can additionally be reA lated to Darcys law. The 1D Darcy relation can then be re-cast in terms of the changes in the injection pressure, Pin, and the position of the bulk ow front, x, with respect to time: hmd i K m ; x l d dt 3

P in where dd m is the slope of the inlet pressure prole with t respect to time for a constant ow rate injection experiment. To enable one to nd the bulk permeability, the ow front velocity, m, is expressed in terms of the Darcy average velocity by dividing it by the available porous volume, /, as dictated by mass conservation:

dx h md i mff . dt /

Eq. (4) is substituted into Eq. (3) and the denition of velocity, Q , is implemented. The rearrangement of terms A results in an explicit expression for the bulk permeability of the preform in the direction of the ow:  2 l Q K . 5 m/ A Clearly only one permeability value can be ascertained from Eq. (5). The micro-scale impregnation is in eect neglected, which is a valid assumption for a single-scale preform or a dual-scale preform in which the tows exhibit the behavior of the two limiting cases mentioned in which the tows are either completely unsaturated or completely saturated. For these cases, one expects that for a constant ow rate injection, the change in pressure at the inlet with respect to time will be linear, resulting in a constant value for the slope, m. It is evident from Eq. (5) that a steeper slope will result in a lower permeability value. When the ber tows of a

N. Kuentzer et al. / Composites: Part A 37 (2006) 20572068

2059

preform remain completely unsaturated as uid ows through a dual-scale fabric, the pressure grows linearly and quickly. The dry tows are providing additional resistance, resulting in a steeper slope and hence a lower permeability value. If the tows are completely saturated, it is suspected that the uid inside the tows is stagnant and only the uid outside the tows is owing. There is less resistance to the ow in this case since the tows are lled with uid and provide lubrication for the moving liquid. The pressure still grows linearly, but not as quickly as in the former case, resulting in a less steep slope and hence a larger permeability value as compared to the unsaturated case. Such ndings were also reported by Breard et al. [3]. The region between the two extreme linear cases represents the true nature of the ow. The tows at the inception may be completely unsaturated; therefore, the pressure prole with respect to time at the inlet follows the linear trajectory of a completely unsaturated preform. As the ber tows start to draw the resin, this prole becomes non-linear and once the tows are completely saturated will approach the linear behavior exhibited by a completely saturated preform. The impregnation of uid into the ber tows during this non-linear phase is addressed by applying mass conservation with a sink term to the bulk ow: r h m d i q p ; s . 6 The ber tows act as sinks, q, which drain away uid from the advancing bulk ow front and these sinks are a function of the pressure, p, and the degree of saturation, s. The saturation, or percent of lled ber tows, is driven by the pressure of the resin surrounding the tow, in addition to capillary wicking eects; the wicking eects are assumed to be negligible over the lling period. If the ber tows ll at the same rate as the regions around them the sink term is zero; therefore, the mass conservation equation is homogeneous and will simplify to the single scale porous media physics of r hmd i 0. 7

Fig. 2. Bulk- and micro-ow interactions are modeled in LIMS by attaching 1D elements (which represent the ber tows) to the 2D or 3D elements (which represent the bulk preform).

impregnation has on the injection pressure, which can be monitored at the inlet injection node. This numerical methodology is completely outlined in the work of Simacek and Advani [4]. This model is chosen, as it allows one to calculate the pressure prole at the inlet by specifying the bulk permeability and a tting parameter related to the tow permeability. In this manuscript, a methodology is introduced that allows one to estimate the bulk permeability and the micro-permeability parameter of dual-scale preforms that can be used in LIMS, by matching the inlet pressure prole from LIMS, to the data obtained from experiments. 2. Background 2.1. Modeling Modeling the dual-scale nature of brous preforms by implementing not only the bulk ow, but also the micro ow, is not a new trend. Parnas and Phelan [1] recognized, early on, the importance of evaluating the global ow front along with the local impregnation of the ber tows. The model incorporated sinks, which take away uid from the advancing ow front, and was a key development since it coupled the ow at dierent length scales by combining 1D radial ow into the tow with 1D linear ow along the length of the preform. Sadiq et al. [2] experimentally investigated the model, and observed key trends, which reinforced the importance of the need for such models. Many further models have been developed, which evaluate the unsaturated ow through porous preforms and

Eq. (7) also describes the mass balance inside individual ber tows, because during the lling of the individual tows no uid sink is present at the micro scale. This important sink term is incorporated into the nite element/control volume based numerical simulation called LIMS that predicts the ow of resin in brous preforms [4]. This is accomplished in the LIMS environment by attaching 1D elements to each node of the 2D or 3D elements. The 1D elements represent the micro-ow regions and 2D or 3D elements represent the bulk regions as shown in Fig. 2. The two types of elements are assigned separate permeability values corresponding to the bulk permeability of the preform and the micro permeability of the ber tow, respectively. Since the 1D elements share the nodes and consequently pressure values with the 2D or 3D mesh, the micro ow is coupled with the bulk ow. The saturation of the tows can be tracked, as well as the impact this

2060

N. Kuentzer et al. / Composites: Part A 37 (2006) 20572068

explore how the bulk ow behavior is inuenced by the impregnation of the ber tows [513]. Such models have the ability to be transformed into ow simulations, in order to numerically incorporate the sink that causes the non-linearity in the pressure prole; however, for accurate predictions one would have to supply two values of the permeability, one for the bulk ow and one for the impregnation of the tows. 2.2. Experimental investigations Experiments to determine the in-plane bulk permeability have been developed and various issues in measurement techniques have been addressed [5,9,1422]. For dual-scale preforms, Parnas et al. [14] conducted 1D linear and 2D radial ow experiments at a constant injection pressure. The unsaturated and saturated permeability values for random mat and woven fabric were compared; a dierence between the two values was observed. Luce et al. [15] extended this work by conducting similar experiments, but utilizing preforms composed of two types of fabric. The impact of compaction of the preform was additionally addressed. Binetruy et al. [5] also conducted constant injection pressure experiments and highlighted the dierence between the saturated and unsaturated regions of the preform. They concluded that the micro ow did not impact the saturated region, but that the unsaturated region is impacted by the impregnation of the tows and accordingly aects the overall permeability of the preform. Slade et al. [9] conducted 1D linear experiments at a constant ow rate, in order to evaluate the resulting pressure prole over time. The pressure results were used to determine both the unsaturated and saturated permeability values of random mat and stitched biaxial fabric. Babu and Pillai [16] also examined the pressure plots resulting from 1D constant ow rate experiments; this work studied what eect compression has on the pressure prole of biaxial stitched, triaxial stitched, biaxial woven, and unidirectional fabrics.

However, no method has been established in which both the bulk and micro permeability values can be determined from one experiment. This paper proposes a methodology to recover both the bulk permeability and the micro-permeability parameter of dierent preforms by conducting one RTM constant ow rate injection experiment. The experimental data is used to analytically determine the bulk permeability and is compared to the numerical results generated by LIMS to determine the micro-permeability parameter. Both values can be used in LIMS environment to accurately predict the ow in closed mold processes. 3. Experimental procedure: bulk permeability determination In order to measure the permeability of dierent preforms, one-dimensional constant ow rate experiments are conducted. A closed mold is used, which consists of one 1/2 in.-thick aluminum lower mold half and one 2 in.-thick clear acrylic top mold half that acts as a viewing window. A 1/8 in.-thick aluminum spacer separates the two halves and provides the cavity in which the layers of fabric are placed. O-rings are tted to grooves machined in both mold halves in order to seal the mold and prevent any uid from leaking. The mold is secured with 26 bolts, which are equally spaced and torqued. The mold is additionally reinforced to minimize any deections by clamping stiening bars, which stand 2 in. on end, to the mold exterior. A schematic of the mold is shown in Fig. 3. In order to achieve an experiment free of race-tracking along the edges of the preform, care is taken when cutting each fabric layer. Race-tracking occurs when a channel of lower resistance is created along the edges of a preform within a mold cavity, as a result of frayed fabric edges or incorrect preform size [17]. A frame that is identical in dimensions to the mold cavity is used as a cutting template to ensure the width of the layers is as exact as possible. The fabric is cut to ensure the inlet port is left uncovered so that the test uid rst lls the empty space within the cavity and

Aluminum spacer plate Aluminum bottom plate Acrylic top plate

Fabric preform

Cross bolts

Inlet port

Steel reinforcing bars

SIDE VIEW

TOP VIEW

Fig. 3. Schematic of mold set-up for 1D ow experiments.

N. Kuentzer et al. / Composites: Part A 37 (2006) 20572068

2061

initiates a planar one-dimensional ow through the preform. The vent port is also not covered by the fabric, so air can escape without any resistance or restriction. The cut preforms are 8 1/8 in. wide and 12 in. long in the ow direction. Each experiment is videotaped to highlight three important criteria: (a) the progression of the ow front is straight, even, and steady; (b) the ow is free of edge eects such as race-tracking; and (c) no layers are washed away by the ow. The transparent top plate facilitates this procedure. The bulk permeability measurements can be conducted in the presence of race-tracking [17], but the numerical LIMS analysis to which the experimental data is compared, assumes no edge eects; therefore, the experiments are conducted to match this assumption and if this criterion is not met, the data is discarded. The data is also discarded if one of the other two criteria is not met. Vegetable oil, either pure or colored by the addition of an oil-based dye, is used as the test uid in this study. The viscosity of the oil is measured before each experiment using a Brookeld viscometer. A Radius brand injection machine moves a piston via a stepper motor, in order to provide a constant ow rate. Once the residual gas is expelled from piston cylinder, the oil is injected into the closed mold through the inlet port. The inlet pressure is recorded using a pressure transducer connected to the Radius injection system. The pressure data is collected over the duration of the experiment, which is complete once the bulk ow reaches the end of the fabric and the vent. The injection is continued and the resin is allowed to bleed through the vent until the pressure reaches a steady state value, signifying the complete saturation of the fabric. A schematic of the 1D controlled ow system is shown in Fig. 4. For each experiment, the inlet pressure data is plotted against time. The plot begins at time zero, which corresponds to when the oil enters the preform; the plot ends when the bulk ow front reaches the end of the preform at time, tf. The plot, an example of which is shown in Fig. 5, can be subdivided into three regions.

Pressure (psi)

(i) t=0 Time (s)

(ii)

(iii) tf

Fig. 5. Dual-scale fabrics exhibit three ow regions; (i) an initial linear region, (ii) a non-linear region, and (iii) a nal linear region, which represents the ow that has a constant partially-saturated length.

Linear region (i) is representative of the time frame during which uid lls the bulk region, yet no micro pores have had time to be impregnated. Non-linear region (ii) is representative of the pressure response as the uid of the bulk ow front begins to impregnate the ber tows. Linear region (iii) is representative of the period during which a constant partially-saturated length ensues. As previously discussed, there are two ow regions in dual-scale fabrics: the ow through the larger pores of the bulk region and the ow through the smaller pores within the tows. In essence, the ow through both of these regions initially develops, but after a certain time frame becomes fully developed. The micro pores are more dicult to impregnate, so the time it takes the ow to fully develop will be longer than the time it takes the ow in the bulk region to fully develop. In order for the constant partially-saturated length to exist in region (iii), the mold must be long enough to ensure that the micro ow front fully develops, ensuring the presence of the fully-saturated front. Once both fronts fully develop, the fully-saturated front and the bulk ow front are then traveling at the same speed. The complete mathematical proof of this phenomenon is available in the Appendix of [23]. The partially-saturated length, which is the distance between the two fronts, is thus constant and the associated pressure drop is linear.

Fig. 4. Schematic of 1D controlled ow system used to determine permeability values of fabric preforms.

2062

N. Kuentzer et al. / Composites: Part A 37 (2006) 20572068

Using an excel template, the nal experimental data point is xed and preceding data points are added as to maximize the t (R2 value) of the linear best t line passing through the data points. The distinction between region (ii) and (iii) can thus be made, since the addition of non-linear points from region (ii) would not improve, but impede the t. Linear region (iii) is used to determine the bulk permeability of the preform. The slope of the best t line in region (iii), which is also determined by the excel template is used directly in Eq. (5) to determine the bulk permeability. Accordingly, a minimum requirement of the experiment is that an appropriate combination of ber volume fraction, viscosity, and ow rate are used in order to yield an experiment in which the tows of the fabric begin to ll and the constant partially-saturated length region is reached. The pressure plot therefore begins linearly, becomes non-linear, and nally reaches a linear slope during the nal portion of the experiment. If this requirement is not met, the bulk permeability value of interest cannot be determined, since region (iii) will not exist. 4. Numerical procedure: micro-permeability parameter determination The 1D constant ow rate RTM experiment is modeled to simulate the experimental conditions and parameters. Pressure data is generated and collected over the course of the injection, just as if a physical experiment were being conducted. A rectangular mesh that is composed of square, 2D elements is generated to replicate the preform contained within the mold used in the experiment. The mesh size is 83 elements in length by 60 elements in width. These 2D elements are assigned a ber volume fraction, V bulk , and a perf meability value, Kbulk, within LIMS, both of which correspond to the bulk-impregnation regions of the preform. One-dimensional bar elements are then attached to every 2D element node. These 1D elements are assigned a ber volume fraction, V tow f , and a permeability value, Ktow, representing the micro-impregnation regions within the ber tows. By the addition of these 1D elements to the
Table 1 2D and 1D element parameters corresponding to LIMS ow simulation Fabric types 2D elements V bulk (%) f

existing 2D elements, the interactions between the bulkand micro-ow regions are enabled. A length, l, is set for each 1D element, each of which also has an associated initial default cross-sectional area, Atow. The area is adjusted to Atow-adj. by the LIMS model, based on the set l and V tow and the input V bulk : f f  bulk   Vf 2 Atowadj: V bulk ; 8 l V tow f is where Vbulk is the volume of a 2D element. Next, V bulk f adj: adjusted to V bulk in order to accommodate for the addif tional volume of the tows, which are attached to the existing mesh; this preserves the overall porous volume of the mesh:    1 V tow bulkadj: bulk f Vf Vf 12 . 9 V tow f The set length of the 1D elements is chosen based on the assumption that the tow permeability is four to ve orders of magnitude smaller than the bulk permeability. The bulk permeability values are of the order 1E9 and 1E10 m2 and accordingly, when l is set to the order 1E3 m, the resulting tow permeability is of the order 1E14 m2. The specic value of l does not matter, since the value of the tow permeability would just need to be accordingly altered to reach the same Ktow/h2 parameter. But, the representative length of the 1D elements is chosen such that the tow permeability is in line with values which make physical sense. The parameters used to run each simulation, as well as the values of the parameters that are modied, are compiled in Table 1. The complete details governing the saturation program approach are explicitly outlined in [4]. For each experiment, a LIMS le of the mesh is created to match the experimental parameters. The viscosity, l, of the test uid is rst input into LIMS. Then, V bulk is set and f Kbulk, which is determined from the experimental pressure prole over time, as discussed in the previous section, is assigned to the 2D elements. The 1D elements are next attached to each node of the 2D element mesh. The constant ow rate injection simulation is nally conducted in LIMS, where the constant ow rate boundary condition

1D elements
adj: V bulk (%) f

V tow (%): 80 f l (m): 0.002 Atow (m2): 0.001 Atow-adj. (1E03 m2) 0.0191 0.0191 0.0167 0.0167 0.0387 0.0387 0.0148 0.0188

Woven E-glassWGa WGb Woven carbonWCa WCb Stitched E-glass 1SG1a SG1b Stitched E-glass 2SG2a SG2b

40 40 35 35 54 54 31 39

60.0 60.0 52.5 52.5 81.0 81.0 46.5 59.1

N. Kuentzer et al. / Composites: Part A 37 (2006) 20572068

2063

is imposed at the edge nodes of the 2D mesh, which are along the width of the preform. In order to determine the micro-permeability parameter, the pressure and corresponding time data that is output from the simulation is matched to the experimental data by changing only a tting parameter in the simulation associated with the tow permeability, Ktow/h2, [4]. This parameter lumps together the ber tow permeability, the cross-section characteristic dimension of the ber tow, h, and possibly a shape factor of the ber tow cross-section. The ber tow is assumed to have a rectangular cross-section; therefore, the rounded corners that are not taken into account for an elliptical cross-section are lumped into the parameter. The designation h is equivalent to l/2. This is done since the 1D elements ll in only one direction in the simulation, whereas lling would occur in two directions in a physical rectangular tow. The actual permeability, Ktow, of the micro-ow regions inside the tows is not evaluated. The parameters set in the simulation for the 1D elements, such as l and thus Atow-adj., only estimate the actual physical parameters of the fabrics used in the experiments. Since a non-dimensional tting parameter is achieved, either h or Ktow can be altered to achieve the best t. But, as long as the same input parameters are used in the simulation environment with the experimentally obtained value of Ktow/h2, the bulk- and micro-ow interactions can be properly modeled. Initially, the Ktow/h2 parameter is estimated and assigned to the 1D bar elements; many simulations are conducted from which a general starting point is established in order to reduce the number of necessary iterations. The corresponding inlet pressure prole that is generated is compared with the experimental data. The next value is selected, by either doubling or halving the guess, depending on how the data from the initial guess compares with the experimental data. Fig. 6 illustrates the two possible scenarios; case (a) shows the simulation plot lies above the experimental plot, which corresponds to an initial guess that was too small and therefore must be doubled, while case (b) shows the opposite case which corresponds to a guess for the tow-permeability parameter that was too large and must be halved. The direction of increasing tow-permeability parameter is shown to clarify the choice,

either doubling or halving, according to this bisection method. This iterative process is continued until two curves corresponding to two dierent tow-permeability parameters ank the experimental data curve. This resulting interval is then subdivided into multiple equally spaced segments. Each region is dened by a tow-permeability parameter and the pressure and time data that corresponds to each value is simulated in LIMS. Multiple segments are generated, as opposed to purely bisecting the interval, so that the minimization location can be graphically veried. The generated time, tj, and the corresponding pressure data, pnum (tj) is input into an excel template, where j begins at the rst data point and increments to the nal number of data points, N. The pressure values are rst interpolated at each experimental data time step, so both sets of data can be compared at the same value of time:     tj1 texp texp tj pnum texp pnum tj pnum tj1 . t j 1 t j t j 1 t j 10 The experimental time, texp, falls between the numerical time steps, tj and tj+1: tj 6 texp 6 tj+1. A simple point wise linear regression scheme is then implemented. The template is used to locally determine the set of numerically generated pressure data that best ts the experimental pressure data, pexp. The dierence in the pressures is taken at each time step, squared, and summed over the length of the experiment from texp = 0 to tf:
tf X texp 0

pnum texp pexp texp .

11

The segment where this error is globally minimized is then identied. The new interval is further subdivided into equally spaced segments. The segments are progressively rened as the procedure is iterated in order to converge on the value of the tow-permeability tting parameter. 5. Validation The model of the one-dimensional constant ow rate RTM experiment is created in LIMS and used as a tool

Pressure (psi)

Increasing Ktow/h2

Case (a): guess too small

Pressure (psi)

Case (b): guess too large

(a)

Time (s)

(b)

Time (s)

Fig. 6. During estimation of the value of Ktow/h2 two scenarios arise: case (a) the guess for Ktow/h2 is too small and case (b) the guess for Ktow/h2 is too large. Bisection method is used to converge on the best match.

2064

N. Kuentzer et al. / Composites: Part A 37 (2006) 20572068

to validate the methodology for determining the bulk permeability and the micro-permeability parameter. Both of these values are chosen and assigned to the 2D and 1D elements in the nite element mesh, respectively. The one-dimensional constant ow rate simulation is then performed using LIMS. The pressure plot is constructed as a function of time, by recording the pressure output from the simulation at an inlet node. This plot is then viewed as if it was obtained from a physical experiment. This experiment is then processed, as if it were any typical physical experiment. The bulk permeability is determined by using the slope of the linear region (iii) shown in Fig. 5, as discussed in the experimental procedure. The micro-permeability parameter is determined by following the sequence of steps outlined by the owchart in Fig. 7. The values obtained for the bulk permeability and the micro-permeability parameter are then compared with the selected values to verify the methodology. The methodology is tested for dierent ratios of permeabilities to ensure that it is suciently robust. The bulk permeability value is selected to be of the order of 1E09 m2, to which permeability values of dierent orders, 1E12 m2, 1E13 m2, 1E14 m2, and 1E15 m2 are assigned to the 1D elements. This is done to ensure the methodology has the ability to recover the permeability values, irrespective of the permeability ratio. Both values can be exactly extracted when imposing the above methodology. One limitation, which is exposed, is that the values cannot be recovered if no tows ll with uid, as in the case when the permeability value of the 1D elements is of the order 1E15 m2. This result corroborates the minimum requirement discussed in the numerical procedure: if the tows of the preform do not even begin to saturate, a saturated bulk permeability cannot be accurately determined. Numerical issues also arise when one goes beyond six orders of mag-

nitude dierence between permeability assigned to 2D elements that represent the bulk perform and the 1D elements that represent the ber tows; but, this is usually unlikely in most performs, as this would preclude lling of the ber tows.

6. Experimental results Four fabrics are evaluated in this study: 9 oz/yd2 biaxial woven carbon (WC), 24 oz/yd2 biaxial woven E-glass (WG), and two types of biaxial stitched E-glass, 96 (SG1) and 18 (SG2) oz /yd2, respectively. Dierent fabrics are evaluated to show how they deviate from a baseline single-scale medium and exhibit dual-scale behavior. Fig. 8 highlights the dierent architectures of the four dierent fabrics. Random mat fabric is included as a reference fabric, since it is known to exhibit a single-scale Darcian ow behavior. The random mat preform has only one length scale and accordingly exhibits a linear pressure prole. The woven and stitched fabrics have slight non-linearities and bend away from the single-scale trend over the duration of the experiment, since these preforms have two length scales. The non-linearity is induced by the presence of the ber tows, which act as sinks into which the uid of the advancing ow front ingresses. The nonlinearity in these preforms is slight, but not insignicant. This transition of the pressure from a linear, to a nonlinear, and back to a linear prole, as is described in Fig. 5 and as can be seen in Fig. 9, enables the determination of the micro-permeability tting parameter that can be used in LIMS to properly model the ow inside the tows along with bulk ow through dierent fabrics. The normalized plots for both the reference random mat fabric

1. INPUT estimated value of Ktow /h2 into LIMS mesh. 2. RUN 1D constant flow rate flow simulation using LIMS. 3. COMPARE simulated pressure plot over time to experimental plot.
No

2. RUNfor all segments. 3. COMPARE for all segments.

Is Ktow /h too small?

No

Yes

Double value.

Does new Ktow /h2 = previous Ktow /h2?

Yes

4. IMPLEMENT interpolation and linear regression scheme for all.

Is Ktow /h2 too large?

Yes

Halve value.

Subdivide region equally.

5. DETERMINE region where Ktow /h2 is minimized.

Fig. 7. Flowchart of procedure to determine the tow-permeability parameter.

N. Kuentzer et al. / Composites: Part A 37 (2006) 20572068

2065

Fig. 8. The four fabrics that are characterized: (i) 9 oz/yd2 biaxial woven carbon, (ii) 24 oz/yd2 biaxial woven E-glass, (iii) 96 and (iv) 18 oz/yd2 biaxial stitched E-glass.

1.0 0.8 P/Pmax 0.6 0.4 0.2 0.0 0.0 0.2 0.4 T/Tfill
Fig. 9. A woven glass fabric has a non-linear pressure prole as compared to a linear pressure prole of random mat.
RM-experimental WGb-experimental WGb-LIMS

sents the ratio of the bulk permeability to the micro-permeability parameter is tabulated:  2  h . 12 K K bulk K tow The partially-saturated length, Ls, generated by LIMS and 2 a non-dimensional constant Ls , which aides in addressing K what parameters are impacting the partially-saturated length, are also incorporated into the results table.
1.0

0.6

0.8

7. Discussion The data sets corresponding to the rst three types of fabric (WG, WC, and SG1) are determined from experiments conducted at the same ber volume fraction. In the case of the stitched glass preforms, the bulk permeability varies approximately 30%. This variation decreases to approximately 15% in the case of the woven glass ber and is only a few percent in the case of the carbon ber preforms. Other researchers have also reported that permeability values exhibit a large scatter from one experiment to the next [18,20]. The way the fabric layers nest together plays an important role in these variances. Layers may end up placed either bundle to bundle or bundle to gap, as shown in Fig. 10. One conguration may preferentially allow uid to more easily permeate through the bulk region as compared to the other, which causes the deviations. In the case of the carbon, seven plies in the mold result in a ber volume fraction of 35%, where as only ve plies of glass occupy 40% of the mold. Deviations in stacking are thus not impacting the bulk region of the carbon to the extent

and an example of one woven E-glass fabric are shown in Fig. 9 to highlight the dierences in the inlet pressure proles. Multiple experiments are performed for each type of fabric to evaluate repeatability. Four types of fabrics that all exhibit dual-scale porosity are characterized and the results of two data sets are included for each fabric. Table 2, which uses the same fabric acronyms as Table 1, outlines the bulk permeability values that are determined, as well as the values of the micro-permeability parameter that are obtained from matching the experimental data to the data output from LIMS. In addition, the value K* that repre-

Table 2 Bulk permeability, micro-permeability parameter, K*, Ls, and Ls2/K* results for each experiment Fabric WGa WGb WCa WCb SG1a SG1b SG2a SG2b Vf (%) 40 40 35 35 54 54 31 39 Kbulk 1E10 (m2) 12.6 10.7 8.6 8.3 20.6 14.2 12.7 7.2 Ktow/h2 1E08 2.3 3.7 1.3 2.3 8.6 15.5 1.8 3.3 K* (m2) 0.055 0.029 0.066 0.036 0.024 0.009 0.071 0.022 Ls (m) 0.257 0.188 0.272 0.199 0.210 0.129 0.267 0.163 Ls/L 0.892 0.653 0.944 0.691 0.729 0.448 0.927 0.566 Ls2/K* 1.21 1.22 1.12 1.10 1.84 1.82 1.01 1.22

Fig. 10. When fabric layers are placed together they stack either (A) bundle to bundle or (B) bundle to gap, creating dierent preferential ow paths.

2066

N. Kuentzer et al. / Composites: Part A 37 (2006) 20572068

the glass does. The thicker and less pliable E-glass plies cannot mesh as well as the carbon plies, and more preferential bulk ow regions are therefore likely to exist. Additionally, more uid is going into the tows in the case of the glass, which impacts the overall bulk permeability; the uid travels more readily through the smaller proportion of bulk porous volume. The glass is thus more permeable than carbon, though there is less porous volume for the uid to ow through. For both stitched fabrics, the tows of one layer are perpendicular to the tows of the preceding layer. The rst type of stitched glass preform (SG1) is made up of seven joined layers, while the second type of stitched glass preform (SG2) is made up of only two layers. Thus, for SG1, only two plies are needed to achieve a ber volume fraction of 54% within the mold, whereas four and ve layers of the SG2 fabric occupy only 31% and 39% of the mold. The individual SG1 plies are accordingly less exible and have much more freedom in how they nest with the subsequently-stacked ply, causing the greater bulk permeability deviations from one experiment to the next. The micro-permeability parameter cannot be directly compared from one fabric to the next, since the parameter is impacted by size of individual bers and ber tows, the shape of ber-tow cross-section and the ber volume fraction of the ber tow. Accordingly, the Atow-adj. values are dierent in the LIMS environment for each fabric. But, the parameter can be compared between the experiments for each type of fabric, since the experiments are conducted at the same ber volume fraction and the simulations are conducted corresponding to the same Atow-adj. parameters. A trend which is exposed is that less permeable bulk regions correspond to a more permeable parameter in terms of the micro pores. A hypothesis for this result is that as the oblong tows are also compressed and become wider and thinner in cross-section, the pores of the bulk region reduce in size. Thus, it is harder to inltrate the pores of the bulk region, but the tows are able to ll more easily, since the uid does not need to travel as far to reach all portions of the tow. The second set of stitched glass fabric data (SG2), which is determined from experiments conducted at two dierent ber volume fractions, also demon-

strates the trend that the micro-permeability parameter increases, as the bulk preform becomes less permeable. As the number of plies increases, the bulk permeability also decreases as anticipated. When comparing the tow-permeability parameter relevant at the micro scale to an equivalent parameter pertinent at the bulk scale, namely Kbulk/L2, where L is the length of the preform, the ratio of the two values is of interest. Table 3 outlines the separate values, in which the nal data column reveals the ratio of the parameter from the micro to the bulk region. This ratio reveals information regarding the dierent times scales in the two ow regions. The tows may need between 1.18 and 9.05 times as long to ll as the bulk regions, depending on the architecture of the fabric. This ratio is important, as it alludes to necessary processing requirements of the dierent fabrics. Though these time requirements are not directly applicable for the characterization of the materials, they must be considered when resin is infused into the fabrics, so as to ensure all regions are saturated with uid before the part cures. As the bulk ow front progresses along the length of the preform, a region near the inlet gate exists in which at least 99% of the micro pores are saturated. Additionally, a region near the vent exists where less than 1% of the micro pores are saturated. The region in between, is the partiallysaturated region. As depicted in Fig. 11, each shade correlates to a varying degree of saturation, and all shades represent the partially-saturated length, Ls. This length identies the region in which ow is continuing to ingress into the ber tows and lies between one and 99% saturation: 1% < Ls < 99%. The degree of saturation output from LIMS has the ability to be mapped to a contour plot; this is depicted in Fig. 11. The partially-saturated length, Ls can be clearly measured from the 99% to the 1% demarcation on the left side and right side, respectively. The values less than 1% are not included in the value of Ls, because they are deemed completely unsaturated, just as the values above 99% are not included, as they are considered completely saturated. This partially-saturated length is generated through LIMS since it is not possible to know exactly which regions are

Table 3 The ratio of the tow-permeability parameter to an equivalent parameter relevant at the bulk scale reveals information regarding the time scales in the two dierent ow regions Fabric WGa WGb WCa WCb SG1a SG1b SG2a SG2b Kbulk (m2) 1.E10 12.6 10.7 8.6 8.3 20.6 14.2 12.7 7.2
K bulk L2

Ktow (m2) 1.E14 2.30 3.70 1.30 2.30 8.60 15.50 1.80 3.30

K tow h2

K tow L2 h2 K bulk

1.5E08 1.3E08 1.0E08 1.0E08 2.5E08 1.7E08 1.5E08 8.7E09

2.3E08 3.7E08 1.3E08 2.3E08 8.6E08 1.6E08 1.8E08 3.3E08

1.51 2.87 1.25 2.30 3.46 9.05 1.18 3.80

N. Kuentzer et al. / Composites: Part A 37 (2006) 20572068

2067

trends arising from Table K* vs. Ls/L and Ls2/K* vs. V bulk f 2 are plotted in Fig. 12. The average summed least squares error for each type of fabric was found to be 4.38 (WC), 1.45 (WG), 0.26 (SG1), and 0.58 (SG2). These results lead to the hypothesis that the experimentally obtained pressure proles of the stitched glass fabrics provide the best t for the model. It can be observed too, that the three fabrics with the lowest error are all comprised of E-glass. The smaller diameter of the individual carbon bers within each tow, which correspond to smaller pores, may alter the sink strength in a manner not captured by the model to the extent the sink strength of the glass tows has the ability to be captured in the current LIMS model. 8. Conclusions
Fig. 11. An example of a contour plot that enables the determination of the partially-saturated length (Ls), which is the region between 1% and 99% saturation.

saturated to which degree from the current experimental set-up. If K*, which is the ratio of the bulk permeability to the tow-permeability parameter, increases, one would expect Ls to increase as well, as Ls will be impacted by changes in either the bulk permeability or the micro-permeability parameter. For example, K* increases if Ktow/h2 is reduced. A lower value of the tow-permeability parameter increases the partially-saturated length Ls. This is an anticipated result, since the partially-saturated region should increase as the micro-permeability parameter moves further away from the value of the bulk permeability. On the other hand, Ls will decrease as the micro-permeability parameter moves closer and closer to the value of the bulk permeability. This result is again anticipated, since no partially-saturated length would be present in a single-scale medium. The notion of the partially-saturated length provides valuable information about the experiments that are conducted. Ls2/K* is an important parameter since it remains steady for experiments conducted on the same type of fabric, at the same ber volume fraction. As the ber volume fraction in the mold increases, results for all four dierent types of fabric that possess very dierent architectures show that the Ls2/K* parameter also increases. Both the
0.08 0.07

By reproducing the injection pressure history at the inlet of a one-dimensional constant ow rate injection experiment using LIMS numerical analysis in which 1D elements represent the ber tows and 2D elements represent the bulk preform, a methodology has been developed, which determines both the bulk permeability and a tow-permeability parameter for that ber preform. This model conrms with experiments that although Darcys law is linear, non-linear behavior is exhibited as a uid injected at a constant rate ows into a dual-scale fabric in which the tow and bulk permeabilities are dierent. This resulting non-linear behavior is due to the way the uid at the bulk ow front enters the ber tows at a delayed pace; the tows are accordingly considered uid sinks. The pores at the micro scale are much smaller and cannot be inltrated as easily as the pores of the bulk region. The notion of the partially-saturated length has also provided valuable information about the experiments that have been conducted. By knowing the partially-saturated length, along with the bulk permeability of the preform, the tow-permeability parameter can be obtained by using the knowledge that the non-dimensional Ls2/K* term is a constant for a specic fabric. In addition, one could gauge the relative time scales to inltrate bulk regions and tows in a fabric at dierent ber volume fractions. Experimental data, along with numerically generated LIMS data, provide a methodology to couple the bulk- and micro-ow
2.0 1.8
Ls 2 / K*

K* (m2)

0.06 0.05 0.04 0.03 0.02 0.01 0.00 0.4

1.6 1.4 1.2 1.0 0.8

0.6 Ls/L

0.8

1.0

30%

35%

40% Vf

45%

50%

55%

Fig. 12. Trends exhibited by parameters K* vs. Ls/L and non-dimensional parameters Ls2/K* vs. Vf.

2068

N. Kuentzer et al. / Composites: Part A 37 (2006) 20572068 [11] Pillai KM. Governing equations for unsaturated ow through woven ber mats. Part 1. Isothermal ows. Compos Part A: Appl Sci Manuf 2002;33(7):1007. [12] Takano N, Zako M, Okazaki T, Terada K. Microstructure-based evaluation of the inuence of woven architecture on permeability by asymptotic homogenization theory. Compos Sci Technol 2002; 62(1011):1347. [13] Grujicic M, Chittajallu KM, Walsh S. Eect of shear, compaction and nesting on permeability of the orthogonal plain-weave fabric preforms. Mater Chem Phys 2004;86(23):358. [14] Parnas RS, Howard JG, Luce TL, Advani SG. Permeability characterization part 1: A proposed standard reference fabric for permeability. Polym Compos 1995;16(6):42945. [15] Luce TL, Advani SG, Howard JG, Parnas RS. Permeability characterization part 2: Flow behavior in multiple-layer preforms. Polym Compos 1995;16(6):44658. [16] Babu BZ, Pillai KM. Experimental investigation of the eect of bermat architecture on the unsaturated ow in liquid composite molding. J Compos Mater 2004;38(1):5779. [17] Lawrence JM, Barr J, Karmakar R, Advani SG. Characterization of preform permeability in the presence of race tracking. Compos Part A: Appl Sci Manuf 2004;35(12):1393. [18] Luo Y, Verpoest I, Hoes K, Vanheule M, Sol H, Cardon A. Permeability measurement of textile reinforcements with several test uids. Compos Part A: Appl Sci Manuf 2001;32(10):1497. [19] Han KK, Lee CW, Rice BP. Measurements of the permeability of ber preforms and applications. Compos Sci Technol 2000;60(1213): 2435. [20] Hoes K, Dinescu D, Sol H, Vanheule M, Parnas RS, Luo Y, Verpoest I. New set-up for measurement of permeability properties of brous reinforcements for RTM. Compos Part A: Appl Sci Manuf 2002; 33(7):959. [21] Ding L, Shih C, Liang Z, Zhang C, Wang B. In situ measurement and monitoring of whole-eld permeability prole of ber preform for liquid composite molding processes. Compos Part A: Appl Sci Manuf 2003;34(8):779. [22] Endruweit A, Ermanni P. The in-plane permeability of sheared textiles. Experimental observations and a predictive conversion model. Compos Part A: Appl Sci Manuf 2004;35(4):439. [23] Zhou F, Kuentzer N, Simacek P, Advani SG, Walsh S. Analytic characterization of the permeability of dual-scale brous porous media. Compos Sci Technol, submitted for publication.

interactions, which are so crucial in dual-scale preforms used in so many LCM applications. Acknowledgements This work was partially funded by the Army Research Laboratory, grant number DAAD 19-01-2-0005, and by Advanced Materials Intelligent Processing Center established at University of Delaware by Oce of Naval Research, grant N00014-04-1-0891. References
[1] Parnas RS, Phelan FR. The eect of heterogeneous porous media on mold lling in resin transfer molding. Sampe Quart 1991;22:5360. [2] Sadiq TAK, Advani SG, Parnas RS. Experimental investigation of transverse ow through aligned cylinders. Int J Multiphase Flow 1995;21(5):755. [3] Breard J, Saouab A, Bouquet G. Numerical simulation of void formation in LCM. Compos Part A: Appl Sci Manuf 2003;34(6):517. [4] Simacek P, Advani SG. A numerical model to predict ber tow saturation during liquid composite molding. Compos Sci Technol 2003;63(12):1725. [5] Binetruy C, Hilaire B, Pabiot J. The interactions between ows occurring inside and outside fabric tows during rtm. Compos Sci Technol 1997;57(5):587. [6] Lekakou C, Bader MG. Mathematical modelling of macro- and micro-inltration in resin transfer moulding (RTM). Compos Part A: Appl Sci Manuf 1998;29(12):29. [7] Binetruy C, Hilaire B. Tow impregnation model and void formation mechanisms during RTM. J Compos Mater 1998;32(3):22345. [8] Pillai KM, Advani SG. A model for unsaturated ow in woven ber preforms during mold lling in resin transfer molding. J Compos Mater 1998;32(19):175383. [9] Slade J, Pillai KM, Advani SG. Investigation of unsaturated ow in woven, braided and stitched ber mats during mold-lling in resin transfer molding. Polym Compos 2001;22(4):491505. [10] Ngo ND, Tamma KK. Microscale permeability predictions of porous brous media. International Journal of Heat and Mass Transfer 2001;44(16):3135.