Version of Record: https://www.sciencedirect.

com/science/article/pii/S1359835X19301186
Manuscript_73f9d5c037fb3d1ba5d9d9afa829dab6

Particle simulation of dual-scale flow in resin transfer molding for process analysis

Shigeki Yashiro1,*, Daichi Nakashima2, Yutaka Oya3, Tomonaga Okabe3, Ryosuke Matsuzaki4,5

1
Department of Aeronautics and Astronautics, Kyushu University

744 Motooka, Nishi-ku, Fukuoka 819-0395, Japan

2
Graduate School of Integrated Science and Technology, Shizuoka University

3-5-1 Johoku, Naka-ku, Hamamatsu 432-8561, Japan

3
Department of Aerospace Engineering, Tohoku University

6-6-01 Aoba-yama, Aoba-ku, Sendai 980-8579, Japan

4
Department of Mechanical Engineering, Tokyo University of Science

Yamazaki 2641, Noda, Chiba 278-8510, Japan

5
Institute of Fluid Science, Tohoku University

2-2-1 Katahira, Aoba-ku, Sendai 980-8577, Japan

* Corresponding author: [email protected] (S. Yashiro)

Abstract

Modeling the inhomogeneous microstructures of fibrous tows is important for analyzing the

process of resin transfer molding because dual-scale pores in a preform can lead to void

formation. This study focused on the development of a microscopic flow analysis method to

predict the impregnation of fiber bundles. The moving particle semi-implicit method was

adopted to model the microstructure of a fiber bundle explicitly and inter-particle potential

force was introduced into the numerical model to take account for the capillary effect. The

1
© 2019 published by Elsevier. This manuscript is made available under the Elsevier user license
https://www.elsevier.com/open-access/userlicense/1.0/
predicted process of impregnation and void formation agreed with empirical observations. The

developed approach was applied to predict the relationship between the modified capillary

number and void content to identify the optimal molding conditions to reduce microvoids. The

obtained relationship reproduced the trends of a reported experiment, which indicates that the

proposed approach will provide information about optimal conditions for minimizing void

content.

Keywords: A. Polymer-matrix composites (PMCs); B. Wettability; C. Process simulation; E.

Resin flow

1. Introduction

Because fiber-reinforced plastics are lightweight and have desirable mechanical properties,

they have been widely adopted in aerospace and automobile structures to improve fuel

efficiency. More cost-effective molding processes compared to prepreg-based autoclave

molding are required for the mass production of automobiles and next-generation aircraft.

Resin transfer molding (RTM) is a promising method based on its high productivity and the

excellent mechanical properties of manufactured composites. However, voids may be formed

in or near fiber tows during the impregnation process, which deteriorates the mechanical

properties of the composites [1]. The formation of such defects is an obstacle to the widespread

2
use of liquid molding processes.

The voids formed during RTM are classified depending on their scale. Macrovoids (also

called dry spots) are large air pockets across fiber bundles, which are caused by incomplete

resin filling. Mesovoids and microvoids are formed based on molding conditions and the

microstructure of the fibrous preform. Dual-scale pores in the preform lead to different

permeability values inside and between fiber tows, which leads to air entrapment during the

resin filling process. Mesovoids are air bubbles between fiber bundles that are generated when

the flow rate inside the fiber bundles is higher than that outside the bundles. Microvoids are tiny

air residuals inside fiber bundles that are formed when the flow rate outside the fiber bundles is

higher than that inside the bundles. The balance of viscous flow and capillary flow governs void

formation and is important for the reduction of void content.

There have been numerous studies focusing on the experiment, analytical modeling and

numerical simulation of void formation and the optimization of the RTM process [2]. Sadiq et

al. [3] experimentally determined that low local permeability in a dense bundle causes void

formation in fiber bundles. Patel and Rohatgi [4,5] observed the impregnation of unidirectional

stitched fiberglass mats and determined that the fraction of mesovoids and microvoids is

correlated with the modified capillary number, which is the ratio of viscous force to capillary

force. Microscopy analysis of a chopped-strand randomly oriented mat revealed significant

3
void variation in the through-thickness (impregnation) direction [6]. In-situ observations of the

filling process of woven fabrics have revealed that mesovoids are formed by the dominant

capillary flow and lead-lag flow front [7]. Mathematical models for analyzing void formation

have been developed by considering microscale flow into inhomogeneous preforms based on

Darcy’s law. Kang et al. [8] predicted the relationship between the void fraction and modified

capillary number by analyzing time differences when filling a bundle and open channel.

Matsuzaki et al. [9,10] developed a similar mathematical model using a mathematical

permeability model [11] and their prediction of the fraction of mesovoids based on flow

velocity agreed with experimental measurements. Lebel et al. [12] applied a mechanistic

imbibition model to fabrics to characterize the impregnation process and determined that two

key parameters, namely the modified capillary number and penetrativity of the fluid, should be

optimized to reduce voids. They also proposed an experimental method to obtain the optimal

conditions for a given resin-fabric pair.

RTM simulations based on the finite-element/control-volume (FE/CV) method [13-15] have

been developed over the past few decades to predict void content. In the standard approach, the

governing equation for single-phase flow is solved on the macroscopic scale, which essentially

neglects microscale pores in the bundle. This approach has been extended to involve the

microscale intra-tow flow and to predict microvoids and mesovoids [16-19]. Although the

4
optimal value of the modified capillary number has been predicted using FE/CV methods [18],

such methods always require a measured value of local permeability or model of permeability

as input data. Proper calibration may be required to apply these permeability models [11,20,21]

to a real preform.

To overcome this difficulty, this study presents an impregnation analysis method from a

microscopic viewpoint. In recent years, microscopic flow simulation in a fiber array has been

investigated by solving the momentum equation using the finite-element method and

finite-volume method, and the effective permeability have been analyzed based on the intra-tow

flow [22-26]. A particle method is employed in this study to explicitly model the

microstructures of fibrous tows and analyze the void formation process based on the flow

through open channels and intra-tow paths. The surface tension effect can be easily introduced

into particle simulations, and this characteristic enables us to predict the entrapment of air

without using permeability. Okabe et al. [27] proposed a method for analyzing microscopic

flows through fibrous tow based on the moving particle semi-implicit (MPS) method [28] and

considered the capillary effect as a potential force. However, the results did not reproduce the

mechanisms of air entrapment during RTM [4,5] accurately because the open channels between

bundles were neglected. Lu et al. [29] presented a flow analysis of the RTM process using

smoothed particle hydrodynamics, but they neglected the capillary effect, which is essential to

5
estimating optimal RTM process conditions.

Therefore, this study focused on the development of a numerical method for analyzing

microscale flows during the RTM process. To this end, we adopted the MPS method to

explicitly model the microstructures of fiber bundles and incorporated inter-particle potential

force into the MPS method to account for the capillary effect. This remainder of this paper is

organized as follows. Section 2 describes experiments on liquid flow using a transparent mold

with simulated fiber bundles. Section 3 briefly introduces the MPS method and an inter-particle

potential force related to surface tension and wettability. In Section 4, flow analysis

corresponding to the experiments is performed to examine the process of microvoid formation.

A numerical study is presented in Section 5, where we estimate optimal molding conditions to

reduce the void fraction without using permeability.

2. Experiment

To observe the process of void formation in a fiber bundle, a transparent polymethyl

methacrylate mold with simulated fiber bundles (Fig. 1) was prepared. The mold cavity space

was 22.0 mm wide and 1.0 mm thick. Quartz rods with a diameter of 1.0 mm were embedded

into the mold and the minimum spacing between adjacent rods was 0.4 mm. The space between

adjacent bundles (blocks with densely arranged rods) was 2.5 mm in the width direction. The

6
local volume fraction of the rods in each bundle was 50% and the nominal rod volume fraction

in the rods arranged area was 27%.

Liquid was injected from one end of the mold (left side in Fig. 1a) and the impregnation

process was observed. In this study, water and olive oil were used as test liquids, which were

chosen by considering their different viscosities and resultant capillary effects. Their properties

are listed in Table 1. The liquids were injected by a micro-syringe pump at a constant flow rate

of 500 ml/h. This flow rate corresponds to a flow-front velocity of 6.3 mm/s in the empty space.

A high-speed digital microscope (VW-9000, Keyence) was used to observe the flow. The

impregnation of the bundle shown in Fig. 1a was observed in detail and the void fraction was

determined as the area of the voids divided by the observed rectangle area, in which the nominal

rod volume fraction was 30%.

Figures 2 presents snapshots of the observed water flow into the bundle, where the dotted

line is the flow front of the liquid. Water impregnated only the outermost layer of bundles and

then flowed between bundles without further impregnation. The flow velocity in the open

channels was higher than that in the bundles and the bundles were surrounded by water before

complete saturation occurred. As a result, air remained inside the bundles and the void fraction

of the observed area was 10.8%. This mechanism of microvoid formation agrees with previous

experiments [3,5]. Figure 3 presents snapshots of the oil injection experiment. In contrast to the

7
results of water injection, the oil flowed into the bundles. Although the flow front in the open

channels still moved faster than that inside the bundles, the velocity difference was small.

Therefore, the bundles were filled completely, and no voids were formed. The capillary effect is

the main cause for this difference between the two liquids [30].

3. Analysis

Microscale flow in fibrous media as well as flow of fiber suspension have been studied using

particle simulation methods owing to their capability to represent microstructures [31]. In

particular, the lattice Boltzmann method [32], the smoothed particle hydrodynamics [29] and

the MPS method [27] have been used for those analyses. Among them, this study employs the

MPS method for the analysis of inter- and intra-tow flow, because it is suitable for solving

incompressible flow and easy to introduce the capillary force into the analysis framework.

3.1 MPS method

The MPS method [28] is a particle-based method in which a continuum is expressed as an

assembly of particles. Each particle is assumed to have an effective radius for particle

interaction and differential models are prepared based on particles within the effective volume.

When a particle i has a scalar variable φi and a vector variable ui, the corresponding differential

8
models are defined as follows:

 φ −φ 
(
 j i ( r j − ri ) w r j − ri )
N
d
∇φ i
= 0
n
∑
j ≠i  r − r
2


(1)
 j i 

 (u − u ) ⋅ (r − r ) 
( )
N
d
∇ ⋅u = ∑  j w r j − ri
i j i
(2)
n0 j ≠i 
i 2

 r j − ri 

i
2d N 
∇ φ = 0 ∑ (φ j − φi ) w rj − ri 
2

λn j ≠i   ( ) (3)

re r  re  re 
λ = ∫ r 2  e −1 2π r d r ∫r0  r −1 2π r d r (4)
r
r 
0

Here, d is the dimensional number, n0 is the constant particle number density under

incompressible conditions, N is the number of particles within the effective radius re, r is the

position vector, and w (= re/r − 1) is the weight function.

The governing equations for incompressible flow are the mass conservation and the

Navier-Stokes equation:

Dρ
= 0, (5)
Dt

Du 1
= − ∇P +ν∇2u + g , (6)
Dt ρ

where t is time, ρ is density, u is velocity, P is pressure, ν is kinematic viscosity, and g is

acceleration caused by an external force, such as gravity. These values are discretized by

differential operator models. In the standard MPS method, the viscous (second) and external

force (third) terms on the right of Eq. (6) are solved explicitly and the Poisson’s equation for

pressure derived from the pressure (first) term and Eq. (5) is solved implicitly. However,

9
analyzed pressure distributions frequently include numerical oscillations. This study employed

an improved solving scheme [33] to obtain a smoother pressure distribution.

3.2 Inter-particle potential force

To analyze the capillary effect, an inter-particle potential force was introduced into the MPS

method. The inter-particle potential force Fs is defined as the gradient of the potential Π. This

study employed the Morse potential, which is an interatomic interaction model.

N
∂Π r j − ri
Fs = ∑ ∂ (r ⋅ (7)
j≠i j − ri )
r j − ri

Π =C e { −2( r / l0 −1)
− 2e
− ( r / l0 −1)
}, (8)

where C is the coefficient, r is the distance between two particles, and l0 is the equilibrium

distance (i.e., initial particle spacing). The Morse potential, Eq. (8), was adopted with

consideration for the stability of analysis.

Wettability is calculated using the same potential model. Wettability can be represented by

using two specific potential coefficients for the interaction between two fluid particles and for

the interaction between a fluid particle and solid particle, respectively. This study used the

following relationship between the two coefficients:

Cfs = α Cff , (10)

where Cff and Cfs are the coefficients for fluid-fluid interaction (i.e., surface tension) and

10
fluid-solid interaction, respectively. Cff is only used in Eq. (8) if the particle of interest i and its

neighboring particle j are both fluid particles. Cfs is used when either i or j are solid (Fig. 4). The

arbitrary positive constant α was calibrated based on droplet collapse simulations to obtain an

appropriate contact angle. The inter-particle potential force was adopted as the external force

term in the Navier-Stokes equation.

The potential model was verified through droplet collapse. The contact angles between the

liquids and a quartz plate were measured by an automatic contact angle meter (DMe-201,

Kyowa Interface Science) and found to be 39.9° and 27.4° for water and olive oil, respectively.

The collapse of a droplet with a diameter of 2 mm was then analyzed using 0.125-mm particles.

The contact angles for analysis were determined using the half-angle method and were

averaged during the period of 0.2–0.5 s, during which the motion of the droplet was mostly

converged. Figure 5 depicts the observed droplet shape and the predicted result with adjusted

parameters. The predicted contact angles (38.3° and 30.7° for water and oil) were matched to

the measured angles by adjusting the parameters Cff and α (Table 2). However, the potential

model needs to be improved in order to reproduce the observed droplet shape.

4. Liquid flow in a fiber bundle

Two-dimensional analysis of liquid flow into a fiber bundle was performed to predict the

11
void formation process. Figure 6 depicts the analytical model that corresponds to this

experiment. The particle size was 0.125 mm and a single rod was represented by 52 particles.

The total number of particles in the initial condition was 8364. Liquid particles were injected

with a velocity of 6.3 mm/s from the left side of the model. Specifically, the left wall moved in

the injection direction at the prescribed velocity. When the travel distance reached the size of a

particle, the wall returned its original position and new fluid particles were generated in the

empty space. A periodic boundary condition was applied to the top and bottom ends of the

model. An outlet boundary condition was imposed on the right end of the model, and particles

passing through the boundary were excluded from the analysis. Water and olive oil were used as

test liquids again with the properties listed in Table 1. The inter-particle potential model was

considered and the parameters listed in Table 2 were used. For the sake of simplicity and

stability of analysis, the density of air was set to the same value as that of water and its kinetic

viscosity was set to 1.515×10-5 m2/s. The inter-particle force was not applied to the air particles.

The void fraction was calculated as the number of air particles divided by the initial number of

particles in the designated region.

Figure 7 depicts snapshots of the predicted water flow into the bundle. When the water

reached the left end of the bundle, it impregnated up to one layer inside the outermost layer of

the bundle (Fig. 7a). The bundle was hardly immersed in the injection direction and water

12
flowed mainly through the open channels, as depicted in Figs. 7b and 7c. The bundle was

finally surrounded by water and air remained inside the bundle (Fig. 7d). This process of

microvoid formation agrees well with the experimental results (Fig. 2). The predicted void

fraction in the final state was 9.4%, which also agrees with the experimental results.

Figure 8 presents snapshots of the predicted flow of oil into the bundle. Similar to the

experimental results, impregnation proceeded from the left end of the bundle until 0.6 s (Figs.

8a and 8b). The bundle was then immersed in the direction perpendicular to injection because

of the capillary effect and a small amount of air was entrapped between the fibers (Fig. 8c). The

flow front velocity in the open channels was slightly higher than the impregnation speed and the

bundle was surrounded by the oil before all air was ejected (Fig. 8d). The amount of entrapped

air differed from the experimental results, but most of the bundle was filled and the void

fraction was 2.9%.

Although there was a small error in the void fraction, the overall impregnation process

agreed with the experimental results in Section 2. Therefore, flow analysis with an adequate

capillary effect model can predict change in the fraction of microvoids in fiber bundles under

varying molding conditions. Dimensionless analysis to address the difference in scale from an

actual fiber bundle and verification of the simplified air viscosity are our future work.

13
5. Discussion

Although the proposed approach reproduced the process of microvoid formation successfully,

it can analyze only a small portion of a real mold based on its high calculation cost. The FE/CV

method [13-15] is realistic for predicting the molding process of a structural member from the

viewpoint of the calculation cost, and the present approach can be used as a convenient

auxiliary tool. The FE/CV method requires the permeability of a fiber preform as input data,

which is typically measured experimentally. Furthermore, optimal molding conditions can be

specified after a number of molding experiments and observations for each combination of

fiber preform and resin. Microscale flow analysis can be used as a substitute for such laborious

experiments. The remainder of this section will demonstrate such an application of the

proposed approach.

The process of air entrapment varies based on molding conditions. The void fraction in an

RTM process can be determined based on the modified capillary number Ca* as follows:

µu , (13)
Ca * =
σ cos θ

where μ is the viscosity, u is the superficial velocity (injection flow velocity in this study), σ is

the surface tension, and θ is the contact angle.

A numerical study was performed to predict the optimal conditions from the perspective of

microvoid formation. Figure 9 depicts the analytical model. The size of each particle was 2 μm

14
and seven particles were arranged in a hexagonal shape to approximate a fiber with a diameter

of 6 μm. There were 3942 particles in the initial state. Fluid particles were injected from the left

end of the model and the periodic boundary condition was applied to the top and bottom of the

model. The void fraction in the fiber bundles was predicted under 27 different conditions with

varying injection speeds (1, 10, and 100 mm/s), viscosities (0.1, 0.5, and 1 Pa·s), and contact

angles (40°, 60°, and 80°). The density and surface tension of the fluid were constant (1000

kg/m3 and 7.28×10-2 N/m, respectively). These conditions correspond to a modified capillary

number range of 1.8×10-2–7.9. The number of air particles remaining in the region shown in Fig.

9 was counted after the major flow front reached the right end of the model.

Figure 10a presents an example of the final particle distribution; air was entrapped in the

bundles under the condition Ca* = 7.9, similar to the observation (Fig. 2). The predicted

relationship between the void fraction and modified capillary number is plotted in Fig. 10b. The

void fraction was zero at Ca* values less than 0.01 and increased monotonically in with Ca*

values greater than 0.02. This trend in the void fraction agrees with reported experimental data

on microvoids [5]. Therefore, flow analysis considering the microstructure of the preform and

capillary effect has possibility of predicting optimal molding conditions to minimize

microvoids. However, this analysis could not reproduce the mesovoids observed in past

experiments [4]. This difference is the result of the analytical model disregarding the woven

15
structures of bundles. Warp yarns should be included in the model to predict the relationship

between the void fraction and Ca* within a low capillary number range. It should be noted that

the optimal Ca* value will depend on the number of fibers in a tow.

6. Conclusions

This study focused on the development of a flow analysis method to predict void formation

during the RTM process using the MPS method, which enabled us to explicitly model the

microstructures of fiber bundles. An inter-particle potential force was introduced into the MPS

method to account for the capillary effect. The impregnation of simulated fiber bundles was

observed in a transparent mold to verify the proposed analytical method. Our conclusions are

summarized below.

1. The developed flow analysis method with microstructure models of fiber preforms

reproduced the process of microvoid formation in fiber bundles observed in experiments

successfully. The predicted impregnation and void formation processes agreed with our

experimental observations.

2. The relationship between the void fraction and modified capillary number was predicted by

the developed analysis method successfully. The predicted results reproduced the trends

observed during past experiments on microvoids [5]. This demonstrates the ability of the

16
 proposed method to predict optimal molding conditions to minimize the void content in

RTM processes.

Acknowledgement

This work was supported by the Cross-ministerial Strategic Innovation Promotion (SIP)

Program, “Structural Materials for Innovation.”

References

[1] Leclerc JS, Ruiz E. Porosity reduction using optimized flow velocity in Resin Transfer

Molding. Compos Part A 2008; 39: 1859-1868.

[2] Park CH, Lee WI. Modeling void formation and unsaturated flow in liquid composite

molding processes: a survey and review. J Reinf Plast Compos 2011; 30: 957-977.

[3] Sadiq TAK, Advani SG, Parnas RS. Experimental investigation of transverse flow through

aligned cylinders. Int J Multiphase Flow 1995; 21: 755-774.

[4] Patel N, Rohatgi V, Lee LJ. Micro scale flow behavior and void formation mechanism

during impregnation through a unidirectional stitched fiberglass mat. Polym Eng Sci 1995;

35: 837-851.

[5] Rohatgi V, Patel N, Lee LJ. Experimental investigation of flow-induced microvoids during

17
 impregnation of unidirectional stitched fiberglass mat. Polym Compos 1996; 17: 161-170.

[6] Hamidi YK, Aktas L, Altan MC. Formation of microscopic voids in resin transfer molded

composites. J Eng Mater Technol 2004; 126: 420-426.

[7] Matsuzaki R, Seto D, Todoroki A, Mizutani Y. Void formation in geometry-anisotropic

woven fabrics in resin transfer molding. Adv Compos Mater 2014; 23: 99-114.

[8] Kang MK, Lee WI, Hahn TH. Formation of microvoids during resin-transfer molding

process. Compos Sci Technol 2000; 60: 2427-2434.

[9] Matuzaki R, Seto D, Naito M, Todoroki A, Mizutani Y. Analytical prediction of void

formation in geometrically anisotropic woven fabrics during resin transfer molding.

Compos Sci Technol 2015; 107: 154-161.

[10] Matsuzaki R, Naito M, Seto D, Todoroki A, Mizutani Y. Analytical prediction of void

distribution and a minimum-void angle in anisotropic fabrics for radial injection resin

transfer molding. Express Polym Lett 2016; 10: 860-872.

[11] Gebart BR. Permeability of unidirectional reinforcements for RTM. J Compos Matter

1992; 26: 1100-1133.

[12] LeBel F, Fanaei AE, Ruiz É, Trochu F. Prediction of optimal flow front velocity to

minimize void formation in dual scale fibrous reinforcements. Int J Mater Form 2014; 7:

93-116.

18
[13] Bruschke MV, Advani SG. A finite element/control volume approach to mold filling in

anisotropic porous media. Polym Compos 1990; 11: 398-405.

[14] Trochu F, Gauvin G, Gao D-M. Numerical analysis of the resin transfer molding process by

the finite element method. Adv Polym Technol 1993; 12: 329-342.

[15] Kang MK, Lee WI. A flow-front refinement technique for the numerical simulation of the

resin-transfer molding process. Compos Sci Technol 1999; 59: 1663-1674.

[16] Simacek P, Advani SG. A numerical model to predict fiber tow saturation during liquid

composite molding. Compos Sci Technol 2003; 63: 1725-1736.

[17] Gourichon B, Binetruy C, Krawczak P. A new numerical procedure to predict dynamic

void content in liquid composite molding. Compos Part A 2006; 37: 1961-1969.

[18] Schell JSU, Deleglise M, Binetruy C, Krawczak P, Ermanni P. Numerical prediction and

experimental characterisation of meso-scale-voids in liquid composite moulding. Compos

Part A 2007; 38: 2460-2470.

[19] Lee DH, Lee WI, Kang MK. Analysis and minimization of void formation during resin

transfer molding process. Compos Sci Technol 2006; 66: 3281-3289.

[20] Berdichevsky AL, Cai Z. Preform permeability predictions by self-consistent method and

finite element simulation. Polym Compos 1993; 14: 132-143.

[21] Xu P, Yu B. Developing a new form of permeability and Kozeny–Carman constant for

19
 homogeneous porous media by means of fractal geometry. Adv Water Res 2008; 31: 74-81.

[22] Liu HL, Hwang WR. Transient filling simulations in unidirectional fibrous porous media.

Korea-Australia Rheology Journal 2009; 21:71-79.

[23] Yazdchi K, Srivastava S, Luding S. Micro–macro relations for flow through random arrays

of cylinders. Compos Part A 2012; 43:2007–2020.

[24] Soltani P, Johari MS, Zarrebini M. Effect of 3D fiber orientation on permeability of

realistic fibrous porous networks. Powder Technol 2014; 254:44–56.

[25] Kundu P, Kumar V, Hoarau Y, Mishra IM. Numerical simulation and analysis of fluid flow

hydrodynamics through a structured array of circular cylinders forming porous medium.

Appl Math Model 2016; 40:9848–9871.

[26] Bodaghi M, Catalanotti G, Correia N. On the statistics of transverse permeability of

randomly distributed fibers. Compos Struct 2016; 158: 323–332.

[27] Okabe T, Matsutani H, Honda T, Yashiro S. Numerical simulation of microscopic flow in a

fiber bundle using the moving particle semi-implicit method. Compos Part A 2012; 43:

1765-1774.

[28] Koshizuka S, Nobe A, Oka Y. Numerical analysis of breaking waves using the moving

particle semi-implicit method. Int J Numer Methods Fluids 1998; 26: 751-769.

[29] Lu G, He L, Chen D, Li W. Smoothed particle hydrodynamics simulation of dual-scale

20
 flow during resin transfer molding. J Reinf Plast Compos 2017; 36: 1431-1438.

[30] Chen Y-T, Davis HT, Macosko CW. Wetting of fiber mats for composites manufacturing: 1.

visualization experiments. AICHE J 1995; 41:2261-2273.

[31] Yashiro S. Application of particle simulation methods to composite materials: a review.

Adv Compos Mater 2017; 26:1-22.

[32] Röding M, Svensson P, Lorén N. Functional regression-based fluid permeability prediction

in monodisperse sphere packings from isotropic two-point correlation functions. Comput

Mater Sci 2017; 134:126–131.

[33] Tanaka M, Matsunaga T. Stabilization and smoothing of pressure calculation for the

moving particle semi-implicit method. J Comput Phys 2010; 229: 4279-4290.

21
Figure captions

Figure 1 Transparent mold with simulated fiber bundles.

Figure 2 Snapshots of the water flow into a bundle. The dotted line indicates the flow front.

Figure 3 Impregnation process of a bundle with oil. The dotted line indicates the flow front.

Figure 4 Schematic of the estimation of inter-particle potential force.

Figure 5 Final shape of the droplet on a quartz plate.

Figure 6 Analytical model of the transverse flow into a fiber bundle.

Figure 7 Predicted process of impregnation of a bundle with water.

Figure 8 Predicted process of impregnation of a bundle with oil.

Figure 9 Analytical model for predicting the void fraction with a transverse flow.

Figure 10 Analytical results of a transverse flow in fiber arrays.

22
 Table 1 Material properties of liquids.

Water Olive oil Measurement method

Density (kg/m3) 1000 (*) 914 Archimedes method

Kinematic viscosity (m2/s) 1.0×10-6 (*) 8.37×10-5 Capillary viscometer method

Surface tension (mN/m) 72.8 (*) 34.6 Pendant drop method

Contact angle to quartz (°) 39.9 27.4 A half-angle method

(*) Values from literature

Table 2 Parameters of the inter-particle potential model.

Water Oil

Cff 20 200

α 0.73 1.215

Effective radius 3.9 l0

23
 10.5

2.5
22
22mm 4.2

Quartz rod (f1) Unit: mm

(a) Schematic

(b) Overview

Fig. 1
 (a) 0.30 s (b) 0.83 s

(c) 1.23 s (d) 1.57 s

Fig. 2
 (a) 0.07 s (b) 0.60 s

(c) 1.20 s (d) 1.53 s

Fig. 3
 Effective radius
of the potential

Fluid Cff
Cfs

Solid

Fig. 4
 Experiment Analysis

(a) Water

(b) Oil

Fig. 5
 Injection wall Fiber

8.5 mm

Air
Area to measure the void fraction
Liquid

Fig. 6
 (a) 0.30 s (b) 0.83 s

(c) 1.23 s (d) 1.57 s

Fig. 7
 (a) 0.07 s (b) 0.60 s

(c) 1.20 s (d) 1.53 s

Fig. 8
 Injection wall
Area to measure
Liquid Fiber
the void fraction

108 μm

Air

140 μm

Fig. 9
 (a) Final particle distribution (Ca* = 7.9)

Analytical results
Experiment (microvoid) [5]
Experiment (mesovoid) [4]
25

20
Void fraction (%)

15

10

0
-4 -3 -2 -1 0 1
10 10 10 10 10 10
*
Modified capillary number, Ca

(b) Void fraction as a function of the modified capillary number

Fig. 10