LCM of thermoset-based fiber reinforced composites for large-scale applications: are process kinetics and structural properties antagonistic?

This article reviews the main methods to manufacture large-scale composite parts, with a focus on Liquid Composite Molding techniques of thermoset-based fiber reinforced structural parts. As this process relies on the impregnation of a dry textile stack, this manufacturing step is crucial in terms of part production rate, and part quality. To increase the process kinetics, a large effort has been devoted to increase the permeability of the textile preforms, while keeping a similar fiber content. An increase of almost two orders of magnitude can be attained if the textile shows a strong separation of scales between densely packed tows and large intra-two spaces. This however leads to a potential degradation in the resulting structural properties, particularly in dynamic mode due to the presence of the resin rich pockets. Alternative solutions emerge, which may help reach a cost-effective compromise.


Introduction
When manufacturing large-scale thermoset-based long fiber reinforced polymer composite (FRP) structural parts such as wind turbine blades, aircraft fuselages, rocket fairings, boat hulls or masts, several additional challenges arise as compared to the manufacturing of smaller parts:(i) the risk of occurrence of a scrap part should be reduced to a minimum due to the large amount of materials involved, (ii) the molds should be as light as possible to avoid extremely heavy weights and costs, (iii) the impregnation process kinetics should be as fast as possible so as to avoid premature cure before impregnation is complete and limit personnel and equipment costs, (iv) the cure temperature and time should be well adjusted to avoid uncontrolled exotherms in thicker parts and uneven cure, and the applied pressure should ideally remain low, to avoid the use of large scale autoclaves or compression presses which also strongly increase the cost, in particular if production volumes remain low.
In early stages of large structural component production, an easy-to-access production method was the wet hand lay-up technique, where a dry fiber reinforcement is placed in the mold and impregnated with the pre-mixed resin using rollers and brushes.The resin is left to polymerize in open air, or under a vacuum bag.This technique requires minimal equipment and knowledge, but requires intensive labor, leads to generally poor-quality composites or lack of reproducibility, and presents environmental issues, due to the contact with resin and evacuation to the open air.As a result, two alternative techniques have reached maturity in the last 10-20 years: (i) prepreg technologies, whereby the fiber reinforcement is initially impregnated with resin (and generally sold as such in rolls), then positioned layer by layer in the mold, by hand or using automated deposition techniques, before curing under vacuum or moderate pressure (ii) liquid composite molding (LCM) technologies, whereby the 2 fiber reinforcement is positioned dry into the mold (again by hand or using automated techniques, sometimes preformed beforehand), and reactive resin is forced to flow into the dry fiber preform, under a pressure difference imposed by a vacuum outlet, or a pressurized inlet, alternatively under a prescribed flow rate imposed by a volumetric pump, before curing.These two methods are currently used in most manufacturing processes for large scale structural parts, although filament winding techniques are also used for pressure vessels or cylindrical parts, and additive manufacturing of large parts is gaining interest, but still limited to pilot scale demonstration purposes [1][2][3].
In aeronautics and space, prepreg use is dominant due to the very high quality and reproducibility requirements, with tight control of the resin content, and due to the existing portfolio of available qualified prepreg materials.In some cases, high pressure autoclaves are still used (for example in the production of airplane wings), although intensive research has been performed over several decades in developing out-of-autoclave techniques to increase cost-efficiency, sometimes with great success, such as for the Ariane and Vega rocket fairings [4].LCM is also gaining importance, with some breakthrough examples, for example the Airbus A380 pressure bulkhead made by resin film infusion [5], wings of the Airbus A220 by Resin Transfer Infusion (previously Bombardier C Series, infused then autoclaved, [6]) or other wing structures [7].
In boat building, both prepreg and LCM technologies are encountered, depending on the costperformance ratio, carbon fiber prepregs being mostly used in high performance racing yachts and in masts, whereas infusion of glass fiber textiles (or combination of textiles and shorter fiber mats) is common for pleasure yachts or speed boats [3].
Wind turbine blade manufacturing is the other obvious example of a large composite part, with some blades reaching over 100 m long.Both prepreg and LCM technologies have been applied successfully to blade skins and spars production, however the trend is now towards an increased use of LCM technologies as they gain further maturity, removing the need to transport and store large amount of reactive prepreg materials.Vacuum Assisted Resin Transfer Molding (VARTM), where the fabric is held within a rigid mold, is currently the most wide-spread and cost effective method for producing wind turbine rotor blades, with room temperature curing resins (and potential higher temperature post-cure, around 120°C) [2,8].
A major difference between prepreg and LCM techniques resides in the impregnation step, which constitutes a major part of the production step.As a result, a large effort has been focused on increasing the kinetics of impregnation, with emphasis on the increase of textile permeability, while keeping the fiber content identical.However, the potential impact on the resulting structural properties needs to be considered, to reach best compromise, but both are seldom evaluated in parallel.This article will provide some insight on the main principles of LCM processes, and of the strategies to increase the permeability of the reinforcement, focusing on structural parts with about 50% fiber content.We will then address the increased risks of reduced part quality, in terms of manufacturing defects, but also in terms of structural properties.Several alternative solutions will finally be addressed.

Liquid composite molding principles
LCM processes are generally described following the approach developed in soil science and petroleum engineering, for the flow of a viscous fluid in a porous medium.Two important differences with flows in soils are that the fluid is generally much more viscous than water (about 500 times) and the porous medium is not homogeneous and very anisotropic, but constituted of fiber bundles, that are assembled into textiles.As a result, the pore size distribution is generally bimodal, with large inter-tow spaces (up to mm), and narrow intra-tow channels (a few microns).Nonetheless, a continuum mechanics approach is generally followed to describe the porous medium, with a representative volume element (RVE) which contains the initially present air, fibers, and infiltrated liquid, in respective volume fractions Va, Vf, Vl, such that: 1293 (2023) 012003 IOP Publishing doi:10.1088/1757-899X/1293/1/0120033 (1) By similarity with soil science, the fluid phase saturation S is defined as: (2) where (1-Vf) is the initial preform open porosity; S varies from 0 to 1 between a fully dry and a fully saturated preform.
The main geometrical descriptors of the dry porous medium are its fiber volume fraction, its specific surface, and the pore size distribution (as well as pore connectivity).For a simple arrangement of aligned cylinders, packed on a square array, the maximum fiber volume fraction that can be attained is Vfmaxs= π/4, whereas for a hexagonal array, Vfmaxh= π/2√3.The pore size distribution is a statistical descriptor of the porous medium that was historically measured through gradual filling of the media with non-wetting fluids [9].As X-Ray Computed tomography became more readily available in the recent years, 3D imaging of an RVE is now becoming routinely accessible, leading to precise geometrical descriptions of the porous medium, including the presence of non-connected pores ( [10][11][12][13][14][15]). As mentioned, textiles are most often formed of tows or yarns that are assembled into a preform, the porous medium is thus generally described by a bi-modal pore size distribution, as illustrated in Figure 1, and the dual scale nature of the pore size distribution can lead to exacerbated capillary/hydrodynamic effects, as compared to more uniform porous media such as encountered in soil science [16,17].The underlying physical phenomena for LCM processes include capillary or surface phenomena, transport of fluid, heat, and mass, the mechanics of preform deformation before and during infiltration, matrix chemical cross-linking, and potential matrix/reinforcement chemical reaction during and after the process.For a complete description, the reader is referred to recent reviews [17,19].In the following, for the sake of simplicity, we consider the general case of infiltration by a liquid of a rigid porous preform (constant volume fraction) in which all initial porosity is interconnected (no closed pores).We do not treat heat or mass transfer (as induced by a chemical reaction) since these often take place after impregnation.We also assume that the densities of liquid and solid phases are constant; in most cases of non-compressible liquid fluids, this is a reasonable assumption.Mass conservation equation is written for the fluid phase as: (3) where ul is the average local velocity of the liquid within the pores.
The momentum equation is generally written using Darcy's law: (4) where η is the fluid viscosity and P is the pressure in the liquid.K is the permeability of the porous medium, it is a tensor, with units: m 2 , describing the ease for the fluid to pass through the pore space.Following usage in soil science for multiphase flow in porous media, it is generally assumed that the influence of saturation S on the value of permeability can be decoupled from the influence of the porous medium intrinsic geometry, by defining two parameters , such that where Ksat is the saturated permeability tensor, with units: m 2 , which depends on the volume fraction and spatial arrangement of the fibers, and kr is called the relative permeability, function of S only, and is a scalar varying from 0 to 1.The left-hand side of Eq. ( 4) is called the superficial velocity, or filtration velocity, which was initially defined by Darcy as the ratio of the volumetric flow rate Q out of a porous medium, over the cross-section of this porous medium, A. Integration of these equations for multiphase flow with the relevant boundary conditions leads to a description of the flow kinetics, and progressive saturation, provided that the hydraulic functions describing pressure and relative permeability versus saturation are identified [20].
For fully saturated flow in a rigid porous medium, the filtration velocity u0 is simply written as: (5) As a result, for unidirectional saturated flow (kr=1) under constant applied fluid pressure at the mold inlet, impregnation time for in plane flow along a given distance L is: (6) where is the pressure difference between outlet and inlet, which in case of a saturated flow assumption, should also include the capillary pressure drop acting at the flow front [20][21][22].To minimize the impregnation time, while keeping the viscosity of the resin and the fiber volume fraction similar, strategies will include increasing the pressure gradient (but this requires heavier equipment and molds, and may lead to fiber movement), decreasing the impregnation length, for example by optimizing the presence of inlet points [23] and favoring through thickness impregnation, as performed in vacuum infusion thanks to high permeability flow grids placed on top of the textile.Another strategy is to increase the textile permeability, where a 10-fold increase leads to a 10-fold reduction of the time for impregnation.Efforts have been numerous to increase permeability, which is the subject of this review.

In-plane permeability measurement
Permeability measurement relies on the measure of the flow kinetics through the porous medium of interest, under well controlled conditions with a well-known Newtonian fluid [24].Even if the method seems very simple, experimental data for permeability tend to vary by almost one order of magnitude, as reported in the first permeability round robin [25], although with a stricter set of guidelines, results fall within about 20% variation, attributed to intrinsic variability of the textile stack, as well as human error [26].In addition, when performing 1D, in plane permeability measurement, two methods can be used, based on the measurement of the flow front progression, or on the flow rate through a fully saturated textile, as shown in figure 2.
When permeability is estimated from the position of the flow front L(t), Eq.( 6) is used to evaluate its value, often neglecting progressive saturation and any capillary effects at the flow front, assuming , where Pi is the inlet pressure, and Pf the pressure in the preform just ahead of the flow front (atmospheric pressure in general).Thus, this value, often named Kunsat to reflect the fact that it measured during flow (and that saturation is taking place during measurement), is computed as: (7) This value is distinguished from the saturated permeability, Ksat, measured from the flow rate Qout at the outlet, after the fluid has filled the whole length of preform Lf: where A is the cross-section area of the fabric stack [27].
Figure 2. Schematic of in-plane permeability measurement set-up, for measurement under constant applied pressure, with unsaturated flow measurement through a transparent top with a camera, and saturated flow measurement through the flow rate from the outlet after full saturation (from [15]).
As shown by Salvatori [15], the relation between these two values depends on the magnitude of the capillary pressure drop acting at the flow front, since , the ratio of unsaturated to saturated permeability is thus: (9)

How to increase permeability while keeping the same fiber volume fraction?
The saturated permeability of textiles is a strong function of the fiber volume fraction and its orientation with respect to flow direction.In packed fiber networks, flow along the direction of fibers is not always easier than across, and early permeability models, based on the concepts of hydraulic radius, or on computing the flow past regular arrays of cylinders, illustrated rather well this relationship, as shown Figure 3, using Gebart's and Happel's models presented in Ref. [28].As a result, for a fiber radius of 8 μm typical of glass fibers, the models predict a permeability for flow transverse to the direction of fibers of about 1.10 -12 m 2 , and for flow parallel to the fibers ranging from 7.10 -14 to 2.10 -12 m 2 .As pointed out by Karaki [29], who also evaluated other permeability models, it is rather difficult obtain realistic values from these models, although they represent a practical and analytically calculated first estimate of permeability ranges.The larger variation in models for flow parallel to the fibers stems from different assumptions leading to pressure build-up during flow.These models are in general not very realistic, except at rather high fiber volume fractions, as the fibers are rarely arranged on a regularly spaced pattern.As mentioned earlier and shown Figure 1, the porous medium is constituted of densely packed yarns and large gaps in between, which form channels that tend to dominate the permeability value.As a result, experimental permeability values are generally higher.Rieber [30] conducted an experimental analysis of 19 woven glass textiles (plain, twill and satin weave), observing values ranging between 1.10 -11 m 2 and 1.10 -10 m 2 for Vf=50%.They highlighted the influence of the weave density and yarn density on the K(Vf) slope, and the role of the weft/warp ratio on the anisotropy of the in-plane permeability tensor, such that an increase in the permeability in one direction does not always directly relate to an increase in the other in-plane direction.In the permeability benchmark, experimental permeabilities were also found in the range 0.5-1 10 -10 m 2 for Vf=50% [31].Ratio of permeability over the square of the fiber radius, for flow parallel or orthogonal to the fibers and for a square or a hexagonal array, as a function of overall fiber volume fraction (models given in Ref. [28]).
To model textile permeability, internal geometry of the pore space, including pore size distribution, must be considered.Analytical geometric models have been proposed, where the unit cell is greatly simplified so that a tractable solution of the equations can be written for micro-scale flow within the tow and meso-scale flow between tows, the two being then assembled into a full permeability model using circuit theories or direct flow models [32][33][34][35][36][37][38][39][40].These models are useful for rapid evaluation of the influence of tow permeability, or intra-tow channel size on the permeability.A first model by Pillai and Advani in 1995 [32] highlighted the role of the dual scale in the permeability estimation, and was one of the first to indicate that the tow permeability can be neglected (porous cylinders can be treated as solid cylinders) in cases where the separation of scales was sufficiently large.This was confirmed in many following studies, for example Syerko et al. [41] performed a parametric study varying the ratio k/h 2 , where k is the permeability in the tow, and h the inter-layer pore size, and Rougier developed an analysis for natural fiber composites [40], illustrated in Figure 4, where a permeability increase by almost two orders of magnitude is predicted when the tow volume fraction increases to 70%, while the overall fiber content remains identical.
As a result from the need to increase impregnation kinetics, in particular if rather high viscosity matrices are used [42] or if large parts are to be manufactured, models have been extended not only to predict permeability of a given reinforcement, but also to engineer new reinforcements with improved permeability [32,35,41].Several modifications to non-crimp fabrics have thus been reported, with the general aim to increase the intra-tow (or intra-layer) spaces and creates flow channels to enhance permeability.Following an optimization scheme, Syerko et al. reported an increase in in-plane permeability, for a volume fraction of 50%, from 7 10 -11 m 2 for a regular quasi-UD NCF up to 1.1 10 -10 m 2 [41].Martin et al. [43] performed an experimental study of the influence of design parameters (stich spacing, pattern and weft tow lineal weight) on the permeability tensor for the same quasi-UD NCF glass fabric, representative of materials used in wind turbine blade applications.The UD tows forming the warp layer were stitched with a PET thread onto a weft backing layer, and three variations were investigated as compared to the reference fabric: "A" with a lower stitch spacing (from 5 to 2.5 mm), "B" with an asymmetric stitch pattern (as compared to symmetric) and "C" with a different backing layer (with more spaces as compared to the reference).The results for the full permeability tensor highlighted that a reduction in stitch length (leading to tighter tows) improved the permeability, only below a threshold volume fraction of about 55%, after which the tow permeability started to play The concept of tight tows separated by large channels was pushed further with the development of highly permeable fabrics for the impregnation of thermoplastic molten matrices, such as G-Ply, as reported by Salvatori et al. [15].As shown in Figure 5, in comparison to a more regular woven fabric, large flow channels were designed along the x direction of the fabric.As a result, permeability values increased from the usual 10 -10 m 2 range around Vf=50%, almost to the goal of 10 -9 m 2 .However, in these cases, the flow front during impregnation was strongly affected by the presence of the flow channels, developing a large unsaturated zone with strong delay of the tow impregnation, and thus increased risks of tow porosity, if care is not taken to fully saturate the part, for example by adding a dam towards the flow outlet.As a result, the ratio Rs as defined in Eq. ( 9) remained close to one above a certain threshold of the flow velocity, where the tow permeability (and capillary effects) did not exert any major influence on the overall permeability value.

Influence on the mechanical properties and possible way forward?
Very few reports focus on a direct comparison between the (in-plane) permeability of fabrics and the resulting composite mechanical properties, and it is difficult to establish a direct relation between these two, since the manufacturing method and the choice of resin strongly influence the part properties.However, some general trends can be highlighted.First, as indicated above, increasing permeability by increasing the separation of scales between tightly packed tows and larger channels, leads to a larger unsaturated region during impregnation, which may increase the risk of porosity in the tow region, if the overall resin velocity is such that viscous flow dominates [20].This can be minimized by a careful control of the processing conditions, with sufficient resin bleeding, or the presence of less permeable regions towards the resin outlet, to fully saturate the part.Another potential risk, created by the presence of large resin pockets, is to promote residual stresses or micro-cracks to take place during cure or mostly during cool-down of the part.Overall, these resin rich regions have been consistently documented to negatively affect the performance of FRP, as recently reviewed by Mahmood et al. [44], who summarized the role of resin rich regions as location for crack initiation and propagation, as well as their negative effects on static as well as dynamic properties.As a conclusion, they recommend exerting caution when modifying the fiber architecture, and to carefully assess the potential change in structural properties.A direct comparison was performed by Ohtani et al [45], for several glass NCF with different constructions and tension in the stitches, on permeability and tensile strength, demonstrating a direct correlation between permeability increase (by a factor 5) and reduction in strength from 550 to 485 MPa.Another thorough comparison was performed by Martin [46], on the same glass NCF as presented in Ref. [43].The fabric with higher permeability "A", due to a reduced stitching length, showed a more uniform rigidity value, but a worse fatigue performance than the reference, due to the resin rich regions which led to higher energy dissipation and damage localization.Rimmel et al. [47] also evaluated the role of stitches on transverse permeability and mechanical properties in textiles positioned by Dry Fiber Placement; results again showed an inverse correlation between increased permeability (from 2.5 10 -14 m 2 up to 10 -12 m 2 for 50% fiber content), and reduced bending modulus and strength (up to 30%), with the observation again that cracks initiate at the location of stitches.It is also worth noting that stitches, since they are formed of drawn polymer fibers, may also change their shape with temperature, and thus lead to a change in permeability if resin flow takes place at a temperature above the Tg of the stitching yarns (which would first contract upon heating, then easily creep to loosen the bundles).This implies that permeability should be measured at a temperature close to the processing temperature.
A potential solution to this dilemma lies in the selection of material used for the stitching yarns.These could be engineered such that the presence of large channels could be ensured during infusion, if they can tightly pack the fiber bundles, and then melt or dissolve into the resin and let the initially compressed fiber tows expand and homogenize the fiber distribution.As an example, Beier et al. [48,49] proposed an elegant solution with the use of phenoxy or polyamide stitches, which melt and phase-separate within the epoxy resin to allow partial rearrangement of the reinforcement fibers, while toughening the matrix phase.They compared biaxial NCF textiles that are stitched either with polyester as reference material or with the novel yarns, processed by RTM with RTM6 aerospace grade epoxy resin.They did not quantify the permeability but performed microstructural observations, indicating that the dissolvable yarns had indeed disappeared; however, some resin rich regions remained, highlighting the potential damage or irreversible fiber movement created by the stitching operation, which cannot be fully recovered by fiber rearrangement, even though the stitches melted into the epoxy resin pool.Fracture toughness of the resin was improved by 20% when using 10 wt% phenoxy polymer added to the RTM6 resin as compared to the neat RTM6, showing the potential benefit of the addition of the thermoplastic phase into the resin.In composites, compression, and tensile moduli, as well as tensile strength were not much affected, while compression strength was improved by about 10% as compared to polyester stitched NCF, and Tg slightly reduced, indicating the potential double effect from homogenization of the fiber structure and matrix modification.All values remained however lower than a non-stitched reference, indicating that fiber misalignment nonetheless remained.This solution seems at this time possibly too costly for wind turbine blade applications, although potentially very promising if dissolvable stitching yarns can be further developed, together with the appropriate processing temperature (to avoid premature yarn melting) and kinetics to ensure that the yarns dissolve and that the bundles have time to expand before the resin viscosity increases too much.
As an alternative, dissolvable second phases could be introduced directly within the fabric stack.Beier [49] also evaluated the combination of dissolvable stitching yarns and powdered binders, so as to limit the negative effect of the increased stitching density.Results were very promising, including fatigue resistance, showing a potential route towards preform optimization by a careful selection of binder and stitching material.Also, as expected, the energy release rate in mode I, GIc, was well improved when using polyamide (3 times higher than reference) or phenoxy stitches (2 times higher than reference) with bindered crack plane, as compared to the reference non-stitched material.A more radical approach was demonstrated by Salvatori who created large flow channels to increase the kinetics of thermoplastic RTM processes [50,51], by the use of 3D printed polymer spacers, which could collapse and melt after impregnation and allow the compressed fabric stack to expand though thickness to form a more homogenous structure.This again required the use of a dam to stop the channel flow towards the outlet, however this approach could form a suitable concept to optimize flow paths by the introduction of fusible spacer lines.Another, less intrusive approach involves the use of spherical polymer beads, which could be inserted into the fabric between layers, and again melt or dissolve into the resin phase after impregnation.It was shown that spherical inclusions can increase permeability of the fabric stack, if their size lies within an adequate range to increase bundle distortion, whereas introduction of smaller beads reduces permeability, as they nest within the spaces between tows [13].
It is worth mentioning that a very common method to increase permeability is to insert a more permeable layer, for example a fiber mat with a reduced volume fraction and thus increased permeability, within the fabric stack, but this comes at the cost of a reduced overall fiber content and mechanical performance in rigidity, hence requires a careful part optimization [52].
Finally, if the fabric is to be left unperturbed, the least perturbing approach to increase the flow kinetics without directly acting on the fiber distribution inside the fabric is to introduce flow channels within the mold [3,53,54], which can be placed in an optimal way to increase the overall impregnation kinetics, depending on the part shape and permeability tensor of the textile stacks.This is of course very similar to the use of highly permeable layers commonly used in vacuum infusion processing of large-scale composite parts, to impregnate mostly through thickness, with the drawback of a heavy use of consumables and some unused resin which remains in this sacrificial layer.

Conclusions
This non-exhaustive review highlighted the antagonism between process kinetics and structural properties that may arise when producing large parts with LCM.On one hand, flow models have demonstrated that a clearly suitable approach to increase permeability of textiles while preserving a similar fiber content is to increase the separation of scales between highly packed tows and large flow channels.On the other hand, the presence of large channels influences the flow patterns, thereby creating large unsaturated zones and increasing the risk of incomplete tow impregnation.In addition, resin rich regions in the final part have a detrimental effect on the strength and fatigue life of the part, by increasing the risk of crack formation, localization, and propagation.As a result, processing and structural properties are strongly linked and should be analyzed together, to reach a suitable compromise for a given application.The creation of preferential channels in the mold, or the use of more permeable layers are current effective methods to adjust the flow kinetics but may also lead to less structural parts.Alternatively, novel fabrics with dissolvable stitches, binders or second phases could allow a permeability increase during the impregnation step, followed by a homogenization of

Figure 1 .
Figure 1.Example of the structure of a carbon UD textile with stabilizing glass yarns.Left, macroscopic view, right top, X Ray-CT cut of a fabric stack, right bottom, the same image after segmentation showing the tows in grey and inter-tow space in white (and glass tow in black) [18].

Figure 3 .
Figure3.Ratio of permeability over the square of the fiber radius, for flow parallel or orthogonal to the fibers and for a square or a hexagonal array, as a function of overall fiber volume fraction (models given in Ref.[28]).

Figure 4 .
Figure 4. Estimate of the permeability Ksat (called here Kgeo ) of a quasi-UD flax fabric as a function of the overall Vf and the local tow volume fraction Vf,μ.The red line shows the permeability increase for an overall Vf =50%, when Vf,μ increases [40].