*Article* **Uniaxial Compressive Behavior of AA5083/SiC Co-Continuous Ceramic Composite Fabricated by Gas Pressure Infiltration for Armour Applications**

**Achuthamenon Sylajakumari Prasanth 1, Vijayan Krishnaraj 2,\*, Jayakrishnan Nampoothiri 2, Ramalingam Sindhumathi 2, Mohamed Raeez Akthar Sadik 2, Juan Pablo Escobedo <sup>3</sup> and Krishna Shankar <sup>3</sup>**


**Abstract:** A novel approach of a gas pressure infiltration technique is presented for the synthesis of Co-Continuous Ceramic Composite (C4). SiC foams of varying pore sizes were infiltrated with aluminium AA5083. Optical examination revealed that the SiC foams contained open cells with a network of triangular voids. The number of pores-per-inch (PPI) in the foams was found to depend on the strut thickness and pore diameter. The compressive strengths of two foam configurations, 10 and 20 PPI, were estimated to lie between 1–2 MPa. After infiltration, the compressive yield strength of the resulting C4 was observed to increase to 126 MPa and 120 MPa, respectively, for the 10 and 20 PPI C4. Additionally, the infiltration of ceramic foam with the AA5083 alloy resulted in an increase in strength of 58–100 times when compared with plain ceramic foam. The failure modes of the composites in compression were analyzed by crack propagation and determining the type of failure. The study revealed that shear failure and vertical splitting were the predominant mechanisms of compression failure, and that the fabricated C4 is advantageous in mechanical properties compared to the plain ceramic foam. This study, therefore, suggests the use of C4 composites in armour applications.

**Keywords:** co-continuous ceramic composites; C4; ceramic foam; gas infiltration; compressive strength; structural characterization

#### **1. Introduction**

Ballistic protection systems such as armour generally consist of several layers of materials. Each layer performs a specific role in attenuating the energy of projectiles [1,2]. Typically, a hard material such as ceramic is positioned on the front striking face and a matrix composite or high strength steel is placed as the backing face. The front face plate retards the striking force of the projectile by actions such as tumble, erosion, and fracture, whereas the backing material absorbs the residual kinetic energy of the projectile to bring the fragments to rest [3,4]. Ballistic protection plates composed of ceramic sticking face with polymer matrix composite (PMC) back plates have been extensively explored and proven for typical ballistic impact conditions. The abrasion resistance and hardness of the ceramic front face enables it to blunt the approaching projectile and absorb its energy to reduce the impact hazard. However, poor ductility limits its potential to take multiple hits. In addition, the processing complexities, cost, and weight of these systems restrict their wide use. Energy absorption studies in both quasi-static and dynamic conditions have revealed that steel–steel composite metal foams are suitable for armour. However, strict desiderata of lightweight materials for the strategic movement of defence personnel and

**Citation:** Prasanth, A.S.; Krishnaraj, V.; Nampoothiri, J.; Sindhumathi, R.; Akthar Sadik, M.R.; Escobedo, J.P.; Shankar, K. Uniaxial Compressive Behavior of AA5083/SiC Co-Continuous Ceramic Composite Fabricated by Gas Pressure Infiltration for Armour Applications. *J. Compos. Sci.* **2022**, *6*, 36. https:// doi.org/10.3390/jcs6020036

Academic Editor: Jinyang Xu

Received: 31 December 2021 Accepted: 17 January 2022 Published: 20 January 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

vehicles restrain the use of such materials [5,6]. Hence, a review of prior studies indicates that weight reduction of ballistic protection plates is paramount in the research of alternative materials for armour.

A study by Chang et al. [6] reported the development of lighter bullet proof material composed of ceramic faced metal–ceramic interpenetrating composites (IPCs). These IPCs exhibited better impact resistance and were less susceptible to abrupt demolition due to the development of stress wave at the interface. The primary reason was attributed to acoustic impedance mismatch between the metal and ceramic. The interlocking type microstructure of the IPCs, also termed as co-continuous ceramic composites (C4), imparts improved fracture toughness, wear resistance, stiffness, and reduced distortions [7,8]. This distinctive combination of enhanced properties of C4 makes it suitable for applications which require high specific modulus, high strength, higher corrosion resistance, and improved abrasion properties, such as brake discs and armour [9,10]. Additionally, these features of C4 can originate new pathways in the design and fabrication of monocoque armour plates exhibiting appreciably less weight with better ballistic protection that may not be attained through a conventional approach. Placing emphasis on weight reduction, which is a requisite for the selection of high specific strength materials in aerospace and armour applications, previous studies have focused on the fabrication of aluminium-based composites by strengthening them with high-strength and rigid particulate ceramics such as B4C, SiC, Si3N4, TiB2 and TiC [10,11]. In C4 composites, the proportion of the ceramic phase can be higher than that of particle-reinforced composites due to their interpenetrated structure. Among the available set of ceramics, SiC is the preferred choice due its mechanical properties, ease of availability, and economic factors. Porous SiC preforms which are interconnected in three dimensions and impregnated with Al alloys, therefore, have the potential for applications in transportation and armour [12,13]. Nong et al. [14] reported the fabrication of 3D SiC/Al co-continuous composite to produce a ventilated shaft disc brake. The wear and friction behaviour of the prepared C4 in this study was comparable to that of cast iron and steel. In addition, better thermal conductivity and better wear resistance were obtained at half the density [15]. Bahrami et al. [16] fabricated bilayer Al/B4C/rice husk ash composites by the pressureless infiltration method. The study revealed that the two factors namely, initial preform porosity and chemical composition of infiltrated alloys, exerted a substantial influence on the electrical resistivity and coefficient of thermal expansion, respectively. Pressureless infiltration was also utilized to fabricate Al/Si3N4 silica composites by Soltani et al. [17]. The results depicted that the processing temperature significantly influenced the modulus of elasticity of the composite. Recent studies by Prasanth et al. [18,19] report that gravity infiltration of SiC co-continuous foam with AA7075 and Al 6063 alloy is effective in enhancing the wear and toughness of the prepared C4 when compared to a monolithic infiltrant material. The compressive strength of such IPCs, produced by squeeze casting, was reported to be 660 MPa. This is higher than that of traditional composites reinforced with SiC particles [20]. Similarly, in another study, infiltration of Ni3Al alloy into porous aluminium oxide by gas pressure infiltration was investigated. Composites with a low volume of Ni3Al showed a fracture strength of 400 MPa. The highest volume fraction of Ni3Al (30 vol.%) displayed a higher fracture strength of 675 MPa [21]. Among the diverse Al alloys, AA5083 is a potential alloy for the regime of C4 composites. Nevertheless, a systematic assessment of AA5083 as an infiltrant material to produce C4 is essential. Though the pressureless infiltration method [22,23] is economical and has the benefits of easy industrialization, pressure infiltration [24,25] is preferred for the manufacture of C4 due to its lower infiltration time, performance, and efficiency. A recent study by Zhang et al. [26] reported that mechanical-pressure infiltration method was effective for producing Al2O3/Al based materials for scaffoldings. The study also reported that the pressure infiltration technique enhanced the interfacial bonding between Al2O3 layers and the infiltrant Al alloy. Though C4 composites have been extensively analyzed for wear applications, studies with an emphasis on their compressive behavior for armour applications are scarce, to the best of the authors' knowledge.

Therefore, the focus of this investigation is on the synthesis, microstructural characterization, and a detailed analysis of the compressive behavior of AA5083/SiC C4 composites manufactured through the gas pressure infiltration technique. The properties of the constituent ceramic foam were also analyzed and discussed in comparison with the C4.

#### **2. Materials and Methods**

The materials utilized to manufacture the C4 composites are delineated in Sections 2.1 and 2.2 hereunder. The method of infiltration employed to manufacture the C4 is detailed in Section 2.3. Sections 2.4 and 2.5 describe the analyses of microstructural and mechanical properties respectively, conducted on the SiC foams and the C4 composites.

#### *2.1. Ceramic Foam*

In this study, SiC ceramic foams consisting of two different pore sizes, namely 10 and 20 pores per linear inches (ppi), were employed as the reinforcement network. The foams were manufactured by the replica method and commercially sourced from M/s Eltech Ceramics, Tamilnadu, India. Prior to fabrication of the C4, the chemical composition and purity of the SiC foams were characterized by X-ray diffraction (Empyrean Malvern Panalytical, Malvern, UK). Copper K-α X-rays of 1.5406 Å were utilized and the diffractions were collected at a scanning rate of 2◦/min. The collected diffractogram was indexed and the peak intensity was used to determine the chemical composition of the SiC foams.

#### *2.2. Infiltrant Alloy*

The present study utilizes a commercially sourced AA5083 alloy from Disha Steels, Maharashtra, India in order to infiltrate the pores of the SiC foam to create the C4. The chemical composition of the as-received alloy, shown in Table 1, was analyzed using Optical emission spectroscopy (AMETEK-SPECTROMAXx, Kleve, Germany).


**Table 1.** Chemical composition of AA5083.

#### *2.3. Manufacturing of C4*

In this study, two configurations of C4 were manufactured by low pressure infiltration of AA5083 into the pores of the 10 and 20 ppi SiC foams respectively. A custom-designed Ar gas based set-up, as depicted in Figure 1, was utilized to perform the low pressure infiltration. The set-up comprises a pressure chamber of dimensions Ø100 mm × 450 mm capable of heating to a maximum temperature of 1200 ◦C and withstanding a maximum pressure of 6 bar. The pressure chamber was designed such that it served as a crucible for the dual purpose of melting and infiltration. The crucible containing SiC foam was preheated to 750 ◦C and a measured quantity of aluminum AA5083 ingot slices were added for melting. This specific sequence aided in reducing thermal shock exerted on the ceramic foam due to the difference between room temperature and the furnace atmosphere. It further prevented sudden clogging of molten Al during the infiltration process. Subsequently, the melt was superheated to 795 ◦C to accelerate the kinetics of infiltration. To facilitate pressure infiltration, the atmospheric air present in the chamber was evacuated by a rotary vacuum pump to a level of 10−<sup>2</sup> bar. Subsequently, Ar inert gas was purged into the sealed chamber to raise the pressure to 4 bar. The composite melt was maintained in this environment for one hour to ensure complete infiltration. Thereafter, the infiltrated composite was allowed to solidify and cool to 400 ◦C in the furnace and eventually, by air cooling. The fabricated C4 was then machined out by a series of facing, turning and grinding operations.

**Figure 1.** Gas pressure infiltration setup.

#### *2.4. Analyses of Microstructural Properties*

Subsequent to manufacturing the C4, the quality of infiltration in both configurations of SiC foam and C4 samples was determined by measuring the porosity levels using Archimedes' principle. The average porosity of three foam samples for each of the two configurations was determined using Equations (1)–(3).

$$\text{Porous Density, PD = } \frac{\text{Mass of porrous foam}}{\text{Volume of porrous foam}} \text{ g/cc} \tag{1}$$

$$\text{Bulk Density, BD} = \frac{\text{Aggregrate mass of SiC particles in the foam}}{\text{Volume of porous foam}} \text{ g/cc} \tag{2}$$

$$\text{Percentage Proosity, PP} = \left(1 - \frac{\text{PD}}{\text{BD}}\right) \times 100\tag{3}$$

The porous density (PD) of the foam in Equation (1) represents the density encompassing the pores in the foam structure. The bulk density (BD) in Equation (2) was estimated using the actual quantity of SiC particles that constitute the foam. The percentage porosity (PP) in Equation (3) is a measure of the air pores available to be completely filled by AA5083 during the infiltration process. The volume fractions of the ceramic and Al phases in the C4 were estimated by Archimedes' principle using Equations (4) and (5).

$$\text{Volume fraction of Al in C4} = \frac{\text{Volume of Al in C4}}{\text{Volume of C4}} \tag{4}$$

$$\text{Volume fraction of SiC in C4} = \frac{\text{Volume of SiC in C4}}{\text{Volume of C4}} \tag{5}$$

Next, microstructural studies were performed on the two SiC foam configurations and the C4 composites thus manufactured. All metallographic samples analyzed in this study, were prepared by standard metallographic sample preparation procedures using a Struers Tegramin-25 grinding and polishing machine. The optical micrographs of un-etched samples were captured using a ZEISS Axio Imager M2m optical microscope (OM).

#### *2.5. Analyses of Mechanical Properties*

The mechanical properties of SiC foams and the C4 were assessed through compression tests. A set of three samples each from SiC foams and C4 of dimensions Ø70 mm × 22 mm and Ø13 mm × 25 mm, respectively, were utilized as compression specimens. The compression tests were performed in accordance with ASTM E9 standard using a universal testing machine (Tinius Olsen, Norway) of capacity 50 kN with a cross head velocity of 0.5 mm/min. Subsequently, fractography analysis was performed in order to substantiate the observations from compression tests. The fractured face and the lateral sides of the specimen were examined using an optical microscope (Dino-lite capture) for indications of crack lines, shear failure, and vertical splitting failure of composite.

#### **3. Results and Discussion**

#### *3.1. Metallographic Analysis of SiC Foams*

The X-Ray diffraction pattern was extracted for the two configurations considered in this study, namely the 10 ppi (F10) and 20 ppi (F20) SiC foams. The identified constituents of the F10 and F20 foams and crystallographic structures along with their Joint Committee on Powder Diffraction Standards (JCPDS) reference patterns are listed in Table 2.


**Table 2.** SiC foam compounds identified by X-ray diffraction.

Next, during quantitative phase analysis, a refinement was performed with the High Score Plus 4.8. The results of the quantitative analysis of phase weight fractions and the percentage of each constituent for F10 is shown in Figure 2.

Figure 3 presents the indexed diffraction pattern for F20 along with the identified phases. All major peaks were assigned to the dominant phases, namely SiC and Al2O3. In addition, small peaks corresponding to CaCO3 and SiO2 were also identified. The compositional details of each compound are listed in Figure 3.

**Figure 2.** XRD pattern of F10.

The peak intensities of F10 and F20 as depicted in Figures 2 and 3 were used to calculate the weight fraction of each phase in a mixture and are tabulated in Table 3. The X-ray diffractograms confirmed that SiC was the major constituent of both the foams. In addition, compounds such as Al2O3, CaCO3, Mg and SiO2 were also observed.


**Table 3.** Weight fraction of phases.

It can be inferred from Table 3 that the SiC content of F20 is higher than that of F10. It is well known that the ceramic phase, namely SiC, is brittle in nature. This can potentially lead to crack initiation and brittle fracture in the ceramic phase when subjected to compression [27].

#### *3.2. Porosity and Structural Analysis of SiC Foams*

The porosity of the foams was estimated using Equations (1)–(3) and is tabulated in Table 4. It can be inferred from Table 4 that the BD and PP of F20 is less than that of F10. This suggests a variation in the morphology of SiC struts between the two foam configurations. Therefore, a detailed study of the morphology of the two foam structures was conducted.



Figure 4a–d represents the morphologies of the F10 and F20 SiC foams respectively. The strut thickness and pore diameters of both foams are tabulated in Table 5.

**Figure 4.** Optical Morphologies of the SiC foams: (**a**); 10 ppi; (**b**) 20 ppi; (**c**) struts; and (**d**) triangular voids in foam.

**Table 5.** Comparison of foam morphologies.


It can be inferred from Table 5 that the strut thickness and pore diameter of F20 is less than that of F10. Figure 4b reveals that F20 is highly interconnected when compared with F10. This signifies that a foam with lower strut thickness and pore diameter will possess a highly interconnected network of SiC. Additionally, it can be deciphered from Table 5 that the pore diameter decreases with an increase in the ppi of the foam.

In addition to the pore diameter and strut thickness, the strength of struts is a crucial factor influencing the mechanical strength of the foam. The SiC foams considered in this study consisted of an open-cell structure with a network of voids. These foams, as shown in the insets of Figure 4a,b, are termed as reticulated ceramics [28]. The extreme porosity, interconnected void volume, adjacent pores and their light weight make reticulated SiC

ceramic foams ideal for the infiltration of molten metal [29]. As evident from Figure 4c,d, these porous structures possess hollow triangular voids that led to a reduction in their mass. Investigation of the strut structure of the F10 and F20 foams revealed that the estimated average side length of the triangular voids was ~466 μm and ~377μm, respectively.

#### *3.3. Microstructural Analysis of C4*

The optical micrographs of the as-cast C4 samples comprising the F10 and F20 foams infiltrated with AA5083 are shown in Figure 5. The ability to bear the compressive load exerted on the composite substantially depends on the morphology of the foam and infiltrant Al [20]. It can be observed that both infiltrated composites exhibit a spherical morphology. In particular, the C4-F10 exhibited a single lobe structure when compared with C4-F20 which reveals a double lobe structure. This distinction can be attributed to the small pore size of the F20 as detailed in Section 3.2. During formation of the composite, the molten Al impregnates and fills the pores of the foam, resulting in the characteristic lobe structures of the F10 and F20 foams. These distinctive morphologies, namely the lobe structures, have a salient effect on the compressive load-bearing capacity of the resulting C4 composite. Figure 5 also depicts through interpenetration of the Al alloy inside the voids of both SiC foams. The volume fractions estimated using Equations (4) and (5) are listed in Table 6. The table denotes that both configurations of C4 have approximately 80% by volume of AA5083 infiltrated into the SiC foams.

**Figure 5.** Microstructure of as-cast composite samples: (**a**) C4-F10; and (**b**) C4-F20.


**Table 6.** Volume Fraction of the C4.

#### *3.4. Mechanical Property Analysis*

As delineated in Section 2.5, the assessment of mechanical properties was performed by conducting compression tests on the two foam configurations, namely F10 and F20, and on the manufactured co-continuous composites, namely C4-F10 and C4-F20. The quasistatic stress-strain response of the SiC foams and three samples each of the C4 composites are depicted in Figure 6. The inferences of the salient outcomes from the compression tests are tabulated in Table 7.

**Figure 6.** Compressive stress-strain behaviors of (**a**) 10 ppi SiC foam and its C4 (**b**) 20 ppi SiC foam and its C4.

**Table 7.** Inferences from Compression Tests.


In the displacement curves in Figure 6, the strength of the C4 and SiC foam are marked on the primary and secondary *Y* axis respectively. From Table 7, the yield strength of F10 and F20 are observed to be ~1 MPa and ~1.3 MPa, respectively. In comparison, the displacement curves of the C4 show that the infiltration of AA5083 melt into the pores of SiC imparts an enhancement in yield strength with values of ~74.3 MPa and ~71.6 MPa for C4-F10 and C4-F20 samples, respectively. As can be deciphered from Table 7, the compressive strength of the SiC foam for the 10 and 20 PPI configuration was ~1.22 MPa and ~2.05 MPa. In contrast, the compressive strength of the C4-F10 composites was ~126 MPa and that of the C4-F20 configuration was ~120 MPa. It can be inferred that the compressive strength of 10 PPI SiC foam was enhanced by about 100 times by infiltrating with ~81 vol.% of AA5083 alloy. In comparison, the compressive strength of the 20 PPI SiC foam improved by close to 58 times when infiltrating with ~77 vol.% of AA5083 alloy.

Table 7 also lists the energy absorbed (EA) per unit volume by the foam and the C4 during compression tests. It was observed that the EA of SiC foam was estimated to be ~1.07 J/mm3 and ~1.68 J/mm3 for 10 PPI and 20 PPI foams respectively. The infiltration of SiC foams with AA5083 enhances the energy absorbed per volume of the C4 samples in both 10 and 20 PPI configurations to ~14.17 and ~13.39 J/mm3, respectively. In addition to the strength and EA, the infiltration of AA5083 alloy remarkably enhanced the elastic modulus from ~0.96 and ~2.3 MPa for the foams to ~2.67 and ~2.69 GPa for the equivalent C4 composites as observed from Table 7. Therefore, it is evident that the infiltration of AA5083 into SiC ceramic foam simultaneously enhances the strength, elastic modulus and toughness, quantified by EA, of the resulting C4.

A salient inference from Table 6 is that the volume fraction of AA5083 in both 10 and 20 PPI composites are nearly the same. However, the compressive strength and EA of the composite is higher in case of the C4-F10 when compared with C4-F20, indicating that the compressive strength depends on the characteristics of the foam. The foam characteristics include PPI, pore diameter and strut thickness.

#### *3.5. Compressive Behavior of C4*

In order to comprehend the behaviour of the C4 when subjected to compression, a typical stress–strain curve obtained for the C4-F10 is shown in Figure 7. The curve can be segregated into four regions. The smooth line in region AB represents the elastic zone of the composite, during which no cracks were observed. Region BC represents the transition from elastic to failure zone. This was characterized by a minor bend in the curve with a reduced slope when compared to region AB. No cracks or strut failures occurred in the region AC. Subsequently, when the C4 traverses the transition region of BC, failure initiates after point C. It was observed during compression tests that the failure of the composite initiated within the foam was due to its brittle nature. The saw-tooth pattern in region CD was therefore attributed to the progressive cracking of the SiC struts. This was characterized by a cracking noise during testing. In this region CD, the AA5083 in the composite bears the load until point D. Finally, in region DE, the saw- tooth pattern was found to occur due to the splitting of the SiC-Al interface of the C4.

**Figure 7.** Compressive behavior of C4-F10.

#### *3.6. Fractography Analysis*

Analysis of the fractured surfaces, referred to as fractography, is essential to reveal the genesis and mechanism of failure [30]. Figure 8a shows the macroscopic image of the specimen of C4 before application of the uniaxial compressive load. It is evident from the figure that the AA5083 (bright phase) has infiltrated the pores of the SiC foam (dark phase) to form the C4. During compression tests, it was observed that, cracks first initiate at the SiC foam and then propagate to the ductile Al phase as evident from Figure 8b. Since both the C4-F10 and C4-F20 comprise approximately 80% of ductile Al and 20% of brittle SiC foam in a bulk form, it is expected to follow the 'shear failure' mode, characteristic of a ductile material. However, as evidenced from Figure 8b, the crack lines deviate from the theoretical line due to disruptions caused by SiC struts. Minor cracks of low intensity were observed to be formed and to propagate for short distances within the specimen. Although the 'shear failure' was expected, other types of failure such as 'Vertical splitting' were found to occur. Vertical splitting refers to failure along the direction in which the compressive load is applied [31]. This resulted in major cracks originating and propagating along the SiC struts, as deciphered from Figure 8c. The study of the fracture surfaces thus revealed that, the failure of C4 composites is contingent on the profile and orientation of the SiC struts. Hence, it can be proposed that, cracks initiate in the region of weak cell-walls of the foam structure. Subsequently, the cracks propagated to the surrounding cell-walls and plateau junctions. This correlates well with prior observations reported in literature [32–34].

**Figure 8.** Macroscopic fractography of C4 composites: (**a**) composite without compression; (**b**) composite crack lines showing shear failure; and (**c**) composite crack lines showing vertical splitting failure.

#### *3.7. Microstructure of Post-Test C4 Composite*

Figure 9a,b exhibit the optical microstructure of the C4-F10 and C4-F20 respectively after compression tests. The microstructure reveals multiple cracks of varied sizes on the SiC struts in both configurations of the C4. In contrast, limited cracks are observed in the AA5083 matrix of C4 composite. This suggests that the quality of interfacial bonding between the Al matrix and foam structure is a paramount factor governing the compressive behavior of C4 composites. The phenomenon of shear failure and vertical splitting discussed in Section 3.6 is also evident from the micrographs.

**Figure 9.** Optical micrograph of post-tested C4 composite (**a**) C4-F10 (**b**) C4-F20.

#### **4. Conclusions**

Two configurations of C4 composites composed of 81% and 77% by volume of AA5083 infiltrated into SiC ceramic foam of 10 and 20 PPI respectively, were synthesized using gas pressure infiltration technique. The foams and C4 were subjected to compression tests. The following conclusions were drawn from the study.


**Author Contributions:** Conceptualization, A.S.P., V.K., J.P.E. and K.S.; methodology, A.S.P., J.N., J.P.E. and K.S.; validation, J.N., J.P.E. and K.S.; formal analysis, R.S. and M.R.A.S.; investigation, J.N.; resources, A.S.P., V.K. and J.N.; data curation, R.S. and M.R.A.S.; writing—original draft preparation, R.S. and M.R.A.S.; writing—review and editing, A.S.P. and J.N.; visualization, V.K. and M.R.A.S.; supervision, V.K., J.P.E. and K.S.; project administration, V.K.; funding acquisition, A.S.P., V.K., J.P.E. and K.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research was funded by The Scheme for Promotion of Academic and Research Collaboration (SPARC), MHRD, India. Project code: P207.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


## *Article* **Organomorphic Silicon Carbide Reinforcing Preform Formation Mechanism**

**Evgeny Bogachev**

JSC Kompozit, 4 Pionerskaya, Korolev, 141070 Moscow, Russia; eug-bogatchev@mail.ru; Tel.: +7-495-513-23-06

**Abstract:** Development of the organomorphic ceramic-matrix composites (CMCs), where the reinforcing preform is built using polymer fibers subject essentially to hot pressing, was motivated by a desire to obtain much higher structural uniformity as well as to reduce the number of the process steps involved in the production of CMCs. This paper addresses the peculiarities of the organomorphic silicon carbide preform formation process. Using X-ray phase analysis, tomography, mass and IR spectroscopy, and thermomechanical and X-ray microanalysis, both the properties of the initial fibers of polycarbosilane (PCS)—the silicon carbide fiber precursor—and their transformation in the preform while heated to 1250 ◦C under constant pressing at 10–100 kPa were studied. Analysis of the data obtained showed the organomorphic SiC preform relative density at a level of 0.3–0.4 to be ensured by self-bonding of the silicon carbide preform, resulting from the fact that during the low-temperature part of pyrolysis, easily polymerizing substances are released leaving a high coke residue, thus cementing the preform. Another possible factor of SiC framework self-bonding is the destruction of the polymer fibers during pyrolysis of various PCS preforms differing in their methylsilane composition (for example, dimethylsilane), where deposition of silicon carbide on the contacting fibers starts as early as at 450–500 ◦C.

**Keywords:** polycarbosilane (PCS); fiber; pyrolysis; self-bonding; silicon carbide; organomorphic preform (OP)

#### **1. Introduction**

At the core of ceramic matrix composites (CMCs) is their reinforcing system. The existing technology for fabricating the reinforcing system involves multiple stages. These stages include a number of operations, namely, the production of elementary organic fibers from the relevant precursor, whereupon they are subject to step-by-step pyrolysis as part of a multifilament (made of thousands of filaments) tow: manufacturing of a fabric or a tape to be impregnated with a binder to obtain the prepreg; molding of the preform by curing; and carbonization that involves high-temperature annealing. There are certainly other ways for obtaining the reinforcing system using the available mechanical binding methods (multi-dimensional weaving, needle punching, and braiding); however, the reinforcing frames obtained using the aforesaid methods show a relatively low density and cannot be used as a basis for structural composites without the use of a binder. The numerosity of the operational processes involved results in the high fabrication cost of the preform as well as the CMC itself.

A special feature that defines the CMCs obtained using conventional methods is their rather non-uniform structure that, for the time being, cannot match the structural uniformity of graphite, metal, and ceramics. This prevents us from expanding the scope of CMC application as an alternative to these materials. It seems that any efforts to solve this problem by modifying a conventional preform, one way or another, are bound to fail. The use of multifilament tows, which are immanent features of the state-of-the-art technology, will result intrinsically in a non-uniform distribution of the reinforcement in relation to the composite volume, since the filament diameter range between 7 μm and 15 μm, and the size

**Citation:** Bogachev, E.

Organomorphic Silicon Carbide Reinforcing Preform Formation Mechanism. *J. Compos. Sci.* **2023**, *7*, 81. https://doi.org/10.3390/jcs7020081

Academic Editor: Jinyang Xu

Received: 30 December 2022 Revised: 21 January 2023 Accepted: 8 February 2023 Published: 15 February 2023

**Copyright:** © 2023 by the author. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

of the voids between the multifilament tows will make up 300–400 μm to 700–1000 μm and between the filaments—0.3–0.7 μm. The notable micro non-uniformity of the reinforcing preform will impose a natural limitation on its minimum thickness. Obtaining a composite material where the elementary fibers alternate uniformly with the matrix, thus bringing the dimensional limit of the material essentially about the diameter of the fiber itself, would bring the lower limit of the thickness of the parts down to 150–200 μm. Such a possibility, along with a notable reduction in the number of the process steps, may expand significantly the scope of application of CMCs through inclusion of many promising applications where CMC can be used instead of graphite, traditional ceramics, or metal.

The search for a solution of this problem resulted in the development of the so-called organomorphic CMCs (C/C, C/SiC, and SiC/SiC) reinforced with a framework based on organic polymer fibers used for the most common carbon and ceramic fibers, namely, polyacrylonitrile (PAN), polycarbosilane (PCS), and polysilazane (PSZ) [1].

Using PAN fibers as an example, hot pressing (up to 1000 ◦C) of PAN preforms was found [2] to provide carbon reinforcing frames, showing a high relative density of up to 0.4 and an open porosity of 50–60% represented by pores of an equivalent diameter ranging from several micrometers to several tens of micrometers. The method used to build the reinforcing carbon preform was found to provide self-bonding (as inherited after pyrolysis) of the filaments. The new quality of the carbon preforms made it possible both to provide an effective substitute for the molybdenum electrodes used for ion thruster accelerating electrodes [3] and to find more effective methods to CVI them with the silicon carbide matrix [4].

Self-bonding during so-called hot pressing (up to 170 ◦C) is a well-known phenomenon typical of natural fibers such as cotton and wood [5–8]. This phenomenon, resulting from a physical contact between the fibers as well as the presence of a bonding-conducive polymer on their surface, makes it possible to fabricate high-strength boards made of natural materials without any binder.

However, the organomorphic CMC preforms made of polymer fibers are subject to annealing at up to 1800 ◦C, and they are not limited by the process temperatures characteristic of the natural fibers. Therefore, it is important to determine the presence of self-bonding in the PCS preforms pyrolized under pressing, including at the low-temperature (up to 400 ◦C) pyrolysis stage. Formerly, self-bonding was observed in pyrolized PAN-based preforms [1,2].

The PAN- and PCS-fibers differ significantly in their chemical composition and structure. At the same time, it seems that the self-bonding discovered previously for PAN fibers during pyrolysis under pressure must be basically of general nature, and the governing laws of obtaining the carbon and SiC preforms, respectively, must be more or less the same.

#### **2. Materials and Methods**

The PCS fibers of Kompozit JSC make, thermally stabilized at 200 ◦C (molecular mass distribution Mn ≥ 1600, polydispersity coefficient D ≤ 1.8, 90% wt. residue after etching treatment) in air, 19–24 μm in diameter, were studied with the Skyscan-2011 nanotomograph (London, United Kingdom) using the following scanning parameters: U = 50 kV, I = 200 mkA, rotation angle—0.3◦ as averaged over 5 frames, and spatial resolution—0.3 μm/pixel.

The PCS fiber behavior, while under load, was studied with the thermomechanical analysis method (TMA) using the TMA Q400 EM (New Castle, Delaware, USA). The test sample in the form of layers of fibers, 7 mm in diameter and 3.4 mm in thickness, was placed in a special pot made of high-density organomorphic C/C, 7 mm in diameter, 11 mm in height, and 1.5 mm in wall thickness. Next, the fiber sample was pressed with a quartz indenter, 2.54 mm in diameter (Figure 1), to be heated up to 400 ◦C in air at a rate of 1 deg/min, under a static load of 0.588 N and a dynamic load of (±0.294 N), keeping a record of the time change in the elasticity modulus and sample thickness.

**Figure 1.** EXPANSION probe.

The structure and functional groups of the PCS fiber macromolecules, both before and after TMA, were analyzed using X-ray tomography with the XT H 320 LC X-ray tomography system (METRIS, Tring, UK) and using IR spectroscopy with the Netzsch STA 449 F3 Jupiter (Selb, Germany).

The PCS fiber thermal decomposition kinetics was studied in high-purity nitrogen flow with the TGA-DSC method using the TA Instruments SDT 650 synchronous differential thermal analyzer (New Castle, DE, USA) at the rate of a temperature rise of 10 degrees per minute.

The PCS fiber gas emission during heating up to 800 ◦C in vacuum was studied using the laboratory mass spectrometer equipped with the CIS300 quadrupole mass analyzer (Sunnyvale, CA, USA). The PCS fibers, 1–5 mg in weight, were placed in a quartz ampoule connected to the quadrupole mass spectrometer vacuum system. The ampoule was pumped out at the room temperature down to ~10−<sup>3</sup> Pa and connected to the mass spectrometer chamber. While keeping continuous record of the mass spectra, the ampoule was subject to heating from the room temperature up to 800 ◦C at a constant rate of 5 deg/min. The mass spectra were recorded every 90 s, over the mass number range of 1–220. Simultaneously, record of the pressure in the mass spectrometer chamber (sensor MID-2) and in front of the mercury diffusion pump behind the nitrogen trap (sensor MID-1) was kept.

To obtain the organomorphic preforms from the silicon carbide fiber, the fibers in the form of a tape composed of 150-filament threads were placed in a graphite container with the internal dimensions of 120 × 60 × 28 mm in the directions of 0◦ × 0◦ and 0◦ × 90◦, whereupon a commensurable massive graphite punch was installed onto the fibrous PCS workpiece, placing an additional load on the top [9]. The total load on the fibrous polymer workpiece (cover + load) was 700 N. The ready assembly was installed in a vacuum resistance furnace to be heated up to 1250 ◦C in non-oxidizing gas (nitrogen) at a rate of 300–400 deg/h. After cooling down, the organomorphic SiC preform was removed from the container to determine the density, whereupon test samples of about 20 × 10 mm in size were cut out of it with a sharp knife for microstructural analysis.

The density and porosity of SiC preforms were measured using the Archimedes method. To study the structural changes in the material, the scanning electron microscope FEI Quanta 600 FEG (Eindhoven, The Netherlands) with field emission and integrated microanalysis system EDAX TRIDENT XM 4 was used. For X-ray phase analysis, the Empyrean diffractometer (PANalytical B.V., Willmington, Delawer, USA) equipped with the specialized HighScore Plus software for phase analysis with a built-in database of reference structures PAN-ICSD (Inorganic Crystal Structure Database) was used.

#### **3. Results**

#### *3.1. Initial PCS Fiber Properties*

Analysis of the initial PCS fibers reveals a completely amorphous and disordered nature of the structure, as evidenced by the fact that the small-angle scattering shows no reflection (Figure 2).

**Figure 2.** Meridional (1) and equatorial (2) scans of PCS fiber bundles in the small-angle area.

Microtomography of the polymer fibers shows no systematic difference in the nearsurface and intra-volume areas (Figure 3).

It is clear, however, that there are more or less dense microvolumes in the fibers, which speaks for certain structural defects of the organic filaments.

#### *3.2. Analysis of the PCS-Fiber Properties during the Pyrolysis*

Thermomechanical analysis of the PCS fiber bundle 3-mm thick reveals complex deformation of the test sample under simultaneous static/dynamic loading with rising temperatures (Figure 4).

**Figure 3.** Interperpendicular sections of PCS fibers.

**Figure 4.** Elasticity modulus vs. temperature curves obtained for the PCS fiber bundle.

What draws attention is the timely coincidence of the sharp decrease in the sample thickness and the rapid increase in the elasticity modulus—both processes start at 215–220 ◦C. The decrease in thickness can be explained both by the partial tangential displacement of the fibers towards the pot walls with rising temperature due to friction reducing between the fibers and the increase in the degree of their contact simultaneously with the deformation of the fibers themselves at a temperature preceding the pyrolysis onset temperature. On the DTG curve, the onset of gas emission from PCS fibers practically corresponds to the start of the events during TMA (Figure 5).

**Figure 5.** Thermogravimetric analysis of PCS fibers.

Mass spectroscopy of the PCS fibers shows that, according to the MID-2 pressure sensor readings, the low-temperature peak of gas emission starts at 215 ◦C, reaches its maximum at 270 ◦C, and ends at 470 ◦C (Figure 6).

The MID-1 pressure sensor fails to record this broad peak, meaning that in the specified temperature range, release of volatile substances with the mass-to-charge ratios of 18 (H2O), 28 (CO), and 44 (CO2) that condense at the liquid nitrogen temperature takes place. These low-molecular weight compounds ensure the loss of about 2% wt. of PCS fibers in the temperature range under study, i.e., 215–470 ◦C (Figure 5). Removal of oxygen-containing compounds from the composition of the fibers results inevitably in the partial breakage of the oxygen cross-link in PCS macromolecules, ensuring an increase in their mobility and ability for autohesive interaction.

The second (higher) peak of gas emission starts at 500 ◦C, reaches its maximum at 660 ◦C, and ends at 740 ◦C. In this case, the MID-1 pressure sensor does show it, meaning that it is associated mainly with release of hydrogen, since it is not frozen at the liquid nitrogen temperature. The mass spectrometry results also correlate with the DTA data of differential thermal analysis (Figure 5). However, according to [10], a significant mass loss (more than 7% wt.) is provided by the evolution of both hydrogen and methane.

Thus, in the range of 215–400 ◦C, transformations occur both in the composition and in the structure of the PCS fibers, leading both to a decrease in the thickness of the compressed sample and an increase in its elasticity modulus.

**Figure 6.** Changes in the total gas release from the PCS fibers when heated in the mass spectrometer.

*3.3. Analysis of the PCS-Based Organomorphic Preform (OP) Microstructure and Properties after Different Stages of the Pyrolysis*

After TMA, the test sample represents a fairly dense tablet (Figure 7).

**Figure 7.** PCS test sample before (**a**) and after (**b**) thermomechanical analysis at up to 400 ◦C.

Analysis of the PCS sample microstructure after TMA shows clearly visible insular buildups on the fiber surface (Figure 8), while in the initial state the fiber surface was free of them.

**Figure 8.** Morphology of PCS fibers before (**a**) and after (**b**) thermomechanical analysis at up to 400 ◦C.

These buildups are not present on all fibers and are characterized by irregular concentrations. It is probably for that reason that, after TMA, the PCS sample shows less irreversible compression in the indenter pressure area than the PAN sample, the entire surface of which is covered with resin-forming compounds that effectively bind the filaments [2] (Figures 9 and 10).

**Figure 9.** Morphology of PAN fibers after pressing at 180 ◦C [2].

**Figure 10.** Tomographic images after TMA: (**a**) PCS fiber and (**b**) PAN fiber [2].

**Figure 11.** Results of IR-spectroscopy of the PCS fibers before (1) and after (2) TMA.

After TMA, signals at 1370, 1750, and 2960–3020 cm−1, part of which are related to oxygen-containing bonds, disappear. These changes also correlate with both the results of mass spectroscopy and the compressed fiber transformation. The persistence of the irreversible deformation of the PCS sample after TMA, albeit to a lesser extent than in the case of the PAN sample (Figure 10), explains the increase in the elastic modulus of the compressed region during TMA; formations protrude onto the surface of the filaments and interact with each other autohesively to fix the gradual transformation of individual fibers into a single whole.

Analysis of the PCS fiber sample composition using IR spectroscopy showed the sample underwent significant changes during TMA; what remains in the IR absorption

Therefore, the insular nature of the self-bonding areas of the fibers does not prevent the formation of dense and highly processible reinforcing silicon carbide frames of various reinforcement structures that proceed from organic to inorganic state, accompanied by their shrinkage as a whole (Figure 12).

(**c**)

**Figure 12.** OP-SiC: (**a**) unidirectional PCS tape in a container before pyrolysis; (**b**) SiC frame, 90 × 45 × 4 mm in size, reinforcement scheme 0◦/0◦ [1]; and (**c**) SiC frame, 90 × 45 × 3.5 mm in size, reinforcement scheme 0◦/90◦.

Characteristics of the organomorphic silicon carbide preforms are given in Table 1.



The diffraction pattern obtained for the SiC frame after PCS preform pyrolysis shows two maxima: one of angular width 2θ of about 10◦ and the other of −20◦ (Figure 13).

**Figure 13.** Diffraction pattern of the organomorphic SiC-preform (wide peak analysis was performed using the specialized HighScore Plus software for phase analysis).

Absence of narrow (angular width 2θ < 2◦) diffraction peaks from the crystalline phase is indicative of a disordered—close to amorphous—state of the frame fibers. This is quite understandable, since the structure of the initial PCS fibers is completely amorphous (Figures 2 and 3) and the maximum temperature for obtaining the organomorphic SiC framework (1250 ◦C) does not reach even half of the incongruent decomposition temperature of silicon carbide [12]. However, the first—less wide—maximum corresponds to the strongest line (111) of β-SiC (reference code 98-002-8389), since this modification is formed during annealing of silicon carbide in a nitrogen atmosphere [13]. The second wide peak is apparently a superposition of other two strong lines, namely, (022) and (113).

Analysis of the silicon carbide preform microstructure confirms presence of some sort of bridges binding the fibers to each other (Figure 14).

According to energy dispersive spectroscopy results (Table 2), the chemical composition of the binding bridge (see Figure 14b; spectrum 1 and spectrum 2) can be represented approximately as Si:C:O = 1:13:1, which indicates enrichment with carbon.

**Element Content, % at. Spectrum 1 Spectrum 2 Spectrum 3 Spectrum 4 Spectrum 5** Si 6.45 6.41 23.42 33.25 26.98 C 84.81 86.71 55.22 48.47 54.51 O 8.74 6.88 21.36 18.28 18.51

**Table 2.** Energy dispersive spectroscopy results of the SiC-filaments within the preform after pyrolysis at 1250 ◦C.

The carbon content decreases sharply in the buildups (spectrum 3, spectrum 4) and their composition practically corresponds to the composition of the surface of the fiber itself (spectrum 5). Since the morphology of the SiC fiber buildup is similar to the buildup's nature of the PCS fibers shown in Figure 8, one can suggest similarity of their origin. The excessive carbon in spectrum 1 and spectrum 2 suggests that the carbon polymer can prevail over the organosilicon one in the buildups and accretions. This reveals possible composition difference among the filaments coming out on the surface to form buildups that are composed of parts of PCS macromolecules in the preform under pressing during the low-temperature (up to 400 ◦C) part of pyrolysis.

#### **4. Discussion**

The results obtained place a new light on the processes occurring during PCS fiber pyrolysis. Annealing of loose fibers results in production of also loose silicon carbide fibers. Pyrolysis of the PCS fiber wound on a spool results in poor unwindability of the resulting SiC fibers, as during carbidization, the PCS fibers when shrinking, compress the underlying layers. Finally, PCS fiber pyrolysis under load produces a dense, highly processible silicon carbide framework. Thus, one should recognize that the main factor affecting self-bonding of the PCS fibers during organic-to-inorganic state transition is their compression while being heated. Gluing of the fibers that not only prevents their unwinding during pyrolysis in the wound state but also contributes to organomorphic frame cementation is the result of the formation of buildups or accretions on the fiber surface from the matter released from the PCS fibers during the low-temperature part of pyrolysis under compressive force.

According to [14], for the half-width of the projection of elastic deformation area ro of two equal parallel cylinders of the same material (e.g., PCS fibers in a unidirectional preform), we have:

$$\mathbf{r}\_0 = 1.52(\mathbf{qR}/2\mathbf{E})^{1/2} \tag{1}$$

where q—specific load per unit fiber length in a unidirectional organomorphic frame; R—PCS fiber radius; and E—PCS fiber elasticity modulus.

During contact, for the maximum stress that develops along the axis of the contact area, we have:

$$
\sigma\_{\text{max}} = 0.418 \text{(2qE/R)} \tag{2}
$$

The maximum tangential stress τ(0, z) falls on the following depth:

$$\mathbf{z}\_{\rm O} / \mathbf{r}\_{\rm O} = \mathbf{0}.8 \tag{3}$$

and makes up about one third of σmax.

Estimates made using Formulas (1)–(3) show that with the ro value of about 4.5 μm, the maximum stress σmax values make up approximately 0.03 MPa and the maximum tangential stress of about 0.01 MPa develops at a depth of 3.6 μm. In addition, it should be

borne in mind that due to an uneven degree of the fiber contact as observed in the actual preforms, the stress values in some contact bundles may be significantly higher.

Thus, both at the contact area and at the depth comparable to the half-width of the contact area, well-marked normal and tangential stresses develop that may affect the state of PCS macromolecules in the polymer fibers during heating and its accompanying shrinkage.

As evidenced by the configuration of the bridge between the filaments (see Figure 14), deformation of the buildups may appear even in the form of a viscous flow. Moreover, the polymer coming out on the fiber surface may be capable of curing, which may cause certain changes in the nature and a sharp increase in the elasticity modulus at above 200 ◦C (see Figure 4). The other reason for modulus increase is the increase in the degree of fiber contact during TMA [15].

It is obvious that these buildups come out on the filament surface from the bulk and correlate with the fiber composition. Since thermal stabilization of the PCS fibers in air is of a diffusive nature, the parts of PCS macromolecules possessed of greater mobility may diffuse through the defects and discontinuities in the more cross-linked surface layer under stress-strain conditions and upon reaching the onset temperature of partial decomposition of oxygen crosslinks. Undoubtedly, they are capable of active autohesive interaction with the surface of adjacent filaments.

Variations in the buildup composition over a wide range of carbon, oxygen, and silicon content, as well as their different concentration on the PCS filament surface, are likely to be associated with different defects in the local areas of the fiber surface and with difference in the degree of stitching of their respective microvolumes. This makes different parts of the PCS macromolecules come out on the surface under stress-strain conditions. Another possible reason for formation of SiC frames is the release of various methylsilanes from the polymer fibers during pyrolysis of the PCS preforms [16], among which dimethylsilane (CH3)2SiH2 was detected by mass spectroscopy. Decomposition of this compound accompanied by deposition of silicon carbide starts as early as 450–500 ◦C.

#### **5. Conclusions**

The carbon organomorphic preform formation mechanism studied earlier using PAN fibers as an example [2] showed that self-bonding of the organic fibers within the preform is due to the presence of a solid polymer film on the fiber surface and a continuous physical contact between the filaments during pyrolysis when combined with hot pressing.

The investigation into the organomorphic silicon carbide preform formation mechanism reveals similarity of its underlying laws with those inherent in the formation of carbon frames. However, the insular nature of the buildups composed of the matter released from the PCS fibers during the low-temperature (up to 400 ◦C) part of pyrolysis reduces the volume-wise continuity of self-bonding in the organomorphic SiC preform. All the while, the relative densities of the silicon carbide preforms are still higher than 40%.

Apparently, the fundamental similarity of the self-bonding nature of the PAN- and PCS-preforms allows us to formulate the following basic prerequisites for origination of the contacts between the fibers within the preforms:

1. Stabilized structure of the polymer fibers to exclude their melting during pyrolysis.

2. Chemical purity of the surface of the contacting polymer filaments (absence of any obstacles for autohesion, i.e., diffusive merging of the polymer fibers).

3. Sufficient continuous mechanical pressing of the fibers against each other during pyrolysis (a constant load).

4. Duration of the contact between the filaments, when in the polymer state, sufficient for the autohesion interaction to form.

However, the presumable deposition of a nanoscale SiC layer from dimethylsilane (CH3)2SiH2 on the fiber surface may complicate, at a later stage, the task of obtaining highstrength SiC–SiC CMCs based on the silicon carbide organomorphic reinforcing framework. **Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** All the data supporting the reported results can be found in this manuscript. No additional data are available in the publicly archived datasets.

**Acknowledgments:** The author extends his great appreciation to the colleagues who assisted in carrying out the research: I.O. Leipunsky, Head of the Laboratory of Nano- and Microstructural Materials Science at the Institute of Energy Problems of Chemical Physics of the Russian Academy of Sciences, D.V. Onuchin, Deputy Vice-Rector for Science at the Mendeleev University of Chemical Technology and D.M. Kiselkov, Senior Scientist at the Institute of Technical Chemistry UB of the Russian Academy of Sciences. All individuals included in this section have consented to the acknowledgement.

**Conflicts of Interest:** The author declares no conflict of interest.

#### **References**


**Disclaimer/Publisher's Note:** The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

## *Article* **Optical Detection of Void Formation Mechanisms during Impregnation of Composites by UV-Reactive Resin Systems**

**Benedikt Neitzel 1,\* and Florian Puch 1,2**


**Abstract:** During the impregnation of reinforcement fabrics in liquid composite molding processes, the flow within fiber bundles and the channels between the fiber bundles usually advances at different velocities. This so-called "dual-scale flow" results in void formation inside the composite material and has a negative effect on its mechanical properties. Semi-empirical models can be applied to calculate the extent of the dual-scale flow. In this study, a methodology is presented that stops the impregnation of reinforcement fabrics at different filling levels by using a photo-reactive resin system. By means of optical evaluation, the theoretical calculation models of the dual-scale flow are validated metrologically. The results show increasingly distinct dual-scale flow effects with increasing pressure gradients. The methodology enables the measurability of microscopic differences in flow front progression to validate renowned theoretical models and compare simulations to measurements of applied injection processes.

**Keywords:** void formation; dual-scale flow; permeability; textile preforms; liquid composite molding; fiber reinforced plastics

#### **1. Introduction**

Liquid composite molding (LCM) is an established industrial production technology for manufacturing thermoset fiber composites. Dry textile preforms are draped in a mold, which is subsequently closed, and the semi-finished fiber product is impregnated with a resin system by means of overpressure. During impregnation of the textile preforms, voids are formed, which result in a reduction of the mechanical properties of the molded component [1–3]. One cause of the formation of voids in fiber composite components is inhomogeneous flow processes at microscopic levels inside the textile preforms. Depending on the process parameters, the resin system flows at different rates within the tows of the reinforcement fabrics and channels between the tows, due to the different permeability of the two areas. As shown in Figure 1, a flow front is formed, which can be divided into a saturated, partially saturated, and unsaturated region of the tows [4,5].

**Citation:** Neitzel, B.; Puch, F. Optical Detection of Void Formation Mechanisms during Impregnation of Composites by UV-Reactive Resin Systems. *J. Compos. Sci.* **2022**, *6*, 351. https://doi.org/10.3390/jcs6110351

Academic Editor: Jinyang Xu

Received: 2 October 2022 Accepted: 11 November 2022 Published: 15 November 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

This effect, known as dual-scale flow, is the focus of several studies since it is a major reason for the formation and transport of voids [6–11]. If the flow velocities within the tows and channels between the tows match, void-free components are produced [12–15] and the lightweight potential of the materials is optimally exploited.

The flow of the liquid resin system through the textile preform as a porous medium is described by Darcy's law (Equation (1)).

$$
\vec{w}\_{\text{ll}} = -\frac{\vec{\bar{K}}}{\eta} \cdot \nabla p \tag{1}
$$

where *vm* is the volume-averaged velocity, <sup>→</sup> *K* the permeability tensor, *η* the resin viscosity, and ∇*p* the pressure gradient from inlet to the flow front position.

Assuming that the resin system is an incompressible medium, the law of conservation of mass applies:

$$
\nabla \cdot \boldsymbol{\upsilon}\_{\rm m} = 0 \tag{2}
$$

The semi-empirical Kozeny-Carman equation can be used to determine the macroscopic permeability *K* of the textile preform [16]:

$$K = \frac{r\_f^2}{4k\_c} \frac{\left(1 - \varphi\_f\right)^3}{\left(\varphi\_f\right)^2} \tag{3}$$

where *rf* is the fiber radius, *ϕ<sup>f</sup>* is the fiber volume fraction, and *kc* is the Kozeny constant. The Kozeny constant is highly dependent on the resin used, impregnation direction, and textile preform [17], and thus is not precisely determined [9,18]. Nevertheless, this model allows calculations on the progression of the flow front in the textile preform and the resulting process duration.

However, this model is not suitable for a more detailed consideration of the impregnation of textile preforms [9] since no information is obtained about the microscopic flow processes within and between the tows. The proportion of resin flowing into the individual fiber bundles of a textile preform during impregnation is described in the numerical simulations by means of an extension by a loss term *q*, which depends on pressure *p* and degree of saturation *s* [6]:

$$\nabla \cdot \left(\frac{\stackrel{\rightarrow}{K}}{\eta} \cdot \nabla p\right) = q(p, s) \tag{4}$$

Analytically, the phenomenon of dual-scale flow can be represented by the modified capillary number *Ca\**. The modified capillary number forms the ratio of viscositydependent and capillary force-dependent flow [8,13] and thus allows conclusions about the proportions of the unsaturated region of the flow front [14].

$$\text{Ca}^\* = \frac{\mu \cdot \overline{u}}{\gamma \cdot \cos \theta} \tag{5}$$

where *Ca*∗ is the modified capillary number, *μ* the dynamic resin viscosity, *u* the averaged macroscopic flow velocity, *γ* the surface tension, and *θ* the contact angle between the resin and fibers.

At high injection pressures, the proportion of viscosity-dependent flow predominates. The channels between the tows fill faster than the areas within the tows. This results in the inclusion of elongated micropores in the tows (Figure 2b). If capillary forces prevail, the flow front within the tows progresses faster. Spherical macropores are formed in the channels (Figure 2a).

**Figure 2.** Formation of voids in the dual-scale model; (**a**) Formation of spherical macrovoids in the channels between the tows; (**b**) Formation of elongated microvoids inside the tows.

With the help of the modified capillary number during injection, in conjunction with the geometric structure of the textile preforms, conclusions can be drawn about the void formation in the component [12–15,19].

Models for calculating the resulting void volume content are based on the ratio of the flow front progress within and between the tows. Gueroult et al. [15] contrast the two time scales of the flow time inside the fiber bundles Δ*tt* in relation to the flow time in the channels Δ*tc*.

$$\frac{\Delta t\_{l}}{\Delta t\_{c}} = \frac{K\_{c}}{K\_{t}} \cdot (1 - \varphi\_{FT}) \cdot \left[1 - \frac{F\_{\text{S}} \cdot K\_{c} \cdot \varphi\_{FT}}{d\_{Fi} \cdot (1 - \varphi\_{FT}) \cdot L\_{l} \cdot \mathbb{C}a^{\*}} \cdot \ln\left(\frac{\mathbb{C}a^{\*} \cdot d\_{Fi} \cdot (1 - \varphi\_{FT}) \cdot L\_{l}}{F\_{\text{S}} \cdot K\_{c} \cdot \varphi\_{FT}} + 1\right)\right] \tag{6}$$

where *Kc* is the permeability of the channel, *Kt* the permeability of the tow, *ϕFT* the fiber volume content in the tow, *FS* a shape factor depending on longitudinal or transversal flow direction, *dFi* the diameter of a single fiber, and *Lt* the length of a tow.

A ratio of <sup>Δ</sup>*tt* <sup>Δ</sup>*tc* < 1 describes the advance of the resin within the fiber bundles and resulting emergence of macropores. <sup>Δ</sup>*tt* <sup>Δ</sup>*tc* > 1 implies a faster advance of resin within the channels and the formation of microvoids. At a ratio of <sup>Δ</sup>*tt* <sup>Δ</sup>*tc* = 1, the resin flows at identical velocities in both sections, which means that no air can be entrapped, and no voids are formed due to the dual-scale flow [15].

Validation of such "dual-scale" computational models usually involves evaluating the resulting void volume contents and classifying them into microvoids and macrovoids [20,21]. An alternative is the optical detection of the different flow velocities. The dual-scale effect was investigated by several studies [10,22,23] using a microscope locally during injection. The challenge in this type of analysis is to find a compromise between maximizing the image section of the flow front while retaining locally high resolution. Another possibility to prove the phenomenon of dual-scale flow is numerical simulation. Godbole et al. [24] describe the differences in the flow velocity within and between the fiber bundles by determining the length of the partially saturated flow front *Lps*, by simulations (Figure 3):

**Figure 3.** Length of the partially saturated flow front *Lps*.

Neglecting capillary forces, it is shown that the length of the partially saturated zone remains constant if the flow path is sufficiently long. This finding agrees with the results of Zhou et al. [25,26], who also calculate a constant length of the partially saturated zone. The resulting length depends on the permeability of the textile preforms and the preform geometry, as well as the volume fraction of the fiber bundles of a unidirectional (UD) unit cell:

$$L\_{ps} = \sqrt{\frac{2a}{K\_{yytow}} \left[\frac{h^3}{3}\right]} \cdot V\_{tow-ply} \tag{7}$$

where *a* is half of the width of a tow, *Kyytow* the transverse permeability of a tow, and *h* half of the width of the channel between the tows. *Vtow*−*ply* is the volume fraction of tows inside a UD unit cell and calculated from the proportion of fiber bundles as closed solids in the cross-section of the laminate:

$$V\_{tow-ply} = \frac{a}{h+a} \tag{8}$$

Detailed information about the microscopic dual-scale flow and associated pore formation becomes possible when the state of the impregnation can be imaged holistically in a cavity. The state of the art in investigating void formation and transport are optical methods [27–29]. Furthermore, ultrasonic measurements, as well as X-ray and micro-CT examinations, were conducted in several renowned studies [1].

However, all the mentioned techniques are currently not suited to analyze big areas of the partially saturated zone, because of their limited resolution or the requirement of additives to increase visibility. Due to the limited field of view, a "snapshot" approach to optical measurement of the void distribution is needed; however, a suitable method has not yet been presented for this issue [30].

The novel approach presented in this article is to freeze the complete impregnation process at different filling levels to investigate the flow front section by section. To obtain snapshots of the component impregnation, a methodology is developed below that uses spontaneous curing of a photopolymerizing resin. Unique to this method is the gathering of specimen with spontaneously cured partially saturated flow fronts, which can be holistically observed via microscopy.

The resin systems used for this purpose include photoreactive functional groups that crosslink when exposed to light [31]. Components of such resin systems are monomers, oligomers, and photo initiators. Upon absorption of high-energy light, mostly in the ultraviolet spectrum, the photo initiators form radicals or ions. These serve as initiators for the crosslinking reaction between oligomers and monomers [32,33].

Depending on the resin system used and film thickness, the time required for complete crosslinking can range from a fraction of a second to several minutes [32]. To minimize interferences caused by changing pressure gradients during crosslinking, it is advantageous to react as quickly as possible. Only in the case of spontaneous crosslinking of the complete molded part are the pore formation as well as the pore transport frozen in situ.

Based on the frozen filling samples produced with photopolymerizing resins, dualscale flow in fiber bundles and channels is investigated sequentially along the complete flow front by means of microscopy. The images obtained are used to compare current computational models with the optical measurements. The evaluation of the flow conditions can be used to verify new calculation models and increase the accuracy of FEM calculations.

#### **2. Materials and Methods**

#### *2.1. Experimental Setup*

An injection mold for linear impregnation of textile preforms was designed. It consists of a bottom side made of aluminum with a linear gate and riser, on which a single-layer textile preform of 300 mm length and 130 mm width is draped. The opposite side of the mold consists of a glass plate fixed with an aluminum frame (Figure 4). By means of defined torque for the screws positioned circumferentially on the frame, a constant compression pressure of 4.4 MPa is set.

**Figure 4.** Experimental setup to produce planar test specimens with a photoreactive resin system.

The glass plate is transparent to light in the ultraviolet (UV) range. UV spotlights are mounted above the mold, which completely and uniformly illuminate the cavity when switched on. A photoreactive resin system is used which, when irradiated with UV light, stops the flow process without any noticeable delay. Three YG-TGD20-405 LED emitters from Shenzhen Creality 3D Technology Co, Ltd., Shenzhen, China, with an emitted waveband of 400 nm to 405 nm, at an overall system power of 7.8 W each, are used at 100% intensity. The resin system is injected under constant injection pressure in various gradations.

The mold is divided into five sections and the flow front is stopped after every 60 mm by switching on the UV lamps. This is followed by an exposure time of 60 s, during which the resin system cures completely.

This methodology allows the examination of the entire flow front of the fabricated specimens after curing. Optical studies were performed using a Keyence VHX-7000 microscope at 100× magnification. The aim of the measurements is a quantified mapping of the dual-scale flow behavior for comparison with the calculation results of the theoretical models. For this purpose, the flow front progress within the tows *lt* and in the channels between the tows *lc*, as well as their flow path difference Δ*l*, is measured (see Figure 5).

The process of impregnation is recorded by a camera to measure the mean flow velocity within the sections and calculate the modified capillary number according to Equation (5).

The resin system is injected with constant injection pressure. The examined parameter combinations are shown in Figure 6. Each combination is repeated three times so that a total of 60 test specimens is produced and evaluated. The last mold section with a flow path length of 300 mm is excluded, since the textile preforms could not be completely saturated over the entire length at pressure levels of 0.05 MPa and 0.1 MPa. The pressure levels were selected to show clearly distinct flow path differences; however, the pronounced dual-scale flow leads to a very high number of overlapping microvoids that cannot be thoroughly evaluated in this study.

**Figure 5.** Schematic course of the flow front and division of the flow path into mold sections denominated from inlet to outlet as one to five.

**Figure 6.** Parameter combinations for injection experiments.

For each parameter combination 20 flow path lengths in the fiber bundle *lt* and between the fiber bundles *lc* are determined, their ratio is calculated, and the values are compared with the models of Gueroult [15] (Equation (6)) and Godbole [24] (Equation (7)).

#### *2.2. Materials*

The resin system used is a photoreactive 3D Printing UV sensitive resin from the manufacturer Shenzhen Anycubic Technology Co., Ltd., Shenzhen, China. It is composed of 30% to 60% oligomers (polyurethane acrylate), 10% to 40% acrylate monomers and 2% to 5% photo initiator. All experiments were conducted with the same batch of the resin system to exclude variations in the components between the experiments.

The textile semi-finished product used is a glass fiber filament fabric type 92130 from Porcher Industries Germany GmbH, Erbach, Germany for which the following data was measured, as depicted in Table 1. A fabric with medium grammage and plain weave with a high number of crossovers was chosen to establish a homogeneous flow while producing samples with conventional thickness. Moreover, the fabric is resilient against fiber displacement in the manual preparation process, which can potentially cause local deviations in permeability.


**Table 1.** Measured and calculated parameters of the fabric.

#### **3. Results and Discussion**

#### *3.1. Experimental Results*

During the entire injection period of all specimens, the uniform, approximately linear progress of the flow front must be ensured in order to avoid volume flows in the transverse direction and comply with the boundary conditions of the continuity equation according to Darcy (Equation (1)). Specimens with irregular flow fronts are excluded from the evaluation and prepared again. Several individual images of the flow front are taken with a 100× magnification factor and assembled to form a complete image of the frozen flow front, Figure 7a,b.

**Figure 7.** (**a**) Frozen flow front in different mold sections and merged microscopy image; (**b**) single flow path difference Δ*l*, shown magnified.

The frozen flow front of the cured specimen shows a distinct edge between the impregnated and dry sections. These definite lines indicate a virtually instant pervasive curing throughout the complete thickness of the transparent composite material and exist at each applied pressure gradient.

As in the literature [24–26], there are no distinct changes in the flow path difference along the mold cavity at constant process parameters, as shown in the comparison in Figure 8. However, there is a clear dependence of the flow path difference on the injection pressure.

**Figure 8.** Flow path difference along the mold sections.

The existing scatter is mainly attributed to fluctuations in permeability caused by the manual insertion of the fabrics into the cavity, since the manual preparation of the preforms may result in local displacements of the tows, and thusly produce irregular gaps and fiber distributions.

The injection pressure has a considerable influence on the flow velocity and thus on the locally prevailing capillary number. Contrary to what was calculated by Godbole [24], the results averaged per experiment show the proportionality of the flow path difference and the capillary number to the injection pressure (Figure 9). This measurable effect supports the model of Gueroult [15], in which the flow time ratio is largely determined by the capillary number. According to Equations (1) and (5), the capillary number is also pressure dependent. The calculations of Godbole [24] imply that a stronger expression of the dual-scale flow is completely compensated for by increased crossflow effects. This assumption is not confirmed by the measurement of the flow path differences for the presented experimental setup.

Considering all local measurements of the flow path difference as a function of the modified capillary number, an increase in flow path difference with an increasing modified capillary number is observed (Figure 10). At capillary numbers above around 0.025, the increase in the flow path difference is less substantial. The flattened slope is in accordance with the natural logarithm of Equation (6), which determines the flow ratio inside tows and channels. Additional experiments must be conducted to verify the shape of the curve for higher capillary numbers.

**Figure 9.** Dependence of the flow path difference and modified capillary number on the injection pressure.

**Figure 10.** Trend of the flow path difference in coherence with the local modified capillary number.

The increase in the flow path difference with increasing capillary number supports the findings of renowned studies [12–14,19] for process conditions with predominantly pressure-induced flow in the dual-scale model. Each specimen is in the region of micropore formation ( <sup>Δ</sup>*tt* <sup>Δ</sup>*tc* > 1). With increasing injection pressure, this flow time ratio also increases.

The measurement results illustrate that this increase is also reflected in increasing flow path differences. The progression of the flow path difference towards a maximum value implies an increase in crossflow effects, which counteracts the further increase in the flow path difference.

The measurements show that the flow velocity in the fiber bundles is slower than in the channels between the bundles. Neglecting crossflow effects, the flow path difference can be inferred from the flow time ratio according to the model of Gueroult [15].

$$
\Delta l = \overline{v} \cdot \Delta t \tag{9}
$$

Since the flow state in the bundle and channel is stopped after the identical injection time Δ*t* when the UV illumination is activated, it follows:

$$
\Delta l\_c(t) = \overline{v\_c} \cdot \Delta t \tag{10}
$$

$$
\Delta l\_t(t) = \overline{\upsilon\_t} \cdot \Delta t \tag{11}
$$

$$\frac{\Delta l\_c(t)}{\Delta l\_t(t)} = \frac{\overline{\upsilon\_c}}{\overline{\upsilon\_t}} \cdot \Delta t = c\_\upsilon \cdot \Delta t \tag{12}$$

However, Gueroult's model [15] is based on the ratio of the flow times of the resin system to saturation of the length Δ*l* of a single unit cell:

$$
\Delta t\_c(l) = \frac{\Delta l}{\overline{\upsilon\_c}} \tag{13}
$$

$$
\Delta t\_l(l) = \frac{\Delta l}{\overline{\sigma\_t}} \tag{14}
$$

$$\frac{\Delta t\_c(l)}{\Delta t\_l(l)} = \frac{\overline{\overline{\upsilon\_t}}}{\overline{\overline{\upsilon\_c}}} \cdot \Delta l = \frac{1}{c\_v} \cdot \Delta l \tag{15}$$

Equations (12) and (15) indicate inverse proportionality of length ratios and time ratios of impregnation of fiber bundles and channels.

$$\frac{\Delta l\_c(l)}{\Delta l\_t(l)} \sim \frac{\Delta t\_t(l)}{\Delta t\_c(l)}\tag{16}$$

If the ratios are plotted on top of each other as a function of the injection pressure, a good agreement of the values is observed for the first mold section, see Figure 11.

**Figure 11.** Comparison of the measured length ratios at 60 mm distance from the sprue with the calculated time ratios according to Equation (6).

Due to crossflow of the resin out of the channels into the fiber bundles in the transverse direction, as shown in Figure 12, the flow path difference Δ*l* stagnates while the macroscopic flow path continues to increase.

As the flow path length progresses, the results of the successive mold sections (Figure 5) therefore show considerable deviations (Figure 13).

**Figure 12.** Division of the total volume flow into crossflow . *Qt* into the tow and longitudinal flow . *Qc* in the channel.

**Figure 13.** Comparison of measured aspect ratios of all tool sections.

The deviations of the flow path ratios from the flow time ratios can be adjusted by correction factors. As Bodaghi et al. [9] describe, a loss term *q* (*p*, *s*) from Equation (4) contributes to the transverse impregnation of tows. The regression analysis of the measurement results confirms the assumption that both the injection pressure and degree of saturation of the macroscopic flow path influence the loss term, as shown in the comparison in Figure 14 While the slopes of the compensation lines are determined by the injection pressure, the intercept is related to the macroscopic flow path and is thus a measure of the degree of saturation of the textile preform. It follows:

$$\frac{\Delta t\_l(l)}{\Delta t\_c(l)} = c\_{corr} \cdot \frac{\Delta l\_c(t)}{\Delta l\_t(t)}\tag{17}$$

$$
\mathcal{L}\_{corr} = a \cdot p\_{\dot{m}\dot{j}} + b(s) \tag{18}
$$

where *ccorr* is a correction factor, *a* is the slope factor, *pinj* the injection pressure, and *b*(*s*) the saturation dependent offset.

This correction factor *ckorr* and its dependence on pressure and saturation proves the occurrence of the loss term *q* (*p*, *s*) described by Bodaghi et al. and quantifies the effect of crossflow for the given material and process parameter combination.

**Figure 14.** Determined correction factors for the adjustment of the measured length ratios to the calculated time ratios according to Gueroult.

#### *3.2. Discussion*

Two distinct aspects must be considered when analyzing the results. First is the applicability of the novel methodology, followed by the observations regarding dual-scale flow experiments.

The presented methodology allows taking snapshots of the flow processes during the impregnation of textile preforms using a photoreactive resin system. Evidence was provided that photoreactive resin systems crosslink sufficiently fast to freeze the microscopic saturation along the entire specimen. The curing occurred sufficiently rapidly for the given glass fiber fabric and applied pressure settings. Upper limitations of flow velocity and specimen thickness must be determined in prospective studies to eliminate the possibility of partial flow progression below the irradiated surface. In contrast to point-wise optical measurements, the complete unsaturated domain of the flow front can be analyzed. The holistic analysis of the flow front represents an improvement compared with the limited field of view of well-established methods [27,28,30]. It is more cost- and time-efficient than x-ray or micro-CT measurements [1]. Depending on the type of microscope, voids with diameters of 5 to 20 μm can be analyzed [1], whereas ultrasonic imaging is only suitable for single defects with a minimum size between 1 and 0.6 mm [34,35]. The simple mold setup and its similarity to live-microscopy makes the introduced method easily accessible. The glass cover allows the elimination of race tracking inside the entire mold during impregnation, which cannot be ruled out with single point-wise observations, as Siddig et al. [36] proved, at least three points of a rectangular shape need to be observed to detect race tracking. However, one downside of this method is that it is not suited to quantify void transport. Freezing the flow produces a single image that can be used to describe the dual flow effect and formation of voids at the very front of the flow, but it contains no information on the history of voids inside the specimen.

The result of the experimental setup provides a multitude of insights into the saturation process of fabrics. A clear dependence on the injection pressure was determined for the measured flow path difference. This effect can be explained by Equations (4)–(6), which show that the flow time ratios of Gueroult [15] depend on the flow velocity caused by the injection pressure gradient. A comparison of the results of the established relationship of the flow paths in the fiber bundle and channels between the bundles with the model of the flow time ratios of Gueroult [15] shows good agreement within the first mold section with a 60 mm flow path. With an increasing flow path, the deviations of the time and path ratios become larger, which is due to the increasing influence of crossflows in the transverse fiber direction. These crossflows are summarized by Bodaghi et al. [9] in a loss term, which depends on the injection pressure as well as the degree of saturation. The results of the measurements show that such a loss term can be described by a correction factor that was determined for each section of the mold and the corresponding injection pressure. Simulations of previous studies [24–26] showed that using constant tool and material parameters, the flow front difference remains unchanged with sufficient tool length. These results could only be partially confirmed. While the flow front difference remains constant along the different mold sections within the individual parameter combinations, in contrast to the studies of Godbole [24] and Zhou [25,26], the influence of injection pressure cannot be neglected. However, there are considerable differences in the experimental setup and simulation constraints. While the simulations [24–26] use perfectly unidirectional tows, the experimental setup was conducted with plain weave fabric and a different type of inlet. Furthermore, the simulated tow permeability *Kyytow* was higher than the determined permeability of the experiments. Both factors result in differences in longitudinal and transverse flow and have an influence on the impact of capillary forces. Further experiments should be conducted that replicate the constraints of the simulations of Godbole and Zhou to be able to compare results.

#### **4. Conclusions**

With the presented methodology, the flow front of the cross-linked specimens can be viewed completely and contiguously at high resolution, which is a significant improvement over locally confined in-situ measurements. The measurement methodology represents a good starting point for the creation and validation of calculation models for the impregnation of textile preforms. Optimization potential lies in the tool design and manual preparation of the textile preforms. The adjustment of the compression pressure of the textile preforms by means of a defined tightening torque of the screw connection is reproducible to a limited extent and leads to fluctuating permeability. This deviation can be eliminated by using a vertically loosely supported plate as the mold surface with pressure sensors underneath. To specify exact physically based models for the calculation of the correction factors, either a more precise adjustability of the compression pressure must be ensured, or the tow permeability must be evaluated individually for each experiment by cutting the cured specimen and determining the surface ratios of fiber bundles and resin of the cross section. Analytical models of the dual-scale flow were confirmed and an approach to quantify the loss term of transverse flow is given. The prevalent increase of flow path differences differs from simulations, which emphasizes the need of further investigations. Since only a limited number of experiments were conducted, additional experiments should be performed to validate the findings. In future experiments, a variety of textile preforms must be analyzed to verify the applicability of the method for diverse types of weaves. Furthermore, higher modified capillary numbers must be examined, by increasing the macroscopic resin velocity via increased injection pressure. After validation of the methodology, future experiments can be conducted to directly evaluate resulting void volume contents and the coherence with the now measurable flow path differences.

**Author Contributions:** Conceptualization, B.N.; methodology, B.N. and F.P.; validation, B.N.; formal analysis, F.P.; investigation, B.N.; resources, F.P.; data curation, B.N.; writing—original draft preparation, B.N.; writing—review and editing, F.P.; visualization, B.N.; supervision, F.P.; project administration, F.P. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Acknowledgments:** We acknowledge support for the publication costs from the Open Access Publication Fund of the Technische Universität Ilmenau and the infrastructural support of the Thüringer Innovationszentrum Mobilität (ThIMo).

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


**Hatim Alotaibi 1, Chamil Abeykoon 2,3, Constantinos Soutis 2,3 and Masoud Jabbari 1,\***


**Abstract:** The filling stage in injection/infusion moulding processes plays a key role in composite manufacturing that can be influenced by the inlet and vent ports. This will affect the production of void-free parts and the desirable process time. Flow control is usually required in experiments to optimise such a stage; however, numerical simulations can be alternatively used to predict manufacturinginduced deficiencies and potentially remove them in the actual experiments. This study uses ANSYS Fluent software to model flow-front advancement during the impregnation of woven fabrics. A developed technique is applied by creating tracking points (e.g., on-line monitor) in the direction of the flow to report/collect data for flow-front positions as a function of time. The study adopts the FVM-VOF-based two-phase flow model together with an implicit time-stepping scheme, i.e., a dual-time formulation solution method with a preconditioned pseudo-time derivative. Initially, three time-step sizes, 5 s (small), 25 s, and 50 s (large), are evaluated to examine their impact on numerical saturation lines at various fabric porosities, 40%, 50%, and 60%, for a two-dimensional (2D) rectangular mould, and predictions are then compared with the well-known analytical Darcy. This is followed by a three-dimensional (3D) curved mould for a fillet L-shaped structure, wherein the degree-of-curvature of fibre preforms is incorporated using a User-Defined Function (UDF) to tailor the impregnation process. The developed approach shows its validation (1–5.7%) with theoretical calculations and experimental data for 2D and 3D cases, respectively. The results also stress that a shorter computational time can be achieved with a large time-step size while maintaining the same level of accuracy.

**Keywords:** liquid composite moulding; flow visualisation; numerical modelling; Volume of Fluid (VOF); multi-phase model

#### **1. Introduction**

During liquid composite moulding (LCM) processes, a liquid resin is injected/infused at a constant pressure/flow rate to impregnate a fibrous reinforcement [1,2]. This serves to fully saturate the placed/laid fibre preforms on the mould. This means that the fibre preforms are dry (unsaturated) during the pre-filling stage (c.f., Figure 1). The dry fibrous reinforcement needs to be processed with a thermosetting resin in order to produce a composite part. This process stage is referred to as the mould-filling stage, in which the injected/infused resin flow starts to impregnate the fibre preforms. The quality of the processed fibre–reinforcement composites would depend on this crucial process stage. This is due to the fact that gelation or early curing, as well as the emergence of voids, could occur within the impregnation process. To avoid these issues, the knowledge and control of filling time and flow-front advancement is required. For this reason, it is important to optimise the mould-filling process in an effective way to obtain good-quality composite parts/components. This optimisation can be done with the traditional trial and error

**Citation:** Alotaibi, H.; Abeykoon, C.; Soutis, C.; Jabbari, M. Numerical Simulation of Two-Phase Flow in Liquid Composite Moulding Using VOF-Based Implicit Time-Stepping Scheme. *J. Compos. Sci.* **2022**, *6*, 330. https://doi.org/10.3390/jcs6110330

Academic Editor: Jinyang Xu

Received: 23 September 2022 Accepted: 1 November 2022 Published: 3 November 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

method; however, this method shows drawbacks from the perspective of time and cost [3]. Recently, there have been alternative proposals to optimise such a process, and one of the most efficient methods is to use the available sophisticated numerical tools. This can include isothermal/non-isothermal conditions, but this study will focus on isothermal mould-filling processing.

**Figure 1.** A schematic diagram illustrating mould-filling process parameters, saturation and unsaturation regions, and filling front advancement with uniform permeability.

Flow-front advancement has been discussed by a considerable amount of research [3–22]. This includes experimental [4–12] and numerical studies [5,8,13,22], in which the latter are subject to validation of the adopted numerical tool with the available experimental data or theoretical models. Experimental works for monitoring/tracking the flow progression used various types of techniques; these would include cameras, ultrasonic sensors, optic sensors, dielectric sensors, and pressure transducers. On the other hand, the developed numerical models involve a wide range of approaches that are based on the control volume finite element method (CVFEM), finite volume method (FVM), finite element method (FEM), and finite difference method (FDM). These discretisation methods allow the solving of multi-phase or multi-physics flow problems, and this could apply to resin flow in LCM processes. Hoes et al. [4] conducted an experimental work that involved filling front tracking at a constant pressure injection. The authors placed electric sensors together with pressure transducers in the bottom flat mould plate, wherein these sensors were located on straight lines and predefined in the connected data collection computer. This method enabled the automatic observation of flow-front positions as a function of time. Binetruy et al. [5] investigated, experimentally and theoretically, the interaction between microand macro-flow through unidirectional woven fabrics, and their impacts on the flow-front progression during the impregnation phase. The authors used a flat Plexiglass (transparent) piece on the top of the mould to allow visual observations of the filling process. The study highlighted that, for unsaturated (dry) tows, a lag of the intra-tow flow progression would lead the macro-flows to cause transverse fluid penetration into the yarns. This study concluded that the micropores could affect flow-front behaviour and slow the filling process. A contribution by Luce et al. [6] was presented by performing mould-filling experiments for multi-ply fibre preforms that consisted of different fabric architectures, such as 3-D woven and random mat (RM). Two cameras were placed on the top and side-view of the mould to monitor and characterise the in-plane and through-thickness advancing flow behaviour. The study showed that the flow-front advancement through RM behaves uniformly with rapid progression. However, this was different for the 3-D woven ply, in which the flow encountered progress resistance caused by the transverse tows, besides the emergence (presence) of out-plane flows resulting from advancing macro-infiltration. The application

of ultrasonics was employed by Schmachtenberg et al. [7] to obtain the on-line monitoring of the flow-front propagation in the RTM manufacturing process. The generated experimental data stressed that signals, a transmission time, and an amplitude could explain the arriving flow-front, and hence achieve the sensitive optimisation of the RTM filling process. The work by Pierpaolo et al. [8] embedded dielectric sensors to capture pressure data of the infused unidirectional flow in LCM, as well as monitoring the progressing flow front. The results were validated against numerical modelling, with satisfaction in terms of pressure profile and flow progression. Their study incorporated multiscale modelling, in which a microstructure of the used fabric (2-D woven), was scanned to obtain a meso-scale image, whereby the meso-permeability was identified and introduced in the macro-scale simulation to validate the approach with the experimental work. At present, a broad range of sophisticated software packages are available to model mould-filling processes in LCM, including RTM and VARTM (see Table 1). These would include, but are not limited to, LIMS [13], PAM-RTM [14,15], OpenFoam [15], and COMSOL [16]. Simacek et al. [13] implemented an algorithmic code based on the finite element/control volume (FE/CV) method using the simulation tool LIMS to model tow saturation during the mould-filling process. The numerical study considered the dual-scale flow, and emphasised that the developed model is able to capture the filling progression as well as the required time to fully impregnate and saturated the fibre preforms. They also added that this approach can be applied to arbitrary complex shapes. Voller et al. [17] followed CVFEM, but using Fortran to track flow-front positions as a function of time. The work analysed the filling front with structured/unstructured mesh grids for fibre–reinforcement mats. The results showed good agreement with the well-known analytical Darcy for various time-step sizes. Multiscale modelling of bi-axial fabrics was developed by Tan et al. [3] to predict the filling front during RTM processes. The numerical model adopted the FE/CV approach to simulate a coupled macro- and micro-flow problem. The PoreFlow program was used in the study, which is based on the Fortran modular package, and hence showed its capability of simulating resin impregnation for dual-scale porous media. This was validated with experimental data obtained by the same authors. Grössing et al. [15] performed a numerical mould-filling analysis via two different simulation tools, i.e., OpenFoam and PAM-RTM. The study evaluated both tools and highlighted that non-porous and porous zones can be both modelled by OpenFoam, whereas this was different with Darcy-based PAM-RTM, whereupon porous zones can only be modelled. The authors also argued that the issue with PAM-RTM could be resolved by assigning non-porous regions as race-tracking zones with %100 porosity value. The numerical study was based on an RTM experimental work that involved two types of fibre preforms, such as NCF and UD. Both numerical results agreed well with the flow-front experimental data.


**Table 1.** Selection of numerical contributions that evaluated simulation tools for flow-front modelling.

The present work is motivated to contribute a simple, accurate technique using ANSYS Fluent to model resin flow advancement in the RTM/VARTM processes. The FVM-VOFbased multiphase flow approach is adopted to monitor the mould-filling process and to report flow-front positions as a function of time. This approach allows the prediction of the required time to fully saturate/impregnate the fibre preforms in simple and complex shapes. During the mould-filling simulation, the saturated, partially saturated, and unsaturated

regions can be observed and located. The numerical simulations fit well with the analytical predictions and the experimental data for a rectilinear/channel flow injection.

#### **2. Numerical Simulation Approach**

#### *2.1. Volume of Fluid (VOF)*

The free-surface flows can be modelled by the prominent technique VOF, whereby the two immiscible fluids' interface position is tracked and located simultaneously. This VOF model can be solved either using implicit or explicit time formulation (see Section 2.2) throughout structure/unstructured fixed meshes—a Eulerian-based approach. Such formulations (interpolation schemes) will allow discretisation of the volume fraction (tracking of the interface) by applying widely adopted options (in ANSYS Fluent)—geometric reconstruction (a piecewise-linear approach) and Modified HRIC (High-Resolution Interface Capturing)—for explicit and implicit, respectively [23–25]. This is to obtain face fluxes (fluid advection) for all filled and empty cells, and also near-interface (partially filled) cells [23–25]. Momentum and continuity equations are applied to each cell throughout the domain, in which the flow motion is solved by Navier–Stokes equations (N–S) based on the FVM discretisation (control-volume-based) approach. For RTM/VARTM filling simulation, the volume fraction in each control volume (computational grid cell) denotes either one of the phases (resin or air), or a mixture of both phases (partial saturation). This approach follows a transient, laminar, incompressible, multiphase flow problem with Newtonian behaviour and an isothermal filling condition. Thus, the continuity/transport equation can be written as follows:

$$\frac{\partial s\_i}{\partial t} + \nabla \cdot (s\_i \mathbf{u}) = 0 \tag{1}$$

where **u** is the volume-averaged velocity, *t* is the time, and *si* denotes a phase volume fraction such that *sf* is the fluid (e.g., resin), while *sa* defines air (or empty) regions. This form assumes a constant fluid density and applies to fixed/stationary control volumes in this study. *sf* varies from 0 to 1 in each grid cell, and this would indicate full saturation for cells with a value of 1, partial saturation for cells with a value ranging from 0 < *sf* < 1, and unsaturation for cells with a value of 0. Since the resin flow impregnates a porous material, the momentum equation used by the simulation tool incorporates a source term to allow porous media modelling. Equation (2) demonstrates the momentum equation after applying the above-mentioned assumptions, while Equation (3) expresses the source/sink term:

$$\frac{\partial}{\partial t}(\rho \mathbf{u}) + \nabla(\rho \mathbf{u} \mathbf{u}) = -\nabla p + \mu \nabla^2 \mathbf{u} + \mathcal{S} \tag{2}$$

$$S = -\frac{\mu}{\mathbf{K}}\mathbf{u} \tag{3}$$

Here, ∇*p* is the pressure gradient, *ρ* is the fluid density, *μ* is the fluid viscosity, *S* is the sink term, and **K** is the permeability tensor. This filling modelling is performed on a rectangular mould (2-D RTM) in a macro-scale flow problem. The porous medium permeability is isotropic; therefore, the in-plane permeability can be denoted as *Kxx* = *Kyy* = *K*. The permeability value was computed in a previous work by the current authors at a dual-scale flow (inter- and intra-tow flow) through a plain-weave fabric unit cell at various aggregate porosity values: 40%, 50%, and 60%. Due to the fact that the unit cell can be taken as a representative volume element (RVE) of the woven fabric ply, the obtained in-plane dual-scale permeability is inputted in the source term to enable true viscous resistance of the woven model at a macroscopic level. The developed approach used the implicit scheme (see Section 2.2) provided by ANSYS Fluent, owing to the fact this method is preferred for slow flow movement problems, i.e., creeping flows, quasi-static motion, and static/dynamic motion [26]. In this method, the time-step size is independent of the results, and can be large to obtain a shorter computational time [17,26]. This work developed a flow-front tracking technique by creating points on a straight line (flow direction) at certain

locations. Each point will report resin flow volume fraction data as a function of time; this would be similar to real-time monitoring RTM/VARTM experiments wherein sensors are embedded/placed (in fabrics or on mould).

As part of this study, three-dimensional flow simulations of complex shapes are included to examine the flow-front prediction model for a real filling process of a composite component. An experimental work by Geng et al. [27] was selected, wherein a curved mould was used with a VARTM process type. Hence, the present work added a set of equations (see Equation (4)) that accounts for the effect of curvature on resin flow through the curved regions during the resin impregnation of fibre preforms. This is done by using the User-Defined Functions (UDF) to interpret/compile these equations with a C-languagebased code.

$$D\_{\mathfrak{c}} = 2 \sin^{-1} \left( \frac{\mathbb{C}}{2R\_{\mathfrak{c}}} \right), \quad \theta = \frac{D\_{\mathfrak{c}}}{2} \tag{4a}$$

$$K\_{\rm H}(\theta) = K\_{\rm H} \cos(\theta) \tag{4b}$$

$$K\_V(\theta) = K\_V \sin(\theta) \tag{4c}$$

The degree of curvature and the deflection angle are *Dc* and *θ*, respectively. Here, *Rc* is the radius of curvature, and *C* is the cord length. Since part of the filling process is vertical, the permeability is set as a function of curvature for both vertical and horizontal resin flow progressions.

#### *2.2. Time-Stepping Scheme*

A temporal discretisation in transient simulations is defined by space and time. This means that an integration is required for each term in the differential equations with every time-step Δ*t*. To apply such a time discretisation, there are two prominent schemes, i.e., implicit and explicit. These methods are to be carefully selected, since a diverse range of restrictions arise in each scheme in certain circumstances. The implicit method is applicable to slow motions (e.g., creeping flows), and it can handle long-duration analyses that vary from seconds to days [26]. On the other hand, the explicit case is suitable for rapid motions (e.g., drop tests, shocks, etc.), and it is restricted by the Courant–Friedrich–Lewy (CFL) condition, in which each grid cell uses the same time-step that depends on the courant number, cell volume, and the sum of outcoming fluxes [26]. Therefore, this work finds the implicit time-stepping a suitable technique for such a flow problem, and an evaluation was conducted to examine the impact of time-stepping size on the flow-front modelling. It is noteworthy that implicit time-stepping is a dual-time formulation that adopts a preconditioned pseudo-time (inner iterations) derivative at each physical timestep (Euler backward) to provide an accurate transient solution [26,28]. Figure 2 describes the time discretisation methods and how the time-dependent equations are computed differently throughout the domain. This shows that a function of a quantity/variable *f*(*ϕ*) is evaluated at a future time level (*n* + 1) for the implicit case, while *f*(*ϕ*) is computed at the current time level (*n*) for the explicit time-stepping method. The general expression used in the current numerical simulation is demonstrated below.

$$\frac{\varphi^{n+1} - \varphi^n}{\Delta t} = f(\varphi) \tag{5a}$$

$$\begin{cases} n+1 = t + \Delta t\\ n = t\\ n-1 = t - \Delta t \end{cases} \tag{5b}$$

**Figure 2.** Unsteady flow solution: temporal discretisation methods and time-stepping schemes.

#### *2.3. Darcy's Law for a Transient Flow*

The general form of Darcy's law (Equation (6)) for an isothermal filling flow problem can be integrated and simplified to suit transient flow conditions [29]. At a constant pressure injection for one-directional flow movement (e.g., *x*-coordinate), Equation (6) can be rewritten as shown in Equation (7). By further simplification and integration of Equation (7), the position of the flow-front can be obtained by Equation (8).

$$\mathbf{u} = -\frac{\mathbf{K}}{\mu} \nabla p \tag{6}$$

$$\frac{d\mu\_x}{d\phi\_0} = \frac{d\chi}{dt} = -\frac{K}{\mu\phi\_0} \frac{\Delta p}{\Delta x} \tag{7}$$

$$\mathbf{x}\_f = \sqrt{\frac{2Kp\_o}{\mu\Phi\_0}}\mathbf{t}\_f\tag{8}$$

where *xf* is the flow-front position, *tf* is the flow-front time, *φ<sup>o</sup>* is the porosity of the medium, and since the pressure at the flow-front can be assumed as zero, the pressure difference Δ*p* becomes equal to the injection pressure (*po*) [29]. Thus, the proposed approach in the present study can be subjected to a comparison analysis with the above-mentioned theoretical solution for validation. On this premise, the developed numerical model will be also feasible for complex structural shapes, and this would contribute to the knowledge and decision of LCM optimisation. This is in terms of the optimal cycle time and location of injection ports/vents, thereby achieving void-free and good-quality composite parts. Figure 3 illustrates the geometry and boundary conditions used in the numerical analysis for the 2-D rectangular mould filling involving the flow-front tracking points.

**Figure 3.** Geometry and boundary conditions used in the numerical simulation, including flow-front tracking points.

#### **3. Results and Discussion**

#### *3.1. Two-Dimensional Rectangular Mould for Regular Shapes*

The numerical simulations were performed at various fibre preform aggregate porosities (*φo*) 40%, 50%, and 60%, with in-plane dual-scale permeabilities (*K*[m2]) 4.89 × <sup>10</sup><sup>−</sup>11, 9.45 × <sup>10</sup>−11, and 2.09 × <sup>10</sup>−10, respectively. Filling front analysis was computed at a constant pressure injection (*po* = 10 [kPa]), and with fluid flow properties *ρ* = 1300 [kg/m3], and *μ* = 0.15 [Pa · s]. Three time step-sizes, 5 s, 25 s, and 50 s, were selected based on the generated grids/meshes for the transient multiphase flow through a 2-D macro-scale geometry. These different step sizes yielded significant factors in terms of the actual flow-front position or the so-called saturation lines and the computational time. It appeared that the partially filled control volumes (i.e., partially saturated region) could indicate the actual flow-front for all three time-step sizes. This can also be seen in similar numerical studies, such as [3], in which the flow-front was observed within the partially filled computational grid cells. Accordingly, the results stressed the fact that the flow-front progression with 40–50% resin volume fraction agreed well with the analytical solutions for all cases. Less computational time was observed at large time-steps such as 50 s; however, smaller step sizes are preferred (e.g., determined by grid cell size divided by fluid velocity value) [26]. By applying this, sharp flow-front behaviour could be seen in which the partial saturation zone was reduced (c.f., Figure 4). Consequently, filling front positions would be more manifested or perceivable, and an accurate on-line flow control could be attained. It should be noted that different implicit-scheme-based time-steps would offer the same required time for the mould-filling process; nonetheless, computational time processing would vary. For instance, to accomplish a fully saturated fibre preform with the geometry dimensions and boundary conditions illustrated in Figure 5 for the 50% porosity value case, 1600*s* is needed. This was achieved by all time-stepping sizes, 5*s*, 25*s*, and 50*s*, but with different numbers of time-steps: 320, 64, and 32 respectively. This means that a shorter computational time can be obtained with the maximum time-stepping size, while maintaining the same accuracy of the output results, and this can be explained by Figure 5 together with Table 2. Therefore, when supposing that the user is interested in a shorter computational time to optimise the mould-filling process, it would be recommended to follow the maximum time-stepping in accordance with the mesh/grid element size. It is worth mentioning that an Intel Core i7-1165G7 Processor (Central Processing Unit CPU) was used to run the present numerical

simulations. Figure 5 presents the computed results for various woven fabric aggregate porosities, 40%, 50%, and 60%, with different time-step sizes, 5*s*, 25*s*, and 50*s*, for each porous medium case. The results showed perfect alignment with ones calculated by transient Darcy for all flow-front simulations. Therefore, the numerical approach proposed in this study shows its reliability and can be confidently used to optimise RTM/VARTM filling processes.

**Figure 4.** Numerical rectilinear/channel flow saturation throughout a porous medium with 50% aggregate porosity (*φo*) at a constant injection pressure for different time-step sizes: 5 s, 25 s, and 50 s. (**a**) 5 s, (**b**) 25 s, and (**c**) 50 s.


**Table 2.** Computational processing time with different time-stepping sizes for each medium porosity.

**Figure 5.** A comparative analysis for prediction of flow-front position as a function of time.

#### *3.2. Three-Dimensional Curved Mould for Complex Shapes*

For a complex shape, a study by Geng et al. [27] was selected to examine the current approach against the populated experimental data. Figure 6 shows a schematic diagram of the experimental setup that was used by [27], in which VARTM was adopted. Geng et al. [27] managed to perform experimental works for curved composite shapes using VARTM, and this was done with different bend angles, i.e., 180, 120, 90, and 60 degrees. Therefore, this paper assesses the developed numerical flow-front prediction for a threedimensional complex shape during the mould-filling process. This is applied to curved non-crimp fabric (NCF) plates with 90 degree bending, and simulations are conducted on single and multiple layers based on the VARTM experiment by Geng et al. [27]. The permeabilities of fibre preforms are set to be 2.98 × <sup>10</sup>−<sup>11</sup> [m2] and 4.6 × <sup>10</sup>−<sup>11</sup> [m2], with fibre volume fractions (*Vf*) 22% and 40% for 1-layer and 6-layer, respectively [27]. This is along with the flow properties 968 [kg/m3] and 0.35 [Pa · <sup>s</sup>] for density and viscosity, and a volume-average velocity range (**u**) 0.00025–0.00035 [m/s] [27]. With such a curved composite part, the curvature region is said to be impacting the resin impregnation, as was thoroughly discussed by Alotaibi et al. [30], and hence it is considered in the present work. In such a case, the resin flow will be affected by the degree of curvature as long as it progresses within the curved zone—see Figure 7. This would show 90 degrees of curvature (*Dc*) with a deflection angle impact range (0◦≤ *θ* ≤ 45◦). Thereby, a set of equations is required to be incorporated with the flow equations—see Equation (7)—and this is done using the User-Defined Function (UDF) in ANSYS Fluent. The results give a good impression of the capability of the current numerical model, in which most of the tracking flow-front points (c.f., Figure 8) fit well, showing ≤ 5.7% discrepancy with the experimental data. This is in addition to the mould-filling time, wherein times of 760 s and 475 s were achieved for 1-ply and 6-ply, respectively, in comparison to 750 s and 470 s experimental fill time. Figure 9 portrays the resin flow progression contours/outlines for 1-ply and 6-ply NCFs, and with a more permeable fibre perform (e.g., 6-ply case), this would hasten the flow advancement. It should be noted that a hexahedron mesh topology is preferred when applying the relevant degree-of-curvature equations through the UDF, since it provides a sequential quadrilateral cell type, and this makes it compatible with the degree-of-curvature concept. The present numerical methodology proves its capability to monitor and predict resin flow advancement in RTM/VARTM processes that include regular and complex shapes, and it also offers a simple and accurate technique that permits efficient computational modelling.

**Figure 6.** A VARTM experimental setup used for a curved or L-shaped composite components [27].

**Figure 7.** A schematic diagram of (**a**) curvature and straight regions of the L-shaped composite part, (**b**) 1-ply fibre preform with 1.46 mm thickness, and (**c**) 6-ply fibre preform with a 4.84 mm thickness.

**Figure 8.** Numerical flow-front predictions vs. experimental observations for a complex shape with single and multiple plies.

#### **4. Conclusions**

A numerical technique was proposed using the data-track-point feature of ANSYS Fluent to predict the flow-front position as a function of time throughout the computational domain, as a means to simultaneously monitor filling progression during LCM processes (e.g., RTM/VARTM). The results stressed that the grid-independent solution (implicit VOF) of structured or unstructured control volumes does not impact the output of flow parameters, while allowing less computational time at larger time-step sizes. The simulation approach followed the FVM-VOF-based multiphase flow problem together with implicit time discretisation. The numerical model has been validated with Darcy's law and an experimental work [27] for a transient flow at various aggregate porosities of dual-scale fabrics. It is noteworthy that the present work assumes no chemical conversion of the fluid flow (i.e., a constant viscosity) during the impregnation process. A future work will include a cure-temperature-time-dependent viscosity to characterise their impacts on the impregnation/saturation of complex (woven) fabric structures with heterogeneous permeability.

**Author Contributions:** Conceptualisation, H.A. and M.J.; Investigation, H.A.; Software, H.A.; Writing—original draft, H.A.; Writing—review and editing, H.A., M.J., C.A. and C.S.; Supervision; M.J., C.A. and C.S. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Data are contained within the article.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


**Felice Rubino 1,\*, Fausto Tucci 2, Vitantonio Esperto <sup>2</sup> and Pierpaolo Carlone <sup>2</sup>**


**Abstract:** The quality of Liquid Composite Molding (LCM) manufactured components is strictly related to the fibrous preform impregnation. As Darcy's law suggests, the resin flow is influenced by the pressure gradient, geometrical features of the reinforcement, and resin viscosity. The former two parameters are dictated by the requirements of the component and other constraints; therefore, they are hardly modifiable during the process. Resin preheating increases its fluency, thus enhancing the impregnation and saturation flow, and reducing the mold filling time. In the present work, a microwave heating system has been integrated within a vacuum bag resin infusion process, to analyze the effect of the online preheating on the fiber impregnation. To monitor the resin flow a dielectric sensors-based system is used. Results from resin infusion tests conducted with and without the resin pre-heating were compared: the outcomes indicated an advance of approximately 190 s of the flow front when microwave heating is applied with respect to the unheated tests.

**Keywords:** liquid composite molding; microwave preheating; dielectric flow monitoring

#### **1. Introduction**

Liquid composite molding (LCM) processes, such as resin transfer molding (RTM) or Seemann composites resin infusion molding (SCRIMP) processes have been addressed by composite industries as a promising technology to manufacture polymeric matrix composites out-of-autoclave. Some of these processes are particularly interesting for the industry involved in the production of large-scale structures, even with complex shapes [1]. However, large scale diffusion of components depends on the possibility to lower the overall costs of the products and scale-up the technology to a mass production, always guaranteeing the quality of the manufactured composite structures [2].

In LCM processes, the final quality of the products is strictly connected to the impregnation and curing phases [3]. Dry spots or excess in resin, delamination or cracks, and residual stress are some of the most common flaws that can occur during the manufacturing compromising the performance and the integrity of the structure if the mentioned aspects are not carefully designed and monitored [4,5]. Impregnation defects can be related to the incompatibility of the two main phases involved. From this point of view, the binder plays a key role [6].

Sensing techniques have been developed to monitor both resin flow front progression and the cure degree in thermoset matrix composite manufacturing processes. They include, but are not limited to, the use of optical FBG sensors [7–9] pressure transducers [10], thermocouples [11–14], SMART weave sensor [15,16], electrical time domain reflectometry (ETDR) [15], ultrasonic, dielectric and piezoelectric sensors [17–22]. In their previous works, authors developed a sensing system based on dielectric analysis (DEA) to monitor the resin flow progression during the Resin Infusion process [19,20]. DEA relies on the measurement of the dielectric properties, i.e., permittivity and ionic conductivity of the test material: dielectric material is placed between electrodes forming a capacitor and an alternating

**Citation:** Rubino, F.; Tucci, F.; Esperto, V.; Carlone, P. Filling Time Reduction in Liquid Composite Molding Processes. *J. Compos. Sci.* **2022**, *6*, 222. https://doi.org/ 10.3390/jcs6080222

Academic Editor: Jinyang Xu

Received: 12 July 2022 Accepted: 2 August 2022 Published: 4 August 2022

**Publisher's Note:** MDPI stays neutral with regard to jurisdictional claims in published maps and institutional affiliations.

**Copyright:** © 2022 by the authors. Licensee MDPI, Basel, Switzerland. This article is an open access article distributed under the terms and conditions of the Creative Commons Attribution (CC BY) license (https:// creativecommons.org/licenses/by/ 4.0/).

electric field is applied across the plates, then permittivity, loss factor and ionic conductivity can be determined from the output current [23]. Dielectric sensors, consisting of parallel plates placed on both sides of a mold, were implemented in a lab-scale LCM apparatus and provide pieces of information on the resin arrival at the sensor position by detecting variation in the dielectric properties of the medium (glass fibers plus resin) [20,24]. Since the electrodes can be placed outside of the composite laminate, the sensing system is less invasive and does not affect either the integrity of the manufactured parts or the surface finishing of the part.

In addition, the parallel plate dielectric sensors are characterized by a higher scanning depth, which makes this design suitable also for thick composite [24,25]. The plane plates dielectric sensors require that the sensing areas of the electrodes must be parallel to each other, and the reciprocal distance must be known and kept constant during the entire infusion process. This limited the usage of these types of dielectric sensors only to specific classes of LCM processes using rigid molds, such as RTM, which ensures that these requirements are observed. LCM processes involving flexible upper mold, such as Vacuum Assisted Resin Infusion (VARI) or SCRIMP processes, were coupled only with co-planar dielectric sensors [16,26]. To the authors' best knowledge, no attempts have yet been made to apply parallel-plates dielectric sensors to monitor these processes.

In addition to the monitoring and the control of impregnation and curing phases to ensure the quality of the composite parts, the scale-up of the composite industry to a massproduction is also limited by the manufacturing time of a component, dictated by the time required to fully impregnate the fibrous reinforcement and the time to complete the curing of the resin. This is especially true in the case of big components, such as boat hulls (usually manufactured by the SCRIMP process), where the impregnation is particularly long and the curing of resin is deliberatively kept slow to avoid gelation prior to the complete wetting [27]. In this scenario, enhancing strategies of the resin flow through the fabric is of paramount relevance to achieve a uniform impregnation of the fibers and optimize the filling time, mitigating the manufacturing flaws and contributing to a reduction in the overall production time of a composite structure. Among the strategies investigated, the reduction in resin viscosity by means of mold temperature increases [28]; preform or resin preheating [28–31] can improve the flow through the preform and, thus, decrease the impregnation time. Dealing with thermosetting resin systems involves time-temperature constraints related to its reactivity and the consequent reduction in pot life. From this point of view, microwave-based heating systems guarantee a high efficiency thermal energy transfer. Reductions in filling time of about 13% and 25% have been obtained in the case of non-reactive and reactive systems, respectively, by optimizing the microwave heating apparatus [32–35].

Previous experiments have been dedicated to the study of the resin flow through a dry fibrous preform sealed between rigid molds [32–35]. In the present article, the author investigated the application of microwave preheating to the SCRIMP process to reduce the filling time. Parallel plate dielectric sensors were also implemented to monitor the resin flow. An ad-hoc system was developed to install the electrodes on the vacuum bag and ensure the correct alignment of the plates during the whole process.

#### **2. Materials and Methods**

The liquid composite molding (LCM) experimental was conducted based on the following preliminary requisite materials along with the ancillary materials. HexForce E-glass twill 2/2 fabric was used as reinforcing material. 12 layers of the glass fiber fabric were cut into rectangular plies with dimensions 300 mm × 240 mm. Main properties of the reinforcement are indicated in Table 1.


**Table 1.** Reinforcement properties.

Epoxy resin SX 10 was utilized upon the room temperature premixing with the Epoxybased hardener in the mixing ratio of 100:26. The resin properties are mentioned in Table 2. Its rheological behavior measured by rheometry testing is depicted in Figure 1 and described by the following equation:

$$\eta = A\_{\eta} \exp\left(\frac{B\_{\eta}}{R} + \mathbb{C}\_{\eta}\alpha\right),\tag{1}$$

where the viscosity *<sup>η</sup>* depends on the pre-exponential term *<sup>A</sup><sup>η</sup>* = 7.093 × <sup>10</sup>−<sup>8</sup> [Pa·s], the calibration coefficients *<sup>B</sup><sup>η</sup>* = 3.999 <sup>J</sup>·mol−<sup>1</sup> and *C<sup>η</sup>* = 1.63, the universal constant of gases *R* = 8.314 J K−1·mol−<sup>1</sup> , the temperature *T* expressed in Kelvin degrees, and the degree of cure of the resin system *α*.

**Table 2.** Resin matrix properties.


**Figure 1.** Viscosity of resin system as function of temperature and heating rate.

The vacuum bagging setup is shown in Figures 2 and 3, and it consists in the following steps:


**Figure 2.** Vacuum bagging setup scheme.

**Figure 3.** (**a**) Lower half-mold embedding the dielectric armatures. (**b**) Upper armatures mounted along with the soldered connecting wires over the vacuum bag during the resin flow progression.

The used ancillary materials, such as resin flow tubes and sealant tapes, were able to withstand high temperature. This selection was taken considering the temperature increase due to microwave preheating.

The lower armatures of the dielectric sensors are embedded in the rigid mold, as illustrated in Figures <sup>2</sup> and 3a. The armatures consist of square copper plates (25 × 25 mm2) located at 60, 150, and 240 mm from the resin inlet, respectively. The three upper armatures of the capacitive sensors were fixed on the vacuum bag in correspondence to the lower ones. The eyelets milled in the rigid lower and sealed by transparent plastic allow to visually monitor the bottom-side flow. Two cameras were focused on the vacuum bag and on the eyelet during the entire test to monitor the position of the resin flow front.

Spiral wraps were inserted for the easy entry and exit of the resin under the vacuum bag. The vacuum bag encompassing the fibrous preform was sealed using the sealant tape avoiding any spot for external air insertion. The vacuum was induced by attaching the resin outlet tube to the vacuum pump. The resin inlet was clamped and negative air pressure of 0.9 bar was maintained to place the entire arrangement under vacuum conditions. Figure 4 shows the two schemes adopted to carry out the laboratory tests: the upper scheme represents the setup without microwave preheating; the lower is the setup involving the microwave preheating.

**Figure 4.** Schematic representation of the experimental setup of the resin infusion mold equipped with the dielectric and thermal acquiring system.

The microwave facility, depicted in Figure 4, consists of a 2 kW microwave generator, stainless steel waveguides, cylindrical resonance cavity, and ancillary tools. The choice of a microwave-based system to preheat the resin is based on the thermo-chemical and rheological behavior of the thermosets. Indeed, these polymeric systems are characterized by short pot-lives, which are further reduced when increasing their temperature. This entails the necessity for an efficient volumetric heating system. More details about the design and optimization of the apparatus can be found in previous articles [32–34]. An intermediate vessel has been placed between the exit of the microwave cavity and the resin inlet into the vacuum bag. The vessel works as a buffer to decouple the heating system and the LCM apparatus to avoid a potential mismatch between the resin flow rate and the amount of energy provided by the microwaves [34], which could lead to an overheating of the catalyzed resin. The resin is driven through the microwave preheating cavity to the buffer vessel by positive pressure. The resin flow through the resonant cavity and the microwave power emitted have been calibrated by performing preliminary heating tests. A resin flow of 0.38l/min was set, while the power emitted by the magnetron was 2 kW.

#### **3. Results and Discussion**

Dielectric measurements and visual analysis of the flow front performed during the two tests are reported in Figures 5 and 6.

**Figure 5.** Dielectric acquisition in conventional resin infusion (**a**) and in resin infusion test with microwave preheating (**b**).

**Figure 6.** Visual resin flow front profiles compared with dielectric signals for resin arrival and preform saturation acquired during the two tests from top side (**a**) and bottom side (**b**).

The dielectric sensors detect variations in the capacitance of medium contained between the two armatures as the resin flow reach the sensor locations [19,20]. The observable variations in the signals during the infusion can be ascribed to the resin flow through the glass fiber fabric. Three distinct trends in the capacitance curves can be detected, which are more evident in the test without preheating where the resin flow is slower than that in the tests with resin preheating. The signal shows an initial increasing step, ranging from 8 to 15% of the saturated signal and is related to the flow of the resin through the flow media: the resin, indeed, promptly impregnated the distribution web covering almost immediately the sensing area due to its high permeability, which can be two orders of magnitude higher that the textile or even more. After that, the resin progressively fills the glass fabric layer below the flow media and the capacitance profile proceeds with a reduced slope until it reaches a plateau. At this point the proportion between resin and fibers stabilizes and the signal remains almost constant or without significant variations. Fluctuations of the signals are due to the continuous flow of the resin through the preform in the sensing areas. The capacitance curves of the three sensors present different slope for each phase of the infusion

and are due to the reduction in the resin velocity as expected in case of applied constant gradient pressure, typical of the SCRIMP process. Therefore, the impregnation is faster at the beginning of the infusion and slows down as the infusion proceeds (it is also visible in the flow front profiles described in Figure 6)

By comparing the profiles of the signals from the two experiments, the effect of the preheating on the behavior of the resin is visible. The signal detected by sensor 1, located at 60 mm from the resin inlet, reaches the plateau with a remarkably advance when compared to the non-heated resin case (Figure 5). Despite that the first step related to the resin flow through the flow media, which does not show remarkable differences, the steeper curve describing the second stage of the infusion evidences how the impregnation of the glass fabric layers proceeds faster in the test with preheated resin reaching the saturation in approximately ten seconds after the arrival of the resin at the sensing location.

A similar trend can be observed at the locations of sensors 2 and 3, where saturation of the preform with preheated resin was achieved in less than half the time required by the non-heated resin. It is worthy of noting that the effect of the resin preheating cannot be appreciated at the very beginning of the infusion when resin first goes through the flow medium: the high permeability has a predominant effect on the resin velocity than the reduction in viscosity from the temperature increasing [36].

The advancement profiles of the resin flow front captured by the cameras during the two experiments are reported in Figure 6. The two profiles in Figure 6a refer to the resin advancing on the flow media, while Figure 6b shows the flow front of the resin acquired on the mold surface from the eyelet (Figure 2). The position of the flow front has been also acquired from the dielectric signals and reported in both graphs for the tests with unheated and preheated resin.

In both tests, the resin flow velocity decreases during the infusion. Indeed, the resin initially advances pushed by a high-pressure gradient and, during the infusion, the gradient decreases in relation to the advancement of the resin describing the conventional flow through a porous medium [36]. Figure 6 shows the difference between the two test cases. Indeed, the microwave preheated resin flow front reaches the vent in less than 70% of the non-heated resin. In the earliest 100 mm the microwave preheated resin flow front is more than twice faster if compared to the non-heated case: 2.3 mm/s for the preheated resin, 1.0 mm/s for the conventional process. The velocity difference decreases as the process continues. In the last 100 mm, the average flow front velocities are 0.34 mm/s and 0.23 mm/s, respectively.

The analysis of the flow-front detections acquired from the eyelets of the bottom halfmold indicate the complete impregnation of the preform, which is delayed with respect to the top flow due to the difference in the permeability of flow media and fiber fabric. Due to the design of the rigid half mold, the bottom flow front can be acquired only on the eyelet, which ranges from 65 to 195 mm from the inlet. Therefore, only data from sensors 1 and 2 can be correlated to the visual analysis, since the sensing area of sensor 3 is not covered by the eyelet. From data reported in Figure 6, it is possible to observe that while the reduction in filling time on the top of the preform is approximately 25% at 195 mm from the inlet, the difference in the bottom flow between the tests with preheating and non-heated resin is around 50%. Clearly, the flow through the distribution medium is less affected by the preheating, thanks to the high permeability of that medium (it is visible also in the dielectric signals, as mentioned before); on the other hand, it has a significant influence on the flow through the fabric where the reduction in the viscosity plays a major role in facilitating the impregnation of the fibers. The beneficial effect of microwave preheating, as reported in previous works [32–34], is related to the rheological behavior of thermoset resins: the temperature reached by the resin at the exit of the microwave cavity was approximately 36 ◦C, while the room temperature at which the test with non-heated resin was conducted was 22 ◦C; at 36 ◦C the viscosity decreases from 0.8 Pa·s up to 0.4 Pa·s, as shown in Figure 1. Nevertheless, the thermal energy conferred to the resin must be carefully tuned to avoid premature gelation of the thermoset. At the operative temperature reached by the resin

in the test with preheating, no gelation occurred for the time required to complete the impregnation of the preform.

Table 3 summarizes the times of the resin flow when it reaches the sensing locations; the arrival on the sensor edges determining the first reaction of the sensors and the time when the resin reached the opposite edge and fully cover the sensor. The saturation time corresponds to the stabilization of the dielectric signals (Figure 5) and it represents the moment when the resin reaches the end of the sensing area on the bottom of the mold, and it has fully impregnated the whole thickness of the preform. The graphs in Figure 6 show the good agreement between the visual and dielectric analyses and the reliability of the latter in detecting the resin flow on both the top and the bottom of the mold.

**Table 3.** Times of resin flow arrival at sensing locations for tests with non-heated and preheated resin acquired by dielectric sensors and the saturation of the dielectric signals (in the parenthesis are reported the distances from the inlet port).


By analyzing the dielectric data, it is possible to appreciate the benefits of the microwave heating method: a remarkable reduction in the infusion time in the test with microwave preheated resin was registered with the shortening of the saturation times of the three sensors by approximately 80%, 65%, and 30%, respectively. The top flow registered a smaller decrease in times of approximately 64 %, 52%, and 7%, respectively, on the three sensors.

Figures 7 and 8 depict a qualitative representation of the flow front derived from the data of the dielectric sensors at specific moments of the infusion.

**Figure 7.** Schematic representation of flow front profile at the arrival of the resin on the sensing area for infusion with non-heated resin.

**Figure 8.** Schematic representation of flow front profile at the arrival of the resin on the sensing area for the infusion with microwave preheated resin.

In the SCRIMP process, the presence of a distribution medium determines the formation of two main flows: the first through the free-fibers region (i.e., the distribution medium) and the second through the fiber preform region (bulk porous medium). The former consists of a longitudinal flow from the inlet to the vent, while the second present a combination of a transverse flow (i.e., the out-of-plane flow through the thickness of the fiber preform) and a longitudinal one. This gives place to two regions characterized by different flow behaviors: a fully saturated zone, where the fluid flows through the preform with a velocity profile constant along the thickness of the preform, and a partially saturated zone, characterized by longitudinal and transverse flows through the thickness (Figure 8, time 450 s).

In the former region, the velocity vector is parallel to the main flow direction representing a fully developed flow, and meaning that the flow is substantially unidirectional with no significant crossed flows.

The flow in the latter zone is bi-directional (visual analysis indicated that in the inplane flow no variations of the resin velocity occurred along the transversal direction) since the resin permeates from the distribution media in the through-the-thickness direction. The higher resin velocity along the longitudinal direction in the distribution medium determined a complex shape flow front. This feature characterizes the unsaturated region [36].

The formation of the unsaturated region, as mentioned before, occurred as the infusion begins due to the high longitudinal permeability of the distribution medium compared to the reinforcement one. In the unsaturated region, part of the liquid resin flows transversally from the distribution medium toward the preform, however, this transversal flow occurs to limited extent also in the saturated region determining the formation of a transition region between the saturated zone and flow front region [36].

Previous work by some of the authors [36] pointed out that the ratios between the thicknesses and between the permeabilities of the distribution medium and preform influence the extension of the unsaturated region and, hence, the delay between the first arrival of the resin to the vent and fully impregnation of the preform while other parameters, such as the fiber volume fraction or the compressibility of textiles, do not play a significant role. Therefore, in the present experimentation, the differences observed have to be ascribed to the change in rheological properties induced by the preheating being the other factors kept equal in the two tests.

In the infusion with non-heated resin, the unbalance in the permeability of distribution medium and fiber bulk determined a large difference in the resin velocity in the tow region resulting in a long unsaturated region (Figure 7). Indeed, when the resin reached the end of sensor 2 the preform in sensing area 1 is only partially filled, the signal is at 30% of its plateau and, by extension, it is possible assuming that the preform is impregnated by the same percentage. On the other hand, in the test with preheated resin the length of the unsaturated zone is far smaller and the preform in sensing area 1 is filled with the resin by almost 70%, consistently with the reduction in the time required by sensor 1 to reach the plateau in the two experiments (Table 3). Furthermore, it is possible to observe that when the resin reaches the end of sensor 3 the preform at sensor 1 location is still not fully impregnated: the resin reached that location almost 250 s in advance with respect to the saturation of the preform, which was approximately 77% (Table 3). The preform at the location of sensor 2, as consequence, result partially filled by only 17%. Conversely, in the test with preheating, the resin filled in a shorter time the preform thanks to the reduced viscosity and the enhanced flow resulting in a full impregnation at sensing area 1 and in a high level of saturation of the preform at the sensing area 2 of almost 80%. The analysis of the present preliminary experimentation indicates that microwave heating is effective to enhance the resin flow not only in the conventional RTM or VARTM processes [34] but also in the case of processes using a flexible half mold, such as SCRIMP, being able to further promote the impregnation and reduce the overall filling time more than that obtainable by using the sole distribution medium. The good agreement between dielectric signals and visual analysis also indicated the validity of the dielectric analysis and parallel-plate sensor for flexible-mold manufacturing processes. Further development involving the numerical analysis of the resin flow will be useful to strengthen the correlation between the actual position of the resin during the infusion and the signals from the dielectric sensors.

#### **4. Conclusions**

This paper compares the performances of the conventional resin infusion process and microwave preheated resin infusion in the case of flexible mold. The experiments conducted and the analysis of the achieved outcomes evidenced that the resin system preheating gives place to beneficial effects in the vacuum infusion processes, with a marked reduction in the cycle time. The reduction in the viscosity provokes an improvement of the flow, in agreement with Darcy's law. Considering the achieved results and what was discussed above, the following remarks can be drawn:


The presented results raise new interrogatives, such as the effects of microwave preheating on the fibrous preform saturation in vacuum bag infusion processes and resin system curing time, the effects related to the geometry and the architecture of the fibrous preform, or the influence of the binder applied, just to mention some of them. These aspects should be investigated by further analyses in future works.

**Author Contributions:** Conceptualization, F.R., F.T., V.E. and P.C.; methodology, F.R., F.T., V.E. and P.C.; formal analysis, F.R., F.T., V.E. and P.C.; investigation, F.R., F.T. and V.E.; resources, P.C.; data curation, F.R., F.T., V.E. and P.C.; writing—original draft preparation, F.R., F.T. and V.E.; writing review and editing, F.R., F.T., V.E. and P.C.; visualization, F.R., F.T. and V.E.; supervision, P.C.; project

administration, P.C.; funding acquisition, P.C. All authors have read and agreed to the published version of the manuscript.

**Funding:** This research received no external funding.

**Institutional Review Board Statement:** Not applicable.

**Informed Consent Statement:** Not applicable.

**Data Availability Statement:** Not applicable.

**Conflicts of Interest:** The authors declare no conflict of interest.

#### **References**


#### MDPI

St. Alban-Anlage 66 4052 Basel Switzerland Tel. +41 61 683 77 34 Fax +41 61 302 89 18 www.mdpi.com

*Journal of Composites Science* Editorial Office E-mail: jcs@mdpi.com www.mdpi.com/journal/jcs

Academic Open Access Publishing

www.mdpi.com ISBN 978-3-0365-8121-7