The GNPs were purchased from First Graphene (UK) Ltd, with an average particle diameter of ∼10 µm and tapped density of 0.124 g cm⁻³. Plain weave woven GF, with an areal weight of 299 g m⁻², was purchased from Easy Composites (UK). Low viscosity Araldite epoxy resin and Aradur hardener were purchased from Huntsman (United States). Ethanol was purchased from Fisher Scientific (United Kingdom).

To disperse the GNPs into the epoxy resin uniformly, high-speed vacuum mixing was applied using the SpeedMixer (DAC 600.2 CM51) under the conditions summarized in Table 1. GNPs were mixed initially with either epoxy resin or hardener, after which the other component was added. Scanning electron microscope (SEM) analysis of the brittle fracture surfaces was then undertaken to assess the GNP distribution and toughing mechanisms, as shown in Figure 1. All images show river lines caused by the brittle fracture of the thermoset resin (Hull, 1999; Olowojoba et al., 2017), with the crack deflection and pull-out contributed by the embedded GNPs (Johnsen et al., 2007; Bindu et al., 2014; Eqra et al., 2015; Domun et al., 2017; Hu et al., 2020), which are beneficial to the composite toughening. No difference was observed whether the GNPs were mixed into the resin or hardener first. Hence, as the weight ratio of the epoxy and hardener was 100:35, the GNPs were initially added to the epoxy for high-loading GNPs to be used.

Once the epoxy–GNP formulation was prepared (Figure 2A), VARI was used to fabricate the composites (Figure 2B), followed by the cure cycle suggested by the supplier (80°C 2 h +140°C 8 h). After demoulding, it was visually observed that few GNPs had penetrated into the laminate, with most of them being filtered and remaining in the infusion mesh. This severe degree of filtering in the mesh suggests that the direct VARI method is not suitable for the GNP-loaded epoxy (Figure 2C).

To solve the GNP distribution problem, a spray-coating method was developed. Initially, GNPs were dispersed in ethanol, with the concentration of ∼5 mg ml⁻¹, by ultra-sonication for 40 min. Eight layers of plain weave GF with a quasi-isotropic layup ([0/90]/[±45])_2s were selected for the sample preparation. All internal surfaces were spray-coated with the dispersed GNPs (Figure 3A), using a Paasche VL airbrush system connected with an Iwata Power Jet Lite compressor. After being left overnight to evaporate all the solvent, the VARI method was used to fabricate the composites. Samples with the GF spray-coated with 0, 0.1, 0.5, 2, and 5 wt% of the GNPs (relative to the fabric) were prepared, where a 0 wt% sample was spray-coated with pure ethanol as a control to ensure this did not affect the mechanical properties of the fibers. Figures 3B, C represent the GF before and after the spray coating, where GNPs spread uniformly on the surface of the fabric. To check the final mass of GNPs, the fabric was weighed before and after the coating procedure, with the values summarized in Table 2. It suggests that there are weight losses caused by overspray and other mechanisms during the spray-coating procedure, and the final GNP loadings were found to be 0, 0.03, 0.15, 0.49, and 1.11 wt% relative to the fabric.

Field emission SEM, SU5000, and TESCAN MIRA3 SC were used to observe the morphology of the GNPs, GF, and composites. An FEI Tecnai G2 20 (LaB₆) transmission electron microscopy (TEM) was used to evaluate the GNPs using bright-field images and diffraction patterns. A Renishaw InVia Raman system was applied to obtain the Raman spectroscopy of the GNPs, using the 633-nm laser. A Thermo Scientific Nicolet iS50-IR with a diamond ATR crystal was used to conduct the Fourier transform infrared (FTIR) spectroscopy. The density of the composites was measured by a Sartorius YDK03 Density Kit through liquid buoyancy, using isopropyl alcohol as the liquid, based on the standard ASTM D792. A NETZSCH STA 449 F5 Jupiter was used for the thermogravimetry analysis.

The mechanical performance of the composites was evaluated through tensile and four-point bending tests, each test with three specimens, according to ASTM D3039 and ASTM D7264 standards. The specimen sizes of 250 × 25 × 2 mm and 100 × 12.7 × 2 mm were selected for the tensile and flexural tests, respectively, following the recommendation from the standards. The tests were undertaken in the environmental lab with a constant temperature of 23°C and relative humidity of 50%. Regarding the tensile tests, the gauge length was calibrated at 50 mm, with a testing rate of 2 mm min⁻¹. During the flexural tests, the support and load span were set at 67.2 and 33.6 mm, respectively, with the testing rate (Y, 3.59 mm/min) calculated based on Equation 1 (ASTM D6272): Y = 0.167 Z L 2 / d (1) where L is the support span (mm), d is the depth (thickness) of the beam (mm), and Z is the straining rate of the outer fibers (0.01 mm/mm min).

The representative Raman spectrum, SEM image, TEM image, and its selected area electron diffraction pattern of the GNPs used are shown in Figure 4. Characteristic G (∼1,581 cm⁻¹), 2D (∼2,672 cm⁻¹), and D (∼1,332 cm⁻¹) bands of graphitic materials are shown in the Raman spectrum (Figure 4A). Here, the G band represents the C-C sp² network (Lin et al., 2015), the broad and asymmetric 2D band suggests the GNP consists of graphene with many layers (Lin et al., 2015; Ferrari et al., 2006), and the D band indicates the structural defects related to the zone-boundary phonons (Ferrari et al., 2006). In addition, FTIR was used to identify functional groups of the GNP flakes. As shown in Figure 4B, the representative C=C stretching band (∼1,586 cm⁻¹), hydroxyl band (C-OH, ∼1,200 cm⁻¹), epoxy vibrational band (C-O-C, ∼1,107 cm⁻¹), and carboxyl band (C=O, ∼1,700 cm⁻¹) are clearly observed (Ţucureanu et al., 2016; Coates, 2006; Li et al., 2018; Li et al., 2013). The existence of these functional groups paves the path for improving interfacial connections through reacting with epoxy matrix. The SEM image and TEM bright-field image (Figures 4C, D) illustrate the GNP flake structure with the lateral size varies from a few to ∼15 microns. Figure 4E shows the corresponding selected area electron diffraction pattern of the GNP in Figure 4D. It indicates the crystalline structure of the GNP, which has many layers of graphene with highly ordered hexagonal arrangements of carbon atoms (Venturi and Hussain, 2020).

Figure 5 shows the SEM images of GFs spray-coated with 0.03, 0.15, 0.49, and 1.11 wt% GNPs. With the increasing loading, more and more GNPs are seen to be connected with each other and form continuous networks. The surface coverage of GNPs was assessed using ImageJ, and it was found that the GNP areal coverage increased from 7.2 to 22.6, 42.1, and 59.9% at 0.03, 0.15, 0.49, and 1.11 wt% GNPs, respectively. This may have significant effects on the interfacial connections and thus mechanical properties of the composites.

To check the fiber volume fraction and void content of the composites, thermogravimetry analysis was used with nitrogen as the atmosphere and temperature ranging from room temperature to 800°C. The results are summarized in Figure 6, from which the weight percentage of the resin can be obtained as 34.7, 34.8, 35.6, 35, and 34% for control, 0.03, 0.15, 0.49, and 1.11% samples, respectively. The residual weight includes both glass fibers (GFs) and GNPs, which can be differentiated as mentioned in Section 2.4. Afterward, the fiber volume fraction ( V f ) could be calculated by: V f = W f × ( ρ c / ρ f ) (2)

whereas the void content ( V v ) could be obtained based on ASTM D2734: V v = 100 − ρ c × ( W r ρ r + W f ρ f + W g ρ g ) (3) where ρ is the density, W is the weight fraction, and V is the volume fraction. The subscript symbols f, c, v, r, and g represent the glass fiber, composite, void, resin, and GNP, respecctively. The density of the composite was measured by a Sartorius YDK03 Density Kit, whereas densities of the fiber (2.54 g/cm³), GNP (2.2 g/cm³), and resin (1.19 g/cm³) are constant. All corresponding results are summarized in Table 3.

The results suggest that with the GNP loading increasing, the fiber volume fraction decreased gradually, and void content tends to increase, particularly at high GNP loadings (0.49 and 1.11%). This could be attributed to the 2D layered structure of the GNPs, which led to more voids trapped between the flakes.

The representative stress–strain curves of the composites with different GNP loadings under tension are shown in Figure 7A. The embedded photo of the samples indicates the uniform distribution of the GNPs. All composites exhibited linear behavior during the tensile tests. In particular, all curves overlapped at the initial stage (0–0.5% strain), where the tensile modulus was determined using the strain range between 0.1 and 0.3%. Hence, with the GNP loading increasing, the tensile modulus, which represents the elastic properties, remained unchanged with the values sitting within error bars, as shown in Figure 7B. However, with the strain further increasing, the curve of the sample with the highest GNP loading (1.11 wt%) separated with all other curves due to the formed continuous GNP networks (Figure 5D), which hindered the stress transfer and triggered the occurrence of the delamination (Kamar et al., 2015; Umer, 2018). As a result, the tensile strength, which is dominated by the fiber reinforcements rather than GNP fillers (Kumar et al., 2020), kept constant with the GNP addition until the loading reached 1.11 wt%, where the strength decreased by 12.9% (Figure 7B and Table 4).

Compared with the tensile properties of the composites, which are governed by the reinforcing GF, flexural properties typically consist of a combination of the fabric, GNPs, matrix, and their interfacial connections (Wang et al., 2016b). Figure 8 shows the flexural stress–strain curves, strength, and modulus for all the composite laminates. All the curves showed linear behavior, indicating elastic deformation, with the slope increasing with loadings up to an optimum loading of 0.15 wt%, above which the properties began to decrease. Accordingly, the trend of both flexural strength and modulus (obtained from the 0.1 to 0.3% strain) initially increased, followed by a decrease with the GNP loading increasing. Both inflection points sit at 0.15 wt%, where the strength and modulus increased by 6.8 and 1.6%, respectively, as shown in Figure 8B and Table 4. This improvement was attributed to the strengthened interfacial connections, consisting of chemical bonding, mechanical anchoring, and interlocking, which can occur between the GNPs and matrix, which, as a result, prevented the crack initiation and increased the crack propagation path. Among them, the chemical bonding was contributed by the functional groups on the GNPs, such as hydroxyl, epoxy, and carboxyl, as evidenced by the FTIR (Figure 4B), which could interact more strongly with the epoxy matrix.

Other groups have also reported similar behaviors (Seretis et al., 2017; Wang et al., 2016b; Topkaya et al., 2020; Jena et al., 2020; Turaka et al., 2021); namely, the flexural properties got improved, followed by weakened with increasing GNP additions, as a small amount of GNPs could contribute to chemical bonding, mechanical anchoring, and interlocking between the interfaces (Pathak et al., 2016; Moaseri et al., 2014), which strengthened the composites under bending. However, the formation of GNP networks under high loadings (Figure 5D) accelerated the delamination and decreased the stress transfer efficiency (Kamar et al., 2015; Umer, 2018), thus weakening the composite performance.