An alternative synthesis route to graphene oxide: influence of surface chemistry on charge transport in epoxy-based composites

Chromium (VI) trioxide was purchased from Fischer scientific (11390939, supplied by Alfa Aesar™), KMnO₄ was purchased from AnalaR (26910 VWR International) and H₂SO₄ from Sigma-Aldrich (79620). The precursor graphite and the hydrogen peroxide (H₂O₂) solution (35%) were also supplied by Sigma-Aldrich (496596 and 349887, respectively). The epoxy resin used was an amine-cured, bisphenol A-based system, namely Epilox™ Infusion System 5300 (LEUNA-Harze GmbH). This resin was chosen for this study because its low viscosity (~ 275 mPas/25 °C) makes it appropriate for structural fibre-reinforced composites in wind energy applications, the application of interest. The two parts used are ER 5300 as epoxy monomer (epoxy equivalent weight: 170 g/eq) and EC 5310 as hardener (amine value: 520 mg KOH/g) components, respectively.

A modified Tour method was used to synthesize GO, to provide a comparator system for the chromium-synthesized mGO. For this, 40 ml of a H₂SO₄/H₃PO₄ mixture (5:2 per volume) was mixed with 1.8 g of KMnO₄ and stirred at 50 °C to give a homogeneous solution. Afterwards, 300 mg of graphite were added to the solution and left for 4 h under mild shaking. Another batch was prepared and left for 30 min to demonstrate the effect of the process duration. The system was neutralized by transferring into 100 ml of cool distilled water, to which, 10 ml of the H₂O₂ solution was added. Centrifugation (Heraeus™ Megafuge™ 8) at 8872 g for 10 min was used to retrieve the GO and remove the supernatant acid. The sediment was re-dispersed and washed in distilled water (three times), ethanol (once) and acetone (twice); after each step, the GO was recovered from the suspension using centrifugation. A final GO/acetone solution was obtained and probe sonicated (UP 200S) for 30 min with a 7-mm-diameter tip; the resulting GO could be incorporated into the epoxy only within this solvent, since subsequent drying led to a material that was too agglomerated to be dispersed. Henceforth, the resulting oxidized graphite will be referred to as GO.

For this synthesis, the graphite/acid ratio was the same as in the previous method, but now, only H₂SO₄ was used as the reaction medium. A saturated solution was produced by adding ~ 20 mg of CrO₃ per ml of H₂SO₄ under continuous magnetic stirring at 50 °C; to assist dissolution of the CrO₃, the as-received crystals were first ground to a fine powder using a mortar and pestle. The resulting orange-coloured solution was poured into a glass flask containing the appropriate amount of graphite and was placed on a shaker for 30 min for the oxidation to proceed. After this time, the system was neutralized, centrifuged, washed and sonicated using the same conditions described above. Finally, the mGO was dried at 100 °C overnight since, in this case, this procedure did not lead to a highly agglomerated product. Visual comparison of suspensions in acetone of GO (yellowish suspension to the left in Fig. 1) and mGO (greyish suspension to the right in Fig. 1) reveals a marked difference in colour, which we interpret in terms of a shift in the band edge resulting from changes in the functional group density on the GO surface [44].

Epoxy/mGO nanocomposites were prepared as follows: the epoxy was first mixed for 5 min with the appropriate amount of dry filler, using a planetary mixer (Speedmixer™ DAC 150.1 FV) operating at 3000 rpm. Although it is more normal initially to distribute GO in a solvent to aid dispersion, this was found to be unnecessary in the case of our mGO, which beneficially obviates the need for subsequent solvent removal. From a technological perspective, direct dispersion without the need for the involvement of additional processing steps may be advantageous [34]. Furthermore, in our experience, the use of solvent-based preparation methods in conjunction with low-viscosity epoxy resins, like the one used in this study, can increase the propensity for particle sedimentation [45] and deteriorated final material performance, particularly when the GO is lightly functionalized. Subsequently, the hardener was added in a stoichiometric weight ratio of 10/2 (epoxy/hardener), as suggested by the supplier. The mixture was then vacuum degassed to remove entrained air and cast into moulds with the required dimensions. Curing was conducted at 70 °C for 6 h, as suggested by the epoxy manufacturer.

The dry graphite, GO and mGO powders were characterized by Raman spectroscopy, in order to evaluate the structural alterations induced by the imposed oxidation process. For this, a Renishaw RM1000 confocal microprobe system was used, using a 780 nm excitation wavelength and operating at 2.5 mW power. A Perkin-Elmer Pyris 1 was used to assess the extent of the graphite oxidation via thermogravimetric analysis (TGA). Graphite, GO and mGO samples were first held at 70 °C for 10 min before being heating at 10 °C/min to 700 °C in a nitrogen atmosphere. Elemental analysis of the examined powders was performed with a monochromated Al Kα X-ray photoelectron spectroscopy (XPS) instrument (Kratos Axis Ultra DLD), with a typical resolution of 0.47 eV. The nanocomposite systems were morphologically characterized via scanning electron microscopy (SEM) using an EVO LS25, Zeiss instrument, and transmission electron microscopy (TEM) using an FEI Tecnai T12, instrument. For TEM, sections were prepared using a Leica Ultracut E ultramicrotome together with a Diatome 45° diamond knife.

The glass transition temperature (T_g) of all the epoxy-based systems (unfilled resin and nanocomposites) was determined by differential scanning calorimetry (DSC), using a Perkin-Elmer DSC7. For this, T_g was evaluated from the point of inflection of the step-like transition of the heat flow while heating up to 120 °C at a rate of 10 °C/min. Prior to this, the respective system’s thermal history was erased [46] by heating and cooling over the same temperature range at the same scan rate. A total of three repeats were used to calculate average and standard deviation values for the T_g of each material system. DC electrical conductivity was determined by means of two-probe methods using two different sample thicknesses for repeatability purposes and to exclude possible geometric factors from influencing the derived values. The thick samples were disc-shaped, 3 mm in thickness, and their surfaces were polished with abrasive paper (grade 1000) and then gold-coated. An Agilent 34401A digital multimeter and guarded, disc-shaped, aluminium, sample holders were used; the applied voltage was 150 V. The thin samples were rectangular shaped, ~ 200 μm in thickness and their surfaces were, again, polished with abrasive paper (grade 1000). A Keithley 6517B electrometer/high resistance metre fitted with an 8009-resistivity fixture was used with the applied voltage being 100 V for 6 min. Both evaluation methods produced comparable results, indicating that conduction at the chosen voltage was Ohmic; the values quoted are the average of all measurements, together with the associated standard deviation.

The research hypothesis that underpins the work reported here is that changing the oxidizing medium from a widely used manganese-based system to one based upon chromium leads to a change in the oxidation state of the oxidized graphite: specifically, the production of material that is characterized by a reduced degree of oxidation/disorder. This issue will therefore be addressed first, using three complementary approaches.

Raman spectra obtained from the graphite precursor, the two GO specimens and the mGO are presented in Fig. 2. A total of four characteristic bands can be observed at 1307, 1569, 1602 and 2610 cm⁻¹, namely D, G, D′ and 2D, respectively. The G band is associated with the first-order scattering of E_2g phonons, representing the in-phase vibrations of the sp² carbon lattice [47, 48], while the D and D′ bands are associated with the extent of disorder [49, 50]. The 2D band, also termed the G′ band, is the overtone of the D band and is associated with the highly ordered stacking sequence of crystalline graphite [50].

The Raman response of the precursor graphite is dominated by vibrations of the carbon atoms (sharp and intense G band) but also contains evidence of defects, as evinced by the presence of the D and D′ bands. In contrast, the GO Raman spectra reveal broad G bands shifting slightly to higher wave numbers with increasing oxidation duration. The sample oxidized for 30 min displayed a G peak at 1583 cm⁻¹, while the respective peak for the sample oxidized for 4 h was located at 1586 cm⁻¹. This indicates resonance at higher frequencies, which has been attributed to the presence of isolated double bonds compared to the graphite [47]. Reasonably, here, this phenomenon parallels changes in oxidation time while, elsewhere, it has been reported to occur as a result of increasing the amount of KMnO₄ [51]. Both GO samples exhibit considerably broadened D bands, which are more pronounced than the respective G bands. This is an indication of oxygen-based groups being attached on the graphitic lattice, decreasing the size of the in-plane sp² domains [47] through sp³ bonding [51]. Also, the 2D band is not visible in either GO sample, which is reasonable given that the stacking order will be heavily disrupted by the inherent structural imperfections [51]. In the case of mGO, the G band is located at 1576 cm⁻¹, a lower wave number than in the GO samples but higher than graphite. Nevertheless, the G band is still broadened and less intense compared to the sample’s D band and, in this respect, the mGO mirrors the Raman behaviour of the GO. This indicates that while the graphitic structure is disrupted, the extent of the disruption is reduced compared with the GO. The latter is also confirmed by the presence of the 2D band (albeit weak) in the mGO Raman spectrum, which suggests that some stacking order is preserved after the oxidation process, contrarily to both GO samples. The I_(D)/_(G) intensity ratios for the precursor graphite, GO (4 h oxidation), GO (30 min oxidation) and mGO were: 0.724, 1.347, 1.312, 1.297, respectively, as portrayed in Fig. 2. This ratio is commonly used as an indication of the induced disorder in the system, and the acquired results clearly align with this, showing smaller I_(D)/_(G) values for the mGO compared to both GO samples. Nevertheless, discrepancies have been observed before [51], leading to the assumption that there is no straightforward relation between the intensity ratio and the oxidation level. Full width at half maximum (FWHM) of the observed G peaks (see Fig. 2) were 18.9, 83.49, 87.64, 71.01 cm⁻¹ for the precursor graphite, GO (4 h oxidation), GO (30 min oxidation) and mGO, respectively. This value has also been used [51] to reflect the oxidation level, by showing increasing dependency with it. It is apparent that the precursor graphite displays the smallest values among the examined samples. While a slight discrepancy was observed for the two GO samples, the mGO shows a considerably lower FWHM value compared to them.

TGA is commonly used to evaluate the concentration of oxygen groups attached to the graphitic structure, since this determines the weight loss at specific temperatures [52]. Figure 3 compares TGA data obtained from the precursor graphite, the mGO and the samples of GO prepared using the two different oxidation times. When heated up to 700 °C, the graphite displays a minimal mass loss; the small observed reduction in mass is ascribed to the presence of impurities/defects and therefore aligns with the evidence of defects provided by the Raman data shown above. In contrast, all the systems that were subjected to oxidation (both GO and mGO) show three weight loss processes. The small mass loss that occurs up to 200 °C is ascribed to removal of adsorbed water [16, 25]. Between 200 and 300 °C, the rate of mass loss is increased, while further heating results in a gradual, monotonic reduction in sample mass; these two processes have, respectively, been related to the removal of labile and stable oxygen-based groups [26]. In the case of both GO powders, removal of labile oxygen-based moieties leads to a mass reduction of more than 30%, a figure that aligns well with published data [25, 26]. Evidently, varying the oxidation time within the range considered here does not lead to very different changes in sample mass between 200 and 300 °C, albeit that the mass loss does appear to be slightly larger in the case of the system oxidized for the longer time. This implies that the graphite surface chemistry is rapidly disrupted by the manganese-based oxidizing agent, since extending the duration of the process by eightfold results in only a minor change in the product. For this reason, only one oxidation time (30 min) was used for the synthesis of the mGO, since it leads to a saving in processing time and aligns with the objective to reduce the extent of oxidation. The TGA data shown above indicate that using CrO₃ in place of KMnO₄ leads to a significant reduction in mass loss (~ 10%) between 200 °C and 300 °C, implying that the mGO structure contains proportionally fewer labile oxygen groups. Indeed, our directly synthesized mGO (i.e. without being subjected to any subsequent reduction process) displays a smaller related mass loss than either chemically reduced GO (via reaction with hydrazine hydrate) [53] or microwave reduced GO [32].

Figure 4 shows XPS data obtained from the precursor graphite, the GO and the mGO powders. Figure 4a shows the overall spectra, while Fig. 4b presents the C1s peak at high resolution. Derived elemental ratios are presented in Table 1.

It can be seen that the precursor graphite contains some minor oxygen impurities, resulting in a C/O ratio of ~ 30, which is consistent with the Raman data presented above. The synthesized GO shows a much lower C/O ratio of ~ 2.8 and an oxygen content of ~ 26%. These values are typical for conventional GO, where C/O ratio values between 2 and 4 have been reported [54]. The C/O ratio for our mGO is ~ 5.1 with an oxygen content of ~ 16%, confirming the above Raman and TGA results, both of which indicate a moderate degree of oxidation. It should be noted that the small S content seen in both GO and mGO is consistent with published data, where the effect has been attributed to covalently bonded sulphates or absorbed sulphuric acid [55]. In comparison, rGO typically shows C/O values of ~ 7.5 [38] but can reach up to ~ 12 [54]. Furthermore, the oxygen content for GNP is usually around 7% [56] or lower [57]. The electrochemical exfoliation approach mentioned earlier has shown, under specific circumstances, similar oxygen contents to mGO, albeit with more complex production [39].

The deconvolution of the C1s core-level spectra (Fig. 4 b) shows 5 different peaks located at: 284.4 eV (sp² hybridized carbon), 285.04 eV (sp³ hybridized carbon), 286.72 eV (C–O bonds from hydroxyls or epoxides), 288 eV (C=O bonds from carbonyls) and 289.06 eV (O=C–O bonds from carboxyls). As expected, the GO shows a significant amount of sp³ hybridized carbon atoms compared to the intact sp² ones, which is a result of severe oxidation by KMnO₄. Furthermore, the pronounced presence of epoxides and hydroxyls, as discussed earlier, can react with the epoxy matrix [29]. The relatively increased carbonyl and carboxyl presence is also a typical result of the permanganate method [20]. Conversely, mGO shows a significantly reduced extent of sp³ carbon hybridization and a decreased amount of functional groups compared to the GO. It is noteworthy that the dominant functionalities present on the mGO surface are epoxide and hydroxyl groups, with the presence of carbonyls and carboxyls being weaker compared to the GO but stronger compared to the graphite. Overall, the chromic method resembles the chlorate method in terms of functionalities present [20], but in a much lower amount, leading to the moderately oxidized graphene.

The evidence obtained above by Raman spectroscopy, TGA and XPS presents a consistent picture, whereby we conclude that the use of our chromium-based oxidant has indeed led to the production of oxidized graphite (termed mGO here) that is characterized by reduced disorder and a reduced level of oxidation, compared with GO produced using the widely used approach first reported by Tour in 2010. The impact of this on the structure and physical properties of a range of mGO/epoxy composite systems will now be described.

SEM images of fracture surfaces of the sample containing 0.5 vol% of mGO can be seen in Fig. 5a–c, from which a fine dispersion within the epoxy matrix is revealed. Image analysis undertaken for 74 individual flakes (Fig. 5d) shows a size distribution between 0.75 and 1.1 μm for 45% of them, while 23% of the flakes exhibit a lateral size between 1.3 and 1.7 μm. Figure 5e, f illustrates TEM images of isolated tactoids found in the 1 vol% mGO-filled sample, with maximum lateral sizes ~ 1.6 μm and ~ 0.8 μm, respectively, aligning with the abovementioned size distribution.

Typical TEM images showing the range of structural states of the mGO when dispersed at 2 vol% within the epoxy matrix are presented in Fig. 6. A satisfactory dispersion of the mGO forming a graphitic network within the matrix is shown. From the variation in contrast from image to image, it is evident that some areas are characterized by well-exfoliated structures (red arrows) while, in others, the mGO retains a graphitic multilayered configuration (red circle). This conforms to the Raman spectra presented above, which implies that some degree of stacking order has been preserved due to the weak oxidation. Analysis of tactoid dimensions reveals a thickness distribution with the highest probability (> 40%) between 5 nm and 6 nm. Assuming a monolayer thickness of ~ 0.8 nm [6] for oxidized graphite, the majority of the examined structures contain less than 10 layers. The aggregated structures, such as that circled in red in Fig. 6, typically appeared to be ~ 60 nm in thickness.

Figure 7 presents the effect of filler content on the T_g of all the nanocomposite samples considered here. In this, both weight fraction (W_f) and volume fraction (V_f) values are presented, the latter being derived from the former using Eq. 1 [19]:where ρ_m is the matrix density (1.16 g/cm³) and ρ_f is the filler density (2.2 g/cm³ [12, 19]).

From the above, it can be seen that the glass transition temperature of the epoxy matrix is not detrimentally affected by the inclusion of mGO, albeit that a statistically significant increase in T_g is observed from 4 to 12.5 wt%. Increased T_g values have previously been reported at small GO filler contents (> 0.5 wt%) and have been ascribed either to strong filler–matrix interactions [30, 34] or to the presence of wrinkled GO morphology [33], both of which serve to constrain matrix chain mobility. Here, comparable effects are seen at higher filler contents, compared to the abovementioned studies. Ultimately, at the highest filler content considered here, the T_g value drops, which has been observed before, albeit at considerably lower filler fractions and ascribed to the high concentration of reactive functional groups that influences the stoichiometry of the epoxy crosslinking reaction [29]. Considering the latter interpretation, the T_g dependence on mGO content presented in Fig. 7 is consistent with the previous conclusion that the density of functional groups grafted onto the mGO surface is limited compared with conventional GO. Evidently, mGO can be incorporated into epoxy matrices at much higher loading levels than conventional GO without adversely affecting the curing reaction of the matrix polymer.

Figure 8 shows the dependence of electrical conductivity on mGO volume fraction. At the lowest loading levels (0.1–0.4 vol%), an increase in conductivity of about two orders of magnitude progressively occurs while, between 0.40 and 0.53 vol%, the conductivity sharply increases by a factor approaching 10⁶.

In general, the conductivity of an insulating matrix containing an electrically conductive phase varies according to [58‐60]:where σ_c and σ_f are the electrical conductivity of the composite and the filler, respectively, t is a constant representing a critical exponent of the electrical conductivity and V_c is the critical fraction for which the composite exhibits a sharp conductivity increase. That is, V_c represents the percolation threshold (P_t). Theoretically, a system containing randomly dispersed spherical, conductive particles would be expected to be characterized by a percolation threshold at about 15 vol% and a value of t ≈ 2 but, in practice, significant divergences occur [61]. Lower percolation thresholds are expected from fillers with high aspect ratios [62]; t values can vary significantly [63‐65]. Also, definitions of percolation threshold vary in the literature, with different studies variously defining this as follows: the volume fraction at which the particles form contiguous paths through physical contact, sharply increasing the electrical conductivity [66]; when the conductivity surpasses the antistatic limit (10⁻⁸ S/cm) [62]; or even when the system starts behaving nonlinearly [61]. Since the testing voltage used in this study was chosen to be low and also because the investigated systems do not display metallic behaviour, we take the percolation threshold to be determined by the first definition given above. By linearly fitting Eq. 2 to the obtained results (log–log plot inset in Fig. 8), with a fixed parameter of V_c = 0.0053, the other two parameters σ_f and t were calculated 10^{−(5.29 ± 0.72)} S/cm and 1.44 ± 0.38, respectively (Table 2).

Comparison of these results with published data reveals that the exponential factor t derived from the mGO-based systems considered here is in reasonable agreement with reported values [65], while the calculated conductivity of the mGO (σ_f) is, unsurprisingly, low compared to pure graphite [67] but is in reasonable agreement when compared to values derived from systems based on GO [12, 68, 69].

The direct usage of conventional GO as a means of tailoring the electrical conductivity of epoxy-based nanocomposites has, to our knowledge, only previously been considered in the following four studies. Mancinelli et al. [70] achieved an increase of almost 2 orders of magnitude (filler content up to 0.5 wt%) only by thermally treating the nanocomposite partially to reduce the GO; before thermal treatment no increase in conductivity was seen, compared with the unfilled epoxy matrix. Similar results were presented in our previous study [29] with the usage of GO in the same epoxy formulation as in the present study. Even though the filler content was extended up to 2 wt%, the electrical conductivity of the respective nanocomposite was the same (slightly reduced) compared to the neat epoxy, while further GO addition reduced the epoxy’s T_g below ambient temperature. Kim et al. [71] did not achieve any significant conductivity increase with the usage of up to 3 wt% GO. Only the study presented by Tang et al. [72] achieved a modest conductivity enhancement (up to ~ 10⁻¹⁰ S/cm) with a percolation threshold of 6 vol%. It is of considerable significance to note that the GO contents mentioned above have been reported [27‐29] to affect the epoxy curing reactions as mentioned earlier. A summary of the abovementioned review is presented in Table 3. Since in most cases the achieved conductivity values were relatively low, the respective data are presented as a σ_c/σ_m ratio, where σ_m is the electrical conductivity of the unfilled matrix.

Thus, previous studies of systems based upon GO in epoxy indicate, at best, percolation onsets at high volume fractions, relatively small improvements in electrical conductivity and adversely perturbed matrix curing. In contrast, the mGO system developed in our work resulted in significantly lower percolation threshold (around 12 times smaller filler in terms of vol%), higher achieved electrical conductivity and had no adverse effects on matrix curing.

In view of the ineffective nature of GO as a means of increasing electrical conduction through epoxy-based materials, alternative strategies with this aim have used more convoluted multistep synthesis procedures to address the limitations of using GO directly. For example, Thachil et al. [32] used microwave exposure to reduce GO and, consequently, achieved a percolation threshold of 0.27 vol%; the maximum conductivity value reported (σ_max) was, however, relatively low (10⁻¹⁰ S/cm). A similar percolation threshold was achieved by Hsiao et al. [73], who employed thermal reduction of GO at 700 °C in an argon atmosphere; in this study the σ_max reached 10⁻⁸ S/cm. Liang et al. [53] utilized an extensive, two-step, GO reduction process involving an initial chemical reaction with hydrazine hydrate, followed by subsequent annealing at 250 °C under N₂. The resultant composite systems exhibited an σ_max of 10⁻¹ S/cm and a percolation threshold of 0.52 vol%.

In the case of functionalized GO, Bao et al. [36] utilized a two-step chemical process to attach glycidol groups to the GO surface and, in this way, increased the electrical conductivity to 10⁻¹¹ S/cm with a percolation of 2 wt%. Tang et al. [69, 72] achieved a much greater increase in conductivity (up to ~ 10⁻² S/cm) with similar, albeit gradual percolation onset, through surface functionalization of GO with polyetheramines. A summary of the abovementioned review is presented in Table 4.

Comparing the above results in their entirety with the behaviour of the epoxy/mGO systems reported here, it is evident that many studies report comparable behaviour with that described here while, in two cases, further processing of GO has led to systems with high bulk conductivity values (0.01–0.1 S/cm). Also, in all of these cases, the rGO/fGO was introduced into the epoxy resin in a solvent, the removal of which may introduce limitations on industrial implementation [34].

Finally, it is worth considering the published literature concerning epoxy matrices including EG and GNP, despite the rather different chemistries and processing methodologies involved. Many studies have considered systems based upon EG, which is prepared by acidic intercalation followed by thermal expansion and then extensive sonication and solvent mixing with epoxy. Such work has reported percolation thresholds of 3 wt% [74], 1.3 vol% [75], 0.5 vol% [76], and 1 phr [77]. The σ_max values reported in these studies were, respectively: 10⁻³ S/cm; 10⁻⁵ S/cm; 10⁻⁶ S/cm; 10⁻³ S/cm. Elsewhere, Li et al. [67] used an additional UV/O₃ surface treatment procedure and achieved a σ_max of 10⁻⁴ S/cm, albeit with gradual percolation. Attempts to disaggregate graphite directly through prolonged sonication in solvent, whilst leading to systems that exhibit high conductivity values approaching 10⁻² S/cm, were shown to be characterized by percolation thresholds > 8 wt% [78], a result we associate with low compatibility with the solvent, limited levels of disaggregation and poor dispersion.

In the case of studies of GNP/epoxy systems, direct dispersion within a solvent prior to introduction of the GNP into the resin has been reported to result in a percolation threshold > 1 wt% and an σ_max of 10⁻⁶ S/cm [79]. More complex dispersion strategies have, however, been described, including three-roll milling [80], encapsulation with poly(vinyl pyrrolidone) combined with freeze-drying [81], spin coating (up to 12,000 rpm) [82] and AC-field induced alignment [56]; the resulting data vary widely, with σ_max values ranging from 10⁻⁹ S/cm to 10⁻⁴ S/cm. A summary of the abovementioned review is presented in Table 5.

Therefore, the usage of mGO as an epoxy filler offers similar or smaller percolation thresholds than EG or GNP. However, the production of EG and GNP includes more than one step and is generally characterized by reduced yields, compared with mGO, which was synthesized with a one-step process and approaching 100% of the starting material being recovered and used, without any fractionation, in preparing the final nanocomposites.