Thermal Properties of Graphite Nanoplatelet-Epoxy Composites Formed Through High Shear Exfoliation of Expanded Graphite

Easily manufactured, low-cost fillers are highly desirable for improving the thermal properties of polymer resins. Graphite-based materials are particularly interesting because of their ability to form high aspect ratio, high thermal conductivity platelets and sheets. It has been shown that graphite vastly outperforms other thermally conductive fillers such as silver, aluminum, or diamond [1]. However, there are many distinctions between types of planer carbon. Single-layer graphite, known as graphene, is estimated to have a thermal conductivity over 2000 W m⁻¹K⁻¹ [2]. Unfortunately, the production of graphene is not currently commercially feasible at large scales. Graphene oxide, a single layer of graphene with oxygen-based surface moieties, can be produced more easily via Hummer's method, which is considered a scalable process [3,4]. However, while graphene oxide can be used to increase composite thermal conductivity, commercially available graphene oxide is still on the order of hundreds of dollars per gram. As a more cost-effective option, multilayer graphene, or graphite nanoplatelets (GNP) have recently become more widely available and have been studied in thermally conductive epoxy composites [5]. However, these particles still cost on the order of a few dollars per gram depending on the quality. As an even lower-cost option, GNPs can be produced from intercalated graphite, which is two orders of magnitude less expensive than GNPs. To form intercalated graphite, graphite flakes are typically treated with sulfuric acid. During heating the intercalation compound breaks down into gas, which creates pressure and increases the interlayer spacing of the graphite by ∼300×, resulting in expanded graphite (EG). These EG particles can then be separated into GNPs through shear forces or sonication. The impact of these particles on thermal conductivity in epoxy was initially explored by Debelak and Lafdi [6]. It was later shown that the temperature of thermal shock controls the degree of expansion, which impacts the final thermal conductivity [7,8]. In general, exfoliated EG outperforms graphite in regard to thermal conductivity, as well as other forms of carbon, such as carbon black, multiwall carbon nanotubes, and pitch-based carbon fiber [9]. In order to increase thermal conductivity further, authors have used functionalization, coupling agents, and numerous other further processing techniques [10–14]. However, increasing processing steps and time increases the cost of the composite. For all of the epoxy composites referenced (excluding Ganguli et al. where micro/nanoscale particles were purchased [10]), expanded graphite is exfoliated into nanoparticles using sonication for 0.5–24 h in solvent. The solvent is either filtered [14] and the particles are added to the epoxy as a powder, or the dispersion is mixed into the epoxy resin followed by solvent evaporation [6,11,15,16]. The only exception is Kim et al. who sonicated in epoxy resin for 30 min, though their particles were expanded using ICP in lieu of a furnace [8]. While EG has been previously exfoliated into nanoparticles from macroscale particles in epoxy using only shear forces [17], the exfoliation method presented here is much more rapid, and we have found no literature reports that also measured the thermal conductivity.

In this work, a GNP composite was formed by the exfoliation of EG within an epoxy matrix using only shear forces applied via a FlackTek SpeedMixer^®. The resulting composites are compared to composites manufactured with sonicated particles, which is the typical method for the production of GNPs from EG. The thermal conductivity of the composites was measured, and they were further characterized using field emission scanning electron microscopy (SEM) and Raman spectroscopy. The resulting larger linear increase in thermal conductivity for the mix-only composites indicates that this facile, low-cost, and rapid fabrication method can be used to produce inexpensive polymer composites with highly tunable thermal properties.

The epoxy system used for the composites was Hexion EPON^™ 815c resin and Hexion EPIKURE^™ 3274 curing agent, mixed at a 10:4 ratio. Expandable graphite, intercalated with sulfuric acid, +50 mesh particle size, was purchased from Sigma Aldrich. The mixing was performed by a benchtop FlackTek DAC 150.1 FVZ-K SpeedMixer using batches smaller than 50 mL.

The processing methodology is shown in Fig. 1. The intercalated graphite was expanded by placing it in a furnace at 800 °C for 90 s. The EG is formed by the intercalated compound decomposing, producing pressure between the carbon layers. This increases the interlayer spacing, making it easier to separate the material into nanosized particles. For the mix-only composites, the EG was placed directly into a mixing cup with the entirety of the epoxy resin without further modification. This was done in two or three steps interspersed by mixing for 1 min at 3500 rpm at high loadings (15–25 wt.%) to create room for more EG. For the sonicated particles, the EG was mixed with ethanol in a glass beaker and sonicated for 5 h. Sonication is the standard method for separating expanded graphite into smaller nanoparticles. Therefore, this step was included as the control for comparison to the mix-only condition. After sonication, the dispersion was vacuum filtered to remove the ethanol and dried at 60 °C overnight before the sonicated particles were added to the epoxy for mixing. For both the sonicated and mix-only conditions, unless specified, the resin and particles were mixed for 2 min at 3500 rpm followed by 10 min at 2250 rpm. The curing agent was then added, followed by the same mixing procedure. An identical mix time was used for both the sonicated and mix-only composites to control for impacts of mixing on dispersion. However, it is expected that for the mix-only condition, the shear mixer did the work of separating the particles. For all composites, depending on the viscosity, the uncured epoxy was poured or scooped into 2 flat bottom cylindrical silicone molds (31.75 mm diameter × 10 mm) and cured at 60 °C for 40 h.

Microscopy images of the samples were taken using a Zeiss field emission scanning electron microscope. The 6 wt.% mix-only sample was fractured in liquid nitrogen and mounted on an aluminum stud using carbon tape. The sample was coated with 5 nm of gold before viewing. Thermal conductivity measurements were taken using a Hot Disk TPS 500 Thermal Constants Analyser (Fig. 2). During the test, the Kapton-encased sensor is sandwiched between two samples where it measures thermal resistance. This measurement is used to find thermal conductivity through an automated curve fitting analysis. Raman data was collected for the expanded graphite, 6 wt.% mix-only sample, and the 6 wt.% sonicate-and-mix sample using a ReniShaw inVia reflex system. During the scans, a 50× microscope lens was used with a 633 nm laser, and the data were fit using the peak analyzer tool in Origin 2019.

While mixing the EG composites in the mix-only condition, a distinct drop in viscosity was noted that appears to be associated with the exfoliation of the particles in the lower filler loadings. Above 8 wt.%, the viscosity of the composites was too high to visually observe this change. It was also observed that large, millimeter length scale particles were no longer visible, and the resulting composites became smoother. In the SEM images of the 6 wt.% composite, the GNP particles can be visualized (Fig. 3). Few-layer flakes can be seen imbedded in epoxy with thicknesses well into the nanometer range (Figs. 3(a) and 3(b). However, larger particles were also observed (Fig. 3(c)). These particles have the typical stacked accordion look of expanded graphite. However, they are still much smaller than the millimeter scale particles initially added to the composite. Overall, the SEM suggests that GNPs have been formed, but every particle is not exfoliated to the same degree.

From the thermal conductivity data, it can be seen that the mix-only EG particles performed as well as or better than the sonicate-and-mix particles for nearly all loadings (Fig. 4). For the mix-only EG composites, the data has been fit using linear regression with a fixed intercept at 0.19 W m⁻¹K⁻¹, which is the conductivity of the unmodified epoxy. This analysis produced a calculated slope of 0.155 ± 0.005 W m⁻¹K⁻¹/wt. %. As seen in Figs. 4(a) and 4(b), the data follows this line closely up to 6 wt.%, excluding the 3 wt.% sample. For example, the 1 wt.% sample showed a 100% increase in conductivity and even the 0.1 wt.% sample demonstrated a 15.8% increase. It is likely that the 3 wt.% sample is underperforming due to a processing error. For example, it is possible that this sample trapped excessive air during molding, or the particles used were not fully expanded. After 6 wt.%, the samples show a dip below the 95% confidence interval and appear to return at 20 wt.%. While this may look like percolation behavior, the less dramatic increase between 20 and 25 wt.%, reaching a maximum conductivity of 4.3 W m⁻¹K⁻¹, seems to confirm that the trend is linear and that the 8–15 wt. % samples are underperforming like the 3 wt.% sample. This may represent a processing transition: After 15 wt.%, the viscosity is so high the sample could be pressed into the molds like clay, potentially removing air bubbles. On the other hand, while the sonicated particles are initially able to match (and in one case surpass) the mix-only composites, this only occurs below 1 wt. %. After 1 wt.%, the sonicated particles do not appear to perform as well as the mix-only composites. This is surprising because the sonicated particles receive the same mix procedure as the mix-only composites, so one would expect the two techniques to work in tandem. While more data and analysis are needed to make the claim that sonication is deleterious to thermal conductivity here, it can be said for certain that the sonication step is not necessary for achieving the thermal conductivities seen.

A further comparison of the mixing methodology was undertaken with the 6 wt.% sample (Fig. 4(c)). As summarized in Table 1, one sample was mixed for 4 min at 3500 rpm: 2 min in the resin; 2 min after the addition of the curing agent. The other sample was mixed for 20 min at 2250 rpm split in the same fashion. Compared to the procedure described in the methods, it appears that the higher speed is more effective at separating and dispersing the nanoparticles; the high-speed sample achieved a conductivity of 1.15 W m⁻¹K⁻¹, which is 96% that of the full mix sample. Meanwhile, the low-speed sample performed worst of all at 0.93 W m⁻¹K⁻¹, which is 78% of the full-mix. However, it was also observed that the high-speed low-time sample trapped more air and appeared less homogenous than the sample mixed for the full procedure indicating that the extra mixing time does provide a benefit.

Figures 4(a) and 4(b) are replotted in Figs. 5(a) and 5(b) together with select thermal composites from the literature for comparison. Vol.% data sets were converted to wt.% when necessary using approximate densities of 1.2 g/ml and 2.26 g/ml for epoxy and filler respectively. In general, the literature demonstrates two types of behavior. Frequently, the particles appear to percolate, with relatively low thermal conductivities at low loadings, followed by a period of increased slope at higher loadings [10,11,14]. However, other reports do show linear enhancement at low loadings [7,15,16]. Interestingly, when Debelak and Lafdi varied particle size, they saw percolating curves for 100 and 150 mesh particles, but their 50 mesh particles appeared to trend closer to linear, similar to the particles presented here [6]. However, dissimilarly, there was still nearly no conductivity enhancement below 4 wt.% for their samples in any size regime. It is therefore all the more interesting that our 50 mesh particles show a strongly linear behavior at low loadings.

The 6 wt.% composites, both from the mix-only and sonicate-and-mix conditions, as well as the unmodified expanded graphite, were analyzed using Raman spectroscopy to reveal the degree of exfoliation and order in the graphite (Fig. 6). The largest peaks were observed at ∼1585 cm⁻¹ corresponding to the G band. A second notable peak was observed at ∼2675 cm⁻¹ corresponding to the two-dimensional (2D) graphite peak. Smaller peaks were observed at ∼1330 cm⁻¹ and ∼1650 cm⁻¹ corresponding to the D band and the defect band (D′), respectively. The most obvious changes between the EG and the composites resulting from exfoliation are: (1) the increase in the D peak relative to the G peak; (2) the appearance of the D′ peak convoluted within the right shoulder of the G peak; and (3) the broadening and flattening of the 2D peak. In general, all these changes are indicative of the appearance of defects, though it is apparent that the sonicated composite has a much higher I_D/I_G ratio as well as a more pronounced D′ peak compared to the mix-only composite. This can be interpreted in two ways: Sonication has been shown to cause defects in the basal plane that can be detected by Raman, while shear mixing does not cause significant basal defects [18]; On the other hand, increases in the ratio of I_D/I_G have also been inversely associated with the size of the crystalline grains, with the laser spot capturing more edge defects when the number of smaller particles increases [19,20]. While the increase in defect peaks and broadening of the 2D peak compared to the EG is an indication of exfoliation for both composites, the difference between conditions is either due to further decreases in particle size in the sonication condition, sonication induced basal defects, or some combination of the two. While the sonication may have resulted in more exfoliation of the particles, it appears that these particles were less optimal for thermal conductivity. One possibility is that the particles may have been thinner, but with a smaller aspect ratio. It is also possible that the average lateral particle size was smaller than the mean free path of the phonon, leading to suboptimal heat conduction [21].

A high-shear mixer was used to exfoliate expanded graphite within an epoxy matrix resulting in a nearly linear thermal conductivity improvement up to 25 wt.% filler loading. While it is possible to achieve higher conductivities, in some cases, through careful exfoliation, or tedious functionalization, our methodology results in significant increases in thermal conductivity relative to time and cost. While our standardized mix time was 24 min total, it was shown that 4 min at 3500 rpm was sufficient to achieve a similar thermal conductivity for the 6 wt.% composite. SEM and Raman spectroscopy showed that a large size distribution of flakes was achieved down to the nanometer scale. This large distribution is the likely cause behind the linear increase in conductivity, which is vital to finely tuning thermal properties for specific applications. While sonication may be more effective for exfoliation, that does not necessarily correlate with thermal conductivity enhancement. Additionally, the mix-only method has the added benefit of being much faster and avoids the use of solvents altogether. To the best of our knowledge, no literature reports exist that measure the thermal conductivity of epoxy composites formed through only shear force exfoliation of macroscale particles. Furthermore, the speed at which exfoliation occurs is unprecedented. Given the availability of mixers, this methodology can be scaled up and the mechanical basis of exfoliation opens the possibility for use with other thermosetting polymer systems.

• Office of Naval Research (Award No. N0001418C2063; Funder ID: 10.13039/100000006).