Composites of epoxy and graphene-related materials: Nanostructure characterization and release quantification

Due to their good electrical and mechanical properties, composites of epoxy and graphene-related materials (GRMs) are applied in sports equipment or gasoline containers. During the use phase, wear and tear may induce the release of composite fragments, which might have different effects on human health than pristine GRMs that have shown to increase inflammatory response when they are inhaled into the trachea bronchus and the alveoli of the respiratory system. Our study aims to determine the fraction of released (protruding and freestanding) GRMs on abraded composite fragments using qualitative, semi quantitative and quantitative methods and to evaluate how the GRM size, the inter-layer strength between graphene sheets and the interaction between the GRM and the epoxy affect the GRM release. Using lead as a marker of GRMs, the release of lead-labeled GRMs was qualitatively evaluated via energy- dispersive X-ray-scanning electron microscopy (EDX-SEM) mapping. Moreover, semi-quantitative evaluation of the GRM surface coverage via Raman spectroscopy mapping and quantitative assessment of the released GRM fraction via inductively coupled plasma-optical emission spectrometry (ICP-OES) showed that the release of GRM depended strongly on the GRM size, the inter-layer strength between graphene sheets and the interaction be- tween the GRM and the epoxy. The detected GRM surface coverage values were comparable to the release fractions determined using ICP-OES. This implies that Raman spectroscopy mapping can be used as a first appraisal to determine the fraction of released GRM, which would accelerate the exposure and risk assessment of GRM-enabled composites.


Introduction
The properties of daily life products such as cosmetics, cement or sports equipment, can be enhanced by the addition of nanomaterials (NMs) (Nowack and Bucheli, 2007;Le et al., 2014;Carolina et al., 2018;Potts et al., 2011;Breuer and Sundararaj, 2004;Samorì et al., 2015). As reported by Mittal et al. (2020) graphene-related materials (GRMs) such as graphene oxide (GO) and graphene nano platelets (GNP), consisting of multiple layers of GO or graphene, are considered to be the super material of future. For single layer graphene extraordinary material properties were reported: for electrical conductivity 10 8 S m − 1 , for thermal conductivity 5000 W m − 1 K − 1 and for the Young's modulus of 1000 GPa (Valorosi et al., 2020;Yang et al., 2020). To profit from these properties, GRMs are widely used in sensors and composites (Kong et al., 2019). As reported by Kong et al. (2019) the GRM production volume increased to about 1200 tons in the year 2019. Polymer composites such as GRM-reinforced epoxy nanocomposite, excel by their mechanical strength (), their high electrical conductivity (Wajid et al., 2013;Netkueakul et al., 2020;Kim et al., 2010;Wang et al., 2015;Chandrasekaran et al., 2013), their thermal conductivity (Chatterjee et al., 2012;Shiu and Tsai, 2014) and the fact that they are extraordinarily lightweight (Carolina et al., 2018;Breuer and Sundararaj, 2004;Anwar et al., 2016;Netkueakul et al., 2020;Wang et al., 2015). Due to the superior physicochemical characteristics, such composites are applied in containers for gasoline, tennis rackets, bicycles, helmets, shoes, or ski equipment (Valorosi et al., 2020;Kong et al., 2019;Froggett et al., 2014;Mittal, 2014).
During all life cycle stages of a composite from production, use phase to end-of-life treatment, nanomaterials (NMs) or composite fragments with associated NMs might be released from the composites. NM release has been studied for several composites and NMs (e.g. carbon nanotubes (CNTs), silica, Ag, iron oxide GRM, etc.) (Netkueakul et al., 2020;Froggett et al., 2014;Duncan, 2015;Doudrick et al., 2015;Schlagenhauf et al., 2015a;Nowack et al., 2016). Results from these studies showed that all three aspects: NM, matrix and applied process influence the NM release. Based on Froggett et al. 2014(Froggett et al., 2014 and Nowack et al. (2016), released NMs often appeared to have different physical and chemical properties than the pristine NMs. For example, incineration at the end-of-life is a strongly transforming process and leads to the release of NMs with transformed size, shape or chemical composition. Wear and tear under normal conditions (e.g. rotary abrasion) or mechanical treatment (e.g. sawing, drilling, etc.) during the product use may induce the release of freestanding NMs, composite fragments with embedded or protruding NMs or matrix fragments into the air (Netkueakul et al., 2020;Faghihi and Morawska, 2016). Workers or consumers might be exposed and adversely affected by inhaling these particles.
The toxic effects of pristine NMs have been extensively studied for more than a decade. Due to a growing number of applications of NMs in daily products as well as in advanced materials in recent years, the focus has been extended towards the investigation of the released NMs, since their morphology and therefore their impact on human health might be different compared to the pristine NMs (Netkueakul et al., 2020;Froggett et al., 2014;Schlagenhauf et al., 2015a;Schlagenhauf et al., 2015b). As described by Froggett et al. (2014) and Nowack et al. (2016), not only released NMs, but also released matrix fragments in the nanometer size range can induce sever health effects. Moreover, the actual toxic effects of a NM or a composite fragment depend on the NM concentration, the exposure duration and the physical and chemical properties of the NM (morphology and elemental composition). To understand the toxicological implications, a complete physical and chemical characterization of the released particles is required.
Abrasion of GRM-epoxy composites leads to aerosolization of the released composite fragments, with partly embedded/ protruding and freestanding GRMs (Netkueakul et al., 2020). In vivo and in vitro toxicity studies on pristine GRMs showed that these particles were able to penetrate into trachea bronchus and the alveoli of the respiratory system of the lung (Su et al., 2016;Singh, 2016;Schinwald et al., 2012;Mendes et al., 2017) and that they could enhance the inflammatory response (Netkueakul et al., 2020;Faghihi and Morawska, 2016;Schinwald et al., 2012;Wick et al., 2014;Lalwani et al., 2016;Ou et al., 2016;Seabra et al., 2014). Liu et al. 2013 showed that GO might induce in vivo and in vitro mutagenesis (Liu et al., 2013). Ou et al. (2016) found that GRMs were toxic, because macrophage cells were not able to take up the GRMs and that the GRMs could even cut cell membranes. Several research studies proposed the relationship between the toxicity of GRMs and their physicochemical properties. For example, Mittal et al. 2016 claimed that the oxidative stress in human lung cells might be related to the physicochemical properties of GO (Mittal et al., 2016). Moreover, Akhavan et al. (2012) and Yue et al. (2012) hypothesized that the lateral dimension of the GNPs might define the toxicity of the GNPs on human mesenchymal stem cells (Akhavan et al., 2012), macrophages (murine macrophage J774A.1 cell line and peritoneal macrophage PMØ) and non-phagocytic cells (human breast cancer MCF-7, human umbilicalvein endothelial cells HUVEC, murine Lewis lung carcinoma LLC and human hepatocarcinoma cells HepG2) (Yue et al., 2012), respectively.
In contrast, there are also studies claiming that the toxic effects of GRMs are not strong. Di Cristo et al. (2020) reported that repeated exposure of human bronchial epithelial Beas-2B cells to aerosolized GO at realistic levels of 21 μg/cm 2 was not strong enough for activation of the inflammatory cascade. Drasler et al. (2018), found that single exposure to aerosolized GO and GNP did not induce any acute biological responses on dendritic cells, monocyte-derived macrophages and human lung carcinoma cell line A549. Kim et al. (2018), who examined a short term in-vivo exposure study on rats with GRM, reported no significant toxic effect induced by GRM exposure after an exposure duration of 5 days and an additional recovery period (1,3 and 21 days). Moreover, Shin et al. (2015) performed in-vivo studies on rats, and found only a minimal toxic effect for 5-day exposure. Based on our previous study Netkueakul et al. (2020), abraded fragments of GRM-epoxy composites do not show any acute cytotoxic effects on human macrophages differentiated from THP-1 cell line. In contrast to pristine GRMs, the abraded fragments have less amount of sharp edges because the GRMs are partly embedded into composite fragments or associated with the polymer matrix. This reduces the ability to cut the cell membrane. However, the fraction of released (protruding and freestanding) GRMs on the surface of the released fragments was significantly high and strongly dependent on the GRM.
Since clear distinction of carbonaceous NMs from carbonaceous matrices in abraded fragments and therefore quantitative determination of the released NM fraction are very challenging, several approaches were developed (Doudrick et al., 2015;Schlagenhauf et al., 2015a;Schlagenhauf et al., 2015b;Rhiem et al., 2016;Schlagenhauf et al., 2012;Hagendorfer et al., 2010;Mantecca et al., 2017). Rhiem et al. (2016), employing the radioactive approach, substituted the carbon atom of CNTs with 14 C and determined the release of 14 C. In contrast, Schlagenhauf et al. (2015a) applied an indirect lead-labeling-andrelease method to detect the fraction of released (freestanding and protruding) CNTs in the abraded particles. In our previous study (Netkueakul et al., 2020), we recently quantified the release of GRMs from GRM-epoxy-composites and evaluated the toxicity of the released composite fragments. As we postulated in Netkueakul et al. (2020), the physical and chemical characteristics of the embedded GRMs could affect the fraction of released GRMs.
In this study, we investigated specifically the influence of the GRM size on the released GRM fraction on abraded composite fragments from rotary abrasion, which is a widely used process to simulate particle release during the use phase since the applied forces under the used configuration are controlled.
To exclude effects coming from differences in the chemical composition, we intentionally used GNP-M5 and GNP-M25 with the same chemical composition from the same manufacturer (XG Sciences, MI, USA) but different diameters of 5 μm and 25 μm. In addition, we assessed how the inter-layer strength between graphene sheets and the bonding strength of the GRM to the epoxy affect the release. To demonstrate the effects, we compared the release of GNP to the release of GO. We used inductively coupled plasma optical spectroscopy (ICP-OES) for the quantitative determination of the released GRM fraction. Since an easy-yet-reliable quantification would be advantageous for a faster determination of the released GRM fraction (Netkueakul et al., 2020), we employed also other analytical techniques such as energydispersive X-ray-scanning electron microscopy (EDX-SEM), wavelength dispersive X-ray fluorescence (WD-XRF) and Raman spectroscopy to assess the release of GRM. We further developed the Raman approach from our previous study (Netkueakul et al., 2020) so that it can be applied as a semi-quantitative method. Since a uniform NM distribution in a composite is important for a representative NM amount in the abraded powders (Schlagenhauf et al., 2014), we determined the GRM distribution in the composite by measuring the refractive index and its variation at multiple positions of a composite by using terahertz spectroscopy (Zolliker et al., 2017;Tonouchi, 2007;Scheller, 2014) prior to the assessment of the GRM release.

Physical and chemical characterization of pristine GRMs
We employed scanning electron microscope (SEM) (Nova NanoSEM 230, OR, USA) to investigate the morphology of the pristine GRMs (GO, GNP-M5 and GNP-M25). To determine the thickness of the GRMs, we performed atomic force microscopy (AFM) analyses using the Solver Nano AFM (NT-MTD Spectrum Instruments, Russia). To analyze the ratio of carbon to oxygen (C/O ratio) on the GRM surface, we used X-ray photoelectron spectroscopy (XPS) similar to Kucki et al. (2016) and Netkueakul et al. (2020) by using the XPS Microprobe system (PHI VersaProbe II spectrometer, Physical Electronics, Germany). Details can be referred to our previous study Netkueakul et al. (2020). The characteristics of the pristine GRM were investigated via transmission electron microscopy (TEM) using a TEM FEI Tecnai G2 F20 (FEI Company, OR, USA).

Lead-labeling of GRMs
Following the metal-ion-labeling approach invented by Schlagenhauf et al., (2015a), GRM powders (GNP-M25 with diameter of 25 μm and GNP-M5 with diameter of 5 μm) (XG Sciences, MI, USA) (Sciences, n.d.) and GO (Cheaptubes, VT, USA) were labeled with lead. For lead ion labeling, 1.5 g of the GRMs were immersed into 400 ml of an aqueous 200 mg/l Pb(II)acetate trihydrate solution. The pH value of the coating solution was 5.92, which facilitated the Pb 2+ ion attachment onto the GRM surface (Schlagenhauf et al., 2015a;Peng et al., 2017;Gu and Fein, 2015). The mixtures of the GRMs and the coating solution were sonicated for 30 min in an ultrasonic bath Bandelin Sonorex RK 100H (Bandelin, Germany) with an ultrasonic nominal power of 80 W (Esmaeili and Entezari, 2014). Bath sonication for 30 min has shown to be sufficient for the lead-labeling of the different GRMs, since an EDX-SEM assessment of the lead concentration on the GRMs for durations of 5, 10, 20, 30, 40, 50 and 60 min showed constant lead levels at a duration of 20 min for all the GRM used. Then the mixtures of the coating solution and the GRMs were stirred for 2.5 h. GRM particles were removed from the remaining filtrate using a Nuclepore filter (Whatman, UK, pore size 0.2 μm), dried for 14 h at 50 mbar and 50 • C.
These batches of lead-labeled GRMs were analyzed using EDX-SEM and later used to produce the composites for the analysis of GRM release via ICP-OES.

Manufacturing of the GRM-epoxy-composites
Adapting the manufacturing method of Schlagenhauf et al. (2015a) mass loadings of 0.1 wt% to 1.0 wt% of lead-labeled GRMs or pristine GRMs were dispersed in epoxy resin Araldite GY-250 (Huntsman, TX, USA) using a mixer and threefold milling with a three-roll mill (Bühler AG, Switzerland). For crosslinking hardener, Jeffamine D-230 (Huntsman,TX, USA) was added. After curing for 12 h at 80 • C, the composite was post cured at for 4 h at 120 • C.

Characterization of GRM embedded composite via terahertz
In THz spectroscopy, a linearly polarized short pulse (half width below 1 ps) was passed through a bulk sample. Spectral optical properties could be derived from the delay and change in the pulse shape. Spectral information was extracted by analyzing the Fourier transform of the pulse. For every measurement set, an additional reference pulse without a sample in the THz beam path was measured. Taking the ratio of the target spectrum and reference spectrum allows compensating influences from the absorption of water and other air molecules in the beam path. Knowing the sample thickness, one could extract a frequency dependent refractive index and extinction coefficient from the measured THz spectrum using the so-called transfer function (Zolliker et al., 2017;Scheller, 2014). The optical properties of GRM-epoxycomposite samples were measured using a THz time-domain spectrometer TeraFlash (Toptica, Germany) in a transmission configuration. The sample could be scanned manually with an x-y-stage. The THz beam was focused to a spot with a width at half maximum of about 2 mm. The polarization dependency was measured using two wire grid polarizers one before and one after the sample. The optical properties were determined for samples with various GNP-M25 contents (0, 0.1 wt%, 1.0 wt%, and 0.1 wt% with lead). A GNP detection limit of 0.05 wt% was calculated. For the 1.0 wt% sample, local variations of the optical properties were also determined by scanning a 5 mm × 10 mm area on a 1 mm × 2 mm grid using three polarizer directions − 45, 0, 45 degrees. The effect of rotation was tested via rotating the sample from 0 • to 360 • in steps of 30 • .

Production and characterization of airborne abraded composite powders
The composites were abraded using the standard instrument to test wear and tear, a Rotary Taber Abrader (Model 5135) (Taber Industries, NY, USA), with a CS-0 abrading wheel with attached S-42 sandpaper strip, and an extra weight of 1 kg. The abrasion was performed for 60,000 cycles at a rotational speed of 60 rpm. The sample dimensions were 10 cm × 10 cm × 0.4 cm (width × length × height). The abraded particles were sampled through a slit inlet (20 mm × 2 mm) with a sampling flow of 11 lpm and collected on Nuclepore filters with a pore size of 0.2 μm for further analyses.
Released aerosol size distributions were measured by aerodynamic particle sizer (APS 3321, TSI, Germany) and scanning mobility particle sizer (SMPS 3081, TSI, Germany) and fitted using the software Origin 2018 (OriginLab Corporation, Northampton, MA, USA) using lognormal fit functions. Control experiments were performed to evaluate the influence of the slit inlet on the sampling of the measured particle size distributions using aerosolized polystyrene latex (PSL) particles similar as in a previous study (Netkueakul et al., 2020). Details are given in the supplementary information. The morphology of the abraded powders was investigated using SEM (Nova NanoSEM 230, OR, USA). To evaluate the characteristic of the GNP in the abraded composite powder, the abraded Epoxy/1.0 wt% GNP-M25 composite powder was analyzed via TEM (TEM FEI Tecnai G2 F20 (FEI Company, OR, USA)).

EDX -SEM and WD -XRF
Chemical elements on the surface of lead-labeled GRMs and abraded composite particles were determined with scanning areas of 500 μm × 500 μm via EDX-SEM using SEM Phenom ProX (Phenom, Germany) with an electron beam acceleration voltage of 15 kV. Since the X-rays had only a short penetration depth of several nanometers into the material, lead could only be detected for released lead-labeled GRMs on the surface of composite particles. Composite plates and composite powders without lead-labeled GRMs were investigated as negative controls. A detection limit of 0.002 wt% lead was calculated. In addition, the lead concentration on the surface of lead-labeled GNP-M5 and GNP-M25 particles was determined by scanning a circular area with a diameter of 23 mm using WD-XRF Rigaku Primus IV (Rigaku, Switzerland). Under the used configuration, a detection limit of 0.5 wt% was calculated.

Raman spectroscopy
Raman spectroscopy measurements were performed on a WITec alpha 300R confocal Raman microscope (AGWITec Wissenschaftliche Instrumente und Technologie GmbH, Germany), equipped with a UHTS 300 Vis Spectrometer and a thermoelectrically cooled CCD. For excitation, a linearly polarized 532 nm wavelength laser was used. To focus the laser beam on the surface of the particles, a 50× objective lens (NA = 0.55) was used. Due to the confocality of the Raman microscope the scattered light is probed only from a limited volume (<1um in Zdirection). An area of 28 μm × 28 μm (84 × 84 points) was scanned with a set laser power of 0.25 mW and an integration time of 3 s per point. Raman spectra of the reference materials including pristine GRMs and abraded epoxy powder were obtained with 0.5 mW laser power (pristine GRMs) and 5 mW laser power (abraded epoxy powder) and with an integration time of 10 s. The G band centered at 1600 cm − 1 was chosen as a representative peak for all GRMs, while the band centered at 3065 cm − 1 was chosen as a representative peak for epoxy. The intensity ratio maps were plotted by taking the ratio of the intensity of the G band to the intensity of the epoxy band (I G /I e ). Areas in the intensity maps that were not in focus were excluded from further assessment.

ICP -OES
The released GRM fraction was determined by measuring the mass of detached lead released as freestanding and protruding lead-labeled GRMs on abraded composite particles using the ICP-OES device, Vistra-PRO (Varian Inc., Palo Alto, CA, USA). First, the release capacity of lead-labeled GO, GNP-M25 and GNP-M5 was determined. 0.3 mg to 2.6 mg of lead-labeled GRMs were immersed into 5 ml of 0.1 M HNO 3 . The dispersion was vortexed and ultra-sonicated for 5 min. Particles were separated from the filtrate by centrifugation at 3000 rpm for 10 min using Amicon Ultra-4 30 kDa filters (Merck Millipore, Germany). SEM images of the filtrate did not show any remaining particles. The lead concentration in the clear filtrate was measured via ICP-OES and utilized to determine the lead release capacity. Lead plasma standard solution, Specpure® Art. 13,853 (Alfa Aesar, Thermo fisher scientific, Germany), a certified standard reference material with a known concentration of 1 g/l, causing characteristic Pb-spectra, was used as quality control. The recovery was 90-110% with a measurement uncertainty of 10%, which was considered to be very good for such measurements. The lead concentrations were determined using the axial Pb-line at 220.353 nm. The limit of detection (LOD) of the ICP-OES instrument was 0.1 mg lead per L.
In order to assess the GRM release in the abraded composite powders, the lead ions released from the abraded lead-labeled GRM/epoxy composite powder was measured. 100 mg of the abraded composite powder was weighed and immersed into 5 ml of 0.1 M HNO 3 and treated in the same way as described above. The mass of the released lead ions from protruding and freestanding lead-labeled GRMs on the surface of the abraded composite particles was measured and the released GRM amount was derived using the lead-release-capacities of each GRMs. The released GRM fraction could be determined by dividing the released GRM amount by the total GRM amount in the abraded composite powder (eq. 1).
Moreover, two control experiments were carried out. The purpose of control 1 was to detect potential Pb 2+ ion leaching from GRM into the epoxy resin. Lead-coated GRMs were immersed into epoxy and sonicated for 1 h. Then the mixture was mixed in acetone to dissolve epoxy resin. Using centrifugal filtration, the GRMs were separated from the remaining filtrate, which then was analyzed via ICP-OES. The purpose of control 2 was to understand how much lead ions could be exposed by the abrasion process under the worst case scenario that all loaded lead ions would detach from GO, which was the GRM with the highest release capacity among all GRMs considered in this study. For control 2, the coating solution (Pb(II)acetate trihydrate) was directly added to neat epoxy resin without any GRMs before curing the sample similar to the other composites. After curing, control 2 was abraded using the Taber abrader as described earlier. The lead amount of the abraded particles was analyzed by ICP-OES as described above. The outcome of control 2 was the potential highest amount of lead ions released from epoxy matrix by abrasion. This value was considered as an upper limit of the uncertainty range, while the lower limit was 0. This range contributed to the final uncertainty of the released lead ions and thus GRM release fraction.

Characterization of GRM embedded composites by terahertz spectroscopy
We employed terahertz spectroscopy to investigate the distribution of GRMs in the bulk composites. Due to its large size, GNP-M25 was assumed to have the highest probability for non-uniform dispersion in the epoxy among all investigated GRMs. Hence, local variations and anisotropic optical properties were determined for the sample Epoxy/ 1.0 wt% GNP-M25, giving insight into local variations of composite composition and orientation of the graphene plates in the composite. A set of samples (Epoxy, Epoxy/0.1 wt% GNP-M25, Epoxy/0.1 wt% GNP-M25-Pb, Epoxy/1.0 wt% GNP-M25) were measured (Table S2). Fig. 1a shows that the pulse delay time as well as the attenuation of the THz pulse increased with increasing amount of GNP in the composite. Fig. 1b gives the corresponding spectra, from which the refractive indices ( Fig. 1c) and extinction coefficients (Fig. 1d) were extracted. The GNP content and the optical properties were highly correlated, indicating that THz spectroscopy was very suitable for measuring relative changes in the GNP content. The specimen with the highest GNP content (1.0 wt % GNP-M25) showed a pronounced increase in the refractive index and also a higher extinction coefficient. No significant information related to additional effects on the THz spectrum, and respectively the refractive index, could be extracted concerning the additional lead content of the Epoxy/0.1 wt% GNP-M25-Pb compared to the Epoxy/0.1 wt% GNP-M25.
Regarding the isotropy and homogeneity of the Epoxy/1.0 wt% GNP-M25 sample, we scanned the sample on an area of 5 mm × 10 mm. Depending on the sample orientation (Fig. 1e), the THz pulse showed a distinct polarization dependency of the time delay of the detected THz beam. This result revealed a birefringence Δn of about 0.02 (which is the difference between the minimal and maximal refractive indices; about 1.84 and 1.82 in Fig. 1f). Δn is defined as the difference of the ordinary and the extraordinary refractive indices for anisotropic materials. The variation of Δn can be explained by a preferred orientation of the GNPs within the sample. Apart from this orientation dependent variation of the refractive index, the sample showed no significant spatial variation of the refractive index, meaning a homogeneous distribution of the GNPs in the sample. There was also only a small spatial variation of the main direction of the birefringence (Fig. 1h). In fact, the local variations of the refractive index were smaller than the birefringence caused by anisotropy (Fig. 1g). Therefore, the results from THz measurement revealed that GNP was homogeneously dispersed in the epoxy composites.

Characterization of pristine GRMs and lead-labeled GRMs
Size and shape of the pristine GRM powders were analyzed using SEM. The analyses showed slightly bigger platelet diameters than the values from the manufacturer (Table S1 and Fig. S2). Based on Fig. S2 all particles possessed a platelet like shape. The analyses of the GRM thickness via AFM showed that the GNP consisted of multiple graphene layers with a total thickness up to 239 ± 174 nm (Table S1). Chemical elements on the surface of lead-labeled GRMs were investigated by EDX-SEM and WD-XRF. EDX-mapping detected carbon, oxygen and lead on lead-labeled GNP-M25 (Table S4) with lead-concentration of 8.07 wt% (Fig. 4a, Table S4). In contrast, WD-XRF detected lead-concentrations of 1.25 ± 0.43 wt% for lead-labeled GNP-M25 and 1.67 ± 0.47 wt% for lead-labeled GNP-M5 in threefold measurements (Table S3). Due to strong dependence on the surface roughness and the measured sample surface area, the lead-concentrations from EDX-SEM and WD-XRF could not be compared.

Characterization of abraded composite particles
The aerosolized abraded particles were measured with SMPS and APS and the measured particles size distributions (PSDs) were fitted using lognormal functions. Fig. 2 depicts the results and Table S6 shows evaluated parameters such as the total concentrations in the SMPS and APS size ranges, aerodynamic diameters (d a ), electrical mobility diameters (d e ) geometric standard deviations σ g and count mean diameters (CMDs). Total concentrations in the SMPS range showed values in the range of 5.0 × 10 4 (#/cm 3 ) to about 1 × 10 5 (#/cm 3 ). The particle concentration of the background air was about 1.3 × 10 3 (#/cm 3 ) with a CMD of about 100 nm. PSDs in the APS size range showed particle concentrations of 3 × 10 3 (#/cm 3 ) to 5 × 10 3 (#/cm 3 ) and background concentrations of about 10 (#/cm 3 ) with a CMD of 0.7 μm. All PSDs in the SMPS range showed one mode at about 330 nm independent on the used GRM. In addition, the PSDs of neat epoxy and Epoxy/1.0 wt% GNP-M25 showed two modes in the APS range, whereas the PSDs of Epoxy/1.0 wt% GNP-M5 and Epoxy/1.0 wt % GO showed three modes.
TEM analysis of the abraded Epoxy/1.0 wt% GNP-M25 composite powder were performed to explain the characteristic of the GNP in the abraded composite powder. GNP-M25 showed transparent features with wrinkles and folded layers (Fig. 3 a, b). These features of GNP observed in the abraded composite as marked by the white dotted line were used to identify the protruding part from the epoxy matrix (dark area). The wrinkled and folded structure can be seen in the magnified images ( Fig. 3 d, e).

Qualitative assessment of the released GRM fraction by EDX-SEM.
EDX-mapping of abraded composite particles detected: carbon, oxygen and lead (Fig. 4b). To avoid interference with lead and carbon, Fig. 4 does not show the oxygen signal. We detected a lead concentration of 0.06 wt% (relative to the other chemical elements; carbon and oxygen) for abraded particles from Epoxy/1.0 wt%GNP-M25-Pb (Table S5). No lead could be detected on the lead-free control samples above the detection limit of 0.002 wt%, which thereby implied that the lead detected on the abraded particles came from the released GRMs.

Semi quantitative assessment of the released GRMs by Raman spectroscopy.
We developed a semi quantitative approach to assess the released GRM fraction in abraded particles by checking whether the GRMs were present on the surface of the abraded composite powders by dividing the scanned area of the abraded particles by the area where Raman signal of GRMs prevailed relative to the epoxy signal in the Raman intensity map. Raman spectra of pristine GRMs and abraded neat epoxy particles are displayed in Fig. 5a). The G band centered at 1600 cm − 1 was chosen to represent the GRMs, because this band appeared in all investigated GRMs and its intensity was not affected by the change in GRMs' properties. The band centered at 3065 cm − 1 represented epoxy, since it did not overlap with other GRM-related bands. By taking the intensity ratio of the G band of GRMs to the epoxy band (I G /I e ), we were able to compare the degree of GRMs on the surface of the abraded particles, indicated by the color scale, among three Epoxy/ Fig. 2. Fitted particle size distributions (PSDs) form abraded neat epoxy, Epoxy/1.0 wt% GNP-M5, Epoxy/1.0 wt% GNP-M25, Epoxy/1.0 wt% GO from 3 measurements per sample: a) measured by scanning mobility particle sizer (SMPS, size range 13-573 nm) and b) measured by aerodynamic particle sizer (APS, size range 0.542-20 μm). For comparison, we show data from Epoxy, Epoxy/1.0 wt% GNP-M25 and Epoxy/1.0 wt% GO, that was also used in our previous study (Netkueakul et al. 2020).
GRM composite materials as illustrated in Fig. 5b) -d). The brighter color in the map signifies higher degree of GRM on the surface of the abraded Epoxy/GRM particles.
Spectra obtained from Raman mapping were manually analyzed whether each spectrum contained useful information i.e. epoxy-related signal and/or GRM-related signal or only noise. To identify the particles/ areas that were completely in focus, boundaries of the abraded particles were evaluated by inspecting each spectrum from the Raman intensity maps together with the corresponding optical images as demonstrated in Fig. S4. At the same time, the areas having spectra containing only noise could be matched with the optical image in which the areas that were not covered by the abraded particles or where the particles which were not in focus of the exciting laser (marked in black in Fig. 5 e) -Fig. 5 g)) and therefore were excluded from the calculation of released GRMs in the abraded particles.
Raman spectra with dominant GRM signal were appointed to the spectra with I G /I e higher than a certain threshold value, which was assigned for individual Epoxy/GRM composites by comparing, spectra with different I G /I e ratios as shown in Fig. S5. When I G /I e increased, the intensity of the epoxy band decreased until it was not noticeable relative to the G band of GRM. The point where the epoxy band appeared indistinguishable from the baseline and the G band appeared as a strong peak was determined as the threshold value. The threshold values of abraded particles from Epoxy/1.0 wt% GO, Epoxy/1.0 wt% GNP-M25 and Epoxy/1.0 wt% GNP-M5 were 0.6, 2.5 and 5.5, respectively. By identifying the abraded particles and calculating the surface area-based percentage of released GRMs on the surface of the particles using the Raman intensity maps, we were able to assess the released GRM fraction in abraded particles in a semi quantitative way. Based on this approach 63%, 23% and 8% of the surface of abraded particles were considered as released GO, GNP-M25 and GNP-M5, respectively. However, the quantification of the surface coverage of the released GRMs using Raman spectroscopy mapping is limited due to its dependency on the threshold values, which relies upon researcher's determination that could vary. Therefore, this method can provide an approximate coverage percentage of GRM on the surface of abraded particles and allows the comparison among different types of GRM composites or different samples of the same type of GRM composite.

Quantitative assessment of the released GRM fraction by ICP-OES.
Quantification of the released GRM fraction was achieved by measuring the lead amount released from freestanding and protruding lead-labeled GRMs in the abraded composite powders using ICP-OES.  Prior to the quantification, control experiments were carried out in order to detect potential Pb 2+ ion leaching from the GRMs into the epoxy resin prior to curing (control 1) or during the curing or the abrasion process (control 2). No Pb 2+ ions were detected in case of control 1, indicating that no Pb 2+ ion leached from GRM into the epoxy resin prior to curing. In control 2, Pb 2+ ion release could be induced by the addition of hardener. In order to simulate the worst-case scenario (maximum possible leaching of lead ions from GRMs into the epoxy) we chose the GRM with the highest lead-ion release capacity. Based on the highest lead-release capacity value (Fig. 6a), GO would potentially release the most lead-ions into the epoxy in the worst-case scenario. To mimic the maximum possible Pb 2+ ion release, we mixed 33.81 mg lead(II) acetate trihydrate (source of Pb 2+ ions) with 100 g epoxy, without any GRMs. After addition of hardener and curing, the cured epoxy resin containing Pb 2+ ions was abraded and the abraded particles were collected for ICP-OES analysis in order to determine the amount of Pb 2+ released from cured epoxy resin by the abrasion process. The outcome of control 2 showed the maximum possible leaching of lead ions from the GRMs into the epoxy during the manufacturing process induced by the addition of hardener, then exposed by abrasion. Maximal 9.0 ± 0.7 wt% of the total added Pb 2+ ion amount was released. In the best-case scenario, this would be 0 wt%. To quantify the released GRM fraction in the abraded powders, first lead release capacities (average release mass of Pb 2+ ions per mg GRM) of leadcoated GRMs were determined. The values were 18.47 ± 0.3 μg Pb 2+ per mg GO, 13.88 ± 0.4 μg Pb 2+ per mg GNP-M25 and 15.23 ± 0.5 μg Pb 2+ per mg GNP-M5 (Fig. 6a).
Based on the lead release capacities of each GRMs, the mass of the released GRMs (freestanding and protruding) was calculated. Then the released GRM fractions were determined with respect to the total mass of GRM in the abraded composite powder as shown in Fig. 6b. The released GRM fractions were 92 ± 15 wt% for GO, 52 ± 16 wt% for GNP-M25 and 18 ± 7 wt% for GNP-M5. The results show that the fraction of released GRM was very high and strongly dependent on the size and the surface characteristics (e.g. oxygen content) of the employed GRM. Fig. 5. a) Raman spectra of reference materials including pristine GRMs (GO, GNP-M25 and GNP-M5) and abraded particles of neat epoxy. The gray boxes indicate the G band of GRMs (centered at 1600 cm − 1 ) as a marker of GRMs and the epoxy-related band (centered at 3065 cm − 1 ). Raman intensity ratio maps (b-d) showing the ratio of the intensity of the G band to the intensity of the epoxy band of the abraded particles (I G /I e ) from b) Epoxy/1.0 wt% GO composite, c) Epoxy/1.0 wt% GNP-M25 composite and d) Epoxy/1.0 wt% GNP-M5 composite. Illustration of the GRM dominating area marked in white (e-g), identified by the I G /I e greater than the threshold values, on the surface of the abraded particles marked in gray of the e) Epoxy/1.0 wt% GO composite, f) Epoxy/1.0 wt% GNP-M25 composite and g) Epoxy/ 1.0 wt% GNP-M5 composite. Black area represented the area where the abraded particles were out of focus or the area without abraded particles.

Discussion
We quantified the released GRM amount in abraded powders from GRM-epoxy composites using fast and accurate analytical techniques and investigated how the GRM size affects the released GRM fraction. As we found out, also the inter-layer strength between graphene sheets and the bonding strength to the epoxy played an important role for the released GRM fraction.
Terahertz spectroscopy mapping confirmed uniform GRM distribution in the composites, which indicated that also the abraded composite powders possessed a representative GRM amount. As confirmed by SEM analyses, the released fragments from Epoxy/GO had a similar random (not spherical or platelet-like) shape as those from Epoxy/GNP (Fig. S2). However, there were some differences in the PSDs of the aerosolized abraded composite powders in the lower APS range from 0.5 μm to 2.0 μm (Fig. 2) which implied that the size of the released fragments depended on the size of the GRM used and bonding strength of the GRM to the epoxy matrix.
Released particles from Epoxy/GO were in average smaller than the released particles from GNP-based composites since Epoxy/1.0 wt% GO released significantly more particles at an aerodynamic diameter of 0.832 ± 0.016 μm (Table S6) in the lower APS range. Due to the higher hydrophilicity (lower C/O ratio in Table S1) and the formation of hydrogen-bonds from carboxyl and hydroxyl groups on the GO surface to the epoxy (Gupta et al., 2017;Dai et al., 2015;Park et al., 2009;Konios et al., 2014), GO showed a stronger bonding to epoxy compared to GNP. Moreover, owing to their higher hydrophilicity GO particles showed a better dispersion in epoxy compared to GNP (Netkueakul et al., 2020;Dai et al., 2015;Konios et al., 2014;Park et al., 2017), which resulted in an enhanced interaction of the GO and the epoxy and thereby in a mechanically stronger composite with higher toughness (Park et al., 2017). As found by Chandrasekaran et al. (2013), GO-epoxy composites showed a higher fracture toughness compared to GNP-epoxy composites. Moreover, Zok and Miserez (2007), who compared the abrasion resistance of various materials, concluded that composites with higher toughness release smaller particles. This implied that the smaller size of the fragments from Epoxy/1.0 wt% GO (Fig. 2) originated from the better binding of the GO particles to the epoxy matrix and the better dispersion of the GO particles within the epoxy matrix. The higher fraction of particles with smaller diameter could lead to higher concern in terms of human exposure. However, the potential adverse implication is not discussed further, because the focus of this paper is not on exposure assessment. Since the GNP particles did not form hydrogen bonds and since their dispersion was not as good, the released fragments from GNP-epoxy composites were bigger.
The abraded particles from Epoxy/1.0 wt% GNP-M25 were slightly larger than those from Epoxy/1.0 wt% GNP-M5. The addition of GNP-M25 to epoxy lead to a stronger reinforcement of the mechanical properties of GNP-epoxy composite than the addition of GNP-M5, which had smaller size than GNP-M25. The bigger lateral size of GNP-M25 prevented the composite from breaking, because the GNP-M25 would have had to be pulled out from the epoxy. Moreover, there was a higher chance for larger GNP-M25 particles, to be hit by the cracks formed due to abrasion. Therefore, the addition of GNP-M25 resulted in a stronger reinforcement of the composite and induced the formation of bigger composite fragments during abrasion.
Based on SEM and airborne analyses, the size of the abraded fragments showed a maximal size of aerodynamic diameter of 4.5 μm with modes at maximal 2.3 μm, which was smaller than the diameter of the pristine GRMs. Consequently, the likelihood for protruding GRMs and a high GRM release fraction was very high for all composite powders. Protruding GRMs could be identified in the abraded powder from the Epoxy/1.0 wt% GNP-M25 composite via TEM analysis (Fig. 3), indicating that the majority of released GRMs was in form of protruding GRMs. EDX-SEM, Raman and ICP-OES were employed to assess the released GRM-fraction in abraded composite powders. Before investigating the released GRM-fraction, we evaluated if the lead labeling of the GRMs was successful via EDX-SEM and WD-XRF mapping. Since both techniques detected lead on lead-labeled GRMs, a successful lead-labeling could be confirmed. Considering scattering effects and uncertain penetration depth of the X-rays into the rough sample surfaces, the X-ray based EDX-SEM and WD-XRF approaches did not show high accuracy and could not be applied for a quantitative determination of the released GRM fraction. Due to big difference in the investigated surface area (0.25 mm 2 in case of EDX-SEM and 415.5 mm 2 in case of WD-XRF) and the aforementioned factors, the detected lead-concentrations could not be compared. Nevertheless, EDX-SEM was able to detect lead ions on the surface of abraded powders, which confirmed the release of GRM and implied that EDX-SEM could be qualitatively applied to assess if there was the release of GRM.
Using Raman spectroscopy mapping, the presence of GRMs on the surface of the abraded composite powders could be confirmed without using lead-labeled GRMs. The assessment showed lower values for GNP-M25 and GNP-M5 (23% and 8%) compared to GO, which showed a value of 63%. Due to its higher accuracy, ICP-OES was used to quantify Fig. 6. a) Lead release capacities of lead-labeled GRMs (GNP-M5, GNP-M25 and GO); points represent measured values and lines represent linear fits of the measured lead release capacities, b) Released GRM fraction in abraded particles and standard deviation from threefold measurements (n = 3) from Epoxy/1.0 wt% GO, Epoxy/ 1.0 wt% GNP-M5 and Epoxy/1.0 wt% GNP-M25. For comparison, we show data from Epoxy/1.0 wt% GNP-M25 and Epoxy/1.0 wt% GO, that was also used in our previous study (Netkueakul et al. 2020).
the released GRM fraction in abraded composite powders (with leadlabeled GRMs). Comparing the measured lead-release capacities of GNP-M25 and GNP-M5, no significant influence originating from the GRM size or the surface characteristics (such as a higher oxygen content) could be concluded. However, the detected GRM release fractions showed high values of 92 wt%, 52 wt% and 18 wt% for GO, GNP-M25 and GNP-M5 which strongly depended on the type of GRM used.
The released GRM fractions from ICP-OES showed a similar trend as the semi quantitative results of GRM surface coverage from Raman intensity mapping implying that Raman intensity mapping could be used for a first appraisal to determine the released GRM fraction, which could accelerate the determination procedure. Moreover, since Raman spectroscopy could identify GRM surface coverage on abraded fragments that did not contain lead-labeled GRMs, this would make a lead labeling of the GRMs unnecessary.
The results Raman spectroscopy and ICP-OES show that the released GRM fraction depends significantly on the type of GRM used. Based on our evaluation, several parameters such as the inter-layer strength between graphene or GO sheets, the bonding of the GRMs to the epoxy and the GRM size simultaneously affect the released GRM fraction. The discussion on the influence of these parameters is provided in the following.
GO showed the highest release fraction among the investigated GRM. Based on Gupta et al. (2017), energy corrugation and shear strength of GO are higher than that of graphene, since the electronic charge of the structures is localized on the functional oxygen groups. This means that the inter-layer bonding strength of the GO layers in a GO particle is stronger than the inter-layer bonding strength of the graphene layers in a GNP particle. Therefore it is more difficult for GO to be laminated (released layer by layer) compared to GNP. Chandrasekaran et al. (2013) reported that GNP-epoxy composites delaminate in-between the graphene layers of the GNP particles due to deflection of propagating cracks in-between the graphene layers. Moreover, the GNPs consisted of multiple layers (Table S1) which facilitated the interlayer delamination in addition, whereas the delamination of GO was impeded due to the compactness of the GO particle (total thickness of 21.6 ± 21.1 nm and a diameter of 1 to 20 μm, Table S1) and the strong inter-layer bonding strength of the GO layers. Thus, the pull-out mechanism was more important for GO-release than for GNP-release, where delamination played an important role. We hypothesize that this difference contributed to a higher release fraction of GO compared to GNP.
GNP-M25 showed a higher release fraction in comparison with the chemically identical GNP-M5. Since the abraded fragments from GNP-M25-epoxy were only slightly larger than those from GNP-M5-epoxy, we conclude that the larger size of the GNP-M25 particles was related to more protruding particles. Considering the smaller diameter and the smaller thickness of GNP-M5 compared to GNP-M25, we hypothesize that more GNP-M5 particles were embedded in the released composite fragments. In addition, there was a higher chance for the larger GNP-M25 particles, to be hit by the cracks formed due to abrasion. Therefore, the release fraction of GNP-M25 was higher than the release fraction of GNP-M5.
In addition, we compared our outcome to the results of Rhiem et al. (2016) who determined release of CNTs from CNT-polycarbonate composites and to Schlagenhauf et al. (2015a) who determined release of CNTs relative to the total mass of CNTs in abraded composite powders from CNT-epoxy composites. Schlagenhauf et al. (2015a), who used a similar lead-labeling approach as us, reported a lead-release capacity of 8.7 ± 0.4 μg Pb 2+ per mg CNT, which was in the same order of magnitude as the lead-release capacity of GRMs used in our study (Fig. 6a), just slightly lower. However, the released nanoparticle fractions were significantly different. In comparison to the reported CNT release fractions of 0.4 wt% from Schlagenhauf et al. (2015a) and 1.03 wt% from Rhiem et al. (2016), we detected significantly higher release fractions for comparable composites with carbonaceous NMs (92 wt%, 52 wt% and 18 wt% for GO, GNP-M25 and GNP-M5).
Based on Schlagenhauf et al. (2015a), abrasion of CNT-epoxy composites induced the release of composite fragments with freestanding and protruding CNTs. The freestanding CNTs were reduced in length in comparison to the pristine CNTs. In case of protruding CNTs, only a small part of a whole CNT fiber was exposed on the surface, whereas the major part remained embedded in the epoxy matrix of the abraded particle. This implied a stronger interaction of the CNT fibers to the epoxy compared to the interaction of GRM platelets to the epoxy. As described by Nguyen et al. (2017), CNTs form interlinked networks within a matrix, which might be the reason for the stronger interaction. The one dimensional CNT fibers possess a higher ability to bend compared to the two dimensional GRMs. The higher ability to bend promotes a stronger entanglement and the formation of interconnected networks. Moreover, a higher amount of carboxyl and hydroxyl groups on the NM promotes the formation of interlinked networks by the formation of hydrogen bonds between the single particles. In addition, a high surface roughness of a NM facilitates the formation of inter connected NM networks, since it leads to a higher friction between the single particles. Due to their two-dimensional shape and their smooth surface, GRMs could not form interlinked networks within the matrix, which promoted a higher release of such kind of particles during the abrasion process. However, the investigation of the ability of NMs to form interlinked networks was not the focus of this study.

Conclusion
The GRM size, the inter-layer strength between graphene sheets as well as the interaction of the GRMs to the epoxy determine the fraction of released (protruding and freestanding) GRMs in abraded fragments from GRM-epoxy composites as demonstrated by our comparison of the released fraction of epoxy composites with GO, GNP-M5 and GNP-M25 nanofillers. Since EDX-SEM mapping detected lead from released leadlabeled GRMs, EDX-SEM mapping can be applied as a qualitative method to determine if there is release of GRMs. The released GRM fractions determined using the quantitative ICP-OES method showed values of 92 wt%, 52 wt%, 18 wt% for GO, GNP-M25 and GNP-M5. In comparison, the results from semi-quantitative Raman spectroscopy mapping showed GRM surface coverage values of 63%, 23% and 8% for GO, GNP-M25 and GNP-M5 on composites with non-lead-labeled GRMs. Since the GRM surface coverage values show a similar trend as the release fractions from ICP-OES, we conclude that Raman spectroscopy mapping can be used for a first appraisal to determine the fraction of released (protruding and freestanding) GRM on the surface of abraded composite fragments. This would accelerate the determination procedure and make a lead labeling of the GRMs unnecessary.

Notes
The authors declare no competing financial interest.

Funding
This study was partially financed by the European project GRACIOUS (project number: 760840) and SNF project "Interaction of graphene related materials and abraded GRM reinforced nanocomposites with 3D lung cell models" (project number 310030_169207).

Declaration of Competing Interest
None.

Appendix A. Supplementary data
Supplementary data to this article can be found online at https://doi. org/10.1016/j.impact.2020.100266.