The dependence of interfacial shear strength on temperature and matrix chemistry in glass fibre epoxy composites

The present work focuses on a fundamental investigation into the influences of the chemistry of epoxy and the testing temperature on the stress-transfer capability of the fibre-matrix interface in a glass fibre reinforced epoxy composite. We discuss how the interfacial shear strength (IFSS) is influenced by the hardener-to-resin ratio, testing temperature and fibre silane coating respectively. It was observed that the IFSS showed a significant inverse dependence on testing temperature for both silanes, with IFSS values dropping as the temperature was increased, for all ratios studied. Notably, it was shown that once the testing temperature was raised above the glass transition temperature that ratios possessing excess amine hardener had larger IFSS values. From the results presented it is hypothesized that residual radial compressive stresses at the interface are influenced by the chemistry of the matrix system and then relax away at the higher testing temperatures. with the results representing the average of individual samples. It is indicated that the IFSS value decreased as the testing temperature was increased for all ratios, with H/R =1.04 initially exhibiting the largest IFSS values for temperatures ≤ 70 °C. A notable gap between H/R >> 1 and H/R ≈ 1 can be seen at 120 °C. The observed gradient of transition appears to occur differently depending on the specifics of the matrix chemistry, with ratios where H/R ≤ 1 appearing to lose performance sharply once above 70 °C whilst ratios where H/R > 1 appear to be less influenced above this temperature.


Introduction
Over the past decades there has been a rapid growth in the development and application of fibrereinforced composite materials for use in high performance applications. Coinciding with this there has been a concerted effort to develop a better understanding of the micro-mechanical parameters that control the structure-property relationships within such composites. Composite properties result from a combination of the properties of the fibre and the matrix as well as the ability to transfer stresses across the fibre-matrix interface. Although tailoring the interfacial stress transfer capability is widely recognized as critical to optimising the performance of a composite, it has been routinely reduced to a discussion focussed on 'adhesion'. In this context 'adhesion' represents a simple term for encompassing the multiple complex mechanisms that exist at the fibre-matrix interface and contribute to its strength. One accepted mechanically measurable value for quantifying fibre-matrix adhesion is the apparent interfacial shear strength (IFSS).
Discussions about the composite interface have primarily focussed on the chemistry of the matrix system and fibre surface coating, to ensure the optimum formation of chemical bonds [1][2][3][4][5][6][7][8]. Extensive studies of fibre sizings, and specifically the use of silane coupling agents, have been undertaken in order to attempt to better understand the specific mechanisms influencing the link between the inorganic glass fibre and the organic polymer matrix system [9][10][11][12][13]. However, a number of authors have also commented on the potential role of shrinkage stresses contributing to the stress transfer capability of the interface [14][15][16][17][18][19][20][21][22]. It has been suggested that residual radial compressive stresses at the interface may contribute significantly to the IFSS. Thermal radial compressive stresses can form during the cooling process due to differences in the thermal expansion coefficients of the matrix polymer and the reinforcement fibre. Thermosetting matrices also undergo volume change during polymerisation, known as cure shrinkage. Due to the increase of the glass transition temperature (T g ) as the thermoset matrix reacts and shrinks it is possible for some level of this cure shrinkage to be 'frozen' into the system, creating additional residual stresses at the interface [15,23]. Hence if the IFSS testing temperature is raised above the matrix T g then these residual stresses can potentially undergo stress relaxation, decreasing IFSS. This was shown to occur by Thomason [15]. Despite this there is still a significant body of opinion that holds that even if residual stresses formed at the interface do contribute to the stress transfer capability, then chemistry and chemical reactions must still play a direct role in determining the IFSS [20,[23][24][25][26].
Our previous studies have shown that the testing temperature does significantly influence the IFSS value [14,15]. This was attributed to the existence of the residual stresses at the interface, although the role of matrix chemistry in forming these stresses was not studied. We previously reported on the influence of the amine (hardener)-to-epoxy (resin) ratio (H/R) of an epoxy system on several thermomechanical properties closely linked to the strength of the interface, such as the T g , linear coefficient of thermal expansion (LCTE) and storage modulus [27]. It was found that all properties were distinctly influenced by the chemistry of the matrix system, with optimum performance occurring around the stoichiometric value. The H/R ratio was also shown to distinctly influence the IFSS with clear correlation shown between the IFSS value and the T g of the matrix. It was shown that the changes in thermomechanical properties influenced the contribution of the thermal residual stresses at the interface as the H/R ratio deviated from the stoichiometric value. However, this value only represented about 20% of the final IFSS value. It has been suggested that cure shrinkage stresses could potentially contribute much more to the IFSS value [23] and vary with the H/R ratio [28]. Additionally, fibre surface coatings have long been associated with increasing adhesion between the reinforcing fibre and the polymer matrix, particularly silane coupling agents [10][11][12][13]29]. Yet the potential difference these coatings may have on the adhesion with a rubbery polymer above T g has never been thoroughly investigated. We have therefore studied how the combination of testing temperature, H/R ratio and silane coating are interrelated to the stress transfer capability of the interface.

Materials
Single boron-free E-glass fibres, coated with γ-aminopropyltriethoxysilane (APS), and glycidoxypropyltrimethoxysilane (GPS), were taken from larger rovings supplied by Owens Corning. The nominal tex was 1200 g/km and the average diameter was 17.5 µm for both. Araldite® 506 epoxy resin and triethylenetetramine (TETA) were purchased from Sigma-Aldrich and used as received.

Microbond sample preparation
The stoichiometric ratio for the system was calculated as 12.0% TETA, equating to a H/R ratio of 1.0. For the APS samples the hardener-to-epoxy ratios varied from 6.8% TETA up to 22.2% TETA, equating to H/R ratios ranging from 0.53 to 2.08. For the epoxy silane (GPS) samples three ratios were studied: 7%, 12.2% and 21.9%, equating to H/R ratios of 0.55, 1.01 and 2.05 respectively. The resin mixtures, with different H/R values, were carefully measured and mixed thoroughly before being degassed for approximated 15 minutes. Minute droplets of the resin system were then applied to a single glass fibre using a thin piece of steel wire. For each resin mixture approximately 30 droplets were placed on different individual fibres before being transferred to a convection oven where they were heated to 60 °C and held isothermally for 1 hour, then further heated to 120 °C and held isothermally for 2 hours. The heating rate was 2 °C/minute for both heating ramps and the samples were left to cool down in the oven overnight. Prior to testing, all samples were examined under 200x magnification using a Nikon Epiphot inverted microscope to obtain values for the fibre diameter (D f ) and the fibre length (L E ) embedded in the resin droplet.

TMA microbond technique
Development of the TMA microbond technique has been reported previously [14,15]. The technique uses a TA Q400 thermo-mechanical analyser, in combination with the cooling accessory MCA270 mounted with a film/fibre probe and microbond setup as shown in Figure 1. The droplet applied to the fibre sits on the shearing plate which itself sits on the stationary quartz platform. The movable probe, installed in the centre of the stationary platform, then rests on the paper tab of the sample. It is this probe which applies the load during the test. The entire microbond setup is then enclosed within the TMA temperature controlled programmable oven. Each sample was initially placed under a very small pre-load of 0.005 N, with the free fibre length between the tab and epoxy droplet maintained at a constant value of 5 mm to match the samples tested in [27].
The notable difference between the TMA and Instron microbond techniques relates to how each technique loads the sample. The Instron technique [27] is carried out by measuring the load generated during the displacement of a droplet at a constant strain rate, however the TMA is unable to operate in this mode. Instead, the TMA was configured to measure the sample displacement during a linear force ramp of 0.15 N/min [14,15]. The testing procedure proceeded as follows: the probe displacement was electronically zeroed and the microbond sample loaded into the shearing plate with the paper tab hanging freely below the movable probe. The movable probe was then carefully lowered onto the paper tab before the furnace was closed. The initial sample length and probe position was then recorded, with the furnace then programmed to equilibrate at the desired test temperature (ranging from 20 °C to 120 °C) with an additional three-minute isothermal segment to ensure a constant equilibrium temperature was achieved.
The force ramp was then initiated at 0.15 N/min. As the test proceeded the probe displacement would increase and be recorded until a successful de-bond occurred. A typical result obtained from a TMAmicrobond test is plotted as a force-displacement curve in Figure 2. The major difference in the plots occurs after de-bonding. Due to the TMA continuing the force ramp after the successful de-bond, there is a rapid downward displacement of the de-bonded fibre. As such, no observations can be made on the friction region that occurs after the de-bonding process [14]. This maximum force value (F max ), along with the measured fibre diameter (D f ) and embedded length (L e ), was used to calculate the apparent IFSS value using Equation 1.

=
(1) All tests were conducted under a nitrogen flow of 50 ml/min with approximately 30 samples per ratio tested. Figure 3 shows a comparison between the TMA microbond results at 20 °C and the previous published data collected using the Instron microbond technique [27] for H/R = 1.22. It can be seen that the results collected using the TMA technique correlate well with the results collected using normal microbond testing. This indicates the microbond test carried out with our TMA and Instron setup is highly consistent in terms of obtaining IFSS. Figure 4 shows a comparison between the different testing temperatures for H/R = 1.04. We can see that as the testing temperature was increased, the max force of the microbond samples decreased, with a notable drop once the temperature was raised above the T g value of 87.3 °C found previously in [27]. It must be noted that this value represented the peak T g value measured in [27] which occurred at H/R ≈ 1.2, with the T g subsequently shown to decrease as H/R ratio deviated from that value. Figures 5 and 6 shows plots where the testing temperature was maintained at constant values of 50 °C and 120 °C respectively whilst the H/R ratio of the matrix system was altered. Figure 5 shows that at 50 °C there is a degree of variation observed in the debonding behaviour of the microdroplets when changing the H/R ratio, with the degree of scatter for H/R = 2.08 observed to be larger than that of the other two ratios. For Figure 6, given the T g value of 87.3 °C [27], 120 °C would represent a temperature that is well above the residual stress-free temperature (i.e. T g ), thus the changes observed would be thought to mainly reflect the effect of the H/R ratio on the degree of chemical bonding. Overall, it can be seen that at 120 °C the effect of changing the matrix chemistry appears greater than at 50 °C. This suggests that the importance of the matrix chemistry to the magnitude of the IFSS may be greater above T g . It can also be seen that the plot shown in Figure 4 does show some correlation with that shown in  where H/R ≤ 1 appearing to lose performance sharply once above 70 °C whilst ratios where H/R > 1 appear to be less influenced above this temperature. The data shown in Figures 7 and 8 can also be plotted as a function of ∆T as shown in Figure 9, where ∆T = T t -T g , with T t representing the testing temperature. Again, using values previously reported for matrix T g [27] we can see that for the more extreme ratios of H/R = 0.53 and H/R = 2.08 the most significant changes in the values measured during the study occur at ΔT values > 0, where the polymer in theory would be in a rubbery state. Again, the gradient of the IFSS decline is clearly different for the H/R = 2.08 ratio where the step transition appears notably reduced.

IFSS dependence on H/R ratio and testing temperature
A comparable series of IFSS tests as a function of temperature and a reduced number of H/R values was conducted using the GPS coated fibres. Figure 10 shows the IFSS plotted against the testing temperature for the H/R ratios studied. Overall, the plot shows a similar correlation to that shown in Figure 7, with the IFSS value decreasing as the testing temperature was increased. Once again, the maximum IFSS value was achieved, (44 MPa), at room temperature at H/R ≈ 1, however this was found to be lower than the equivalent IFSS value (49.1 MPa) recorded for the APS fibre under those conditions.
As with the APS fibres, it can be seen that at temperatures > 80 °C, the ratio where H/R >> 1 retained a higher level of IFSS, with performance at 120 °C appearing comparable to the results shown in Figure 7.
One difference between the results shown in Figures Figure 11 plots the IFSS results for the GPS coated fibres against the H/R ratio similarly to Figure 8. It can be seen again that, for temperatures up to 70 °C, the IFSS was maximum at H/R ≈ 1. However, for temperatures > 70 °C performance increased as more amine was added to the system. Figure 12 highlights the smaller decline in IFSS previously discussed for the GPS fibre samples when compared to the APS fibre samples shown in Figure 8. Again, IFSS drops less significantly for R>>1. Since this occurred for both silanes it suggests that there may be other non-silane related factors influencing the IFSS as the testing temperature is increased and that these must be associated specifically with the differences in the matrix system amine content.
Overall, it appears that the IFSS was clearly influenced by the chemistry of the matrix system as well as the testing temperature, with performance deteriorating consistently as the temperature was increased for both silanes. The APS fibres appear to possess consistently higher IFSS values at lower temperatures, suggesting there may be a stronger link between the matrix and the amine sized fibres than for the epoxy sized fibres. This stronger link equated to an extra 4 MPa at H/R ≈ 1 ratio at room temperature. The ratios where APS fibres with H/R < 1 were shown to possess comparatively higher IFSS values at lower temperatures than for the GPS fibres, as shown in Figure 13. Whilst this may be explained through chemical bonding theory to some degree [5,30,31], it would be expected that the extra amine provided by the silane coating of the fibre surface should result in more bonds forming at the interface with the excess of reactive epoxy located within the matrix system since H/R < 1. The extra amine would also aid in the formation of the crosslink network within the matrix. Yet for the GPS fibres, the coating would provide further excess epoxy groups to the system, resulting in less bonding and weaker thermomechanical properties [27]. This may contribute to the difference in stress transfer capability observed between the APS and GPS coated fibres.
According to this hypothesis we might also expect the GPS fibres to perform better for ratios where H/R > 1 since the extra reactive epoxy groups provided by the silane coating would be expected to bond with the extra amine groups within the matrix system. This should facilitate more bonds formed and in theory larger IFSS values. Yet this does not seem to be the case, with the APS fibres consistently possessing higher IFSS values for these ratios. However, at temperatures > T g the GPS fibres appear to perform comparably to the APS fibres when H/R > 1. It is interesting to note that the overall decrease in IFSS with temperature appears consistent for both silanes, suggesting that they are undergoing a similar transition despite the chemical differences at the fibre surface.
With IFSS values appearing similar for both silanes at 120 °C for all ratios, it suggests that the mixture chemistry of the matrix system may play a more important role than the chemistry of fibre coating in defining the stress transfer capability of the interface at this elevated temperature. This hypothesis, that the chemistry of the matrix has more influence over the IFSS value than the fibre coating, agrees with what can be seen in Figures 5 and 6 where at 120 °C the effect of changing the matrix chemistry appears greater than at 50 °C. This would suggest that adhesion mechanisms other than chemical bonding are contributing more to the drop in IFSS as shown in Figures 7 and 10, particularly when the temperature is elevated close to, or above, the matrix T g . A potential explanation is that the residual radial compressive stresses (σ R ) formed at the interface during the curing process are being influenced by the properties of the matrix and relaxing away as the IFSS test temperature is increased.

Residual stress contribution to IFSS
On reviewing the results in Figures 7-13, it can be seen that the IFSS showed a significant inverse dependence on the testing temperature for both the APS and GPS fibres. With that in mind it would suggest that the deterioration in IFSS was due to factors other than a decrease in chemical bonding at the interface. One potential explanation is the existence of residual radial compressive stresses that may contribute significantly to the stress transfer capability of the fibre-matrix interface [14,15,23]. Several models have been proposed [16,17,20,32,33] to calculate these residual stresses with Nairn [16] having proposed a model showing significant normal stresses at the interface as shown in Equation 2. The benefit of Nairn's model is that it accounts for the effects of differences in axial and transverse fibre properties exhibited by many reinforcements.
where A 1 is the result of the calculation shown in Equation 3.
The model shown in Equation 3 represents a simplification of the composite cylinder model proposed in [16]. Nairn assumed the volume fraction of the interphase to be zero which allowed for the equation to be reduced to a two-by-two linear system. The X terms represent elements of the matrix that can be calculated if certain material properties of the fibre and matrix are known, such as modulus, Poisson's ratio, and volume fraction. These expressions are detailed in [16].
Of the variables presented, b is a function of the fibre volume ratio (V f ), r is the radius of the fibre, α is the linear coefficient of thermal expansion, ∆T represents the difference between the stress-free temperature (T s ) and the testing temperature (T t ), and f, m, L and T represent subscripts for the fibre, matrix, longitudinal and transverse respectively. From the form of Equation 3 the interfacial thermal residual stresses are directly influenced by changes in ∆T, thus we would expect the thermal residual stresses to relax away as T t increase and surpasses T s which itself is highly related to the T g of the system.
Our previous work [27] showed that both the T g and LCTE of the polymer are highly influenced by the epoxy-to-hardener ratio and thus thermal residual stresses were similarly affected. Using this data in combination with a coefficient of static friction of µ s = 0.6 [12,30] allowed for the residual stress contribution to the interfacial stress transfer capability ( R ) to be obtained using Equation 4.

= (4)
Taking this previously published thermal residual stress data [27] and combining it with the TMA microbond data presented here allows us to plot the IFSS value for each ratio against the residual stress contribution to the interfacial stress transfer capability ( R ). Figure 14 shows and H/R = 2.06 respectively. It can be seen that for each H/R ratio there is some level of correlation between the IFSS decrease with temperature and the relaxation of the residual stresses due to thermal shrinkage as the testing temperature was raised. However, it is also clear that the residual stresses due to thermal shrinkage could only be making a small contribution to the overall IFSS values observed in the system.
One significant difference between thermosetting and thermoplastic systems is that thermosets will undergo chemical shrinkage during the curing process due to the polymerization reaction. This cure shrinkage can result in significant volumetric changes, applying a large degree of strain to the system. Di Landro and Pegoraro [20] have previously discussed how cure shrinkage may provide the bulk of the residual stresses when compared to thermal shrinkage stresses in thermoset systems. However, any cure shrinkage that occurs before the gel point of the system can potentially relax away. This can be attributed to the low molecular weight and liquid nature of the matrix in this state. Gelation is defined as the liquidsolid transition of an epoxy resin that occurs when the average molecular weight approaches infinity. One can reasonably assume that any strain induced by cure shrinkage will require more time to relax away as the system approaches the gel point. Thus, once the gel point is exceeded then any further cure shrinkage related residual strain will be locked into the epoxy network. As the system is cooled back through T g the matrix modulus will increase and consequently the residual stress from the cure shrinkage strain will increase in proportion to the matrix modulus. This hypothesis fits with the relationships observed in Figures 7 and 10 where the increase in temperature would lead towards this stress potentially relaxing away. Applying this hypothesis, using our modulus data from [27] and applying an isothermal volumetric cure shrinkage value of -6% [15,23,34,35] produces the additional stress curves (marked as cure shrinkage and total shrinkage) shown in Figures 14-16. It can be seen that the residual stress contribution due to cure shrinkage appears to be significantly larger than that provided by the thermal shrinkage for each ratio. This is in agreement with the results discussed by both Thomason [15] and Jakobsen [23]. Figures 14-16 show that the residual radial compressive stresses at the interface could indeed be contributing significantly to the stress transfer capability of the interface at room temperature. The sum of the thermal and cure shrinkage contributions produces a value comparable in magnitude to that measured using the microbond technique. However, it appears that the decrease of IFSS at temperatures above the T g cannot be due to just the residual stresses at the interface relaxing away. Specifically, at 120 °C there is still a distinct level of adhesion shown to remain between the fibre and matrix, at a point where it would have been expected that all the residual stresses have relaxed away. As discussed previously, at these higher temperatures the IFSS value appears to be dictated by the chemistry of the matrix, not the surface coating of the fibre, and scales linearly with the level of amine in the matrix system. Applying chemical theory, one would expect the IFSS value to decrease as the ratio deviated from stoichiometry, with the number of bonds across the interface decreasing and the level of crosslinking within the matrix also decreasing [36,37]. That statement would apply to a glassy state polymer and match what was shown below T g , yet at 120 °C the epoxy matrix would be in a rubbery state, with perhaps other mechanisms governing the stress transfer capability. One hypothesis would be that hydrogen bonding within the matrix increases as more amine hardener is added to the system, leading to the observed lessened decrease in IFSS observed in Figures 7 and 10. This would occur since the excess hardener within the system would result in an abundance of reactive hydrogen atoms remaining within the matrix structure, contributing to the level of hydrogen bonding present. This would also explain the linear increase shown in Figures 7 and 10 as the ratio was increased, with the hydrogen bonding now potentially providing a significant contribution to the adhesion remaining in the system.
Another hypothesis is that at these higher temperatures, due to the polymer now being a rubber, the nature of the debonding process is being influenced by the R ratio. D'Almeida has discussed [38][39][40][41] how epoxy matrices containing excess amine can possess greater deformation capacity than at lower H/R ratios.
Thus, at these higher temperatures the higher H/R ratio matrices may be superior at maintaining the stress transfer capability at the interface whilst under load due to a better stress distribution across the droplet during the microbond test. Zhao [42] also observed that variations in the crosslink density could significantly impact the shear failure behavior in amorphous polymers. Zhao found that the ultimate stresses and the broken ratios, (the bond number to all polymer chain number ratios), increased with increasing crosslink density under shear, but that the ultimate strains decreased with increasing crosslink density. Furthermore Zhao found that when the temperature was varied, the broken ratio varied significantly due to the potential reorganization of the polymer structure [42]. These observations could partially explain the data shown in Figures 7 and 10 and why excess amine ratios perform superior to those ratios where H/R ≈ 1 and temperature > T g . However, ratios where H/R < 1 would also possess smaller crosslink density values [36] and thus would be expected to benefit in the same manner; yet they do not. As such it could be hypothesized that although this mechanism may be contributing to the stress transfer capability of the interface, its contribution is less than other mechanisms. This, combined with the hydrogen bonding theory, may potentially explain the IFSS value trends shown at temperatures above T g .
Analyzing Figures 14-16, it can be seen that whilst the assumptions for Nairn's model [16]  before the resin reached gelation. Thus, the degree of shrinkage frozen into the matrix structure would be influenced by the gel point of the system, as well as the degree of shrinkage that occurred. The level of cure shrinkage in epoxy matrices has been shown to change with different curing schedules [26,43,44] and with the influence of different H/R ratios on cure kinetics [45,46]. It can be hypothesized that a similar relationship may exist between matrix chemistry, cure shrinkage, residual interfacial stress and interfacial stress transfer capability. This hypothesis led to the investigation using a novel hot-stage microscope technique in [28] where it was found that the degree of chemical cure shrinkage increased as the H/R ratio was increased.
Taking the data produced in [28] Figures 14-16 can be adjusted accordingly to take account of the different degrees of cure shrinkage found to occur for the different H/R ratios. Figure 17 shows that the results of Nairn's model do correlate well with the measured IFSS results for the TMA microbond technique for ratios near to the stoichiometric value. Since the level of cure shrinkage measured was smaller than the original 8% assumed for Figure 14, the contribution values appear to be smaller.
However, if the coefficient of static friction is slightly increased from 0.6 up to 0.75 as shown in Figure   20 then the two plots are shown to correlate to a greater degree. Another potential reason for the lack of correlation in the temperature region above T g may be due to the assumption made between the residual stress and the gel point by the model. In reality the transition of residual stresses being "frozen-in" once the gel point is crossed may not be clear cut. Given the visco-elastic nature of the polymer (or polymerizing polymer) the "frozen-in" stress may start to build up even before the gel point, to give a more gradual increase in residual stress as observed in the IFSS data. Overall, these results would support the hypothesis that residual stresses can contribute significantly to the stress transfer capability of the interface. Figures 17 and 20 suggest that there does exist a clear link between the IFSS and the residual stress contribution.
It can be seen that the contributions predicted by the model for H/R ≈ 0.5 and H/R ≈ 2.0 again appear to correlate poorly with the results produced by the TMA microbond method. This was also the case in Figures 15 and 16, although the decrease in cure shrinkage for H/R ≈ 0.5 appears to have been impacted the most. It can be seen that the contribution of the residual stresses to the IFSS equates to only ≈ 20 MPa maximum. Compared to the 46 MPa measured at 20 °C this would confirm that other adhesion mechanisms must be at work in order to produce this value. In addition to this, both plots show that the residual stress contribution decreases before any notable drop in IFSS occurs. For H/R ≈ 0.5 the value for IFSS drops ≈ 70 °C, while the residual stress contribution is shown to have relaxed away ≈ 50 °C. In the case of H/R ≈ 2.0, the suggested contribution at 20 °C still appears notably larger than the measured IFSS.
This was also the case in the original model in Figure 16. The contribution is then shown to relax away at ≈ 60 °C while the IFSS maintains a gradual decline. Both would suggest that the residual stress contribution is not the key contributor of the complex mechanisms at work at the interface, highlighting the limits of the model currently used. Compared to the results for H/R ≈ 1.0, it appears that the contribution of the residual stresses plays a more central role in defining the value for IFSS the nearer the H/R value is to stoichiometry. Once the ratio begins to deviate from this value, the significance of the residual stresses appears to decrease. However, the apparent limits of the model are directly linked to our lack of understanding regarding the T g of an epoxy microdroplet. This is due to the model assuming that the T g of the droplet is the same as the bulk value measured using DSC in [27], where it was shown that the T g could vary significantly for H/R >> 1 or << 1. It has been shown in the past by Zinck [47] that the T g of a thin film of resin differed from that of the bulk matrix. They also suggested that this affected their microbond results. Both Thomason [48] and Bryce [49] have found that the results of the microbond test can be highly dependent on the properties of the microdroplet, with the issue of scaling being highlighted as a challenging factor when it comes to assuming T g values that reflect the properties of the macroscale composite [48]. Given the importance of the T g value towards the stress-free temperature T s and the residual stress created from the cure shrinkage strain, this would provide an explanation for why the model works for H/R ≈ 1.0 but not for off-stoichiometric values. The possibility of this T g scaling issue also being influenced by the H/R value would also explain why the results of the model decrease in correlation the further the H/R value deviates from stoichiometry. Nevertheless, it can be seen that there exists a relationship between the IFSS value, the H/R value, the testing temperature and potential residual stresses "frozen in" at the fibre-matrix interface. Further study of the T g value of epoxy microdroplets, the effect of the H/R ratio on the scaling factor discussed in [48,49] and an investigation into the specific friction coefficient of the fibres studied would be beneficial for use in Nairn's model used for this paper.

Conclusions and concluding remarks
The results presented in this paper show that the hardener-to-epoxy ratio, testing temperature and fibre silane coating can each influence the stress transfer capability of the interface in a glass fibre reinforced epoxy composite. Each hardener-to-epoxy ratio tested was shown to possess different IFSS values at each testing temperature. Below the epoxy polymer T g , the highest IFSS values were shown to occur at the stoichiometric value, with IFSS decreasing as the ratio deviated further from this value for both types of silane coated fibres investigated. However, as the temperature was increased a transition in IFSS was observed when the testing temperature exceeded 70 °C. Whilst interfacial performance continued to decrease overall after this point, the optimum relative IFSS value was shown to improve linearly when more amine was added to the epoxy system. This was even the case when the testing temperature was raised further to 120 °C. It was shown that there was some correlation between the deterioration in IFSS and the potential relaxation of the "frozen" in residual stresses at the interface as the temperature was increased. This was shown particularly for H/R ratios close to the stoichiometric value. It was also observed that at temperatures well above T g a small value of IFSS still existed, despite the residual stresses having likely relaxed away. It was hypothesized that an increased level of hydrogen bonding, due to the increasing amine content of the matrix system, combined with the variation in shear failure behavior of the matrix due to the differing crosslink densities, was potentially responsible for this. One