How epoxy cure profiles (may) impact performance of non-crimp fabric composites

Glass fibre reinforced polymers made from quasi uni-directional non-crimp fabrics have recently been shown to suffer from a specific damage mechanism. This damage mechanism is dominated by off-axis cracks in the off-axis backing bundles of the fabrics. In addition, research results have indicated that the onset and propagation of such off-axis cracks may be influenced by the residual stresses that arising from the manufacturing process used to produce the Glass fibre reinforced composites. The current article investigates the potential effect of residual stresses on the fatigue life through an experimental set-up that involves tailored cure profiles and image based evaluation of tensile test experiments.


Introduction
Non-crimp fabric reinforced polymer matrix composites are used in many structural applications due to a low material cost, good stiffness properties and high resistance to fatigue degradation.These properties makes it beneficial to use non-crimp fabrics in a quasi uni-directional (quasi-UD) form, especially in structures subjected to loading conditions where one loading direction is particularly dominant.An example of such an application with high stiffness and fatigue requirement is the load-carrying laminates in the spar cap of a modern large wind turbine blade, in which UD Non-crimp fabrics are heavily used.
Brøndsted et al. [1] found that the most critical loading condition for Non-crimp UD fabrics was tension/tension fatigue.For tension/tension fatigue, Zangenberg et al. [2] postulated a fatigue damage evolution mechanism based on an extensive microscopy study of a quasi-UD non-crimp glass fiber fabric reinforced composites.This fatigue damage evolution mechanism which was also observed by Jespersen et al. [3,4].Furthermore, a study using Large-Field optical microscopy by Mortensen et.al [5] showed both qualitative and quantitative results that supports the existence of this mechanism.In these studies, the fatigue damage in tension/tension has been found to be controlled by the off-axis tunnel cracks.Tunnel cracks which was found to appear rather early during the fatigue life testing [5] and which was resulting in fiber breakage located in alignment with those.Therefore, a delay of these off-axis cracks may have the potential of a better fatigue resistance of the composite materials.
In addition to the fiber-bundle architecture, the fatigue resistance of polymer matrix composites has recently been found to be influenced by the chosen cure profile [6,7].This study aims to combine recent developments knowledge about epoxy cure profiles [8,9], and experimental procedures for quantifying the crack-density of off-axis cracks in tension/tension   [10].The goal of the study is to investigate the potential effect of residual stresses on off-axis cracks in backing bundles of UD Non-crimp fabrics.

Materials and Manufacturing
The glass fibre composite specimens were manufactured using Vacuum Assisted Resin Transfer Moulding (VARTM).The details of the fabric is presented in Table 1, showing that the area weight of the fabric is dominated by axially oriented UD fibre bundles, resulting in high axial stiffness of the final composite.The focus in this study is , however, the off-axis cracks that form in the backing bundles of the fabric.Non-crimp fabrics of this sort are sometimes referred to as quasi-UD fabrics in the literature [4] because only around ≈ 10 percent of the fabrics area weight is made up from off-axis bundles.Quasi-UD non-crimp fabrics are commonly used in the wind turbine industry where handling of the fabrics are aided by the backing bundles, while stiffness in the axial direction of the blade remains high [11].The backing bundles of a non-impregnated fabric sheet are depicted in Figure 1a where the axial orientation is aligned horizontally, and the backing bundle is presented.It is worth noting here that even though quasi-UD non-crimp fabrics contains off-axis bundles, the lay-up notation for a quasi-UD composite often ignores the presence of the backing bundles.A lay-up of 10 non-crimp quasi-UD fabrics could be described with a notation like this [0 • ] 5s , where the subscript s refers creating a symmetric lay-up (2 time 5) wrt. to whether the backing side faces up or down .For the lay-ups in this work a notation that is similar to the one used by Jespersen [4] will be applied.The axial direction will be noted by the 0 • angle, an individual backing layer will be noted by /b θ where the /b represents that a backing layer is part of a fabric sheet, and the superscript denotes the angle orientation.To mark the separation of two non-crimp fabric sheets a semi-colon ';' will be used.
The crack-counting method, described in further details below, counts cracks in a plane field of view with no perception of depth.The plane view means that only a single layer of backing in each of the backing orientation may be on the field of view for crack counting.If more than one layer of backing is present in the material, then no information is available in the images to confirm to which layer the crack belongs.Manufacturing a symmetric composite plate with only a single layer of backing is not possible without manipulating the fabric sheets.Figure 1b shows the lay-up modification made in order to have a gauge area in the plate that is symmetric yet only have one layer of each of the backing orientations.The symmetric lay-up with only one layer of backing bundles has been build by defining a gauge area of roughly 10 cm in width, where the backing from one fabric sheet has been removed.Stacking the modified fabric sheet on top of another (un-modified) fabric sheet results in a composite plate lay-up that is slightly asymmetric in most of the plate -with a lay-up of [ 0  The polymer matrix used in the VARTM process was a commercially avaible epoxy (Huntsman Araldite LY 1568 / Aradur 3489) described in the literature by Mortensen et al. [9].The ultimate glass transition temperature of the epoxy is T ult g = 84 • C. As mentioned previously the plates that tested in this work was manufactured using VARTM.Two plates were made with the same manufacturing set-up, but with different temperature profiles applied for curing.The two temperature profiles -a one step (fast) and a two-step (slow) -was used, and temperature data from embedded thermocouples are shown in Figure 2. The temperature profiles were based on research papers presented by Mortensen et al. [8] and Mortensen et al. [9] which suggests that the magnitude of residual stresses induced from curing the epoxy matrix can be substantially lowered by using a two-step cure profile instead of a one-step cure profile.The outcome of using a two-step cure profile, which requires significantly longer time for curing, compared to a one-step cure profile, is a significant reduction in cure induced residual stresses.The cure kinetic model presented by Mortensen et al. [9] confirms that the cure profile for both plates ensures that the epoxy matrix in the composite plates have reached full cure.Testing of the composite material was performed by cutting the plates and mounting tabs material using the improved tensile fatigue specimen geometry from DTU Wind Energy [12] of 410 mm length.All tests were conducted in load control with R=0.1.The load were determined from an initial measurement of specimen stiffness conducted with two extenso-meters.The frequencies used for the specimens were varied based on the maximum strain at which the specimens were tested.Varying the frequencies allowed the strain rate of the specimens to be kept at identical levels for all specimens.The strain rate was set to 141 s based on a reference test with 0.5% maximum strain at 5 Hz.During testing a digital camera was used to capture images of specimen gauge areas, which were later used as data in an automated crack counting scheme.The image series captured were taken such that roughly 1 100 images were captured for each log-scale decade.All captured images had a 10 ms exposure time, and were captured at the peak load of the sinusoidal load curve exerted by the testing machine.The test machine and camera set-up are displayed in Figure 3b.

Automated Crack Counting and X-ray image masking
Images captured during testing have been used for Automated Crack Counting (ACC) using the procedure described by Glud et al [10] and used by Jespersen et al [4] for crack counting in a quasi-UD non-crimp fabric material system.In a material system like the one investigated in this work the off-axis cracks that can be counted by ACC is limited to the areas where off-axis backing bundles are present.Jespersen et al [4] used 3D X-ray tomography to extract the fibre bundle structure that makes up this area.A similar approach to defining the appropriate area was used by Jespersen et al. [4], except only a single x-ray exposure was used rather than the full x-ray scan applied by Jespersen.
The single exposure X-ray images were captured using an Xradia Versa 520 µCT scanner with a Low Energy filter.An exmaple of the test set-up used for capturing an x-ray image exposure is shown in Figure 3a.An example (shown horizontally) of a single exposure image is shown in Figure 4 where it is combined with an image by the digital camera during tensile testing.The backing bundles in the middle layer of the specimen are visible in the x-ray images as areas that are darker than the surrounding areas of the images.The x-ray images are darker in the regions where there is backing because fewer gamma particles penetrates these areas during the x-ray exposure, causing fewer detections on the x-ray CCD detector, and eventually resulting in dark areas of the image.The optical image and the x-ray image can ultimately be scaled and translated such that the images covers the same physical image space.The pixel size p -with units mm pixel -for the x-ray images are given by the tomography equipment, while the pixel size for the optical images are calibrated based on the measured width of the specimen and the width observed in the optical image.The scaling of the x-ray image is performed automatically based in the known pixel sizes of each image, while manual positioning of the x-ray image wrt. to the reference optical image have been performed base on the small paint-marks -not pen-marksthat are visible in the x-ray and optical reference image.The manual position of the x-ray image wrt. the optical image is only needed for the first image -the reference image -in used in ACC because the algorithm automatically aligns all subsequent images wrt. to the reference image position.As portrayed in Figure 1a, and documented in Table 1, the backing bundles are sown onto the The overlay of the x-ray exposure allows a to be overlayed the optical image for crack counting.
main UD bundles in three distinct orientations: +45 • , 90 • and -45 • .These three distinct orientations can not be automatically extracted from the x-ray images.Thus the backing areas from the single exposure x-ray images are segmented manually and the area related to a particular orientation θ are computed as the product of the pixel size p and the number of foreground pixels n mask in the manual segmentation, as written in eq. ( 1).The relation for the (2)

S-N curve and tensile specimens failure modes
The results of the tensile test in terms of final failure is shown in Figure 5.The test series with the fast cure have lower fatigue resistance than the test series with the slower two-step cure.The relative difference in performance is relatively low, and for the tests conducted at a 1.0 % strain level the specimens failed at nearly identical amount of load cycles.There are, however, no instances in which the specimens cure with the fast cure can sustain more loading cycles than the slow cure.

Crack Counting Results
A representative example of cracks detected in an image is displayed in Figure 6.The opaque red areas in the image marks the areas of the gauge ares where 90 • is present and the red lines mark the detected cracks.Similarly, the opaque blue areas and opaque green areas marks the areas of the gauge ares with +45 • and 45 • is present.The cracks are colored in the same representative colours.The masked areas functions both as a representation of the area, and as mask that removes cracks that are falsely detected outside the area with backing bundles in the same direction as the detected crack. Figure 7 compares the development of the crack densities in the specimens manufactured with the fast one-step cure cycle and the slow two step cure cycle.Before addressing the differences that arise as a result of the cure cycles it is worth addressing how the crack density development is influenced by the off-axis backing fibre bundle orientation.While no ±45 • cracks were observed for the 0.5% strain level, the general trend is that the crack density develops at a much faster rate in the 90 • backing bundles than in the ±45 • backing bundles.The difference is not surprising giving that the fracture mode for the 90 • cracks is pure mode I, while the fracture mode for the ±45 • is mixed mode, and dominated by mode II, which means that crack growth in the 45 • direction requires a higher fracture energy to propagate.The stress state in the ±45 • bundles are likely to cause the cracks to open less than for the 90 • bundles.Furthermore, the crack density of the ±45 backing bundles are expected to be less sensitive to the residual stresses, and the actual effect of residual stresses on the ±45 • bundles are not as clear as for the 90 • bundles that are loaded in pure mode I.Therefore the impact of residual stresses will be based on the crack density in the 90 • backing layers.
For strain levels ε = 0.5 and ε = 0.8 there is no significant difference between the development of the crack densities for the 90 • backing bundles.For strain levels of ε = 0.9 and ε = 1.0 the crack density for the 90 • backing bundles grows faster for the fast cure, and importantly achiieves levels of crack density that is 1.5 to 2 times higher than what is reached for the slow cure.

Conclusion
A crack counting scheme was applied to measure the development of crack densities in two test series of quasi UD NCF composites.The two test series had been manufactured with the same manufacturing technique, but with two different cure cycles, designed to create variation in the magnitude of residual stresses in the plates from which the test series specimens were cut.The S-N curve for the two test series confirmed a relative performance difference between the two test series.The expected faster development of the crack density for the fast cure, with the high magnitude of residual stresses, was observed for tests conducted at two specific test strain levels (ε = 0.9%,ε = 1.0%).For two other strain levels the observed effect was small or non-existent.

( a )
Structure of non-crimp fabric backing showing the +45 • , 90 • , − 45 • fabric bundles and the stiching thread.(b) Lay-up used for manufacturing of composite plates with only a single layer of backing.

Figure 1 :]Figure 2 :
Figure 1: Illustrations of the fabric used in the laminate lay-up, and the special gauge-area with a single layer of non-crimp fabric.

Figure 4 :
Figure 4: Combination of image obtained with optical digital camera and X-ray exposure of gauge area.The overlay of the x-ray exposure allows a to be overlayed the optical image for crack counting.

Figure 5 :
Figure 5: S-N Curve for fast and slow cure