Failure of dissimilar material bonded joints

1.1 Adhesively bonded joints

“Adhesive bonding” was introduced into structural design approximately 70 years ago, starting with metal/wood bonding in aircraft (phenolic resin, UK) and metal/metal bonding (epoxy resin, Germany) in the early 1940s, which allowed more flexibility in structural lightweight design. In the late 1960s rubber-like, hyperelastic polyurethane adhesives were first used for bonding glass to steel structures in automotive applications. These types of adhesives recently became, in new and improved formulations, good candidates for FRP/metal structural bonding. To date there are several structural adhesive types tailored for specific applications. The list of those adhesives is exhaustive, and providing detailed information on those is beyond the scope of this chapter.

The use of adhesive fracture mechanics is one of the most common and useful type of analysing a single-lap joint, taking into consideration the main parameters that influence the failure mode, like the joint geometry and the properties of adherends and adhesives. One of the analytical models was first proposed by Volkersen in 1938 [1] also known as the “shear-lag model”. The adhesive shear distribution τ is given by

where ω = GaEttta(1 + tttb).

The reciprocal of ω has units of length and is the characteristic shear-lag distance, a measure of how quickly the load is transferred from one adherend to the other. t_t is the top adherend thickness, b_t is the bottom adherend thickness, a_t is the adhesive thickness, b is the bonded area width, l is the bonded area length, E is the adherend modulus, a G is the adhesive shear modulus, and P is the force applied to the inner adherend. The origin of x is the middle of the overlap and is shown in Fig. 1.

Fig. 1

Volkersen model.

The model was then further extended by Goland and Reissner [2], who were the first to consider the effects due to rotation of the adherends; see Fig. 2.

Fig. 2

Goland & Reissner model.

The adhesive shear distribution τ according to Goland and Reissner is given by

where P is the applied tensile load per unit width, c is half of the overlap length, t is the adherend thickness, and k is the bending moment factor given by

where u2 = 3(1 −v2)2 1t P¯tE; β2 = 8GaEtta and ν is the Poisson’s ratio.

Although the work of Volkersen and Goland and Reissner was a major step forward in stress analysis of adhesively bonded joint, it faces several limitations. Some analytical models are difficult to be incorporated because the analysis becomes very complex, so most of the analyses are linear elastic for both adherends and adhesive. Regarding the conditions of applicability and the stresses considered, the suitable model can be applied. In a case where the joint bending is not neglected and the adhesive is brittle, Volkersen’s analysis is sufficient, but when yield and peel stresses of the adhesive and/or the adherends are under consideration, a more complex model is necessary.

The analysis by Goland and Reissner for the single lap joint has been generalized by Bigwood and Crocombe [3, 4], so that it can be applied to an arbitrary end loaded single overlap configuration and take account of adhesive nonlinear behavior. The resulting analyses can be applied to single lap joints as well as to many other configurations, as shown in Fig. 3.

Fig. 3

Arbitrarily end loaded single overlap joint.

The resulting expressions for the adhesive shear and transverse direct stress in terms of the adhesive strains and hence substrate displacements are:

Fig. 4

Goland & Reissner’s adhesive shear and peel stress distribution for aluminum alloy adherends and an epoxy adhesive.

Although the analyses of adhesively bonded joint failure by Volkersen and by Goland and Reissner are significant, they do have a number of limitations:

1. the models do not take into account the variation of stress in the through thickness direction, especially the interface stresses which become significant when failure takes place close to the interface;

2. the stress-free condition at the ends of the overlap are ignored, which results in overestimation of the stress at the overlap ends and tends to yield conservative failure predictions;

3. the adherends are considered as thin beams, ignoring the through thickness shear and normal deformations. Adherend shear is particularly important in polymer composite adherends.

These limitations were addressed by Ojalvo and Eidinoff [5], who investigated the shear and normal deformations in the adherends. They extended the work of Goland and Reissner by using a complete shear strain/displacement equation for the adhesive layer to investigate the influence of adhesive thickness on the stress distribution. The adhesive shear stress was allowed to vary across the thickness, irrespective of the thickness of the adhesive, but the adhesive peel stress was maintained constant across the thickness. The adhesive longitudinal normal stress was neglected when compared with the adherend longitudinal normal stress. They concluded that the main difference between the theories that include and those that ignore adhesive thickness effects occurs at the ends of the overlap: the maximum shear stress increases and the peel stress decreases with the inclusion of this effect. The effect of the adhesive thickness is more significant with short overlaps, thick adherends, and stiff adhesives.

One of the drawbacks associated with adhesive joints is owed to the stress singularity which develops at the interface corner due to elastic mismatches, which may initiate failure at the interfaces. Failure can often initiate unexpectedly in a catastrophic manner under relatively low mechanical or thermal service loads.

It is, however, important to understand how cracks propagate, in terms of both their direction and directional stability, which requires more information about the stress state at the crack tip. Williams, in 1957 [6], developed an asymptotic stress expansion equation which characterises the stress state in the viscinity of the crack tip as follows:

where r and q are the polar coordinates, and K_I and K_II are the mode I and mode II stress intensity factors at the crack tip, respectively. The third term is nonsingular and acts parallel to the crack plane. This term is normally referred to as the “T-stress”. A Cartesian coordinate is set at the crack tip and the x direction is that of the crack plane.

The stress state at the crack tip dictates the direction of crack propagation in homogeneous materials, and, following several different investigations over the past few years, three main criteria for the direction of crack propagation have received much attention:

1. maximum opening stress criterion which dictates that the direction of cracking is perpendicular to the direction of maximum opening stress [7];

2. mode I fracture criterion, which states that a crack will propagate along a path such that pure mode I fracture is maintained at the crack tip [8, 9];

3. maximum energy release rate criterion, which states that the direction of crack propagation can be obtained by maximizing the energy release rate as a function of the angle of crack kinking [10].

Although these criteria were developed for homogeneous materials, they are also applicable to adhesively bonded joints, but care should be taken in dissimilar material joint interfaces, due to differences in fracture toughness in the vicinity of the interface.

In the Williams equation shown above, the T-stress is the most important term in determining the directional stability of crack propagation. If the T-stress in negative (compressive), a stable crack growth occurs. If the T-stress is positive (tensile), the crack growth is unstable, and the crack will deviate away from the original path.

The first criterion above related to the loading mode is most commonly used, since they all yield similar experimental results. This states that the direction of cracking depends on the ratio of mode II versus mode I fracture components at the crack tip. This in turn means that the crack path will be straight under mode I, but will deviate from a straight path under mixed mode loading.

The discussion of directional stability of cracks in adhesively bonded joints was described by Chai [11–13], who described the crack trajectory in the mode I delamination failure of graphite reinforced epoxy composite laminates and aluminum/epoxy bonds. In his work, the crack periodically alternated between the two interfaces with a characteristic length 3–4 times the thickness of the adhesive layer. More specifically, as the crack advanced, the crack propagated along one interface and then gradually deviated away with an increasing slope until the other interface was approached. An abrupt kink then occurred when the crack approached the opposite interface, and the crack stayed at the interface for a distance about 2–3 times the thickness of the adhesive layer before deviating from the interface again. This trajectory obviously reflects very directionally unstable crack propagation. However, it was later shown by other workers that the crack trajectory can be affected by the residual stress state in adhesive bonds [14] and the T-stress level [15, 16]

The J integral (a path independent integral) method to calculate the energy release rate (ERR), first presented by Rice in 1968 [17], has been used to assess the fracture behavior of adhesively bonded joints. For a monotonically loaded process, large-scale bulk inelasticities and nonlinear interfacial cohesive separations might be considered by the J-integral method. This approach has given rise to the development of cohesive zone models (CZM), and over the past 30 years numerous studies to investigate the nonlinear fracture of adhesively bonded joints have been carried out [18].

Parrinello et al. [19] developed an interface constitutive model in order to investigate the mechanical behavior of the internal adhesive layers in bonded substrates. Based on the damage mechanics theory this model captured the transition of the adhesive from the sound elastic condition (elastic behavior) to the entirely cracked one (frictional behavior). The interface intermediate mechanical properties during the formation and development of microcracks to macrocracks were calculated. Ouyang et al. [20] proposed a natural boundary condition based model for the local damage evolution of double cantilever beam specimens. Crack initiation and propagation, local deformation, and interfacial stress distribution were modelled as a function of the remote peel load using the bilinear cohesive laws. The nonlinear response during loading and unloading stages was predicted and verified using both experimental and numerical results.

Högberg [21] presented a traction-separation relation. The fracture process for different fracture parameters, such as fracture energy, strength, and critical separation in different mode mixities for unsymmetrical and mixed-mode double cantilever specimens, was modelled. The advantage of the proposed cohesive law is that it can be applied for mode I (peel), mode II (shear), and mode III (mixed mode) and model the fracture process without being altered for the different mode mixities.

Mixed mode double cantilever beam specimens were tested in [22] under shear and/or peel forces. Two different approaches of the J-integral expression were used to evaluate the energy dissipation in the failure process zone. The constitutive properties of the adhesive layers obtained by an inverse method were found to be coupled, i.e. the peel and shear forces depend on the peel and shear deformations.

1.2 Effect of adhesive thickness

The effect of thickness on the fracture performance of adhesively bonded joints has received limited attention over the past few years although there are some interesting pieces of work found in the literature where the global behavior of joints with different adhesive thicknesses have been carried out [23].

Fleck et al. [24] also showed that the T-stress is closely related to the specimen geometry i.e. the thickness of the adhesive layer and the thickness of adherends for DCB specimens. The T-stress increases with the thickness of the adhesive but decreases with the thickness of the adherend for this bonding geometry. This specimen geometry dependence of the T-stress level suggested a variation of the directional stability of cracks as the specimen geometry changes.

Chai [25] studied the mode II and III fracture of three different adhesive systems, i.e. two thermosetting resins, i.e. H3502 and BP-907, and PEEK, which is a semicrystalline thermoplastic. The adhesive bondline thickness varied up to a maximum value of a few microns, so that the bulk fracture behavior was exposed. The adherends were either an aluminum alloy or structural steel. For the thin bondline joints and for all three adhesives, the main source of energy consumption is due to shear yielding or distortion. As the adhesive thickness decreased, the fracture work, Gsc, decreased monotonically after it had reached a plateau. Another important conclusion extracted from this work was the coincidence of the forward and anti-plane shear components for all three cases of adhesive. Chai also incorporated particles into the epoxy adhesive aiming at the alteration of the fracture resistance [26]. DCB specimens with aluminum substrates bonded with two different types of adhesive under mode I, II, and III were evaluated. Shear strain increased with decreasing the bondline thickness, and it was higher for the FM1000 adhesive (elastic-plastic behavior) than for the F185 adhesive (strain hardening). G_IIC and G_IIIC, which were not affected by the bondline thickness, were found to coincide. In addition, when the bondline thickness was decreased as to approach the micrometre range, the three fracture energies obtained the same value.

Ikeda et al. [27] studied the J-integral and the near-tip stress of a crack in a joint with hard substrates, which were adhesively bonded with a ductile adhesive. Edge cracked plate (ECP) and tapered double cantilever beam (TDCB) types of specimens were tested. It was determined that stress distributions near a crack tip for both ECP and TDCB agreed with each other when the bond thickness was the same. However, a decrease of the bond thickness increased the stress ahead of a crack tip, resulting to the decrease of fracture toughness. Therefore, it was concluded that stress distribution did not depend on the shape of the specimens, but on the bond thickness and the J-integral.

Kafkalidis et al. [28] used an embedded-process zone (EPZ) to investigate the mode I parameters for plastically deforming adhesively bonded joints. Steel and aluminum substrates were bonded together with a thin layer of adhesive in order to manufacture symmetrical joints. It was found that the “intrinsic” toughness of these systems was not affected by either the adherend properties or the adhesive thickness. Cohesive zone elements [11] were also embedded in order to simulate the fracture process in adhesively bonded joints with steel substrates while varying the adhesive and adherend thickness. Plastic dissipation was found to be affected by the variation of the thicknesses, i.e. it increased when the adhesive thickness increased, until it decreased to reach a plateau for very thick adhesives. It also increased with the decrease of the adherend thickness.

Recently, work by Ji et al. [30] assessed the bondline thickness effect through simple DCB tests.

Fig. 5

Schematic of DCB specimen.

The analysis is based on the J-integral approach as developed by Rice [17]:

where W(x,y) is the strain energy density, x and y are the coordinate directions, T = nσ is the traction vector, n is the normal to the curve or path Γ, σ is the Cauchy stress, and u is the displacement vector. For the geometry of a DCB, Anderson and Stigh [31] developed an expression for the J-integral as follows:

where J is the energy release rate of the DCB specimen during crack initiation process, P is the global peel load at the loadline, D is the adherend’s bending stiffness, θ₀ and θ_P are the relative rotations between the upper and lower adherends at the crack tip and at the loadline, respectively, and d is the crack tip opening or separation. The results of that work have shown that the fracture energy and the characteristic energy release rate both increase with increasing bondline thickness. However, it was found that the interfacial strength is high at small bondline thicknesses (< 0.1 mm) and then drops rapidly at a bondline thickness of 0.2 mm, while thereafter the increase of the adhesive thickness marginally reduces strength.

In contrast, work by Afendi et al. [32], although it demonstrated the effect of bond line thickness on failure stress, failed to show any effect on the fracture performance of bonded joints where JIC remained constant for bond line thicknesses from 0.1 mm to 1.2 mm. Afendi et al. based their work on the analysis carried out by Akisanya et al. [33], who introduced the effects of the interface corner toughness H_C , which is described through a parametric expression of the form

where Q is a nondimensional constant function of the material elastic parameters, t is the interface thickness, σ_C is the applied stress at the free corner, and q is a constant.

Despite extensive work being carried out, the mechanisms of the dependency of joint strength and fracture on bondline thickness are not yet clarified.

Fig. 6

Variation of J_o and J_c with bondline thickness [31].

1.3 Carbon nanotube-enhanced adhesively bonded joints

In the past decades several research works have been carried out focusing on the effects of incorporating nanoparticles in polymers. By tailoring the adhesive properties, a stronger interfacial bonding between the matrix and the filler can be achieved, and this potentially leads to an increased load-bearing capability of the joint [34–37].

Various nanofillers have been used up to now in order to improve the adhesive properties of the joint. In [38] the effect of nano-SiC particles was evaluated in terms of mechanical and thermal properties of both lap joints and bulk matrix. A comparison of the micro- and nano-SiC was also investigated, with the nano-SiC composites having better properties. Lap shear strength was improved, when the adhesive was reinforced with silane treated SiC particles reaching a maximum value at 20 wt.%. A further increase of the particle content though led to a decrease of the lap shear strength.

Yu and Wang [39] added 10 wt.% of various particles, such as Al, Ag, and Ni (microsized) and Ag and Al (nanosized) into the neat resin in order to investigate their effect on wear rate and the several mechanisms of wear, such as deformation, delamination, and adhesion. The presence of Ag particles increased the wear rate of polyethylene (LPDE) when compared to neat resin, whereas the Al particles decreased. Nanoparticles seemed to be again more effective than microparticles.

Another type of nanofiller which is also used as reinforcement is carbon nanotubes (CNTs), due to their remarkable mechanical, thermal, electrical, and optical properties. In general, polymer-based nanocomposites exhibit much better mechanical, thermal, and multifunctional properties compared to those of the polymer matrices reinforced with microparticles. However, CNTs have also been used in adhesively bonded joints to enhance their mechanical behavior. Srivastava [40] bonded carbon/carbon and carbon-silicon carbide composites using both pure epoxy resin and resin reinforced with 3 wt.% of CNTs. He reported that CNTs increased the joint strength and fracture toughness of the joints when compared to values obtained from the joints bonded with neat resin.

1.3.1 Toughening mechanisms

The addition of nanofillers to base adhesive formulations generally improves the modulus and mechanical strength. However, the main objective in these cases is to increase fracture toughness without any decrease of the adhesive characteristics. The increase in fracture toughness is often due to different toughening mechanisms which take place during the test. According to Gojny et al. [41], the toughening mechanisms that arise when CNTs are added into the matrix act at two different dimensional levels, i.e. the micro- and nanolevels. Crack pinning, crack deflection, and crack blunting are considered to be micromechanical mechanisms. Other mechanisms, such as pullout, interfacial debonding, void nucleation, and crack bridging, do exist in the microlevel; however, they can occur at the nanolevel as well.

Many researchers have investigated these toughening mechanisms in order to determine the relationship between the microstructure and fracture behavior. A brief review to explain how the CNT composite performance is affected is described below.

During the crack-pinning mechanism particles arrange in lines and act as obstacles for the crack front. Therefore, the crack front has to bow locally between the filler particles in order to pass through the line they form (Fig. 7). The particles are considered as toughening agents, because the crack front remains pinned at them resulting to an increase of the crack length. Secondary cracks can also be generated which coalesce after passing the particles [42]. As the strain energy increases, local step fracture occurs, and the pinned points are released, creating a “tail-like” feature on the fracture surface. During this process the propagation rate is slowed down, which finally leads to an increase of the fracture toughness due to the absorbed amount of energy.

Fig. 7

Crack pinning process.

Faber and Evans [43] proposed another mechanism, called the crack-path deflection mechanism, which can also contribute to the improvement of the fracture toughness. According to this mechanism, the particles cause the crack to deviate from its original path. As the crack front approaches the particle/matrix interface, the crack is forced to change direction and pass around them along the interface. This deflection continuously changes the local stress state from mode I (crack opening) to mixed-mode. Therefore, more energy is absorbed to propagate a crack under mixed mode conditions than under pure mode I, which results in a higher fracture toughness of the material. Although this mechanism does not seem to depend on the particle size, it is believed that uneven spacing provides better results than uniform spacing [44, 45].

A key factor that greatly affects the toughening mechanisms is the nature of the interface region between the particle and the matrix. The interface must be of sufficiently low toughness, so that the particle will not be able to slide both too easily and with too much difficulty. In the case of a very strong bonding between the matrix and the CNT, fracture of the outer layer of the tube or even a complete rupture of the CNT can occur.

When the CNT/matrix bonding is weaker, another toughening mechanism called pullout (Fig. 8(a)) may be favored. In this case the CNT is pulled out of the matrix leading to a partial interfacial debonding and enabling the particle bridging mechanism (Fig. 8(b)), where the particles are stretched between the edged of the propagating crack, and therefore, more energy is absorbed prior to failure.

Fig. 8

(a) Pullout (i) Initial state of the CNT (ii) Pullout caused by CNT/matrix debonding in case of weak interfacial adhesion, and (b) Crack Bridging

Microcracks, which can be created close to the crack tip, are able to reduce the stress intensity, because they allow the residual stress release resulting to the fracture toughness improvement. Certain size and spatial distributions of microcracks in the vicinity of the main crack tip act as a hindrance to the crack and reduce the crack propagation rate.

Another mechanism that might occur is crack blunting (Fig. 9). During crack propagation, macromolecular chains in the vicinity of the crack tip are stretched and broken. The initial sharp crack becomes more and more blunted as a result of the formation of a plastic zone and decohesion of particles. The stress concentration effect at the crack tip becomes lower and the crack is slowed down.

Fig. 9

Crack Blunting; (a) the crack at the beginning, (b) the blunted crack

Particles may also cause localized plastic deformation of the polymer matrix [46] (i.e. shear banding and crazing) and as a result enhance the fracture toughness. Shear banding is a narrow zone of intense shearing strain, usually of plastic nature, developed during severe deformation of a glassy polymer and results in partial orientation of the polymer chains [47]. Crazes are microscopic regions of highly localized plastic deformation similar to those developed on a macroscopic scale in glassy polymers. Crazing occurs in regions of high hydrostatic tension, or in regions of high localized yielding, which leads to the formation of microvoids (Fig. 10) oriented parallel to the tensile direction. If an applied tensile load is sufficient, these crazed regions elongate and break, causing the microvoids to grow and coalesce and cracks begin to form. Inclusion of rigid particles induces stress concentrations and alters the local stress state, which favors local plastic deformation. Because of the large number of particles in nanocomposites, potentially more plastic deformation may exist in these systems than in the unfilled polymer, leading to higher fracture toughness.

Fig. 10

De-bonding and Void-Growth mechanism.

Despite the various toughening mechanisms proposed by different researchers, it is difficult to explain an experimental result based only on one theory. Fracture is a complex phenomenon, and more than one of the mechanisms that were described above may take place at the same time.

The aforementioned toughening mechanisms that change the properties of the CNT/epoxy polymer can also affect the joint performance, when this polymer is used as the adhesive. Hsiao et al. [48] investigated the mechanical properties of a CFRP (carbon fiber reinforced polymer) joint with CNT/epoxy adhesive. It was found that the shear strength was increased by 31.2 % and 45.6 %, when 1 % and 5 wt.% MWCNTs, respectively, were added into the epoxy. This increase is related to the enhanced mechanical properties of the nanoreinforced adhesives and the change of the failure mode of joints. The failure mode of the bonded joints with nonmodified epoxy adhesive was adhesive, i.e. occurred at the epoxy along the bonding interface and no significant damages were observed on the composite adherends. In contrast, the failure mode changed to cohesive in the case of nanoreinforced adhesive. The nanotubes effectively transferred the load to the adherends, leading to the composite failure.

1.3.2 Functionalization and dispersion

It has been reported in many studies that after the addition of CNTs into the matrix, many properties of the material are improved. In [46], the average fracture toughness of 1 wt.% and 3 wt.% MWCNT/epoxy composites is 1.29 and 1.62 times that of pure epoxy respectively, and the 0.5 wt.% MWCNT/epoxy composite fatigue lives are 10.5 and 9.3 times of the average fatigue life of neat epoxy. However, CNT addition into polymers does not always improve the properties of the composite material. There are also studies confirming that the increase of some of the properties is not that significant, or it does not exist at all, especially when the weight percentage of the CNTs is increased more than a certain value. In [49], the experimental results reveal that varying the weight percentage of the nanofillers into the epoxy adhesive favorably influences the shear strength and modulus. On the other hand, the results also indicate that increasing the amount of the nanofillers beyond a certain weight fraction, a drop in the values of the properties is observed.

This inconsistency among the results in different studies may be attributed to the factors that determine the CNT performance. These factors are the state of dispersion and orientation of CNTs into the matrix. The interaction of CNT with the matrix as well as the interface between the nanoparticles and the matrix also play an important role. Uniform CNT dispersion is very challenging, because as-produced CNTs are often held together in bundles by strong van der Waals forces. Many methods have been developed to assist dispersion, and they can be generally classified [29] into three categories.

1. Direct mixing: CNTs are dispersed in the polymer matrix by a mechanical force, such as shear-intensive mechanical stirring using a dissolver disk or ultrasonication [50]. However, the quality of dispersion is not always satisfactory and if either too aggressive or long sonication takes place, CNTs can be seriously damaged.

2. Chemical surface modification: treatment of CNTs with strong acids, such as nitric acid, or with other strong oxidizing agents, such as KMnO₄/H₂SO₄. The downside of this method is that the oxidative treatment tends to disrupt the conjugated electronic structure, shorten the CNTs, and deteriorate their electrical and mechanical properties.

3. A third component is added to assist the dispersion of CNTs in solvents and polymer matrices. The third component might be surfactants, polyelectrolytes, or surfactant-like block copolymers [54]. These chemicals are beeing adsorbed onto the walls of CNTs during sonication, and the dispersion is stabilized by repulsive electrostatic interactions between the surfactants adsorbed on nanotubes. The use of block copolymer Disperbyk 2150 (alkylammonium salt of a low-molecular weight polycarboxylic acid polymer) was investigated in [51] and [52] and found to improve dispersion.

Konstantakopoulou and Kotsikos [55] have also reported a beneficial effect of introducing multiwall carbon nanotubes (MWCNT) in epoxy resin to bond steel to glass-reinforced composite joints. They evaluated several wt.% contents of MWCNTs in epoxy, and an improvement on shear strength was found in all cases. An optimum strength was achieved with a MWCNT content of 0.1 % (Fig. 11).

Fig. 11

Effect of addition of MWCNT (%wt) on the failure strength of lap shear joints.

By examining the fracture surfaces of the single lap joints, a change in the failure mode was also observed. In the case where the adhesive is pure epoxy resin, the joints failed adhesively (Fig. 12(a)), but after the addition of MWCNTs, this failure mode changed to cohesive where parts of the adhesive are on both substrate surfaces (Fig. 12(b,c)). This suggests that CNTs change the failure mechanism leading the crack to move from one adhesive/adherend interface to the opposite interface by propagating through the adhesive layer.

Fig. 12

Adherend surfaces after failure: (a) epoxy resin, (b) 0.1 CNT wt.%, (c) 0.3 CNT wt.%.

It was noted, however, that there is a large scatter observed in the experimental results. This was attributed to insufficient dispersion of the MWCNTs of the results.

Examination of samples of adhesive in a scanning electron microscope (SEM) have shown regions of agglomeration of MWCNTs that act as initiators of cracks (Fig. 13).

Fig. 13

Scanning electron micrographs of MWCNT filled epoxy resin (RS-L135) fracture surface showing aggregated area (0.3 wt.% and sonication time = 30 min): (a) ×3,500 magnification, (b) ×10,000 magnification.