Effect of Stacking Sequence on Fatigue Performance of CFRP–Al Single-Lap Adhesive Joints: Experimental Study

This study aimed to explore failure mechanisms of carbon fibre-reinforced plastic (CFRP)–aluminium (Al) single-lap adhesive joints which CFRP adherends had different stacking sequences. These results showed that fatigue performance of CFRP decreased as the number of 45° plies increased, which caused the initial failure location to gradually move from the adhesive layer towards the CFRP. Under high load levels, joint-failure models were influenced by the stacking sequence of CFRP; large-area cohesive failure occurred in joints when the CFRP stacking sequence was [0/90]4s and [0/45/−45/90]2s, and delamination failure occurred when the CFRP stacking sequence was [45/−45]4s, due to the weak interlaminar properties of CFRP. However, under low load levels, the stacking sequence of CFRP had little effect on the failure model of the joint, with interfacial failure being the main failure mode for all joints due to weakening of the mechanical interlock.


Introduction
Employing lightweight materials to reduce structure weight is one of the most effective ways to solve the problems of rising carbon emissions and environmental degradation [1][2][3][4]. Carbon fibre-reinforced plastic (CFRP) is a material with exceptional properties such as high strength, light weight, and corrosion resistance. It has been extensively used in the building, aerospace, and automotive industries to solve problems related to structure repair [5][6][7] and reinforcement [8][9][10][11][12][13] or to form new structures for engineering applications [14].
However, in practical engineering applications of composite connections, composite materials cannot completely replace metal materials, so the connection between CFRP and Al has become a hot topic [15]. Compared with traditional connection methods (bolting and riveting), adhesive technology is widely adopted in wings, fuselages, and multi-materialized vehicle bodies due to its great structural integrity and rational design flexibility [16][17][18][19]. Numerous factors affect mechanical properties of CFRP adhesive joints, such as stacking sequence, adhesive thickness, and overlap length. [20][21][22][23][24][25]. Among these, stacking sequence can change the mechanical properties of the composite by altering interlaminar stress, which provides a reference for the design of braided CFRP connection structures [25]. Therefore, many scholars are focused on the study of the effect of stacking sequence on mechanical properties of joints.
Hitherto, Jiang et al. [20] investigated quasi-static failure behaviour of CFRP-CFRP adhesive joints. They revealed that the direction of a fibre contact with adhesive has a significant influence on overall structural strength. Compared with the ±45 plies, when fibre contact with the adhesive is 0/90 plies, the joint had greater strength. The numerical and experimental studies of Ozel et al. [21] explored how CFRP adherends with different stacking sequences significantly affect joint-failure load and stress distribution. The results of that study showed that stresses formed at the ends of the overlap region gradually increased when the stacking sequence of the CFRP adherend changed from [0/90] 8 to [0/45/−45/90] 4 , then to [45/−45] 8 . Zhang et al. [22] studied the tensile performances of adhesive joints by designing three stacking sequences Many studies have focused on effects of the stacking sequence on quasi-static mechanical properties of adhesive joints. However, during long-term service, structures experience cyclic loads that result in 50-90% of mechanical failures and threaten their service life severely [26,27]. At the same time, owing to excellent sealing performance of the adhesive, it is difficult to check for internal failure of an adhesive joint in time. This can result in damage accumulation and overall structure destruction [28]. Therefore, fatigue resistance of adhesive joints is an essential property in connecting CFRP and Al [29,30]. In this regard, Mariam M. et al. [31] investigated tensile fatigue properties of aluminium alloy (AA7075) and GFRP single-lap adhesive joints. Their results indicated that fatigue damage mechanisms in composites were considerably more complicated than those in metal adherends. Fibre breakage, extensive delamination, and matrix cracks led to composite failure, as metal surfaces only experienced mixed mode and adhesive failure. S. Azari et al. [32] investigated the effect of adherend thickness on the fatigue performance of joints under cyclic loads. They revealed that due to greater global stiffness and larger crack-tip stresses of thicker joints, fatigue properties of joints decreased when adherend thickness (h) increased, and when h > 12.7 mm, joint fatigue behaviour was independent of h. Kara E et al. [33] studied the effect of overlap length on static and fatigue behaviours of adhesive joints. They found that increasing overlap length increased static strength but decreased fatigue life. Shin et al. [34] investigated the effect of the stacking sequence on fatigue performance of the steel-composite co-cured joint. They found that under a high load level, a joint with a composite stacking sequence of [±45] 4s had more transverse shear stress than a joint with the [0] 16 [36] performed an experimental study of failure mechanisms in adhesive joints that composited adherends with different stacking sequences under cyclic loads. They revealed that compared with 0 • plies, presence of 45 • plies could slightly extend crack propagation life, but that effect tended to disappear when total fatigue life was considered.
To sum up, compared with influencing factors such as overlap length, adhesive thickness, etc., there are fewer studies on effects of stacking sequences on fatigue performance of adhesive joints, and most mainly focus on the influence of joint fatigue life. Meanwhile, limited attempts have been made to reveal failure mechanisms of joints under cyclic loads.
In this paper, fatigue performance and the failure mechanisms of CFRP-Al singlelap adhesive joints that CFRP adherends with different stacking sequences were studied. For this purpose, first, CFRP laminates with three typical stacking sequences ([0/90] 4s , [0/45/−45/90] 2s , and [45/−45] 4s (where subscript number n indicates repetition of n plies and subscript s indicates symmetrical arrangement of plies) were selected to fabricate adhesive joints, and a series of quasi-static and fatigue tests was carried out for those joints. Second, based on experimental data and the Weibull method, joint mean fatigue life was calculated and SN curves were fitted using multiple function models. Finally, microscopic fracture morphologies were measured by SEM in order to analyse differences in failure modes of the joints, thereby further revealing failure mechanisms of the joints under cyclic loads.  Tables 1 and 2, respectively. Furthermore, Araldite 2015 (Huntsman, SLC, US)), a two-component epoxy resin adhesive that cures at room temperature, was used to bond the adherends. To simplify expression of the joints which CFRP with different stacking sequences, the specimens are abbreviated as J-stacking sequences, as shown in Table 3.

Preparation of Specimens
According to the standard of ASTM D3166 [39], the CFRP laminates were cut into rectangular plates with a size of 120 × 25 × 1.65 mm each. Similarly, Al was cut into plates with dimensions of 120 × 25 × 1.5 mm for joints. The joint overlap length and adhesive thickness were designed to be 15 mm and 0.2 mm, respectively, and end tabs with the same thickness as the adherends were bonded at the free ends of each joint to make the joint better align with the gripper during the fatigue test process. Specific structural parameters are shown in Figure 1.
Before bonding of specimens, waterproof abrasive papers (80 mesh) were used to polish each adherend bonding surface in order to increase its roughness. During the polishing process, polishing direction, polishing intensity, and polishing frequency were kept consistent for all specimens. Acetone was subsequently used to remove detritus and oil contamination from each adherend surface.

Quasi-Static Tensile Test
Quasi-static tensile tests were conducted in a universal testing machine (Instron 9520, USA) equipped with a 100 kN load cell, as shown in Figure 3a. The joints were stretched at a constant speed of 2 mm/min in accordance with ASTM D5868-01 [40]. The Al adherend was fixed and the joint was stretched at the end of the CFRP. Failure loads and displacements of the joints were recorded.

Fatigue Test
The electro-hydraulic servo fatigue testing machine (100 kN) was used to conduct fatigue tests of the joints under different load levels, as shown in Figure 3b. In the fatigue test, the Al adherend was fixed and the joint was cyclically stretched at the end of the CFRP in the same way as in the quasi-static tensile test. According to the experimental results from S. Ç avdar et al. [41], when load levels reached 80% of peak load, specimens To obtain valid test results, 0.2 mm calibration spacers and adherend spacers with the same thickness as substrates were used to control thickness of the adhesive layer. After fixing of the spacers, the joints were pressurized in 0.6 MPa by the upper and lower pressure plates and cured at room temperature. The schematic diagram of the control method of adhesive thickness is shown in Figure 2.

Quasi-Static Tensile Test
Quasi-static tensile tests were conducted in a universal testing machine (Instron 9520, USA) equipped with a 100 kN load cell, as shown in Figure 3a. The joints were stretched at a constant speed of 2 mm/min in accordance with ASTM D5868-01 [40]. The Al adherend was fixed and the joint was stretched at the end of the CFRP. Failure loads and displacements of the joints were recorded.

Fatigue Test
The electro-hydraulic servo fatigue testing machine (100 kN) was used to conduct fatigue tests of the joints under different load levels, as shown in Figure 3b. In the fatigue test, the Al adherend was fixed and the joint was cyclically stretched at the end of the

Quasi-Static Tensile Test
Quasi-static tensile tests were conducted in a universal testing machine (Instron 9520, USA) equipped with a 100 kN load cell, as shown in Figure 3a. The joints were stretched at a constant speed of 2 mm/min in accordance with ASTM D5868-01 [40]. The Al adherend was fixed and the joint was stretched at the end of the CFRP. Failure loads and displacements of the joints were recorded.

Quasi-Static Tensile Test
Quasi-static tensile tests were conducted in a universal testing machine (Instron 9520, USA) equipped with a 100 kN load cell, as shown in Figure 3a. The joints were stretched at a constant speed of 2 mm/min in accordance with ASTM D5868-01 [40]. The Al adherend was fixed and the joint was stretched at the end of the CFRP. Failure loads and displacements of the joints were recorded.

Fatigue Test
The electro-hydraulic servo fatigue testing machine (100 kN) was used to conduct fatigue tests of the joints under different load levels, as shown in Figure 3b. In the fatigue test, the Al adherend was fixed and the joint was cyclically stretched at the end of the

Fatigue Test
The electro-hydraulic servo fatigue testing machine (100 kN) was used to conduct fatigue tests of the joints under different load levels, as shown in Figure 3b. In the fatigue test, the Al adherend was fixed and the joint was cyclically stretched at the end of the CFRP in the same way as in the quasi-static tensile test. According to the experimental results from S. Çavdar et al. [41], when load levels reached 80% of peak load, specimens will immediately fracture, making data collection extremely challenging. Therefore, load levels were reduced from 75% of peak load, and four different load levels were applied to obtain effective fatigue life of the joints (less than 10 6 cycles) for analysis. According to the standard of ASTM D3166 and the numerous literatures that study the fatigue behaviour of Araldite 2015, fatigue tests were carried out under 30 Hz [39,[42][43][44]. The load ratio was set to 0.1, and four samples were tested under each load level to ensure that the test results were repeatable and reliable.

Fatigue Analysis
Weibull distribution, normal distribution, and logarithmic normal distribution are often used as statistical analysis methods to describe the distribution law of test data from the fatigue test [45]. Compared with other methods, Weibull distribution is more extensively applied. In this study, two-parameter Weibull distribution was adopted to deal with joint fatigue life. Probability density and cumulative distribution functions (namely reliability) in the Weibull model can be represented as [43,46]: where t is a random variable that refers to the joint fatigue life, α is shape parameters, and β is scale parameters. Based on Equation (1), reliability R(t) can be estimated as Taking the logarithm of both sides of Equation (3) can obtain the following formula: It can be seen from Equation (4) that lnln [1/R(t)] has a linear relationship with ln(t), and α and β can be obtained through the intercept. However, F(t) cannot be calculated through Equation (2), so the definition replaces reliability in order to describe reliability as where i is the serial number of the specimen under a particular load level and n is the total number of fatigue samples under a specific load level. Shape parameters α and scale parameters β could be calculated through Equation (4), and the mean life (mean time to fatigue, or MTTF) of joints was determined as Polymers 2022, 14, 5088 6 of 17 To evaluate relative dispersion of fatigue life under different load levels, the standard deviation (SD) and coefficient of variation (CV) were used for respective assessment: where Γ is the gamma function [47].  , the failure loads of the joints decreased gradually. Since the fibres bear the main load in CFRP, it can be presumed that the more 0 • plies are in the same direction as the load, the higher the failure load of the joints will be. Meanwhile, it was found that J-[45/−45] 4s exhibited a plastic deformation stage during the test; that is, the load continued at around 3.5 kN, but joint displacement increased from 1.82 mm to 2.08 mm, while J-[0/90] 4s and J-[0/45/−45/90] 2s exhibited a linear elastic stage until a brittle fracture occurred. The main reason for this is that ±45 plies can rotate and deform in the loading direction when the ply angle is changed. Along with tensile stress, the matrix is subjected to compression stress caused by deformation, and eventually, shear or crush failure will occur [48]. The matrix will lose support and restraint of the CFRP adherend, thereby reducing bearing capacity of the CFRP adherend and causing more deformation [49].

Quasi-Static Tensile Behaviour
Polymers 2022, 14, x FOR PEER REVIEW 6 of 17 Since the fibres bear the main load in CFRP, it can be presumed that the more 0° plies are in the same direction as the load, the higher the failure load of the joints will be. Meanwhile, it was found that J-[45/-45]4s exhibited a plastic deformation stage during the test; that is, the load continued at around 3.5 kN, but joint displacement increased from 1.82 mm to 2.08 mm, while J-[0/90]4s and J-[0/45/-45/90]2s exhibited a linear elastic stage until a brittle fracture occurred. The main reason for this is that ±45 plies can rotate and deform in the loading direction when the ply angle is changed. Along with tensile stress, the matrix is subjected to compression stress caused by deformation, and eventually, shear or crush failure will occur [48]. The matrix will lose support and restraint of the CFRP adherend, thereby reducing bearing capacity of the CFRP adherend and causing more deformation [49].

Fatigue Life
Two-parameter Weibull distribution was used to analyse fatigue life of the joints, and SN curves were performed to explore the characterization function suitable for joint fatigue life.  4s , the load level which joint fatigue life exceeding 10 6 cycles increased from 15% to 25% to 35%, respectively. Cycles greater than 10 6 were generally considered to be the infinite cycle of each joint, which meant that no failure would be experienced when a cycle was greater than 10 6 . There was an indication that as the number of 45 • plies increased, the joints became more resistant to damage.  Equation (4) shows a linear relationship between lnln [1/R(t)] and ln(t). Linear fitting was performed to obtain the shape parameter (α) and scale parameter (β) of the Weibull distribution mathematical model, as shown in Figure 5. Based on Equations (6)-(8), MTTF and the CVs of the joints are presented in Table 6. Equation (4) shows a linear relationship between lnln [1/R(t)] and ln(t). Linear fitting was performed to obtain the shape parameter (α) and scale parameter (β) of the Weibull distribution mathematical model, as shown in Figure 5. Based on Equations (6)- (8), MTTF and the CVs of the joints are presented in Table 6.   Through analysis of MTTF and the CVs of the joints under different load levels, it was found that CVs showed an upward trend with an increase of MTTF (as shown in  Through analysis of MTTF and the CVs of the joints under different load levels, it was found that CVs showed an upward trend with an increase of MTTF (as shown in Figure 6a (11) According to the SN curves, the MTTF of the joints did not change significantly under high loads, and the MTTF gradually increased with a decrease in load levels. This means that joints are more sensitive under low loads, which resulted in the CVs of the samples increasing gradually with the decrease in load levels.   Table 7. With increasing load levels for J-[0/90]4s, area of cohesive failure increased, area of interfacial failure decreased, and complete cohesive failure occurred at 100% of the peak load. On one hand, under a high cyclic load, the adhesive defect had a more significant effect on joint failure than did interface defect [51]. On the other hand, due to the larger displacement per cycle at the high load level, crack propagation velocity was higher in the adhesive along the overlap length. These factors resulted in a larger cohesive failure area with increasing load levels. Moreover, under 100% of the failure load, there were a lot of slight cracks in the lap end of the CFRP adherend. This occurred because the lower elastic modulus of the CFRP was the first to deform in the tension process [51].

Failure Mode
For J-[0/45/−45/90]2s, the variation pattern of the joint-failure model at different load levels was similar to that of J-[0/90]4s; with increasing of load levels from 35% to 45% to 55% to 75%, the proportion of cohesive failure areas increased from 20% to 34% to 62% to 85%, respectively. When the load level reached 100% of the failure load, CFRP exhibited 0.0 5.0x10 4 1.0x10 5   According to the SN curves, the MTTF of the joints did not change significantly under high loads, and the MTTF gradually increased with a decrease in load levels. This means that joints are more sensitive under low loads, which resulted in the CVs of the samples increasing gradually with the decrease in load levels.  Table 7. With increasing load levels for J-[0/90] 4s , area of cohesive failure increased, area of interfacial failure decreased, and complete cohesive failure occurred at 100% of the peak load. On one hand, under a high cyclic load, the adhesive defect had a more significant effect on joint failure than did interface defect [51]. On the other hand, due to the larger displacement per cycle at the high load level, crack propagation velocity was higher in the adhesive along the overlap length. These factors resulted in a larger cohesive failure area with increasing load levels. Moreover, under 100% of the failure load, there were a lot of slight cracks in the lap end of the CFRP adherend. This occurred because the lower elastic modulus of the CFRP was the first to deform in the tension process [51].

Failure Mode
For J- [45/−45]4s, with the increase of load levels from 45% to 65%, the proportion of cohesive failure areas increased from 23% to 49%. However, except for in interfacial failure and cohesive failure, delamination failure occurs when load levels reach 75% and 100% of the failure load. On one hand, there were no fibres in the same direction as the cyclic load, causing the matrix to carry the main load. On the other hand, J-[45/−45]4s had weak interlaminar properties at high load level due to fewer intersections between its different layers. Therefore, it was easy for J- [45/−45]4s to delaminate under high loads.    For J-[0/45/−45/90] 2s , the variation pattern of the joint-failure model at different load levels was similar to that of J-[0/90] 4s ; with increasing of load levels from 35% to 45% to 55% to 75%, the proportion of cohesive failure areas increased from 20% to 34% to 62% to 85%, respectively. When the load level reached 100% of the failure load, CFRP exhibited slight fibre failure, indicating a weaker performance when the stacking sequence was [0/45/−45/90] 2s as opposed to [0/90] 4s . In general, extension of cracks in the matrix is prevented when there are more intersections between different plies, but this factor was not applicable to this phenomenon, indicating that the number of fibres in the same direction as the cyclic load has a greater effect on joint fatigue performance than the number of intersections between multiple layers [50].
For J- [45/−45] 4s , with the increase of load levels from 45% to 65%, the proportion of cohesive failure areas increased from 23% to 49%. However, except for in interfacial failure and cohesive failure, delamination failure occurs when load levels reach 75% and 100% of the failure load. On one hand, there were no fibres in the same direction as the cyclic load, causing the matrix to carry the main load. On the other hand, J-[45/−45] 4s had weak interlaminar properties at high load level due to fewer intersections between its different layers. Therefore, it was easy for J- [45/−45] 4s to delaminate under high loads. Figure 9 shows typical microscopic fracture morphologies of the joints, obtained through SEM. In the enlarged view of Figure 9a,c,e, it can be seen that compared with J-[0/90] 4s , the failure location of J-[0/45/−45/90] 2s was close to the CFRP adherend. For J- [45/−45] 4s , it could be observed that the matrix partially failed, and some parabolic shape cracks existed on the fibres, revealing that the CFRP adherend was subjected to shear stress [52]. It was shown that the initial failure occurred at the CFRP adherend near the adhesive. This phenomenon indicates that from J-[0/90] 4s to J-[0/45/−45/90] 2s , then to J- [45/−45] 4s , the location of initial damage gradually moved from the adhesive to the CFRP. The main reason for this is that the location of initial failure was biased toward the weakest part of the joint, and overall fatigue properties of CFRP gradually decreased as the number of 0 • plies declined.

Fracture Analysis
Under the low load level, some holes (Figure 9a,c) in the adhesive could be seen due to thermal expansion of the cavities in the adhesive, indicating that temperature inside the joints had an upward trend after massive cycles [53]. In addition, the Al surface was smooth, without any adhesive residue, as shown in Figure 8, demonstrating that adhesion strength decreased at the low load level: a phenomenon caused by weakening of mechanical interlocking of the interface. The cross-sections of the adhesive joints as shown in Figure 10 could explain the weakening phenomenon of mechanical interlock. Figure 10a shows interface morphology of the Al and of the adhesive layer of the joint not subjected to cyclic load, while Figure 10b displays those of the joint subjected to cyclic load at 35% load level. This image shows that there were almost no gaps at the interface of the joint not subjected to cyclic load, resulting in more compact mechanical locking. However, at low load level, extensive gaps were found at interface, indicating the poor penetration of adhesive into grooves. This occurred because with an increase of temperature, adhesives enter high-elasticity instead of glassy states, which causes Al and adhesive surfaces to fail to bond closely; meanwhile, strength of adhesion produced by the mechanical interlock between the adhesive and the Al surface declines [54][55][56]. Therefore, fatigue cracks propagate to the Al-adhesive interface through the adhesive layer, and interfacial failure will eventually occur. At the same time, there were many fish-scale patterns along the direction of crack propagation, fracture surfaces of each layer were relatively smooth under low load level, and edges of each layer were slightly raised (Figure 9a,c), illustrating that plastic deformation occurred in the joint due to the highelasticity-state adhesive layer. Figure 9b,d,f present micro-failure morphologies of the joint under high load level (75% of failure load). For J-[0/90] 4s and J-[0/45/−45/90] 2s , in this condition, the joints broke after fewer cycles, the temperature inside the joints was insufficient to degrade the interface bonding strength, and adhesive strength was less than interface bonding strength, causing cohesive failure. In addition, it was observed that some sharp particles existed on the fracture surface under the high load level, as shown in Figure 9b,d. This implies that high cyclic loads result in adhesive brittle fractures. For J- [45/−45] 4s, the matrix failed and the fracture surface was relatively smooth, and the fibres were pulled out. This phenomenon can be explained by the fact that the matrix of a CFRP adherend does not act as strong as the adhesive, and when a failure to the joint occurs, fibres are pulled out under great tension [57].   Under low load, damage gradually moves from the initial damage location to the Al-adhesive interface due to a decrease in bonded strength of the Al-adhesive interface, and eventually, interface damage will occur. Therefore, matrix modification could be considered as a method to increase matrix strength and avoid delamination failure of J-[45/−45] 4s under high load level, and the Al-adhesive interface property should be heightened when the joints need to be used in service under low load level [58].   Under low load, damage gradually moves from the initial damage location to the Al-adhesive interface due to a decrease in bonded strength of the Al-adhesive interface, and eventually, interface damage will occur. Therefore, matrix modification could be considered as a method to increase matrix strength and avoid delamination failure of J- [45/−45]4s under high load level, and the Al-adhesive interface property should be heightened when the joints need to be used in service under low load level [58].

Conclusions
In this paper, three types of CFRP-Al single-lap adhesive joint (J-[0/90] 4s , J-[0/45/ −45/90] 2s , and J-[45/−45] 4s ) were prepared for quasi-static and fatigue tests. Based on the Weibull statistical analysis method, effects of different stacking sequences on fatigue life of the joints were studied. Additionally, failure modes and microscopic fracture morphology were obtained in order to reveal the joint-failure mechanism under cyclic loads. Within limitations, some conclusions can be drawn, as follows: