Effects of cellulose nanofibers on flexural behavior of carbon-fiber-reinforced polymer composites with delamination

Understanding the influence of delamination defects on the damage evolution behavior of carbonfiber-reinforced polymers (CFRPs) is crucial to improve their engineering applications. This study examined the flexural damage behaviors of CFRP composites by using a combination of acoustic emission (AE) and X-ray microcomputed tomography (micro-CT). Four specimens with different delamination defects and 0.1 wt% cellulose nanofibers (CNFs) were subjected to three-point bending tests. AE was employed to monitor the loading process, and then, micro-CT was utilized to detect the internal damage. The results showed that for the specimens with preset delamination defects near the surface, CNF-reinforced specimen exhibited no obvious enhancement effect on bending strength, and its cumulative acoustic energy decreased by 28% compared with that of CFRP specimens. For the specimen with preset delamination damage in the middle position, CNFs had an obvious enhancement effect on mechanical behavior, and the cumulative acoustic energy decreased by 43%. No obvious kink band was observed in the CNF-reinforced specimens, and during crack propagation, causing cracking and delamination damage was difficult. The results of micro-CT are consistent with those of AE. The results combined the combination of AE and micro-CT reflect the superiority of the hybrid detection system.


Introduction
Carbon-fiber-reinforced polymer (CFRP) composites have been used to realize the lightweight of ships, automobiles, and aircrafts as well as in other applications due to their advantages of lightweight high strength, and high corrosion resistance, and they have shown strong irreplaceability [1]. These composites have been successfully applied to the supporting structure of roller coasters and glass viewing facilities. However, there are still some limitations in basic research. CFRP laminates are prone to delamination in the matrix region because of the weak interface properties, which seriously lower the stiffness and strength of the structure [2]. Amaro et al. [3] studied the bending failure behavior of CFRP laminates with delamination defects in different positions. They found that the existence of defects reduced the maximum load by approximately 17-24% on average compared with that in the undamaged case. Bolotin classified delamination into internal delamination, near-surface delamination, and multiple cracking [4,5]. Moreover, when the depth of delamination is <25% of material thickness, delamination is highly likely to expand [6]. A challenging task in this field is to increase the resistance of laminated composites against delamination. Three different techniques have been proposed to enhance the interface performance, namely three-dimensional braided structure, fiber surface treatment, and modification of a polymer matrix by using nanoparticles. Gao et al. [7] co-grafted carbon nanotubes and graphene oxide onto the surface of carbon fibers to prepare a carbon fiber/graphene oxide/carbon nanotube multi-scale reinforcement material. The interlaminar shear strength and interfacial shear strength of the resulting composite increased by 48.12% and 83.39%, respectively. The introduction of graphene oxide and carbon nanotubes improved the wettability and contact area of fibers and resins, thereby considerably improving the mechanical properties of the graphene oxide/carbon nanotubes/carbonfiber-reinforced composite. Borowski et al. [8] verified the reinforcing effect of multiwalled carbon nanotubes (MWCNTs)modified epoxy resin on CFRP laminates through experiments and numerical simulations. By adding 0.5, 1.0, and 1.5 wt% MWCNTs, the maximum interlaminar fracture toughness of the CFRP composites increased by 25,20, and 17%, respectively. Zhou et al. [9] investigated the effects of MWCNTs-, graphene-, and cellulose nanofibers (CNFs)modified epoxy resin on the damage evolution of CFRP laminates. The results showed that CNFs had the best reinforcing effect on laminates because their high specific surface areas increased the mechanical interlock between the fiber and matrix, preventing sliding between them and eventually improving interface properties. The aforementioned studies have confirmed that nanoparticles enhance the mechanical properties of composites. However, CFRP causes all types of damage during practical applications. Under the influence of internal damage, the enhancement effect and mechanism of nanoparticles remain fields to be studied.
Acoustic emission (AE) technology is suitable for nondestructive testing and can be used continuously to test the dynamic response of structures. During loading, the material releases strain energy due to microstructural changes, leading to the propagation of stress waves [10]. AE signals are generated not only through structural changes but also from other sources, such as electrical interference and friction of testing machines [11]. The cluster analysis of the AE signals generated from structural changes can well reflect the evolution of composite structural damage [12][13][14]. Fotouhi et al. [15] adopted the AE technique to investigate the effects of nanofiber ply interface reinforcement on mode I properties and failure behavior when subjected to fatigue and static loads. The delamination toughness of nanomodified specimens increased by 130% under static loading, and relevant crack growth resistance ability improved under fatigue loading. Saeedifar et al. [12] utilized AE technology to study the clustering of indentation-induced inter-and intra-laminar damages in CFRP. Then, the AE signals were clustered to track the initiation and evolution of different damage mechanisms. This was achieved using six clustering methods, namely k-Means, Genetic k-Means, Fuzzy C-Means, Self-Organising Map, Gaussian Mixture Model, and hierarchical model. The results showed AE as an effective tool to determine the structural health of CFRP laminates when appropriate clustering methods are used. To visualize the internal microstructure and damage of composite materials, X-ray micro-computed tomography (micro-CT) is used in the field of structural health monitoring [16][17][18]. Duan et al. [19] used AE, digital image correlation, and micro-CT to study the bending progressive damage behavior of 2.5D SiC f /SiC braided composites. Qin et al. [17] investigated the influence of nearsurface delamination on the buckling response behavior of CFRP composites under different bending modes by using AE and micro-CT techniques. Matrix failure, interface failure, and fiber failure were the main damage mechanisms. Zhang et al. [20] combined AE and micro-CT techniques to characterize the progressive bending damage of three-dimensional braided composites. The results showed a strong correlation between the morphology of microdamage observed by micro-CT and the variation characteristics of AE parameters. The combination of AE and micro-CT techniques can provide detailed mechanical data and internal damage images, which is critical for understanding the failure behavior of composite materials.
Currently, many investigations are aimed at the damage initiation and evolution of CFRP composites embedded with different nanoparticles. However, fewer studies have explored the addition of CNFs to CFRP composites with prefabricated delamination defects. This study uses a combination of AE and micro-CT via three-point bending tests to explore flexural damage behaviors of CFRP composites. Four specimens with different prefabricated delamination defects were used. AE was used to monitor the damage behaviors. Micro-CT was used to identify the internal damage of the composites. Application of the combination of AE and micro-CT can provide valuable insights for the health monitoring of CFRP composites.

Materials and specimens
The CNF-reinforced composite material used in the experiment was prepared through hand lay-up and vacuum pouring. The reinforced fiber had 10 layers of a T300-3k carbon fiber plain fabric that were arranged in parallel in the same direction (0°), and the mass ratio of the 3329A epoxy resin to 3329B curing agent was 100:40. The length and diameter of CNFs were 1-5 μm and approximately 37 nm, respectively. The scanning electron microscopy (SEM) images of CNFs are shown in Figure 1. A 20 mm × 5 mm PTFE film, which was preset between layers 2 and 3 and layers 5 and 6 of carbon fiber fabric, respectively, was used to simulate the delamination defects of near and internal surfaces ( Figure 2).
First, AL-Turaif [21] found that the ultimate tensile stress, ultimate strain, modulus, and toughness of epoxy resin with 0.1% CNF were increased by 121, 73, 64, and 300%, respectively, which was greater than that for any other nanomaterial-reinforced composites in the literature [22][23][24]. Therefore, 0.5 g of CNFs (mass fraction: 0.1%) was placed in a beaker and dissolved in a small amount of alcohol. This solution was treated ultrasonically in a water bath for 1 h to ensure uniform dispersion of CNFs in alcohol. Then, epoxy resin (3329A) was added, and the resulting solution was treated ultrasonically in the water bath at 85°C for 10 h to ensure the uniform dispersion of CNFs in the resin and complete removal of the residual alcohol solvent [25]. After cooling the mixed solution to room temperature, a curing agent (3329B) with a predetermined mass ratio was added. The resulting mixture was completely stirred and defoamed in a vacuum-drying oven. After the preparation of the resin matrix, the CNF-reinforced composite was prepared through hand lay-up and vacuum pouring. After curing at room temperature for 8 h, post-curing in the vacuum-drying oven at 130°C for 6 h [26], and cooling to room temperature, the composite thickness was 3 + 0.1 mm. Finally, the composites were cut into 60 mm × 20 mm specimens. A schematic of the production process of laminate formation is shown in Figure 3. To study the enhancement effect of CNFs on the flexural mechanical properties of composites with delamination, the composite specimens with near-surface and internal delamination were designed and prepared because of the randomness of the delamination position. Near the surface, the delamination depth was 20% of the thickness, and delamination was easy to expand, which was convenient for observing the enhancement effect of CNFs particles. At the delamination position of 50% thickness, internal delamination was selected as a control. The four types of specimens are as follows: specimen A was the carbon-fiber-reinforced composite, and the delamination defect was between layers 2 and 3, as illustrated in Figure 2(a); specimen B was the CFRP composite and had delamination between layers 5 and 6, as shown in Figure 2(b). Specimen C was the CFRP-reinforced by CNFs with delamination between layers 2 and 3, as illustrated in Figure 2(a); specimen D was the CFRP reinforced by CNFs with  delamination between layers 5 and 6, as shown in Figure 2 Although the positions of the prefabricated delamination defects were different, no difference was observed in thickness among the four types of specimens after curing; the measured thickness was 3.0 ± 0.1 mm. Table 1 shows the detailed characteristics of the specimens.

Experimental setup and procedure
According to the ASTM D790 standard, the three-point bending test of the specimen was performed on an LD24 mechanical testing machine, and the loading rate was set to 2 mm/min. According to the standard, the span-tothickness ratio of the specimen was 16:1, and the span was set to 48 mm. The AE signal of specimen damage was monitored in real time by using the AE instrument (DS2-8A). Two AE sensors (RS54A) were used, as shown in Figure 4(a), and the frequency was 100-900 kHz. The sensors were attached to the specimen surface by using vacuum silicone grease to ensure strong acoustic coupling. The gain of the AE preamplifier of the sensor was 40 dB, and the signal acquisition threshold was set to 10 mV (40 dB) to effectively eliminate the electromagnetic and mechanical noise [27]. The sampling frequency of the AE signal was 3 MHz; the parameters of peak definition time, hit definition time, and hit lock time were set to 30, 150, and 300 μs, respectively [28]. The Skyscan1172 X-ray system was used for micro-CT scanning and three-dimensional reconstruction of the specimen. After the 136-μA beam current generated through 75 kV accelerating voltage penetrated the specimen, its transmission signal was received by the detector with a resolution of 6.8 μm. The specimens were scanned through X-ray at a rotation step of 0.2°, and the scanning was finished after 180°rotation. The collected tomographic images were reconstructed using CTvox software to realize the visual representation of the internal damage of the composite specimens. The working principle of micro-CT is illustrated in Figure 4(b). Figure 5 displays the representative load-displacement curves and the average flexural strength of the four specimens. Initially, each curve showed linear growth; however, with increasing load, all four curves showed obvious differences. The load of specimen A decreased suddenly when it was loaded to 1.1 kN, which indicated that the first and second sub-boards of specimen A were destabilized. This was because under the three-point bending loading, upper fiber of the specimen was subjected to compressive stress. When the first and second sub-boards were pressed, the delamination expanded continuously, and the delamination surface was subjected to additional shear load [17]. Due to  the strengthening effect of CNFs, the first and second subboards of specimen C were destabilized and damaged when it was loaded to 1.3 kN, which led to decrease in the load. Delamination defects affected the stress transfer between sub-boards; for specimens A and C, the first and second boards were subjected to bending load until the break. Subsequently, the eight-layer sub-boards under the delamination defect were subjected to bending load, and the load continued to increase. The maximum failure loads of specimens A and C were 1.63 and 1.71 kN, respectively. The flexural strength of the CNFs-modified composite (683 MPa) increased slightly compared with that of specimen A (652 MPa). The average failure loads and flexure strengths of laminates with different types of prefabricated defects are listed in Table 2. It can be concluded that although CNFs can enhance the interlaminar fracture toughness, the flexure strength of specimen C was only 4.7% higher than that of specimen A due to the presence of near-surface delamination defects.

Mechanical characterization
For specimens B and D with delamination defects between layers 5 and 6, the load increased continuously with increasing displacement and slightly fluctuated when the load reached the maximum value. The ultimate load and bending strength of specimen B were 1.84 kN and 736 MPa, respectively, whereas specimen D were 2.17 kN and 868 MPa, respectively, showing an increase of 17.9%. A remarkable improvement in the mechanical properties of CFRP composites with delamination was achieved by the addition of CNFs. The reasons can be summarized as follows: CNFs addition can effectively improve the surface roughness of the carbon fiber, thus increasing the contact area between epoxy resin and carbon fiber, and further improving the interfacial bonding strength between the fiber and epoxy resin [29]. Moreover, CNFs can be used as a bridging element between the carbon fiber and matrix for the transmission of stress between them [30][31][32]. The high specific surface area of CNFs can be converted into an extremely high interface area in CFRPs, which can further  promote stress transfer. The mechanical properties of laminates are related to the location of delamination defects. Composite materials with near-surface delamination defects are broken by sub-boards under low stress, and the stiffness decreases, which can affect the structural integrity and lead to premature failure.

AE characteristics
Flexural tests with simultaneous AE monitoring offer useful information about different aspects of damage evolution in composite specimens. AE hits and amplitudes provide information for damage characterization. The curves of load, amplitude, and cumulative hit of the specimens versus time are shown in Figure 6. In the linear loading stage, a small amount of AE signals are observed in all four specimens, which can be regarded as noise signals. The main AE signals of specimens A and C appeared much earlier than those of CNFs-modified specimens, which indicated that the introduction of CNFs could effectively mitigate damage initiation. Before reaching the maximum load, specimens A and C had more AE signals than specimens B and D, indicating that serious damage had occurred during loading. In addition, there was a strong correlation was observed between mechanical behavior and AE activity, and every decrease in load caused a sharp increase in cumulative hits. When specimen A was loaded for approximately 90 s, the load decreased once. As mentioned earlier, this may be because the delamination defects were between layers 2 and 3, and the 1 and 2 sub-boards broke first under flexural load, generating more AE signals. High-amplitude signals appeared at the same time, and there were various failure modes inside the specimen [33]. Therefore, the amplitude can reflect the damage signal intensity. The damage evolution of  specimens B, C, and D could be divided into two stages: in the linear loading stage, the distribution of AE signals with time was sparse, and a few high-amplitude signals were observed. This stage can be considered the damage accumulation stage; after the maximum load was reached, the AE signals became denser, the high-amplitude signals increased, and the specimen damage became more serious. Due to the presence of delamination defects, the failures of the four specimens were not explosive. The specimens still had some residual bearing capacity after failure, and consequently, the composite structure continued to absorb energy [17]. This phenomenon should be focused on the health monitoring of composite structures in engineering applications. Although the specimen has the bearing capacity, it continuously generates high-amplitude signals, which indicates that delamination damage may have occurred inside. The energy of the AE signal reflects the strain energy released by materials during damage evolution; thus, it is also widely used to evaluate AE activity [34]. The relationship between load and individual AE energy was different for the four specimens versus time (plotted in Figure 7). Combined with Figure 6, specimens A and C had low amplitude AE signals at the initial stage, and these signals had the characteristics of low energy. These signals comprised noise signals, and they were mainly the damage signals caused by process defects such as microcracks and voids in the composite materials under small loads and a small amount of matrix cracking. When specimen A was loaded for 90 s, a large number of highenergy AE signals appeared, implying that the energy accumulated in the elastic stage was partially released at this stage, and irreversible damage evolution began; serious damage such as delamination, debonding, and fiber breakage also occurred in the specimen. Thereafter, the energy of AE signals was <2,000 mV*ms. However, for specimen B, the AE activity after reaching the peak load was relatively weak, and the energy of the AE signal generated by specimen B was lower than that of specimen A. For the same delamination size, the closer the defect position is to the specimen surface, the faster the damage evolution process and the more active the AE signal. Eventually, the AE energy of specimens A and B was 6457.5 and 5503.05 mV*ms, respectively, at the end of the fracture. The load of specimen C decreased slightly when it was loaded for 90 s, and the energy of the acoustic signal was 2239.11 mV*ms. Figure 6(c) shows that the signal also had a high amplitude. With the continuous increase in load, the high-energy signal increased gradually. The energy of the AE signal was 4314.14 mV*ms at the end of the fracture and that of specimen D was 3624.01 mV*ms, which was far lower than the acoustic energy of specimens A and B. However, the energy of the AE signal is positively correlated with the strain energy released by the material when it breaks; thus, the reinforcing effect of CNFs results in less damage and less defects in the composite materials.
However, the relationship between the damage mode of composite specimens and AE signals cannot be accurately determined using only a single AE parameter such as amplitude or energy. Therefore, k-means clustering algorithm was used to reliably classify AE signals in the bending tests, and the correlation between AE signals and damage modes was analyzed. Similar investigations have been reported [35][36][37]. K-means algorithm finds the optimal clustering center through finite iterations and then completes clustering according to Euclidean distance. Data differences between different types of specimens cause differences in clustering centers, thus leading to differences in frequency ranges. First, the dimensions of multidimensional AE data were deduced via principal component analysis. The results showed that the information contained in peak frequency, amplitude, and rising time/amplitude (RA) value accounted for >80% of all parameter information. According to the fact that the low Davies-Bouldin index and high Silhouette represent well-separated and dense clusters, the optimal clustering number was determined to be 3.
The variation of bending load and peak frequency with time is depicted in Figure 8. Peak frequency can be divided into three stages: the low-, medium-, and high-frequency signals, which correspond to matrix cracking, fiber/matrix debonding, and fiber breakage, respectively [11][12][13]. The frequency ranges of different failure modes of the four specimens are summarized in Table 3. Under low loads, the matrix cracking and fiber/ matrix debonding first appeared in specimens A and C. Before reaching the maximum load, the delamination defect continued to evolve and propagate rapidly because it was close to the surface, and a few fiber breakage signals appeared. Due to the low compressive strength of the fiber, the high-pressure stress concentration in the contact area with the indenter promoted the compressive fracture of the longitudinal fiber of the first and second sub-boards. For specimens B and D with delamination defects between layers 5 and 6, the matrix cracking damage occurred slightly earlier than debonding damage. In addition, the fiber fracture for the specimens with delamination defects near the surface occurred earlier than that for the specimens with delamination defects in layers 5 and 6, irrespective of CNF addition. Table 3 shows that the location of delamination and the strengthening effect of CNFs also affect the frequency range of the failure modes of composite specimens. The frequency range of fiber/matrix debonding and fiber breakage of CNF-reinforced specimens was slightly lower than that of CFRP specimens.
The cumulative energy of the AE signals is a suitable parameter to represent the overall state and integrity of the specimens. Cumulative AE energy of the four specimens in  different failure modes is plotted in Figure 9. With reference to Figure 7, specimen A had a cluster of high-energy AE signals at 95 s, and the accumulated energy of matrix cracking, fiber/matrix debonding, and fiber breakage corresponding to these signals was 24683.8, 3106.1, and 0.24 mV*ms, respectively. Whereas the accumulated energy of matrix cracking of specimen A continued to increase but at a slow rate, the accumulated energy of fiber/matrix debonding and fiber breakage changed only slightly. The cumulative acoustic energy of the three types of damage at the end of fracture was 57918.3, 3284.43, and 23.27 mV*ms, respectively. The cumulative energy of matrix cracking of specimens A and C presented two forms in the process of damage evolution. The cumulative energy of specimen A increased rapidly in approximately 90 s, whereas that of specimen C increased gradually in approximately 150 s. The cumulative acoustic energy of matrix cracking of specimen C was 32% lower than that of specimen A. The delamination in specimen A limited the load transfer between different sub-boards, and the bridging effect of CNFs in specimen C provided another approach for load transfer [38], showing that CNFs improved the integrity of the laminated structure. The cumulative energy curves of matrix cracking of specimens B and D were almost the same, and these both showed stepwise growth. Finally, the cumulative energy of specimens B and D was 50392.17 and 24291.84 mV*ms, respectively. However, for CNF-reinforced specimens, the cumulative acoustic energy of debonding or fiber breakage slightly increased, which indicated that the damage was small but serious. This phenomenon occurred possibly because small CNF particles effectively fill the pores between fiber fabrics, and the fiber/matrix interface was complete. In addition, CNFs increased mechanical interlock between the fiber and matrix and improved interface bonding. The synergistic effect of these two phenomena rendered the performance of the fiber/matrix interface stronger; thus, it absorbed more energy and released more strain energy after being damaged. More importantly, AE technique, as a mature detection technology, can provide detailed information on damage detection, location, and classification, but it does not provide information about the severity of the damage. The cumulative energy parameters used in this study make efforts in this respect.

SEM images
To compare the difference in damage modes caused by the addition of CNFs, SEM images of specimens A and C are presented in Figure 10. Several damage modes, such as matrix damage, fiber/matrix debonding, and fiber breakage, were observed. The carbon fiber of the CFRP composite specimen randomly breaks during bending loading; thus, specimen A had considerable fiber pullout damage [39]. The surfaces of the pulled-out fibers were smooth, and the number of fiber/matrix peelings was large, indicating that the bonding strength between fiber and matrix was low. Therefore, in CFRP composites, the fragile fiber/matrix interaction is an important cause of specimen failure [38]. In contrast, the fracture morphology of CNF-reinforced specimens is different. By imaging the cross-section of specimen C after fracture, it could be observed that after partial fiber fracture, the remaining fibers were still firmly combined with the adjacent fibers; the surface of the fibers was rough, and the epoxy resin matrix adhered to them. The results of high magnification show that although a small amount of fiber/matrix interface debonding can still be observed, the fiber/matrix interface is firmly bonded to a large extent. Besides the residual resin on the fiber surface, there is also a small amount of CNFs, which increases the roughness of the fiber surface and further expands the contact area between the fiber and the matrix, thus forming a more cohesive interface. Therefore, the fracture failure mode of specimen C is different from that of specimen A to some extent. Uniform dispersion of CNFs and strong adhesion to epoxy matrix are necessary to improve the fracture toughness of materials. An optimal amount of CNFs can substantially improve the mechanical properties of materials and inhibit crack growth. However, excessive CNFs can lead to agglomeration in epoxy resin, adversely affecting the mechanical properties of the composites [39,40].

Internal visualization based on micro-CT
AE can well describe the damage evolution process. To verify the findings of AE, micro-CT was used to visualize the damage inside the specimens. The micro-CT images of failure modes of the section of specimen A at 25, 50, and 75% in the width direction are shown in Figure 11. The main failure modes of specimen A were kink bands near the indenter, along with a large number of matrix cracking and fiber fracture damages. Due to poor adhesion between the fiber bundle and the matrix, debonding occurred; fiber buckling also affected the adjacent fibers; therefore, a kink band was formed along the width direction of the specimen. The second sub-board was subjected to additional shear stress at the same time when it was pressed. The existence of prefabricated delamination caused stress concentration at the end of the delamination defect, from which, the defect began to cause damage and evolve, and the microcracks continued to expand upward and downward until the first and second sub-boards broke. As shown in the image, at 25% in the thickness direction, some black areas, which should be holes caused by processing defects, are observed. These defects are mainly interlayer holes. Epoxy resin may contain moisture and air inclusions. At a certain temperature, when the pressure of volatile gas exceeds the resin pressure, holes will be formed between fiber layers. From 25% of the width of the cross-section to 75% of the cross-section, it can be seen that the microcracks can induce cracking damage of the matrix in the process of propagation along the thickness direction. Cracks always propagate along the direction of the least resistance, and when encountering high-strength fibers, these cracks can deflect along the fiber direction, reducing the interface properties between the fibers and the matrix. When the shear stress exceeded the bearing capacity of the interface between the fibers and the matrix, the fibers slipped relative to each other, resulting in debonding between the fibers and the matrix [41]. At this time, the fiber/matrix interface could no longer transfer stress well, especially after causing more matrix cracking and delamination damage, and the laminate lost its integrity. In contrast, the fibers far away from the indenter exhibited tensile stress, and fiber breakage damage occurred to a large extent at the bottom of the specimens.
The micro-CT images of the longitudinal failure modes of specimen B at 25, 50, and 75% positions in the width direction are shown in Figure 12. Specimens B and A showed the same failure modes, and the kink band appeared in the area near the indenter due to the fiber buckling caused by bending and shear coupling loads. From the cross sections at 25 and 75%, it can be concluded that the kink band damage easily occurs near the indenter end, and the transverse crack starts to propagate from the kink band. After the crack propagated to the preset delamination defect in the middle, it formed a macroscopic crack along the width direction, that is, along the propagation path with less resistance. Similarly, the tensile stress of the sub-boards of the specimen away from the indenter resulted in delamination and fiber breakage. Observation of the three sections indicated that the most serious delamination damage was around the prefabricated defects, where the crack length was much longer than those at other positions, which could be confirmed from the 50% section image. In addition, comparing specimens A and B, the damage of specimen A, especially the propagation of microcracks, was mainly concentrated under the indenter. However, the range of microcrack propagation in specimen B was more macroscopic. The crack length in the middle of specimen B was long, whereas the number of cracks in specimen A was large. Figure 13 displays the micro-CT images of the width cross-section at the location of 25, 50, and 75% for specimens C. As depicted in the figure, the matrix cracking damage of specimen C was less than that of specimen A, which was also confirmed by a cluster of high-energy AE signals generated when specimen A broke in Figure 6. Similarly, the damage in the area below the end of the indenter was increasingly serious. Under the coupling action of bending load and shear stress, the surface of specimen C formed kink bands; however, in the process of damage propagation, there was no matrix cracking along the fiber layer similar to specimen A. This may be attributed to a vast, rigid network formed by CNFs inside the material, which effectively impeded the propagation of damage. In addition, from the sectional images at 25, 50, and 75%, it could be seen that the extent of damage to specimen C at the prefabricated delamination defect was less than that to specimens A and B. The delamination did not extend along the delamination defect. Figure 14 clearly shows the micro-CT images of the width cross-section at 25, 50, and 75% locations for specimen D. Macroscopically, the damage of specimen D was much less than that of the other three types of specimens. The cross-sectional image at 25% shows that the main failure mode of specimen D was the matrix cracking near the end of the indenter, and the microcracks spread downward to form a small delamination area. Interestingly, the crack stopped propagating when it reached the vicinity of the prefabricated delamination defect. At the same time, there were many fiber breakage and subboards fracture damages were more in the area far away from the indenter. The 50% cross-section was located directly under the indenter, and the damage on this cross-section was less; the fiber buckling lead to the roughness of the upper surface of the specimen. In addition, matrix cracking and fiber breakage were observed. The 75% cross-section images show that the microcracks extended to the vicinity of the prefabricated defects and began to extend along the width direction. Generally, due to the specific position of the prefabricated delamination and the strengthening effect of CNFs, specimen D had less matrix cracking damage, and thus, its cumulative AE energy was the lowest among the four specimens. The larger aspect ratio of CNFs could fill the hole of fiber cloth, making the fiber and matrix bond stronger. Therefore, specimens C and D had fewer manufacturing defects than specimens A and B. Additionally, when microcracks were generated in the matrix, their propagation was reflected and impeded by nanoparticles, which consumed energy, thereby limiting the growth and propagation of cracks. The blocking of nanoparticles will leads to shrinkage and plastic deformation of the crack tip [42]. This could also improve the fracture toughness of the resin matrix composites. Moreover, CNFs could be used as a bridge element to transfer the load between  the fiber and matrix, which could ensure laminate integrity and reduce the occurrence of damage.

Conclusions
In this study, we investigate the damage evolution behavior of CFRP laminates with prefabricated delamination defects under three-point bending tests. The effects of CNF addition and delamination position on mechanical behavior and acoustic response are characterized via mechanical experimental data and AE monitoring results. Visualization of internal damage is realized using micro-CT. The main conclusions are summarized as follows. 1. The mechanical properties of laminates are related to the location of delamination. Near-surface delamination reduces the stiffness and toughness of laminates, and the sub-boards near the indenter fails in advance, affecting the structural integrity. In the presence of delamination, CNFs can also effectively improve interlaminar fracture toughness. For the composite specimens with delamination defects between layers 5 and 6, the addition of CNFs increased the bending strength by 17.9%. 2. There is a strong correlation between the results of AE and mechanical behavior. The increase in AE energy is related to the sudden change of load. When the delamination was near the surface, the fiber was more likely to break under lower load. The CNF-reinforced specimens showed low frequency range and cumulative acoustic energy during damage evolution. The cumulative energy parameter of AE provides new research ideas for evaluating the severity of material damage. 3. The observation results of micro-CT are in agreement with the findings of AE. The damage of CFRP materials with delamination is mainly kink band near the indenter area, matrix cracking, delamination, and fiber breakage. The propagation of cracks in the matrix easily induces delamination and matrix damage. No obvious kink band is observed in the CNF-reinforced specimens; the strong interface around delamination inhibited the propagation of transverse cracks, and delamination damage was less.
Future studies should investigate the damage evolution behavior of large composite specimens under actual working conditions. In-situ CT or in-situ SEM techniques are used to monitor the progressive mechanical response and damage evolution process of nanoparticles in composite materials with delamination defects in real time. Finite element method may also be an ideal method to study these problems. These insights are useful for understanding the damage modes of large structures.