A Comparative Study between Knocked-Down Aligned Carbon Nanotubes and Buckypaper-Based Strain Sensors

Carbon nanotubes (CNTs) are one of the most promising materials in sensing applications due to their electrical and mechanical properties. This paper presents a comparative study between CNT Buckypaper (BP) and aligned CNT-based strain sensors. The Buckypapers were produced by vacuum filtration of commercial CNTs dispersed in two different solvents, N,N-Dimethylformamide (DMF) and ethanol, forming freestanding sheets, which were cut in 10 × 10 mm squares and transferred to polyimide (PI) films. The morphology of the BP was characterized by scanning electron microscopy (SEM). The initial electrical resistivity of the samples was measured, and then relative electrical resistance versus strain measurements were obtained. The results were compared with the knocked-down vertically aligned CNT/PI based sensors previously reported. Although both types of sensors were sensitive to strain, the aligned CNT/PI samples had better mechanical performance and the advantage of inferring strain direction due to their electrical resistivity anisotropic behavior.


Introduction
From aerospace to microelectronic applications, the growing demand for multifunctional materials with a set of outstanding properties put carbon nanotubes (CNTs) on the map of the most promising ones [1,2]. While CNTs mechanical performance and piezoresistive response make them suitable for sensing applications, the electric anisotropy of aligned CNTs can be used to infer strain directions [3]. This nanomaterial can be incorporated into polymer-based sensors in several ways: (1) As conductive fillers in polymer nanocomposites [4,5], through mechanical mixing and deposition methods, such as coating, dip casting, and filtration, among others; (2) As Buckypaper films (BP), produced by dispersion and deposition methods, such as vacuum filtration (the most common method), drop casting, or hot-press compression, among others [6,7]; (3) As vertically aligned CNT forests (VA-CNTs), synthesized by chemical vapor deposition (CVD), incorporated into polymeric matrices and substrates [8,9]; and (4) As vertically aligned CNT forests (VA-CNTs), synthesized by laser-oriented deposition (LOD) method directly into the matrices, allowing the incorporation of CNTs due to the covalent bounds formed during the process [10]. Furthermore, a dramatic change of materials resistivity via LOD application is reported in the literature [11].
• Buckypapers, in which CNTs are randomly dispersed with two different solvents and show an isotropic electrical behavior, and; • Knocked-down VA-CNTs, which are highly aligned in one direction and show anisotropic electrical properties (and were developed in our previous work [3]).
The aim is to assess the effect of the isotropic/anisotropic electrical properties on strain sensor behavior. For this purpose, commercial multi-wall carbon nanotubes (MWCNTs) were dispersed in N,N-Dimethylformamide (DMF) and ethanol, which were chosen in order to avoid the use of organic solvents (e.g., Triton X) and, thus, extra washing steps, and then vacuum filtrated to form freestanding Buckypapers, BP DMF and BP ETOH , respectively. After being transferred to polyimide (PI) films, relative electrical resistance and Gauge factor versus strain measurements were obtained. The results are compared with the knocked-down VA-CNT/PI-based sensor.

Aligned CNT/PI Samples Preparation
VA-CNTs were synthesized via CVD, at 750 • C, with a flown gas mixture of ethylene/hydrogen/helium (100/200/55 sccm), in a 10 × 10 mm size silicon patch, previously patterned with Fe/Al 2 O 3 catalyst. The CNT forests were manually knocked down onto PI films by a 10 mm diameter rod. Also, a silver conductive epoxy adhesive (8330S from MG Chemicals) was used as electrodes between the samples and the copper wires and placed at the corners of the CNT patch. A detailed description of this procedure can be found can elsewhere [3].

CNT Buckypaper Samples Preparation
Vacuum filtration was the method used to produce the CNT Buckypapers with different dispersing agents. First, 0.025 g of commercial multiwall carbon nanotubes (NC7000TM from Nanocyl) were dispersed in 100 mL of DMF. The solution was stirred for 5 h and left in an ultrasonic bath (CREST ultrasonics, 240 V, 50/60 Hz) for 2 h [6]. Also, 0.025 g of CNTs were dispersed in 50 mL of ethanol with an ultrasonic tip (Hielscher UP200Ht) at 50% speed during 30 min [18]. The dispersions were filtrated through a porous nylon membrane (45 µm), washed with Millipore water, and then dried at 60 • C. The freestanding Buckypapers, BP DMF and BP ETOH , respectively, were peeled off from the membrane and 10 × 10 mm squares were cut off and transferred onto the center of a polyimide film, PI (75 µm Kapton MP film). To measure the electrical resistivity of the samples, a silver conductive epoxy adhesive (8330S from MG Chemicals) was used as electrodes, as shown in Figure 1. dispersions were filtrated through a porous nylon membrane (45 µm), washed with Millipore water, and then dried at 60 °C. The freestanding Buckypapers, BPDMF and BPETOH, respectively, were peeled off from the membrane and 10 × 10 mm squares were cut off and transferred onto the center of a polyimide film, PI (75 µm Kapton MP film). To measure the electrical resistivity of the samples, a silver conductive epoxy adhesive (8330S from MG Chemicals) was used as electrodes, as shown in Figure 1.

SEM Analysis
In order to characterize the surface morphology of the obtained CNT Buckypapers and the alignment of the knocked-down VA-CNTs, a scanning electron microscopy (SEM) analysis was carried out in a NanoSEM-200 apparatus from FEI Nova (FEI Europe, Eindhoven, Netherlands).

Electrical Resistivity Versus Strain Measurements
A MATLAB software (R2018a, Mathworks, Natick, MA, USA) was used to determine the electrical properties of the samples with an adapted Van der Pauw method. The electrical resistivity, ρ, of the BP and aligned CNT-based sensors were calculated by Equation (1): where d is the average thickness of the BP or of the aligned CNT patch, and the RS is a sheet electrical resistance experimental value obtained by a simple adaptation of a Van der Pauw equation (Equation (2)): where and are the means of the electrical resistance experimental values obtained in the strain and opposite directions, respectively. The electrical resistance values in axial strain and opposite (transverse) directions, Rstrain and Ropp, respectively, were determined using = and ⁄ values. Specifically, ⁄ was obtained from Equation (3): where a and b are the CNT patches square dimensions, which varies with the deformation of the polymeric film. Initially, a0 = b0 = 10 mm, and upon deformation increment, in strain direction: a = a0 + ∆l (∆l is the elongation of the film considering the perfect adhesion between the electrodes and the film), and in the transverse direction: b = b0 − vεlt0 considering the Poisson ratio, v, of 0.34 for PI film, where ε is the mechanical strain and lt0 is the initial width of the film. The k value is obtained from Equation (4):

SEM Analysis
In order to characterize the surface morphology of the obtained CNT Buckypapers and the alignment of the knocked-down VA-CNTs, a scanning electron microscopy (SEM) analysis was carried out in a NanoSEM-200 apparatus from FEI Nova (FEI Europe, Eindhoven, The Netherlands).

Electrical Resistivity versus Strain Measurements
A MATLAB software (R2018a, Mathworks, Natick, MA, USA) was used to determine the electrical properties of the samples with an adapted Van der Pauw method. The electrical resistivity, ρ, of the BP and aligned CNT-based sensors were calculated by Equation (1): where d is the average thickness of the BP or of the aligned CNT patch, and the R S is a sheet electrical resistance experimental value obtained by a simple adaptation of a Van der Pauw equation (Equation (2)): where R vertical and R horizontal are the means of the electrical resistance experimental values obtained in the strain and opposite directions, respectively. The electrical resistance values in axial strain and opposite (transverse) directions, R strain and R opp , respectively, were determined using R strain R opp = R s 2 and R strain /R opp values. Specifically, R strain /R opp was obtained from Equation (3): where a and b are the CNT patches square dimensions, which varies with the deformation of the polymeric film. Initially, a 0 = b 0 = 10 mm, and upon deformation increment, in strain direction: a = a 0 + ∆l (∆l is the elongation of the film considering the perfect adhesion between the electrodes and the film), and in the transverse direction: b = b 0 − vεl t0 considering the Poisson ratio, v, of 0.34 for PI film, where ε is the mechanical strain and l t0 is the initial width of the film. The k value is obtained from Equation (4): where I AB and I AD are the injected currents in the CNT patch in two different directions and V DC and V BC are the respective voltages measured. A more detailed description can be found elsewhere [3]. The samples' process of measuring the electrical resistivity versus strain was quite similar to that described previously [3]. The BP DMF /PI and BP ETOH /PI samples were strained using a manual microtester and the elongation of the samples (∆l) was measured by a digital calliper (Mitutoyo). The relative electrical resistance, ∆R/R 0 , and gauge factor, GF (Equation (5)), which is the ratio between the relative electrical resistance and mechanical deformation, ε = ∆l/l 0 , were evaluated upon increased strain levels: where R 0 is the initial (unstrained sample) electrical resistance. It is important to note that the GF values in the opposite direction of the applied strain were calculated using transverse deformation values and plotted versus axial strain deformation.

SEM Analysis
From the SEM images of the BP DMF and BP ETOH samples shown in Figure 2a,b, the CNTs appeared to be randomly dispersed within the Buckypapers. However, considerable CNT agglomerates can be seen in the BP DMF samples in comparison to the more homogenous surface of the BP ETOH samples, which was also confirmed visually. The nonhomogeneous CNT dispersion observed in those BP DMF samples was what led to the production of Buckpapers using ethanol, BP ETOH .
where IAB and IAD are the injected currents in the CNT patch in two different directions and VDC and VBC are the respective voltages measured. A more detailed description can be found elsewhere [3]. The samples' process of measuring the electrical resistivity versus strain was quite similar to that described previously [3]. The BPDMF/PI and BPETOH/PI samples were strained using a manual microtester and the elongation of the samples (∆l) was measured by a digital calliper (Mitutoyo). The relative electrical resistance, ∆R/R0, and gauge factor, GF (Equation (5)), which is the ratio between the relative electrical resistance and mechanical deformation, ε = ∆l/l0, were evaluated upon increased strain levels: where R0 is the initial (unstrained sample) electrical resistance. It is important to note that the GF values in the opposite direction of the applied strain were calculated using transverse deformation values and plotted versus axial strain deformation.

SEM Analysis
From the SEM images of the BPDMF and BPETOH samples shown in Figure 2a,b, the CNTs appeared to be randomly dispersed within the Buckypapers. However, considerable CNT agglomerates can be seen in the BPDMF samples in comparison to the more homogenous surface of the BPETOH samples, which was also confirmed visually. The nonhomogeneous CNT dispersion observed in those BPDMF samples was what led to the production of Buckpapers using ethanol, BPETOH.  To assess the alignment of the knocked-down CNTs, an SEM analysis was also performed. As shown in Figure 2c, the CNTs were aligned almost in the horizontal direction (parallel to the PI substrate), despite some squashing of the sample due to the knocked down process.

Electrical Resistivity versus Strain Measurements
As plotted in Figure 3, the initial electrical resistivity of the BP DMF /PI was higher than BP ETOH /PI samples and showed more variability, probably due to the nonhomogeneous CNT distribution (as revealed by the SEM images), which compromises the number of conductive paths in the CNT network, thus increasing the electrical resistance. As expected, the knocked-down VA-CNT/PI samples showed the lowest electrical resistivity value due to the high CNT alignment. All these values were according to the ones reported elsewhere [6,18,25]. To assess the alignment of the knocked-down CNTs, an SEM analysis was also performed. As shown in Figure 2c, the CNTs were aligned almost in the horizontal direction (parallel to the PI substrate), despite some squashing of the sample due to the knocked down process.

Electrical Resistivity Versus Strain Measurements
As plotted in Figure 3, the initial electrical resistivity of the BPDMF/PI was higher than BPETOH/PI samples and showed more variability, probably due to the nonhomogeneous CNT distribution (as revealed by the SEM images), which compromises the number of conductive paths in the CNT network, thus increasing the electrical resistance. As expected, the knocked-down VA-CNT/PI samples showed the lowest electrical resistivity value due to the high CNT alignment. All these values were according to the ones reported elsewhere [6,18,25]. As shown in Figures 4 and 5, the relative electrical resistance values in the strain direction of the BPDMF/PI and BPETOH/PI samples always increased, almost linearly, with the strain increments, while they decreased in the opposite direction. The BPDMF/PI samples that reached a mechanical breaking point at approximately 2% of deformation with a sudden steeper slope presented relative electrical resistance values between 17% and 22% in the strain direction and had similar behavior in the opposite direction. The relative electrical resistance of one of the BPDMF/PI samples reached, with a steady slope, values of 40% and −20% in strain and opposite directions, respectively, and higher deformations at break of 5%. These differences between relative electrical resistances in opposite directions are probably due to the Poisson contraction effect. In a similar way, most of the BPETOH/PI samples reached a deformation at break below 2.5%, with a sudden steeper slope, presenting relative electrical resistance values between 14% and 24% in strain direction and had also similar behavior in the opposite direction. One sample showed higher deformations at break of almost 4%, reaching a steady slope, approximately, 23% and −19% of relative electrical resistance in strain and opposite directions, respectively. As shown in Figures 4 and 5, the relative electrical resistance values in the strain direction of the BP DMF /PI and BP ETOH /PI samples always increased, almost linearly, with the strain increments, while they decreased in the opposite direction. The BP DMF /PI samples that reached a mechanical breaking point at approximately 2% of deformation with a sudden steeper slope presented relative electrical resistance values between 17% and 22% in the strain direction and had similar behavior in the opposite direction. The relative electrical resistance of one of the BP DMF /PI samples reached, with a steady slope, values of 40% and −20% in strain and opposite directions, respectively, and higher deformations at break of 5%. These differences between relative electrical resistances in opposite directions are probably due to the Poisson contraction effect. In a similar way, most of the BP ETOH /PI samples reached a deformation at break below 2.5%, with a sudden steeper slope, presenting relative electrical resistance values between 14% and 24% in strain direction and had also similar behavior in the opposite direction. One sample showed higher deformations at break of almost 4%, reaching a steady slope, approximately, 23% and −19% of relative electrical resistance in strain and opposite directions, respectively.
These results are according to the CNT-CNT junctions model for electrical conduction [26]: When stretched in the strain direction, the number of CNT-CNT junctions decreases with strain, resulting in a linear increase of the electrical resistance, while in the opposite direction, this number increases due to the Poisson contraction effect and the electrical resistance decreases.  These results are according to the CNT-CNT junctions model for electrical conduction [26]: When stretched in the strain direction, the number of CNT-CNT junctions decreases with strain, resulting in a linear increase of the electrical resistance, while in the opposite direction, this number increases due to the Poisson contraction effect and the electrical resistance decreases.
As shown in Figure 6, the values of the gauge factor, GF, of the BPDMF/PI samples are almost constant between 7.7 and 9.4 in the strain direction, while in the opposite direction, a higher variability was observed, with values approximately between −24.3 and −16 (Table 1). Regarding BPETOH/PI samples (Figure 7), the values of the gauge factor, GF, are also almost constant in the axial strain direction, varying between 7.2 and 8.2, while in the opposite direction, a higher variability was observed, with values between −17.9 and −15.5 (Table 1). Both BP sensors showed higher sensitivity   These results are according to the CNT-CNT junctions model for electrical conduction [26]: When stretched in the strain direction, the number of CNT-CNT junctions decreases with strain, resulting in a linear increase of the electrical resistance, while in the opposite direction, this number increases due to the Poisson contraction effect and the electrical resistance decreases.
As shown in Figure 6, the values of the gauge factor, GF, of the BPDMF/PI samples are almost constant between 7.7 and 9.4 in the strain direction, while in the opposite direction, a higher variability was observed, with values approximately between −24.3 and −16 (Table 1). Regarding BPETOH/PI samples (Figure 7), the values of the gauge factor, GF, are also almost constant in the axial strain direction, varying between 7.2 and 8.2, while in the opposite direction, a higher variability was observed, with values between −17.9 and −15.5 (Table 1). Both BP sensors showed higher sensitivity As shown in Figure 6, the values of the gauge factor, GF, of the BP DMF /PI samples are almost constant between 7.7 and 9.4 in the strain direction, while in the opposite direction, a higher variability was observed, with values approximately between −24.3 and −16 (Table 1). Regarding BP ETOH /PI samples (Figure 7), the values of the gauge factor, GF, are also almost constant in the axial strain direction, varying between 7.2 and 8.2, while in the opposite direction, a higher variability was observed, with values between −17.9 and −15.5 (Table 1). Both BP sensors showed higher sensitivity in opposite direction of strain (transverse) due to the fact that the GF values in that direction were calculated using transverse deformation despite being plotted against axial deformation. Therefore, a correction of these values with the Poisson ratio, ν, was presented in Table 1 and a similar sensitivity is then observed in the two different directions. Although, in the BP ETOH /PI samples, the CNTs were much more homogeneously dispersed than in BP DMF /PI, the slight differences observed in the electrical properties are probably due to the presence of random agglomerates in both types of Buckypapers, as already reported elsewhere [19]. is then observed in the two different directions. Although, in the BPETOH/PI samples, the CNTs were much more homogeneously dispersed than in BPDMF/PI, the slight differences observed in the electrical properties are probably due to the presence of random agglomerates in both types of Buckypapers, as already reported elsewhere [19].    in opposite direction of strain (transverse) due to the fact that the GF values in that direction were calculated using transverse deformation despite being plotted against axial deformation. Therefore, a correction of these values with the Poisson ratio, ν, was presented in Table 1 and a similar sensitivity is then observed in the two different directions. Although, in the BPETOH/PI samples, the CNTs were much more homogeneously dispersed than in BPDMF/PI, the slight differences observed in the electrical properties are probably due to the presence of random agglomerates in both types of Buckypapers, as already reported elsewhere [19].   In Figure 8, the previous results are compared with the values for the knocked-down VA-CNT/PI-based sensor. Both sensor types are strain sensitive even at low strains, presenting Materials 2019, 12, 2013 8 of 10 relative electrical resistance values higher than some reported in the literature [6] (see also Table 2). The Buckypapers-based sensors show, with a linear trend, higher relative electric resistance values and, consequently, a higher sensitivity between approximately 1.5% and 2.5% of deformation (inset graph in Figure 8). However, the aligned CNT/PI sensors present an almost exponential trend, due to tunneling effect conductive mechanisms that become dominant, and thus, higher sensitivity to higher strains, and as seen in Table 1, their GF values in strain and opposite directions highlighted the sensitive differences due to CNT alignment. * correction of GF values with the Poisson ratio, ν.
In Figure 8, the previous results are compared with the values for the knocked-down VA-CNT/PI-based sensor. Both sensor types are strain sensitive even at low strains, presenting relative electrical resistance values higher than some reported in the literature [6] (see also Table 2). The Buckypapers-based sensors show, with a linear trend, higher relative electric resistance values and, consequently, a higher sensitivity between approximately 1.5% and 2.5% of deformation (inset graph in Figure 8). However, the aligned CNT/PI sensors present an almost exponential trend, due to tunneling effect conductive mechanisms that become dominant, and thus, higher sensitivity to higher strains, and as seen in Table 1, their GF values in strain and opposite directions highlighted the sensitive differences due to CNT alignment.    [30] In Table 2, the GF of the CNT-based sensors of this work are compared with others in the literature. The different fabrication methods and polymer matrices are also referred to.

Conclusions
Carbon nanotube Buckypaper-based strain sensors were successfully produced. Morphological differences between BP DMF and BP ETOH were observed by SEM analysis, specifically the presence of CNT agglomerates in BP DMF . However, these CNT non-homogeneities did not seem to significantly influence the electrical properties, which did not present considerable variability between the two types of BP. Nevertheless, BP DMF -based sensors showed slightly higher GF values (7.7-9.4) compared with BP ETOH -based sensors (7.2-8.2). These values are somewhat lower in the transverse direction, showing the quasi-isotropic electrical behavior of BP-based sensors. Despite showing higher relative electrical resistances (at low strain level) compared to the knocked-down VA-CNT/PI-based strain sensors, these latter showed higher mechanical performance and improved electrical properties. The deformation at break are much higher (8.4%) as compared with BP sensors (2.5%-2.7%). The GF of the knocked-down VA-CNT-based sensors are the highest (16.4) when stretched in the CNT direction. On the opposite direction, GF values are remarkably lower, evidencing the unique electrical anisotropic behavior of VA-CNT/PI-based sensors. Although the electrical conduction mechanisms in BP and knocked-down CNTs are dependent on the number of CNT-CNT junctions, the alignment of CNTs causes a variation of these numbers in opposite directions. Moreover, these aligned CNT sensors also show a tunneling effect that becomes predominant at higher strains. In the case of the isotropic BP sensors, the number of CNT-CNT junctions are identical in orthogonal directions, resulting in similar relative electrical resistance. Specifically, for VA-CNT sensors, in the CNT alignment direction, this number decreases with strain, increasing the electrical resistance, whereas in the opposed direction, this number increases, decreasing the electrical resistance at a slower rate.
This electrical anisotropic behavior of VA-CNT/PI sensors can be a huge advantage, potentially allowing us to identify the direction of applied strain. Their higher deformation capabilities also allow their use as large strain sensors, despite the loss of the linearity behavior.