Keywords

1 Introduction

Self-sensing cementitious composites change their electrical properties in response to mechanical loadings and, therefore, are becoming a novel and useful class of material for structural health monitoring when compared with conventional technologies [1,2,3]. An essential element in these composites is the functional filler that forms an interconnected conductive network across the cement matrix. In this context, carbon nanotubes (CNT) are particularly promising when compared with other conductive fillers, given their excellent electrical, physical, and mechanical properties [4]. Once incorporated into the composites, in an adequate and well-dispersed concentration, it is possible to obtain nanostructured cementitious materials, electrically sensitive to stress–strain that can ultimately be applied as excellent piezoresistive sensors for monitoring structures.

Yoo et al. [5] investigated the performance of cement pastes containing different carbon-based fillers, and CNT were found to be the most effective filler for increasing conductivity when compared with other particles at the same concentration. Han et al. reported the CNT/cement-based materials’ ability to detect both stress–strain in the elastic regime, as well as structural damages based on irreversible change in their electrical resistance responses [6]. Nalon et al. verified the ability of mortars to detect fire damage and exhibit a residual piezoresistive response, even after their exposure to high temperatures [7]. Le et al. [8] produced ultrahigh-performance concretes with improved self-sensing properties by combining micro- and nanoscale fibers and functional aggregates. In that case, CNT improved the conductive network at the nano level and, consequently, the conductivity and piezoresistivity of the corresponding composites.

However, the effective application of this material in structural monitoring requires more investigation, which motivated the present study. Several authors report that aggregates, especially the coarse, create obstacles to electrical current flow, extending or interrupting the conductive paths and negatively affecting the self-sensing response [1, 2, 9]. Tian et al. reported that from the paste to the concrete, both the number of publications and the self-sensing performance decreased [3]. On the other hand, the claims regarding the use of aggregates are not unanimous in the literature. Additionally, other researchers suggest that aggregates do not significantly affect the percolation threshold of CNT composites and that their proper incorporation can reduce the matrix voids and promote more contact points between the fillers [4, 10]. Thus, the electromechanical performance of the CNT composites can be improved. The effects of aggregates on the properties of self-sensing composites are complex and therefore require further investigation for better understanding.

Both Han et al. [9] and Wang and Aslani [4] point out that CNT composites with aggregates are much more profitable in real applications than those made of cement paste. According to Dong et al., both the conductive and non-conductive phases are key factors to obtaining self-sensing composites with higher performance [2]. Han et al. reinforced that the proper selection of constituents and the determination of their proportions are crucial for designing self-sensing composites [1]. Han et al. consider that one of the challenges for the development and future implementation of self-sensing composites is their fabrication and, therefore, suggest that upcoming research should include investigations related to the design, optimization, and production of this type of composite, especially those containing aggregates [9].

In this context, aiming to contribute to the smart monitoring of concrete structures, we present the self-sensing properties of paste and mortars containing CNT and evaluate the effects of both conductive filler content and fine aggregates. In addition, the self-sensing test settings and their influence on the quality of the acquired electrical signals were investigated.

2 Methods

2.1 Materials and Sample Preparation

The materials used for this work were high initial strength Portland cement (LafargeHolcim), NC7000 Multi-Walled Carbon Nanotubes (MWCNT; Nanocyl), MasterRheobuild 1000 naphthalene sulfonate-based superplasticizer (BASF Chemicals), crystalline silica flour (S325; Mineração Jundu), and MasterMatrix UW 410 cellulose-based viscosity-modifying agent (VMA; BASF Chemicals). As the fine aggregate, a mixture of fine (#16) and coarse (#100) quartz sands was used in the volumetric fractions of 61.53% and 38.47% respectively.

The MWCNT used in this study consisted of a powder formed by CNT agglomerates, with an average diameter of 9.5 nm, an average length of 1.5 µm, and a surface area between 250 and 300 m2/g, according to the manufacturer. They were dispersed in deionized water using ultrasonication (Sonics Vibra cell, VCX 500) and superplasticizer as the dispersing agent. In the dispersion procedure, 30 g of suspension with a 2 wt% CNT concentration and 2 wt% of superplasticizer solids was subjected to 20 sonication cycles with an average energy of 110 J/g at an amplitude of 40% and pulses of 20 s. In each cycle, the suspension of CNT was homogenized in a low-temperature bath to cool the mixture. The UV–Vis absorbance spectra of the CNT/superplasticizer aqueous solutions were measured before and after the dispersion procedure. For the measurements, an aliquot of the suspensions was diluted in deionized water in the ratio of 1:2000 (% v/v).

To investigate the influence of sand addition on the self-sensing properties of the composites, samples containing 0.75% of CNT were produced, without and with sand (P1 and M1, respectively) at a 1.5 sand/cement (or s/c) ratio, as shown in Table 1. In addition, to evaluate the effect of the CNT content on the mortars’ piezoresistivity, a sample of the same mix design was produced incorporating 0.5% of CNT dispersed by mass of cement (M2). In the second part of the study, to investigate the effects of the test configurations and the sand content, the M3 trait was defined, with an s/c ratio of 1.0.

Table 1 Mix designs of the self-sensing cementitious composites

The P1, M1, and M2 samples were produced as cubic specimens with 50 mm edge and 4 copper electrodes with dimensions of 0.3 × 30 × 50 mm (thickness x width x length) embedded equidistantly (Fig. 1). The M3 sample was cubic with 40 mm edge and copper electrodes with 1050 mm2 (30 × 35 mm) of embedded area. All mixtures were mixed manually, using a glass rod, into which the previously homogenized fine solids (cement, mineral admixture, and VMA) were incorporated into the aqueous dispersion of CNT and additional water and mixed for 10 min. The sand, when included, was incorporated into the mixture after the previous process and homogenized for another 3 min. Finally, compaction was performed by vibration, before and after the copper electrode insertions. After demolding, the samples were cured in a humid chamber at room temperature.

Fig. 1
Three photographs of four copper electrodes, the setup of a mold, and the sample produced as cubic specimens.

Copper electrodes (a) and views of b mold setup and c a typical produced sample

2.2 Self-sensing Test

Self-sensing tests were executed by applying cyclic compressive loadings with simultaneous acquisition of electrical signals using the embedded copper plates as electrodes. Before the self-sensing tests, all samples were dried at 60 °C for 3 days to eliminate the effect of humidity. In addition, the specimens were instrumented with strain gauges and electrically isolated from the machine using insulating tape. The electrical resistance of the matrices was evaluated using the four-probe method, in which the DC voltage is applied to the samples through the external electrodes, and the voltage measured in the sample is recorded by a data acquisition system connected to the internal electrodes. The test setup is shown in Fig. 2.

Fig. 2
Two photographs of the self-sensing test setup. They indicate the strain gauge, loading, and shunt resistor.

Experimental setup for the self-sensing test

Prior to the mechanical loading, DC voltage was applied to each sample for 20 min, to achieve electrical signal stabilization and to mitigate the sensor capacitive effects. During the test, the electric current of the circuit was acquired by using a shunt resistor with a known electrical resistance connected in series. Using the voltage of the shunt recorded throughout the test and applying Ohm’s 1st law (Eq. 1), the electrical current was obtained. With the current and voltage between the internal electrodes, the electrical resistance of the matrix was determined.

For the correlation with the strains, the fractional change in resistance (FCR) in response to the applied loads were obtained according to Eq. 2. The initial electrical resistance of the sample R0 was measured after a pre-load of 0.5 kN at the end of the electrification time of 20 min. On the other hand, electrical resistance R corresponds to the electrical resistance value over time, under a given compressive loading cycle in the linear elastic regimen of material. Finally, the electrical resistivity ρ of the samples, in Ω.cm, was obtained through Ohm’s 2nd law (Eq. 3), considering the distance between the electrodes and their contact area with the cement composite.

$$V=R\times i$$
(1)
$$FCR=\frac{R-{R}_{0}}{{R}_{0}}$$
(2)
$$\rho =\frac{{R}_{0}\times A}{L}$$
(3)

where: V is the voltage (V); R is the electrical resistance (Ω); i is the intensity of the DC current (A); R0 is the initial electrical resistance of the sample, in Ω, which corresponds to the value recorded at the end of the electrification period and before the loading cycles; L is the distance between the copper electrodes (cm); and A is the embedded area of the electrodes (cm2).

In the first sequence of tests, in which samples P1, M1, and M2 were evaluated, the described test setup was used, applying an electrical voltage of 5 V and shunt resistance of 100 Ω. In the second part, using the sample M3 and aiming to investigate the effects of the test configurations, both the applied DC voltage and shunt resistance were varied, respectively, from 4 to 12 V and from 6 to 100 kΩ.

3 Results and Discussion

3.1 Materials Characterization

Figure 3 shows the particle size distribution, obtained by laser diffraction analysis, of the cement, silica flour (S325), and sand fractions. Table 2 lists the chemical composition of the raw materials and their main physical properties. After CNT dispersion by sonication, there was a relative increase in the area under the UV–Vis spectrum and an increase in the absorbance peak, corresponding to the wavelength range around 260 nm (Fig. 4). Before sonication, the absorbance intensity of 0.16 at the peak demonstrated a low degree of dispersion of the nanotubes. After ≈66,000 J of sonication energy, the highest value of 1.31 at the peak suggested effective CNT dispersion in the water/superplasticizer solution.

Fig. 3
A graph of cumulative volume in percentage versus particle size in micrometers for cement, S 325, fine sand, and coarse sand. Their cumulative volumes are constant at first, then increase, and finally remain constant.

Particle size distributions of the raw materials

Table 2 Chemical composition and main physical properties of the raw materials
Fig. 4
A graph of absorbance versus wavelength in nanometers before and after the sonication process. It indicates 1, 31 after sonication, and 0, 16 before sonication. The absorbance after sonication is high when compared to the absorbance before sonication.

UV–Vis spectra of carbon nanotubes in the water/superplasticizer solution before and after sonication process

3.2 Effect of Fine Aggregate and CNT Content on Electrical and Piezoresistive Properties

The initial resistivity ρ0 of the composites P1, M1, and M2 was 100, 600, and 700 kΩ.cm, respectively. Figure 5 shows the strain and FCR variations as a function of time for these samples, under cyclic compressive loadings from 2 to 8; and 2 to 12 kN (0.8–3.2 and 4.8 MPa). Figure 6 shows the relationship between strain and FCR values during the self-sensing tests of samples P1, M1, and M2. No smoothing was applied to the obtained data.

Fig. 5
Three graphs depict stress in megapascals, F C R in ohm over ohm, and strain in microseconds versus time in seconds for P 1 paste with 0.75 percent of C N T, M 1 mortar with 0.75 percent of C N T and s over c equals 1.5, and M 1 mortar with 0.5 percent of C N T and s over c equals 1.5.

Fractional change in resistance (FCR), stress, and strain responses as a function of time of a P1, b M1, and c M2 under cyclic compressive loading. CNT, carbon nanotubes; s/c, sand to cement ratio

Fig. 6
Three scatter plots of F C R in ohm over ohm and linear fit versus strain in microseconds for P 1, M 1, and M2. The three scatter plots tend towards negative correlations. R square equals 0.91, as indicated in P 1. R square equals 0.52, as indicated in M 1. R square equals 0.73, as indicated in M 2.

Fractional change in resistance (FCR) versus strain diagrams of samples a P1, b M1, and c M2 under cyclic compressive loads

A piezoresistive response of all cementitious composites can be seen in Fig. 5, once the FCR values decreased with the increase in both loading and strain, and increased with the reduction of strain, during the unloading step. During cyclic loading, under compression, the CNT came closer and conductive paths formed due to the contact and tunneling conductivities, allowing more electric current flow and reducing the electrical resistance of the sample. Upon unloading, the conductive network returned to its initial state and recovered its electrical resistance, increasing the FCR value [2, 11].

However, it should be noted that there was an increase in the electrical resistivity ρ0, as well as a change in the amplitude and quality of the piezoresistive response with the addition of fine aggregate and the reduction in CNT content. Comparing paste P1 and mortar M1 (Fig. 5a, b), both with 0.75% of CNT content, it is notable that the aggregate addition affected the FCR variation. In P1, the load cycles generated a higher maximum amplitude of FCR, equal to 0.33, while for M1 a less sensitive response to loading was obtained, with a maximum FCR value of 0.15, besides a greater degree of noise. Naturally, this also affected the gauge factor (GF) values, equal to 1076 for P1 and 375 for M1, which is equivalent to the slope of the FCR–strain curve (Fig. 6) and represents the sensitivity of the composites. In addition, the highest coefficient of determination (R2) was obtained in paste P1 (0.91), meaning that the FCR–strain curve was less scattered compared with mortar M1—noisier and with a lower R2 of 0.52.

This behavior is possibly explained by the addition of the fine aggregate, which besides modifying the mechanical response of the composite to loading, disturbed the conductive paths and electron mobility due its insulating nature. Previous evidence from the literature suggests that aggregates can constitute obstacles to current flow, negatively affecting the sensitivity and noise of the self-sensing response [2,3,4, 9, 12]. Therefore, the results obtained by this study reiterate that both the conductive (CNT) and non-conductive phases (aggregate) affect the self-sensing performance of the composite.

Chiarello and Zinno observed that the magnitude of the conductivity of the system containing conductive fibers decreases exponentially with the increase of the s/c ratio [13]. García-Macías et al. [14] found variations in the resistivity of paste tenfold greater than those found for mortar and concrete. Dong et al., in a literature review, explained that the presence of the aggregate weakens the sensitivity of the composite, due to the imperfect crack controlling capacity of conductive fibers under the interaction with aggregates [2]. Moreover, due to the tendency of aggregates to separate the connectivity of the CNT conductive network. Both D'Alessandro et al. [12] and Han et al. [15] reported other adverse effects with the use of aggregates, such as reduced sensitivity, repeatability, and signal quality of self-sensing composites containing CNT.

The detection ability of the composites was also affected by the CNT concentration, as can be seen when comparing the mortars M1 and M2, with 0.75% and 0.5% of CNT, respectively. The maximum FCR amplitude obtained for these samples, corresponding to the compressive stress of 4.8 MPa, was equal to 0.15 and 0.29 respectively, suggesting a higher sensitivity of the composite containing 0.5% of CNT. It can be seen that, compared with M1, the self-sensing response of sample M2 showed higher linearity (with R2 = 0.73) and detection sensitivity (GF 883).

This behavior is probably due to the CNT content of M2 being closer to the percolation threshold, which generated a conductive network with a greater capacity to rearrange under loading, resulting in more significant changes in electrical resistivity and greater sensitivity. In the case of samples with 0.75% of CNT, the concentration of conductive filler was closer to the conductivity zone, in which conductive networks are more stable and denser, leading to a lower sensitivity [1,2,3,4, 9]. Thus, the self-sensing properties are also dependent on the CNT concentration in the matrix and its distribution, which is negatively affected by the distribution of the aggregates, once they are the non-conductive phase.

3.3 Effect of Test Setup on Electrical and Piezoresistive Properties

Figure 7 shows the variation in FCR for sample M3, under 10 kN of cyclic compressive loading, when both the applied DC voltage and shunt resistance were varied from 4 to 12 V and from 6 to 100 kΩ, respectively. In Fig. 7a the lower values of electrical resistance of the shunt resistor led to noisier, intensified, and even distorted responses. From 14 kΩ of shunt resistance, as seen in Fig. 7b, the amplitude of the FCR variation over time became approximately equal, regardless of the voltage and shunt value adopted, suggesting that the response to loading in those cases was solely dependent on the matrix properties.

Fig. 7
Two graphs of F C R in ohm over ohm versus time in seconds for the voltages of 4 and 8 volts and shut resistance of 6 and 9 kilo ohms, and the voltages of 4, 8, and 12 volts and shut resistance of 14, 17.5, and 100 kilo ohms.

Fractional change in resistance (FCR) response under 10 kN cyclic compressive load with a varying voltage from 4 to 8 V and shunt resistance from 6 to 9 kΩ and b varying voltage from 4 to 12 V and shunt resistance from 14 to 100 kΩ

Supposedly, lower shunt resistances interfere with their acquired voltages, negatively affecting the current and matrix resistance values derived from these measurements. Higher shunt resistances, compatible with the magnitude of resistance variations observed in the cement matrix, seemed to cause a better resolution of the acquired data. Therefore, the present results suggested that the higher noise level observed was not only a property of the matrix but also dependent on the test configuration. The latter is easily adjustable to allow a more efficient acquisition of electrical signals.

Previous investigations suggest that factors related to the measurement of electrical signals in the self-sensing test can affect the intensity and stability of the piezoresistive response—for example, the configuration of the electrodes or the current type and its magnitude—and therefore should be properly adjusted [2, 10, 16, 17]. Galao et al. applied varying electrical currents to the self-sensing composites (0.1, 1.0, and 10 mA) and found that the piezoresistive response was better with increasing current intensity, achieving better signal stability and correlation with the strain of material [17]. On the other hand, among the applied electrical voltages of 10, 20, and 30 V, Konsta-Gdoutos and Aza obtained an optimal voltage of 20 V, suggesting that high electrical current intensities can also be harmful [16]. Ding et al. evaluated self-sensing cementitious composites using reference resistors with 1000 Ω of electrical resistance [18], and other researchers used the same circuit model but without mentioning the value of shunt resistance [19].

Based on these results, our setup was defined with a shunt resistor of 27 kΩ and a DC voltage of 4 V to evaluate mortar M3. Figure 8a shows the results under cyclic loading of 3.2 MPa for these conditions. Comparing an excerpt of the M2 test using the first setup (Fig. 8b), a clear change in the self-sensing response can be seen, with considerable noise reduction of the signal. The better self-sensing response suggested that not only the concentrations of conductive filler and fine aggregate affect the electrical signals acquired in self-sensing tests. The improved response can be explained in part by the decrease in the s/c ratio but seems to be mainly due to the change in the test settings.

Fig. 8
Two graphs depict stress in megapascals, F C R in ohm over ohm, and strain in microseconds versus time in seconds for M 3 of 4 volts and 27 kiloohms and M 2 of 5 volts and 100 ohms. The stress and strain of both cases trend in triangular waveforms.

Fractional change in resistance (FCR), stress and strain responses as a function of time of samples a M3 and b M2 under cyclic compressive loading

The R2 coefficient observed in Fig. 9b, equal to 0.96, was even higher than that obtained for the paste P1, equal to 0.91, which had more CNT and no aggregate. The mortar M3 also showed good detection sensitivity, represented by GF of 693.

Fig. 9
Two scatter plots of F C R in ohm over ohm and linear fit versus strain in microseconds for M 3 of 4 volts and 27 kiloohms and M 2 of 5 volts and 100 ohms. They tend toward negative correlations. R square equals 0, 96 is indicated in M 3. R square equals 0, 57 is indicated in M 2.

Fractional change in resistance (FCR) versus strain diagrams of a M3 and b M2 mortars under cyclic compressive loading

Yoo et al. [5] investigated the self-sensing performance of cement pastes and verified that the sample containing 1.0% by volume of CNT presented a maximum amplitude of the FCR equal to 0.26 under a compressive load of 40 kN. The GF of the composites ranged from 77.2 to 95.5 with a minimum R2 of 0.9382. Yin et al. [11] produced cement pastes with 1.7% CNT, by volume, with a high coefficient of sensitivity to deformation (1500), high linearity (R2 = 0.97), and maximum FCR equal to 0.19 under cyclic compressive loading of 10 MPa. The hybrid combination of CNT and nickel nanofibers resulted in the best piezoresistive sensitivity and response linearity under the same load, with a maximum FCR equal to 0.24, GF of 1880, and R2 of 0.99.

4 Conclusions

In this study, the self-sensing properties of cementitious composites containing fine aggregate and varied content of CNT were evaluated under compressive cyclic loading. In the first step, the paste P1 with 0.75% CNT led to higher linearity and sensitivity of the self-sensing response and a low degree of noise in the electrical signal, when compared with mortars M1 and M2. The electrically inert aggregate possibly modified the mechanical response of the composite to compressive load and constituted an obstacle to electron flow, increasing tortuosity or even interrupting the conductive paths of the CNT. Comparing the mortars M1 and M2, with 0.75% and 0.5% of CNT respectively, higher detection sensitivity and linearity were verified for the composite with lower CNT content—closer to the percolation threshold. At this CNT concentration, a conductive network formed with a greater capacity to rearrange under loading, resulting in more significant changes in electrical resistivity. On the other hand, higher concentrations of CNT, especially closer to the conductivity zone, formed denser and more stable conductive networks, leading to lower detection sensitivities. These results confirmed that both the conductive and non-conductive phases affect the electrical and self-sensing behavior of cementitious composites containing CNT. The conductive network is, therefore, formed by an appropriate filler concentration, and the choice of aggregate, as well as the appropriate proportions, is extremely important for the fabrication of self-sensing cementitious composites with good detection sensitivity and signal stability.

Additionally, the test setup configuration should be also taken into account when investigating the self-sensing performance of cementitious composites. We verified that the test settings can allow more efficient acquisition of electrical signals and even reduce the noise caused by the addition of aggregates. In a circuit with a reference resistor, lower values of shunt resistance led to noisy, scattered, and distorted electrical responses, whereas shunt resistances from 14 to 100 kΩ led to a better self-sensing response signal, with reduced noise and higher linearity. Under these conditions, the FCR under loading was always less scattered and with the same amplitude, reinforcing its capability of reflecting the piezoresistive properties of the matrix. This suggests that strategic adjustments in the self-sensing test settings may be particularly useful to reduce the deleterious effects of added sand in the matrix, enabling its use as aggregate in self-sensing composites while maintaining satisfactory detection performance. This approach may benefit the implementation of smart mortars in structural health monitoring. In addition, it can contribute to the massive use of aggregates in this type of composite, consequently lowering the consumption of raw materials with higher cost and environmental impacts such as CNT and Portland cement.