The cement used in this study was standard P.O. 42.5 from Hailuo Ltd. conforming with the Chinese standard (GB175-2007, 2007). The river sand (0.0075–5 mm) and natural granite coarse aggregates (5–10 mm) were washed to remove silt content. This was done in order to remove the dust from coarse and fine aggregates because the dust particles may influence the hydration products. Normal tap water was used for mixing concrete.

Three different sizes of conductive materials were prepared to improve the piezoresistivity and mechanical properties of the intelligent concrete. The graphene obtained from Jicang Nano Tech Ltd. (Nanjing, China) was used as conductive nanoparticles. Carbon black was also adopted, which is one kind of superconductive acetylene carbon powder produced by DENKA, Japan. Copper-coated steel fibers were selected due to great durability.

The fundamental properties of graphene are listed in Table 1, according to which the average thickness and diameter of the graphene nano-platelets used are 2–3 nm and 5–15 μm, respectively. To facilitate the dispersion of graphene in an aqueous suspension, a polycarboxylate superplasticizer bought from Subote (Nanjing, China) was selected.

To obtain the micro-morphology of graphene, a Hitachi S-4800 field emission scanning electron microscope (SEM) was used. First, with the assistance of superplasticizer and ultrasonic water bath, graphene was dispersed in water. Then, one drop of graphene suspension was placed on an aluminum sheet. After the water evaporates, a thin graphene film was left. The aluminum sheet carrying the graphene was placed in the chamber of the SEM to capture images. The acceleration voltage is 30 kV, and the resolution can reach 10 nm. According to the SEM images, see Figure 1, the graphene particles show a layered and wrinkled appearance and their in-plane diameter and thickness are consistent with those reported in Table 1.

The images of carbon black and steel fibers under an optical microscope are shown in Figures 2A,B, and the size and morphology are consistent with the manufacturer’s reports, which are also shown in Table 1.

Graphene has a very high surface area; therefore, it requires ultrasonic treatment and the assistance of superplasticizer to disperse it in water. The graphene aqueous suspension was treated with ultrasonic waves for half an hour before being added into the mixer, as shown in the first step in Figure 3. The coarse aggregate, sand, carbon black, steel fibers, and cement were mixed in a planetary mixer for 30 s, as illustrated in the second step in Figure 3. While the mixer was running, the premixed graphene aqueous suspension was added into the mix. After 3 min of thorough mixing, the fresh concrete was ready to be cast in molds with the dimensions of 40 × 40 × 160 mm, in which two copper probes at a distance of 100 mm were pre-placed. The probe was fabricated by using the copper wires (diameter = 1 mm). The fresh concrete was filled into the molds and then compacted using a vibrating table. The specimens were demolded 24 h after casting and then stored in a standard curing room with a constant temperature of 20°C and humidity of 98%. The whole preparation procedure is shown in Figure 3.

The mix proportions of the intelligent concrete and contrast normal concrete are shown in Table 2, according to which the water/cement ratio is 0.665. For the contrast group, which is named “Con” in Table 2, there are no conductive materials or superplasticizer used in the mix. In Table 2, the mix proportion is defined in kg/m³. The density of the contrast concrete is 2,400 kg/m³. The intelligent concrete was produced by adding extra conductive materials. For the intelligent concrete containing multi-scale conductive materials, which is named “M,” the content of graphene is 3% of the cement by weight. The content of carbon black is 1.5% of the volume of concrete, and the content of steel fibers is 0.6% of the volume of concrete. For comparison, group G was prepared by adopting graphene alone as the conductive filler, and the graphene content is 6% of the cement by weight. According to previous studies, the intelligent concrete incorporating graphene by 6% of the cement weight was endowed with stable piezoresistivity (Liu et al., 2016).

This experiment was performed on a universal testing machine (SANS100) with a capacity of 100 kN. For applying even compressive load, the specimens were preloaded to 30% of the estimated peak load three times. After that, a cyclic load was applied on the specimen. The cyclic load started to increase from 0 to 2 kN; then, it dropped to 1 kN, increased to 4 kN, and went down to 1 kN again. The force went up to different levels, including 6 kN, 8 kN, and 10 kN, and it went back to 1 kN after reaching each level. The ultimate force of the specimens was supposed to be 30 kN. Therefore, 10 kN was applied to ensure the specimens are still within the elastic range. Polytetrafluoroethylene (PTFE) sheets that were settled on the top and bottom surfaces of the specimen were used to dismiss the friction between the platens and the specimen.

In order to obtain full stress–strain curves, a hybrid loading procedure containing a force-control phase (10 kN/min) and a displacement-control phase (0.1 mm/min) was adopted. The force-control loading was applied within 70% compressive capacity, which was estimated with the samples’ mix proportion. After that, the displacement-control loading was used until the test was completed. An extensometer fixed on the backside (as to probes) of the samples was used to measure the strain. The loading setup is shown in Figure 4.

After 28 days of curing, the specimens were stored at room temperature for one day to dry the surface. Then, they were treated by wrapping the ends with copper foil; as a result, the two ends of the specimens can be used as electrodes. The four-electrode method was selected to measure the electrical resistance of the specimens.

The electrical resistance was recorded while applying a uniaxial compressive load. It was measured with a four-probe method to dismiss the contact resistance between electrodes and concrete. The outer two electrodes were used for applying a constant electrical voltage of 1 V, and the inner two probes were used for the current measurement. The electrical current was captured with a universal meter (Fluke 17B). The electrical resistance can be gained by dividing the constant voltage by the captured current. Then, the resistivity ρ is calculated with the following equation: ρ = V A / I L , (1) where V is the constant voltage, A is the area of the cross section, I is the captured electrical current, and L is the distance between the two inner probes.

The strength, resistivity, and failure mode of the three groups of specimens are shown in Table 3. It can be seen that the strength of group M is higher than the strength of the contrast group, 23.4 MPa. However, the strength of group G only incorporating graphene is lower. This may be because a large amount of superplasticizer is added for dispersing graphene and the superplasticizer has also a certain air-entraining effect (Liu et al., 2016). Therefore, there are more bubbles in group G intelligent concrete; as a result, the strength is weakened. Group M contains steel fibers, which restrain the propagation of cracks. This improves the strength of group M and improves its ductility more obviously. The resistivity of group M is similar to that of group G. They are 0.940 Ωm and 0.900 Ωm, respectively.

According to the above analysis, it can be found that the resistivity of groups M and G is similar. And group M has a higher strength and better ductility. In addition, the cost of group M is also lower. The amount of graphene in group M is 3% of the weight of cement, while that in group G is 6%. Considering the expensive price of graphene, the cost of group M is lower than that of group G.

When the specimens of group M and group G are subjected to a cyclic load within the elastic stage, the resistivity of the intelligent concrete also changes cyclically, as shown in Figure 5. As the load increases, the resistivity decreases, and the reduction in load leads to an increasing resistivity. The amplitude change of the resistivity is roughly consistent with the force. From Figure 5A, it can be seen that when the load is 0kN, the resistivity of group M is 0.940 Ωm. The peak values of each cyclic load are 2 kN, 4 kN, 6 kN, 8 kN, and 10 kN, respectively. The axial compressive stress of the specimen is calculated as 1.25 MPa, 2.5 MPa, 3.75 MPa, 5 MPa, and 6.25 MPa, and the corresponding resistivity of group M is 0.929 Ωm, 0.914 Ωm, 0.908 Ωm, 0.905 Ωm, and 0.898 Ωm, respectively, which shows a downward trend. According to the difference, it can be seen that the resistivity of the intelligent concrete is approximately proportional to its axial pressure. Also from Figure 5, it is seen that when the load returns back to 1 kN, the resistivity of the intelligent concrete is 0.935 Ωm, 0.934Ωm, 0.932 Ωm, and 0.938 Ωm, respectively. This shows that the resistivity of the intelligent concrete in this study has a great performance of repeatability. Group G has a very similar trend of resistivity when a cyclic load is applied, see Figure 5B.

The linearity of a sensor is an essential property to define the degree to which the actual output curve of the sensor resembles the expected linear plot. To see the linearity of the piezoresistivity of the intelligent concrete, Figure 6 shows the relation between the stress and the resistance captured during testing and also illustrates the best-fitting straight line and its equation. The fitting equation for group M is shown in Figure 6A as follows: r = 0.94 − 0.0076 s , (2) where r is the resistivity and s is the applied stress on the specimen.

The fitting equation for group G is shown in Figure 6B as follows: r = 0.90 − 0.0087 s . (3)

The coefficient of determination (R ² ) is used to evaluate the feasibility of the prediction equation as reported above. Its formula is shown as follows: R 2 = 1 − ∑ i ( r i − f i ) 2 ∑ i ( r i − r ¯ ) 2 , (4) where r i   is the resistivity measured, r ¯ i is the average resistivity, and f i   is the prediction result. The coefficient of determination between the fitting curve and the captured values can be calculated according to the above formula. For groups M and G, the coefficient is 0.79 and 0.95, respectively. This illustrates that the piezoresistivity of the intelligent concrete group G has a great linear relationship with the fitting function that can predict the stress performance. However, group M has an acceptable linear relationship with the fitting curve.

After cyclic loading, a uniaxial compression until failure was applied on the specimen, and the resistivity recorded is shown in Figure 7. For group M, see Figure 7A, it can be seen that the resistivity continues to decline during the increased loading process. When the stress reaches the peak value of 25.1 MPa, the resistivity of the intelligent concrete decreases to the minimum value of 0.808 Ωm. After the peak, with the gradual development of cracks, the load on the specimen drops and the resistivity of the specimen increases gradually. Because of the steel fibers presented, the specimen shows ductile failure and the stress–strain curve shows a relatively complete declining section. When the stress of the specimen decreases to 7 MPa, the resistivity increases to 1.139 Ωm.

The resistivity of group G also has a trend of first decreasing and then increasing as illustrated in Figure 7B. However, the lowest point is at about 40% of the peak load, which is the end of the elastic regime for concrete.

The damage of concrete is defined according to the effective area and the initial area of the material as follows (Mazars and Pijaudier-Cabot, 1989): D = 1 − A / A 0 , (5) where A is the effective area and A ₀ is the initial area.

As the concrete damage develops, consequently, the secant modulus of it degrades. The damage of concrete is defined with the degradation of secant modulus (Chaboche, 1981; Lemaitre, 1985), as follows: D = 1 − E / E 0 , (6) where E is the present secant modulus and E ₀ is the initial secant modulus. It is adopted to calculate the damage evolution curves as shown in Figure 8, along with strain–stress curves for comparison.

According to the classical definition of damage in Eq. 5, as the effective area decreases, the concrete damage develops from 0 to 1 until the concrete cannot withstand external forces. In this study, a constant voltage is applied on the intelligent concrete; therefore, the electrical current through the concrete drops with the decrease in the effective concrete area. Therefore, the damage can be defined as a function of the current as follows: D = 1 − I 2 I 0 2 , (7) where I represents the real-time current and I ₀ is the initial current.

Before the peak stress, the electrical current through the intelligent concrete increases gradually, which leads to a negative damage calculated with Eq. 7. This is because the conductive particles in the intelligent concrete are getting close to each other under pressure. Therefore, the resistivity decreases and the current increases (Chen and Liu, 2008). But the negative damage is not meaningful for the definition of damage. Because of this, this study does not consider the descending stage of the resistivity curve to calculate the damage. As a result, the highest electrical current is regarded as the initial current.

Based on the classical damage definition, different calculating methods are proposed. In this paper, the damage evolution curves obtained with the strain–stress relationship and the electrical current passing the concrete are plotted in Figure 8. In Figure 8A, two curves of damage evolution are drawn in the same coordination, and the strain–stress curve is also provided for comparison. It can be seen from the figure that the damage calculated with the secant modulus is zero in the elastic stage and then accumulates. The damage climbs to 0.1 at peak load and then increases rapidly to 0.7at the concave of the descending section of the stress–strain curve. After that, the damage continuously increases to 0.85, slowly.

The damage calculated with the electrical current through the specimen of group M is negative before the peak load. It increases gradually after the peak load with a small amplitude and similar trend to the damage curve calculated with secant modulus. Finally, the damage obtained with the current is about 0.32. The damage calculated with the current through the specimen is relatively small, which may result from the fact that the current can still pass through the specimen even if the specimen has splitting cracks or inclined cracks under pressure and gradually loses its bearing capacity. Although the damage characterized with the current has a disadvantage as illustrated above, it still has an obvious advantage. The current-increasing process means the stage before the peak load, and the current-decreasing process is the stage after the peak load.

For group G, the damage calculated with the secant modulus is showing a great damage evolution regular pattern, see Figure 8B. However, the damage determined with the resistivity has only a very small amplitude from 0 to 0.03. This may be because the failure of group G occurs between the first electrode and the second one rather than between the two inner probes, which was observed during testing.

The intelligent concrete is observed under the scanning electron microscope, as shown in Figure 9. It can be seen from the figure that the graphene is folded and flaky, which is distributed in the hardened cement paste. Carbon black can also be found in the hardened paste. The graphene nano-sheets and carbon black particles form a conductive network. Under pressure, the tunnel effect among graphene nano-sheets and carbon black can form pressure-sensitive characteristics, so as to make the concrete have intelligent performance, which can be used for strain detection and damage identification of concrete. The size of steel fibers is much larger than the scope shown in the figure, which is mainly used to prevent the brittle failure of intelligent concrete.