A study of the effects of MWCNTs on the fresh and hardened state properties of 3D printable concrete

Much research in 3D concrete printing focuses on material design and testing new additives to improve performance. However, there are very few examples of studies using carbon nano-materials to improve mechanical performance and create smart mortars with self-sensing capabilities. 3D concrete printing is a new technology that focuses on optimizing the building industry, a composition able to monitor structural health can provide important insights into stress levels and microcrack concentration in critical structural parts. In this study, multi-walled carbon nanotubes (MWCNTs) were embedded in a printable concrete, and its self-sensing ability was investigated. Four compositions with different concentrations of MWCNTs were prepared to assess the porosity distribution and mechanical properties in the fresh and hardened states. To evaluate the self-sensing ability of the material, the change in conductivity was measured using a resistivity meter with a four-point Wenner probe, while the material was tested under compression and tension. The results showed that porosity, strongly influenced by the distribution of nanomaterials in the matrix, had a significant effect on the fresh and hardened state behavior and self-sensing ability of the compositions, and the composition with better self-sensing capabilities was the one with a MWCNTs content of 0.5% per binder content.


Introduction
3D concrete printing is a new building technique that is gaining relevance in the construction industry due to its advantages related to the reduction of labor costs, production time and material waste [1].It consists of a concrete composition that is mixed, pumped through a hose and extruded through a nozzle layer by layer in a process controlled by a robot to obtain the wanted geometry.This technique has the potential to optimize the building process by using less material, having more geometry freedom and potentially tuning the material's composition depending on the location and specific needs [2,3].Current research focuses on system design and how to increase large-scale production efficiency [4][5][6] and materials design and how to improve rheological properties and mechanical performance in fresh and hardened states [7][8][9][10].The performance of 3D printable concrete (3DPC) depends on many factors, such as thixotropic behavior, structuration rate [10], yield stress [10,11], amount of porosity trapped during the printing process [12], mechanical strength behavior of the material as a result of low aggregate content and small size of the aggregates compared to conventional concrete [11].
To improve the performance of 3D printable concrete, various additives, such as VMA, nanopowders, and fibers, are added to the composition [11].In this study, multi-walled carbon nanotubes (MWCNTs) are added to a printable concrete composition (employed in previous studies [13]) to produce self-sensing smart concrete with enhanced mechanical performance in order to incorporate a monitoring system that can optimize the lifespan of a printed structure, requiring less material but also more fines than conventional concrete, making it potentially sensitive to shrinkage and high stresses.
MWCNTs are highly conductive carbon nanomaterials with outstanding mechanical performance and thermal properties, which make them an interesting additive in construction materials [14].
In some studies, MWCNTs, when well dispersed in the matrix, are used to improve the mechanical strength of cementitious materials [15,16].The main reinforcement mechanism of MWCNTs is their ability to bridge microcracks and delay their propagation [15].
MWCNTs are also used for their highly conductive behavior to modify the conductivity of the matrix in which they are mixed to create a smart material that reacts to stresses and strains by changing its conductivity, thus obtaining a self-sensing material [17].Self-sensing materials are used to monitor their structural health without the need for external sensors, identifying risks and damages in order to avoid failure [9].
MWCNTs are made of carbon atoms bonded in hexagonal forms by sp 2 bonds to create a tubular structure with multiple layers.Sp 2 bonds are very strong and are the cause of the high modulus and strength of MWCNTs [15,18].The conductivity is determined by the chirality of the tubes (an armchair structure gives very high conductivity values, and the nanotube is considered metallic) and the number of defects they present; MWCNTs have generally lower conductivity than single-walled carbon nanotubes due to their layered structure and higher presence of defects [15].
In the concrete industry, there has been much research related to the study of self-sensing concrete composites, as it is a brittle material that would benefit from constant monitoring [19][20][21][22].There are multiple studies [23][24][25][26][27] that investigate various structural health monitoring systems for real-life applications, such as for masonry structures [24], concrete beams [25] or traffic detection systems [26,27].However, real-life applications of self-sensing smart concrete are still limited due to a decrease in workability when nanotubes are added to the matrix, a high cost associated with such nanomaterials, safety issues associated with handling nanomaterials, and difficulties in obtaining a homogeneous dispersion of nanotubes [28,29].The dispersion process of functional nanomaterials can be tricky due the high Van der Waal attraction forces between the nanomaterials and large surface area, decreasing the sensing efficiency for structural health monitoring applications [30].Furthermore, for conventional concrete, the presence of large aggregates and the presence of steel reinforcements hinder the electrical conductivity measurements, complicating the measurement process: the presence of large aggregates interrupts the electrical path of the electrons traveling through the matrix, decreasing the sensitivity of the measurements; on the other hand, the presence of steel reinforcements interferes with the path of the electrons, making the measurements unreliable [31].
In 3D-printed concrete, however, the lack of large aggregates and of a traditional steel reinforcing system in favor of new nonmetallic reinforcement strategies [32] makes resistivity measurement more feasible, and the presence of MWCNTs could be beneficial in terms of mechanical properties (depending on the dispersion quality) by improving the microstructure of the binder and the mechanical strength.
In this study, a dispersion of MWCNTs in water and surfactant, manufactured by Nanocyl, was used to incorporate MWCNTs into a printable concrete composition at different concentrations.The printable concrete was used as a reference to which compositions containing nanomaterials were compared.The objective was to investigate the effects of the nanomaterials on porosity by performing a mercury intrusion porosimetry (MIP) test, fresh and hardened state mechanical properties by performing unconfined uniaxial compression tests (UUCT) and compression at 28 days, rheological behavior and microstructure using a scanning electron microscope (SEM), followed by a study of the self-sensing response of the composite to both compressive and tensile stresses using a four-point Wenner probe.MWCNTs solution Aquacyl AQ030X was selected as a result of a previous study [13] and was suitable for incorporation into a printing system and safe to handle, as the nanomaterials were safely incorporated into the solution.

Materials and samples preparation
The concrete mix used in this study was developed to be printable and have good thixotropic behavior.This mix was used in previous studies [13] with different concentrations of multi-walled carbon nanotubes (MWCNTs).The details of the compositions are reported in Table 1.
Table 1 shows the amount of additives and MWCNTs broken down by weight per binder content.The MWCNTs, produced by NANOCYL, were supplied dispersed in water (AQUACYL AQ030X) at a concentration of 3% by weight.For this reason, during the preparation of the different mixtures, the amount of water was calculated taking into account the water present in AQUACYL AQ030X, and the percentage of MWCNTs was calculated taking into account their concentration inside AQUACYL AQ030X.
To prepare the compositions, the water was mixed with AQUACYL AQ030X for 5 min using a KitchenAid® Classic Plus 4.5 qt mixer, then the superplasticizer, accelerator and retarder were added one at a time while mixing for 5 min after each addition, for a total of 20 min of liquid mixing.The liquid part was subsequently added to the dry materials, which were pre-mixed with an industrial mixer, and the resulting composition was mixed for a further 5 min inside the industrial mixer.
Cast samples were prepared using 40x40x40 mm 3 molds for hard-state compression tests and 1600x40x40 mm 3 molds for hardstate resistivity tests under compression.
For the hard-state resistivity tests under tension, printed samples were made with an ABB IRB 1200 robot and a Makita DCG 140 caulking gun (Fig. 1).
The tests performed on the prepared samples are summarized in Table 2.

Table 2
Composition of the printable concrete.

Mercury Intrusion Porosimetry test
A Mercury Intrusion Porosimetry test (MIP) was performed using a Micrometrics AutoPore IV 9500 to investigate the influence of MWCNTs on the porosity of the compositions.
The samples were aged for 28 days in water at approximately 20 • C and then oven-dried for 24 h at 60 • C. The analyzed sections were taken from the inner core of the samples and sieved with a sieve diameter between 3 and 5 mm.
The contact angle between the mercury and the pore surface θ was 130 degrees and the surface tension γ was 485 mN/m.Knowing the pressure P (MPa) applied to the pores, their radius (r, in meters) was calculated by Eq. (1). (1)

Hard state compression test
A compression test was performed following EN 1015-11:1999.The instrument used was an automatic compression tester from Controls Group with a head speed of 2400 N/s.Five samples of each composition, with dimensions of 4x4x4 cm 3 , were prepared.The samples were aged for 28 days in water at 20 • C.

Rheology study
The rheological study on the fresh compositions was performed using a Viskomat XL from Schleibinger, to evaluate the effect caused by the MWCNTs on the concrete' thixotropic behavior.
The rheology test protocol and the model to define the static yield stress evolution in time were developed by Kruger et al. [33].The model describes the behavior of fresh concrete between 1 min and 80 min after its deposition in the rheometer's bucket, which is used to evaluate the thixotropic behavior and structuration rate of the 3D printed concrete.
The rotating speed was fixed at 12 RPM for 60 s, then the material was left to rest at different resting times from 0 to 30 min.Every time a rotation started the static yield stress was calculated from the highest measured torque to obtain the material's thixotropic behavior and structuration rate.The test profile is listed in Table 3.
According to Kruger's model, within minutes from the material deposition and before the structuration begins, the concrete's particles flocculate due to surface forces and physical interactions; when the composition is agitated by the blades of the rheometer these forces are broken while, when the composition is at rest, the particles re-flocculate.The re-flocculation rate is an indication of the concrete's thixotropy.The static yield stress can be described by Eq. (2).
With τ s (t) = static yield stress of the material at time t after agitation, τ D,i = initial dynamic yield stress of the material measured from the first rheological test, R thix = re-flocculation rate, and t = time from the cessation of agitation.
After the first few minutes, the particles start to chemically react, the first hydration products are created, and the structuration rate can be measured.In this phase, the static yield stress can be described by Eq. (3).
With τ s (t) = static yield stress at time t after agitation, τ S,i = initial static yield stress of the material, measured from the first rheological test, A thix = structuration rate, t = time since cessation of agitation, and t rf = period of time in which re-flocculation occurs and dominates.The yield stress (static or dynamic) can be calculated from the torque measurements [34] using Eq. ( 4).
With M = measured torque, a = radius of the vane probe and l = length of the blades.The static yield stress was plotted as a function of the resting time for each composition, and the re-flocculation rate and structuration rate were calculated from the slope of the function and were then compared for each composition to assess the effect of the nanotubes on the fresh state behavior of the reference material.

Unconfined uniaxial compression test
A uniaxial unconfined compression test (UUCT) [13,35] was performed on fresh concrete to assess the material's strength and stiffness development rate.The instrument used was Instron 5967 with a 5 kN load cell and a head speed of 30 mm/s.
The test was performed at 0, 15, 30, 60 and 90 min from preparation and the samples were prepared using cylindrical steel molds of 140 mm in height and 70 mm in diameter.The fresh material was compacted with a metallic rod after being poured inside the molds.
A camera Canon, type EOS 700D, with a resolution of 18 megapixels and a 2-second interval self-timer was used to record the lateral deformation.The pictures were processed with the Vision Builder from National Institute software to measure the lateral deformation via optical analysis.

Electronic microscope
A scanning electron microscopy (SEM) analysis was performed to investigate the distribution of MWCNTs and if it was influenced by their concentration, using Quanta3D from Thermo Fisher Scientific in low vacuum mode.The samples were aged for 28 days in water at approximately 20 • C and then oven-dried for 24 h at 60 • C and crushed to collect internal parts less than 1 mm wide and less than 0.5 mm thick.

Resistivity test
The resistivity tests were performed using Resipod from Proceq, a resistivity meter that operates on the principle of the Wenner probe.Four probes equally spaced with a distance (a) of 3.8 cm from one another were put in contact with the material's surface.An AC (alternate current) (I) of 50 μA with a frequency of 40 Hz was applied to the outer probes while the inner probes measured the electrical potential difference (V).The apparent resistivity (ρ APP ) was calculated by the instrument in kΩcm using Eq. ( 5) [36].
The apparent resistivity was calculated for samples when at rest, under compression and tension.To be tested under compression, the samples were cast in molds with dimensions of 16x4x4 cm 3 , cured in water for 28 days, and then tested under compression strain using an Instron 5985 with a head speed of 1 mm/min and a 250 kN loadcell.The resistivity was measured by placing the Resipod on the surface of the sample while compressing up until failure (Fig. 2) [37].Only the composition with the highest self-sensing response (highest resistivity change) was then selected to be tested in tension.
To be tested under tension, both cast and printed samples with a dogbone shape were prepared [38,39].This shape was selected because it was easily adaptable to printed specimens.For the printed samples, the central part (2x5x25 cm 3 ) was printed as a beam with a rectangular nozzle of dimensions 2 × 1 cm 2 and placed inside the mold after 1 day of curing, covered in plastic sheet.The rest of the mold was filled with cast material of the same composition for support, as shown in Fig. 3, and covered in a plastic sheet to cure for 24 h.After 24 h the samples were de-molded and put in water to cure for the remaining time, up to 28 days.The cast samples were prepared by filling the entire mold with mortar, which was then covered in a plastic sheet for 24 h.The samples were then demolded and put in water for the remaining 27 days of curing.
The samples were glued to two steel cubes (using glue Pedikit 860 (Permacol BV)) and then clamped to the instrument Instron 5985.Two metallic plates were glued to the samples and screwed to the cubes for additional support.Before the start of the test, the samples were pre-compressed at a constant force of 50 N for around 30 min, until the glue was completely dry and stable.Two LVDTs at opposite faces were attached to the samples to record the deformation.The samples were tested at a head speed of 0.01 mm/s and the resistivity was recorded every 3 s using Resipod.

Energy dispersive X-Ray analysis
An Energy Dispersive X-Ray (EDX) analysis was performed to analyze the elemental composition of a printed layer of material, assessing the difference between the core and the surface of a layer, using SEM Phenom Table Top from Thermo Fisher Scientific operating at a voltage of 15 kV.The samples were cured for 28 days, covered in plastic, and stored at 20 • C.

Mercury intrusion porosimetry test
The MIP test results are presented in Figs. 4 and 5, and Table 4.
The presence and different concentrations of MWCNTs significantly impact the porosity values and distribution.Compared to the reference mix, the porosity decreases by around 10% for the mix with the lowest concentration of nanotubes (CNT01) while it increases by approximately 19% and 13% for the mixes with higher concentrations of nanotubes (CNT02 and CNT05).This behavior is in line with previous studies [40,41], which report a threshold value for the nanotube's concentration below which the nanotubes have a positive influence on the porosity amount, and above which the porosity increases.The threshold value depends on the diameter of the nanotubes and the quality of their dispersion in the matrix.
The MWCNTs used in this study had an average diameter of 9.5 nm with a high surface area, between 250 and 300 m 2 /g, which promoted aggregation and cluster formation.MWCNT clusters entrap water, causing a delay in hydration; once all the water has left the clusters, voids remain in the cement matrix.The size and number of clusters depend on the concentration of MWCNTs in the matrix   and determine the different porosity distribution among the analyzed samples [40,41].
Table 5 shows the porosity distribution in percentage per pore diameter.According to Figs. 4 and 5 and Table 5, MWCNTs mainly affect the porosity distribution of pores with diameter smaller than 200 nm: CNT01 has the lowest percentage of porosity, but 55.9% of the pores have a diameter greater than 167 nm; CNT02 instead has the highest percentage of porosity, but 56.6% of the pores have a diameter of less than 167 nm; CNT05 has the highest amount of pores ranging in diameter from 56 to 167 nm.
The median pore diameter and total pore area per porosity percentage are shown in Fig. 6.The median pore diameter shows a descending trend with increasing porosity and, consequently, the total pore area shows a rising trend.This indicates that higher porosity increases the number of smaller pores in the matrix, which is in line with previous studies [42] using composites of MWCNTs with small diameters (below 20 nm).

Hard state compression test
The compression test results are reported in Fig. 7.All the mixes exhibit comparable strength values within the range of 121.4 MPa to 123.3 MPa.The reference material has a marginally higher compressive strength of 123.3 MPa than the mixes with carbon nanotubes, and a lower standard deviation of 3.3 MPa, but the overall difference is neglectable.The results are in agreement with previous studies [43][44][45].The MWCNTs used in this study had a small diameter of 9.5 nm which should have improved the hardened state mechanical performance.However, the high surface area of the nanotubes caused agglomeration that resulted in no improvement of the compressive strength [43].
It is expected that the different porosity content and distribution among the compositions, caused by the different MWCNT concentrations, will have a stronger impact on the compressive strength.A study from 2021 by Jin et al. [46] shows how to calculate the influence of the porosity content and distribution on the compression strength and what values of strength to expect based on them.
The compressive strength σ is calculated as per Eq. 6 [46]: Where σ 0 is the strength value at zero porosity, k i and p i are the influence factor and the percentage of porosity in the i-th interval.
The influence factor can be calculated for each pore diameter D i by Eq. 7 [46]: The experimentally measured strength of Ref, 77.1 MPa (Fig. 7) is used to calculate the strength value of Ref at zero porosity σ 0 , of 94.5 MPa.
If we assume that the presence of nanotubes has no effects on the strength at zero porosity σ 0 , the value σ 0 of Ref, 151.2 MPa, is used to calculate the strength of all the other compositions using and Eqs.6 and 7, and the results are shown in Table 6.It can be observed that the difference between the obtained calculated values of σ and the experimentally measured values is very small.The porosity doesn't seem to have a big effect on the compression strength at the hardened state since the amount contained in the samples is very small (around 4%) and the variations inside the different compositions are not enough to determine a strong change in compressive performance.This changes when the material is in its fresh state, as will be explained in Sections 3.4 and 3.5.
In the case considered above, it is presumed that MWCNT's presence doesn't affect the values of σ 0 .However, if we assume that the nanotubes do influence the strength of the material, as stated in other studies [47], it is interesting to calculate σ 0 using the experimentally measured values of σ.In Table 7, σ 0 values are reported: it appears that the presence of MWCNTs has a positive effect on the calculated strength at zero porosity, increasing for high nanotubes' concentration.The dispersed nanotubes in the matrix improve the strength of the material, delaying crack propagation and densifying the microstructure [47], however, the nanotubes aggregated in clusters have the opposite effect, acting as a crack initiation site and influencing the porosity [43][44][45], which explains the lower σ 0 value of CNT01.
The compressive strength values are a direct result of these two contrasting phenomena, and their dependency on the porosity content can be observed in Fig. 8.When the MWCNTs are mixed with water, their distribution is uniform, as shown in Fig. 10a.Once the liquid suspension is mixed with the dry cementitious material, the nanotubes are distributed among the binder particles, creating agglomerates (Fig. 10b-d) [48][49][50].During the mixing process, low concentrations of MWCNTs tend to form large agglomerates due to the attractive Vand der Waal forces (Figs.9b and 10b).These agglomerates were difficult to detect during the SEM analysis, as almost all the nanotubes appeared to be agglomerated into a small number of bundles scattered within the matrix.When the concentration of nanotubes increases, the resulting bundles are more numerous, are formed by a higher concentration of MWCNTs and have smaller dimensions (Fig. 10c and 10d) [50].CNT02 shows a larger number of bundles with a high concentration of nanotubes inside them and some scattered nanotubes distributed in the matrix.CNT05 shows many scattered nanotubes, a high number of bundles easily detectable by the SEM analysis, often close to each other, and with variable sizes due to the high concentration of nanotubes [50,51].

Electronic microscope
The observation of the microstructure from the SEM analysis can help understand how the microstructural characteristics and changes influence the macrostructure and macro-properties of the material [23,52,53].In this case, it can be observed from Fig. 9 that all the samples show the presence of micropores, but when MWCNTs are present inside the matrix, many of the pores are concentrated near or inside the nanotube bundles, with their size being dictated by the agglomerated nanotubes size.CNT01 shows a big pore at the center of a big bundle, while smaller pores are outside the bundle (Fig. 9b).CNT02 and CNT05 show smaller pores at the center of the smaller bundles and pores with varying sizes around them.The reference material shows a compact matrix with CH and some CSH hydration products, while the compositions with nanotubes show a concentration of CSH around the MWCNTs; the nanotubes act as a nucleation site for CSH formation [40].These observations are in agreement with the trend observed in Fig. 6   compositions with a high amount of MWCNTs (CNT02 and CNT05) have a smaller cluster size and, consequently, a smaller pore size than compositions with a lower MWCNTs content (CNT01), which influences the hard state macroscopic behavior, as observed in Section 3.2, and the fresh state behavior investigated in the following sections.

Rheology study
The results of the rheology test for the different compositions are shown in Fig. 11.The mixes show comparable yield stresses over time, except for composition CNT02, which clearly shows a lower static yield stress over time.MWCNTs agglomerate in small clusters that are spread all over the matrix.The clusters trap part of the free water present in the system, causing weaker forces between particles and a delay in hydration that keeps the yield stress evolution low over time [33,54], causing a slower yield stress evolution.This effect is especially visible in CNT02 samples because they have a larger number of clusters, which results in a larger amount of porosity once the material is hydrated, as explained in Sections 3.1 and 3.3.

14.
Structuration rate and porosity content per MWCNTs concentration in the matrix.
To better understand the effects of nanotubes and porosity on the yield stress evolution over time of the fresh compositions, its trend is analyzed and Kruger's model [33] is employed.
The outcomes of Kruger's model are illustrated in Fig. 12 and Table 8.The presence of MWCNTs increases the static and dynamic yield stress of the material since their presence increases the interactions between particles, and it significantly impacts the reflocculation rate of the mixes.For the mix with the lowest concentration of nanotubes (CNT01), the re-flocculation rate increases by approximately 60% compared to the reference mix.This observation suggests that the presence of nanotubes has a favorable impact on the early fresh state of the material before the start of the hydration process.However, as the concentration of nanotubes increases, R thix decreases by about 70% (CNT02) and 60% (CNT05).This phenomenon can be attributed to the fact that CNT01 has a low number of clusters compared to CNT02 and CNT05, with some dispersed nanotubes in the matrix.The dispersed nanotubes have a positive effect on the re-flocculation rate, interacting with the surrounding particles, while the clusters have a negative effect that decreases the Van der Waals forces between the particles.At low concentrations of nanotubes, the clustering effect is counterbalanced by the dispersed nanotubes effect, which results in lower porosity in the hardened state since the nanotubes act as nucleation sites for CSH products, as seen in Section 3.3.At higher concentrations of clusters, however, their effect dominates and the higher number of agglomerated MWCNTs causes a decrease in R thix , as seen in Fig. 12.This phenomenon is even more clear when the hydration starts: the structuration rate A thix decreases for all the mixes with nanotubes since the hydration is slowed by the presence of the clusters that trap part of the free water in the system, as observed in Table 8.CNT01 and CNT05 have, respectively, a 20% and 26% lower A thix than Ref, while CNT02 has a structuration rate of 69% lower than Ref.CNT02 displays the lowest A thix since it has the highest amount of clusters (resulting in the highest porosity at hardened state), while CNT05 has a lower amount of clusters since more nanotubes are dispersed in the matrix, resulting in lower porosity at hardened state and a higher structuration rate compared to CNT02.
The time at which the structuration rate starts (and the first hydration products appear) also demonstrates the delayed hydration: Ref begins hydrating after 44 s, CNT01 after 43 s, CNT02 after 537 s, and CNT05 after 317 s In Figs. 13 and 14 the relationship between porosity, re-flocculation rate and structuration rate per MWCNTs concentration, respectively, is presented.Since the amount of porosity is an indication of the clusters amount in the fresh state, it can be observed a  perfect matching trend between the values, in which increasing one decreases the porosity and vice versa.
Although the structuration rate decreases with the presence of MWCNTs, according to Roussel [55], mixes with A thix greater than 0.5 Pa/s are to be considered highly thixotropic.In this study, all the mixes maintain their thixotropic behavior.

Unconfined uniaxial compression test
Fig. 15 and Table 9 show the compressive strength evolution over time for the fresh mixes.
The initial compressive strength of samples with MWCNTs is higher than Ref, in agreement with the results in Section 3.4.CNT01 has the highest value of initial strength, while this value decreases in CNT02 and CNT05.The behavior depends on the amount of clusters that affects the porosity once the material hardens and on the distribution of MWCNTs in the matrix.CNT01 has a smaller number of clusters due to the lower concentration of nanotubes and the fact that most of the nanotubes are agglomerated, resulting in a higher initial compressive strength in the fresh state and a lower porosity in the hardened state.CNT02 and CNT05 have a higher quantity of clusters, resulting in a lower initial compressive strength in the fresh state and higher porosity in the hardened state, but still show a higher initial strength than Ref, thanks to the presence of dispersed nanotubes that strengthen the matrix.The trend of the initial strength per MWNCTs content is compared to the porosity content, and the results are shown in Fig. 16.The porosity trend is opposite to the initial strength trend; in accordance with the results in Sections 3.2, 3.3 and 3.4, the microstructure is affected by the MWCNTs dispersion, which in turn affects both the fresh and hardened states of the material.The Young's modulus evolution over time for the different mixes is presented in Fig. 17 and Table 10.CNT01 CNT02 and CNT05 exhibit a higher initial stiffness compared to Ref: the presence of nanotubes makes the material stiffer, which is in accordance with the results in Section 3.4, in which the initial static yield stress was higher for the mixes with nanotubes than Ref, indicating a higher stiffness.The initial stiffness is, as expected, based on the results in the previous tests, affected by the clusters size and distribution, and the trend is compared to the porosity trend in Fig. 18.The porosity trend is opposite to the stiffness trend, in accordance to previous results.

Comparison between UUCT and the rheology test
The compressive strength evolution over time presented in Section 3.5 is not consistent with previous tests in Section 3.4.It is expected to see a slower compressive strength evolution over time in the samples containing MWCNTs, due to the effects of the clusters on the hydration [56], as explained in Section 3.4.The results, however, indicate a faster strength evolution for CNT01 and CNT02 compared to Ref.This inconsistency between the results from the two tests was also found in other studies [57,58].
One simple explanation could be that the standard deviation between the samples of the UUCT is very high since the samples were handled manually to be de-molded and placed in the Instron, which could have cause variations in the measurements, making them less trustworthy.However, since similar results were found in other studies, other explanations regarding this behavior are possible.
When comparing the two tests, the testing timeline is different: the rheological test was conducted at different resting times, from 0 up to 30 min, while the UUCT was conducted at different times from preparation, from 0 up to 90 min.During the UUCT, in the first 30 min from preparation, the material was still soft resulting in very few samples tested in the time window from 0 to 30 min from preparation, with inconsistent results: no sample was analyzed at time 0 and very few were analyzed at time 15 min from preparation.This suggests that most of the data come from samples older than 30 min, making the comparison between the rheological behavior and UUCT difficult.Finally, it was assumed that the strength development over time followed a linear trend, while this may not be the case when considering a longer period of time (above 30 min from preparation).
To better understand the strength development and the relationship with the rheological behavior of the material, we need to address what chemical reactions occur from the first contact with water up to several minutes after the mixing process.Within the first minutes, the material undergoes a rapid hydration process, during which there is the dissolution period [59] of alite, where the first CSH products are formed.This is followed by an induction period that can last from one to a few hours, during which the hydration process slows down [59][60][61][62].
As observed in Fig. 11, the first rapid hydration process for Ref and CNT01 starts at 1.5 and 1.2 min from preparation, respectively, Fig. 19.Compressive strength values at different resting times of the fresh state mixes, exponential fit.

Table 11
Exponential evolution over time of the compressive strength for all the studied compositions.and shows a steep yield strength evolution until they enter the induction period [59][60][61][62], which can be observed in the graph as a slow down in evolution as a consequence of a slow down in hydration, which happens after 15 min and 10 min for Ref and CNT01 respectively.
CNT02 and CNT05 on the other hand, have a delayed start of hydration compared to CNT01 and Ref, due to the presence of a high number of clusters.The hydration starts at around 10 and 5 min from preparation, for CNT02 and CNT05 respectively, after which a steady hydration rate is maintained.The samples don't show any change in yield strength evolution trend within the 30 resting minutes in which the rheology test is performed, so it is assumed that they don't enter the induction period withing these 30 min.
To understand what happens after 30 min, we can look at the UUCT data proposed in Fig. 19.The data reported are the ones

Table 12
Gauge factor under compression for all the studied compositions.A. Dulaj et al. collected between 30 and 90 min from preparation, and the model proposed is not linear but exponential.In Table 11 the exponential fit data are reported, showing a very high R 2 .
The exponential fit shown in Table 11 is more in agreement with the results obtained from the rheological test, than the linear fit proposed in Fig. 12.The samples with nanotubes show a slower trend than Ref, and C02 and C05 show a faster strength development than CNT01 due to the presence of dispersed nanotubes in the matrix that act as nucleation points for CSH products, as explained in Section 3.3.It is not clear, however, why CNT02 has a faster strength development than CNT05: according to Fig. 11, after the start of the hydration, CNT05 shows a faster rate, which is expected due to a higher number of dispersed nanotubes.This result does not match the results in Fig. 19 and Table 11.Further research is needed to understand how hydration is affected by the presence of nanotubes and how this translates to UUCT and rheological results.

Resistivity test
The results of the resistivity test on cast specimens subjected to compressive stress are presented in Fig. 20.Except for mix CNT02, the resistivity change increases as the concentration of MWCNTs in the matrix increases.The compression stress applied to the samples till failure prompts a decrease in resistivity promoting a better electron flow through the material [18,63].This effect is enhanced by the presence of MWCNTs.
The resistivity change doesn't increase linearly with the increased presence of MWCNTs as could be expected, but the behavior is highly affected by the porosity content of the matrix, as shown in Fig. 21.The resistivity change increases when the porosity decreases and vice versa.CNT01 has the lowest amount of porosity which results in a high resistivity change.However, CNT05 has the highest number of conductive MWCNTs, resulting in the highest resistivity change even if the porosity content is higher than CNT01.
In Table 12 the gauge factor of the compositions is reported.CNT05 shows the highest gauge factor.Moving forward, only the CNT05 mixes will be tested and compared to the reference as they have shown the best sensing capability with the lowest standard deviation.
In Fig. 22 the tension stress-displacement graphs of Ref and CNT05 mixes are presented.For this test, both cast and printed specimens were analyzed.The cast specimens exhibit similar behavior for both Ref and CNT05 mixes.The printed samples, on the other hand, show a lower Young's modulus.CNT05 printed mixes show the lowest Young's modulus.From these results, it appears that the presence of MWCNTs affects the ductility of the material [64] when printed.The printing process may influence the orientation of the nanotubes due to high shear forces exercised by the nozzle on the material's surface and a lubrication layer forms as a consequence [43,65,66].
Figs. 23 and 24 depict the resistivity behavior of cast and printed Ref and CNT05 mixtures in response to stress and displacement.There is no discernible resistivity response observed in the reference mixes due to low levels of applied deformation and stress.However, the CNT05 mixes showcase a significant change in resistivity when subjected to such conditions.Cast and printed samples exhibit similar increase in resistivity per unit tensile deformation, whereas printed samples display a greater increase in resistivity in response to applied stress.This indicates that the printed samples have a more elastic behavior, with a lower Young's modulus, as confirmed in Fig. 21, and higher sensitivity to stresses compared to applied deformation.
In Table 13, the gauge factor of Ref and CNT05 is reported.The values are very high for CNT05, demonstrating a high sensitivity to tensile strains.As seen in other studies [67][68][69], the gauge factor is very high when the stain is within 100 με.In this case, the values are comparable to the results obtained by Zhan et al. [69].Printed samples exhibit the highest gauge factor, suggesting again an orientation of the nanotubes during the printing process.However, due to the high standard deviation in the results, further studies are needed to establish the orientation process of the nanotubes.A. Dulaj et al.

Energy dispersive x-ray analysis
EDX analysis results of printed layers are displayed in Fig. 25.The test was performed analyzing the surface and the inner part of printed layers, and the findings indicate that there is a higher concentration of carbon atoms on the surface compared to the inner part of the printed layer [66].These atoms come partially from the superplasticizer and partially from MWCNTs.
The concentration of carbon atoms is higher in CNT05 mixes, which contain nanotubes, compared to the reference.This suggests that the distribution of MWCNTs may not be uniform.Based on the results presented in Section 3.6 and considering the high shear forces exerted on the extruded material from the nozzle walls, it is possible that the nanotubes are subjected to some degree of alignment, at least on the surface [43,65,66].However, due to the high standard deviation, these hypotheses should be further investigated in future studies, particularly through a detailed characterization of the microstructure of both the surface and the inner part of a printed layer.
To understand the significance of the data, a statistical analysis was performed.The data were tested for normality with the Shapiro-Wilk normality test and the results are reported in Table 14.The significance level of the data was selected at 5% and, as seen in Table 14, the p-value of the test is far above 5% for all the data.
In Figs. 26 and 27 the histogram distribution and the density distribution of the collected data for C05 are reported.As confirmed by the test for normality, the data have a normal distribution.The data collected from the surface are closer to normality, as shown by Fig. 27a and thus have a higher p-value than the data collected from the inner part.
In Fig. 28 and Fig. 29 the histogram and the density of the data distribution for Ref is reported.Similarly to C05, the data distribution is normal.However, in this case the normality is not straightforward as for C05, especially for the data collected in the inner printed layer, having a lower p-value (10%) and two peaks instead of one.
Since all the data resulted normally distributed, a t-test could be performed to assess if there is significant difference between the collected data (which means a significant difference in carbon concentration between the surface and the inner part of a printed layer).The t-test was performed assuming the data had different variances and the results are reported in Table 15.
The variance resulted different for each data set, as shown in Table 15.The significance level of the data was selected at 5% and the main hypothesis was that the values of the carbon concentration between the inner part and the surface of a printed layer were the same, which means there was no significant difference between the two.
The p-value for Ref resulted way above 5%, confirming that the data don't show a significant difference between core and surface of a printed layer.
The p value for C05, however, resulted lower than 5%: 1.9%.This indicates that there is a significant difference in carbon concentration between the surface and the core for samples containing MWCNTs.A p-value of 1.9%, however, is still quite close to the significance level, which means that a difference exists but more data need to be collected to lower the p-value and the standard deviation, and/or different tests are required to investigate the microstructure and better understand what is happening at the surface when the material is printed.

Conclusions
In this study, MWCNTs were added at different concentrations to a 3D printable concrete to assess the effects of the carbon nanomaterial on the porosity distribution, fresh and hardened state mechanical properties, rheological behavior, microstructure and self-sensing ability under compression and tension.
The results show that the distribution of MWCNTs in the matrix depends on the concentration of the carbon nanomaterial and has a strong impact on the microstructure and porosity distribution of the composites.
At the fresh state, MWCNTs aggregate in clusters of different sizes, that, once the material hardens, remain as pores in the matrix, influencing the porosity distribution.The clusters amount influenced the rheological behavior (the re-flocculation and structuration rate increased when the porosity amount decreased in hardened samples, Fig. 13 and Fig. 14) the fresh and hardened state mechanical behavior (the strength increased when the porosity decreased, Fig. 16 and Fig. 8) and the resistivity measurements (the resistivity   change in compression increased when the porosity decreased, Fig. 20).
The composition with the worst fresh and hardened state performance and self-sensing ability was the one with 0.2% of MWCNTs by weight (CNT02), while the composition with overall better performance was the one with 0.5% of MWCNTs by weight (CNT05).CNT05 was printed and the self-sensing capability was tested in tension: the results show high sensitivity to applied strain, with a sharp increase in resistivity (Fig. 23 and Fig. 24).
Finally, an EDX analysis was performed on printed CNT05 samples with the results suggesting that the distribution of carbon atoms (and subsequently of MWCNTs) might not be uniform in printed samples, with a higher concentration on the surface of a printed layer (Fig. 25).

Declaration of Competing Interest
The authors declare the following financial interests/personal relationships which may be considered as potential competing interests: Albanela Dulaj reports financial support was provided by Dutch Research Council.

Fig. 2 .
Fig. 2. Representation of a sample under compression while the resistivity is being measured with Resipod.

Fig. 4 .
Fig. 4. Cumulative pore volume for the reference mix and the mixes with MWCNTs.

Fig. 5 .
Fig. 5. Log differential intrusion for the reference mix and the mixes with MWCNTs.

Fig. 6 .
Fig. 6. a) Median pore diameter per porosity percentage, and b) total pore area per porosity percentage for the different compositions.

Fig. 8 .
Fig. 8. Porosity values and compressive strength values per MWCNTs concentration in the compositions.

Fig. 9
Fig.9shows SEM pictures of the different compositions.The images agree with the findings from Section 3.1.MWCNTs form bundles of different sizes due to their high surface energy, and the cluster sizes appear to vary at different nanotube concentrations: CNT01 shows bigger clusters compared to CNT02 and CNT05, as seen in Figs.9b, 9c and 9d.The size of the bundles depends on the concentration of nanotubes in the compositions.When the MWCNTs are mixed with water, their distribution is uniform, as shown in Fig.10a.Once the liquid suspension is mixed with the dry cementitious material, the nanotubes are distributed among the binder particles, creating agglomerates (Fig.10b-d)[48][49][50].During the mixing process, low concentrations of MWCNTs tend to form large agglomerates due to the attractive Vand der Waal forces (Figs.9b and 10b).These agglomerates were difficult to detect during the SEM analysis, as almost all the nanotubes appeared to be agglomerated into a small number of bundles scattered within the matrix.When the concentration of nanotubes increases, the resulting bundles are more numerous, are formed by a higher concentration of MWCNTs and have smaller dimensions (Fig.10c and 10d)[50].CNT02 shows a larger number of bundles with a high concentration of nanotubes inside them and some scattered nanotubes distributed in the matrix.CNT05 shows many scattered nanotubes, a high number of bundles easily detectable by the SEM analysis, often close to each other, and with variable sizes due to the high concentration of nanotubes[50,51].The observation of the microstructure from the SEM analysis can help understand how the microstructural characteristics and changes influence the macrostructure and macro-properties of the material[23,52,53].In this case, it can be observed from Fig.9that all the samples show the presence of micropores, but when MWCNTs are present inside the matrix, many of the pores are concentrated near or inside the nanotube bundles, with their size being dictated by the agglomerated nanotubes size.CNT01 shows a big pore at the center of a big bundle, while smaller pores are outside the bundle (Fig.9b).CNT02 and CNT05 show smaller pores at the center of the smaller bundles and pores with varying sizes around them.The reference material shows a compact matrix with CH and some CSH hydration products, while the compositions with nanotubes show a concentration of CSH around the MWCNTs; the nanotubes act as a nucleation site for CSH formation[40].These observations are in agreement with the trend observed in Fig.6in Section 3.1:

Fig. 9 .
Fig.9shows SEM pictures of the different compositions.The images agree with the findings from Section 3.1.MWCNTs form bundles of different sizes due to their high surface energy, and the cluster sizes appear to vary at different nanotube concentrations: CNT01 shows bigger clusters compared to CNT02 and CNT05, as seen in Figs.9b, 9c and 9d.The size of the bundles depends on the concentration of nanotubes in the compositions.When the MWCNTs are mixed with water, their distribution is uniform, as shown in Fig.10a.Once the liquid suspension is mixed with the dry cementitious material, the nanotubes are distributed among the binder particles, creating agglomerates (Fig.10b-d)[48][49][50].During the mixing process, low concentrations of MWCNTs tend to form large agglomerates due to the attractive Vand der Waal forces (Figs.9b and 10b).These agglomerates were difficult to detect during the SEM analysis, as almost all the nanotubes appeared to be agglomerated into a small number of bundles scattered within the matrix.When the concentration of nanotubes increases, the resulting bundles are more numerous, are formed by a higher concentration of MWCNTs and have smaller dimensions (Fig.10c and 10d)[50].CNT02 shows a larger number of bundles with a high concentration of nanotubes inside them and some scattered nanotubes distributed in the matrix.CNT05 shows many scattered nanotubes, a high number of bundles easily detectable by the SEM analysis, often close to each other, and with variable sizes due to the high concentration of nanotubes[50,51].The observation of the microstructure from the SEM analysis can help understand how the microstructural characteristics and changes influence the macrostructure and macro-properties of the material[23,52,53].In this case, it can be observed from Fig.9that all the samples show the presence of micropores, but when MWCNTs are present inside the matrix, many of the pores are concentrated near or inside the nanotube bundles, with their size being dictated by the agglomerated nanotubes size.CNT01 shows a big pore at the center of a big bundle, while smaller pores are outside the bundle (Fig.9b).CNT02 and CNT05 show smaller pores at the center of the smaller bundles and pores with varying sizes around them.The reference material shows a compact matrix with CH and some CSH hydration products, while the compositions with nanotubes show a concentration of CSH around the MWCNTs; the nanotubes act as a nucleation site for CSH formation[40].These observations are in agreement with the trend observed in Fig.6in Section 3.1:

Fig. 10 .
Fig. 10.a) SEM analysis of AQUACYL AQ030X dispersed in water and superplasticizer, with a MWCNT concentration of 0.5%wt, and a schematic representation of the distribution of MWCNTs in b) CNT01, c) CNT02 and d) CNT05.

Fig. 11 .
Fig. 11.Static yield stress resting time gap the different compositions.

Fig. 12 .
Fig. 12. Static yield stress evolution over resting time of the mixes.

Fig. 13 .
Fig. 13.Re-flocculation rate and porosity content per MWCNTs concentration in the matrix.

Fig. 15 .
Fig. 15.Compressive strength values at different resting times of the fresh state mixes.

Fig.
Fig. Initial compressive strength values and porosity values for different MWCNTs concentrations.

Fig. 17 .
Fig. 17.Young's modulus values at different resting times of the fresh state mixes.

Fig. 20 .
Fig. 20.Resistivity change of the cast mixes subjected to compressive stress with a rate of 0.5 mm/s.

Fig. 21 .
Fig. 21.Resistivity change and porosity of the mixes at different MWCNTs concentrations.

Fig. 22 .
Fig. 22. Stress-displacement curves of Ref and CNT05 mixes of cast and printed specimens.

Fig. 23 .
Fig. 23.Resistivity-displacement curves of Ref and CNT05 mixes of cast and printed specimens.

Fig. 24 .
Fig. 24.Resistivity-stress curves of Ref and CNT05 mixes of cast and printed specimens.

Fig. 25 .
Fig. 25.Carbon atoms concentration on the surface and inner part of printed layers of Ref and CNT05 specimen.

Fig. 26 .
Fig. 26.Data distribution of the carbon concentration of C05 on a) the surface of a printed layer, and b) the inner part of a printed layer.

Fig. 27 .
Fig. 27.Data density distribution of the carbon concentration of C05 on a) the surface of a printed layer, and b) the inner part of a printed layer.

Fig. 28 .
Fig. 28.Data distribution of the carbon concentration of Ref on a) the surface of a printed layer, and b) the inner part of a printed layer.

Fig. 29 .
Fig. 29.Data density distribution of the carbon concentration of Ref on a) the surface of a printed layer, and b) the inner part of a printed layer.

Table 1
Composition of the printable concrete.

Table 3
Rheometer test profile.

Table 4
Mercury intrusion porosimetry test results for all the compositions.

Table 5
Porosity distribution per pore diameter, for all the compositions.

Table 6
Measured strengths values and calculated strengths values depending on the porosity content, for the compositions.

Table 7
Calculated values of the strength at zero porosity for all the compositions.

Table 8
Initial static and dynamic yield stress, re-flocculation and structuration rate of the mixes.

Table 9
Compressive strength evolution over time for all the studied compositions.

Table 10
Stiffness evolution over time for all the studied compositions.

Table 13
Gauge factor under tension for Ref and C05.

Table 14
Carbon concentration inside and on the surface of printed layers for Ref and C05, and test of normality of the collected data.

Table 15
Variance and p-value of the t-test for Ref and C05.