# Multi-scale numerical simulations on piezoresistivity of CNT/polymer nanocomposites

- Bin Hu
^{1}, - Ning Hu
^{2}Email author, - Yuan Li
^{3}, - Kentaro Akagi
^{2}, - Weifeng Yuan
^{4}, - Tomonori Watanabe
^{2}and - Yong Cai
^{4}

**7**:402

**DOI: **10.1186/1556-276X-7-402

© Hu et al.; licensee Springer. 2012

**Received: **29 May 2012

**Accepted: **30 June 2012

**Published: **17 July 2012

## Abstract

In this work, we propose a comprehensive multi-scale three-dimensional (3D) resistor network numerical model to predict the piezoresistivity behavior of a nanocomposite material composed of an insulating polymer matrix and conductive carbon nanotubes (CNTs). This material is expected to be used as highly sensitive resistance-type strain sensors due to its high piezoresistivity defined as the resistance change ratio divided by the mechanical strain. In this multi-scale 3D numerical model, three main working mechanisms, which are well known to induce the piezoresistivity of strain sensors fabricated from nanocomposites, are *for the first time* considered systematically. They are (a) the change of the internal conductive network formed by the CNTs, (b) the tunneling effect among neighboring CNTs, and (c) the CNTs’ piezoresistivity. Comparisons between the present numerical results and our previous experimental ones were also performed to validate the present numerical model. The influence of the CNTs’ piezoresistivity on the total piezoresistivity of nanocomposite strain sensors is explored in detail and further compared with that of the other two mechanisms. It is found that the first two working mechanisms (i.e., the change of the internal conductive network and the tunneling effect) play a major role on the piezoresistivity of the nanocomposite strain sensors, whereas the contribution from the CNTs’ piezoresistivity is quite small. The present numerical results can provide valuable information for designing highly sensitive resistance-type strain sensors made from various nanocomposites composed of an insulating polymer matrix and conductive nanofillers.

### Keywords

Carbon nanotube Conductive network Nanocomposite Piezoresistivity Tunneling effect## Background

Due to their excellent electrical properties, carbon nanotubes (CNTs) of high aspect ratio are predicted to open up a whole range of smart structural applications [1]. In particular, it has been confirmed that the conductance of a CNT could be dramatically changed by introducing a mechanical strain as a consequence of structural change (e.g., chirality change [2]). Due to this piezoresistivity of CNTs, great interest has recently been aroused in building strain sensors not only using CNTs themselves (either single-walled carbon nanotube (SWCNT) or multi-walled carbon nanotube (MWCNT)) [3–5] but also incorporating them into an insulating polymer matrix [6–9], as summarized in a recent review paper [10]. As compared to conventional metal-foil strain gauges, higher sensitivities have been observed in these novel sensors at least at a macro-scale [5–9]. Moreover, a linear piezoresistivity has been identified within very small strain ranges (e.g., 200 με [4] or 1,300 με [5]), whereas nonlinear piezoresistivity has also been reported for a quite large strain range (e.g., 6,000 με [6, 7]). In spite of these exciting results, the fundamental understanding of piezoresistivity behavior in a CNT/polymer nanocomposite is still lacking, largely due to less effort being put into theoretical and numerical investigations.

Based on a three-dimensional (3D) statistical resistor network model [11], we have previously considered the change of internal CNT conductive network and tunneling effect in exploring the piezoresistivity behavior of the above CNT/polymer nanocomposite strain sensors [12]. The movement of CNTs in the polymer under a given strain is firstly predicted using a rigid-body fiber reorientation model. Then, by integrating this fiber reorientation model into the 3D resistor network model in an iterative way, the resistance change of the nanocomposites caused by the applied strain is estimated. Moreover, the predicted electrical conductivity and piezoresistivity of CNT/polymer nanocomposites are verified using our previous experimental data [6–8]. Furthermore, the influence of various parameters (i.e., cross-sectional area of tunnel current, height of barrier, alignment of CNTs, electrical conductivity, and nanofiller electrical conductivity) are systematically investigated [12]. However, the effect of the piezoresistivity of CNTs themselves is not mentioned.

*for the first time*, the following three working mechanisms are incorporated simultaneously:

- (a)
The change of the internal conductive network formed by the CNTs themselves.

- (b)
The tunneling effects among neighboring CNTs.

- (c)
The piezoresistivity of CNTs themselves.

To include the CNTs’ piezoresistivity effect in a multi-scale way, we firstly calculate the strain of an embedded CNT by applying a mechanical tensile strain on a micro-mechanics-based cylindrical representative volume element (RVE), which simulates the CNT deformation for the case of multiple one-directionally aligned CNTs in a polymer matrix. Then, we transform the strain of the CNT into that of an arbitrary-oriented CNT dispersed in macroscopic CNT/polymer nanocomposites. By employing this strain information and the piezoresistivity of CNTs themselves predicted from the first-principle computations at the atomistic scale [13, 14], we update the electrical conductance of the CNTs in the internal conductive network of the macroscopic nanocomposite sensors under the applied strain. By using this updated electrical conductance of CNTs and incorporating the internal network change of CNTs and tunneling effect simultaneously, we finally comprehensively evaluate the piezoresistivity of the nanocomposite strain sensors.

## Methods

So far, the resistor network model has been well documented [15] to predict the electrical conductivity of some conductive composites. However, because of its high computational cost, the corresponding numerical studies have been limited [15, 16], even for that of conventional composites with short carbon fibers. There has been no work about its extension for predicting the piezoresistivity of CNT/polymer nanocomposites, except for our previous work [12]. To maintain the integrity of the present work, our previously improved 3D resistor network model [12] is briefly described.

### An improved 3D resistor network model without the effect of CNT’s piezoresistivity

^{−10}S/m. Moreover, the aggregation of CNTs is also neglected, according to our previous experimental studies [6, 7] in which no obvious aggregation is observed in the nanocomposites filled with straight MWCNTs of a comparatively large diameter (≥50 nm).

*L*and a diameter of

*D*. For the ideal state of uniformly dispersed straight CNTs in Figure 2, the simulations were carried out using a procedure to identify those CNTs being in contact, until the CNT loading reaches the required volume fraction of the CNTs in the 3D unit cell to form a global conductive network [11]. To construct the 3D resistor network model (a simple 2D model is given in Figure 2), for a CNT with two points,

*i*and

*j*, contacting with two other neighboring CNTs, the conductance

*g*

_{ij}between

*i*and

*j*(the inverse of resistance

*R*

_{ ij }) can be evaluated as:

*l*

_{ij}is the length between the points

*i*and

*j*, and

*σ*

_{CNT}and

*S*

_{CNT}are the electrical conductivity and the cross-sectional area of the CNTs, respectively.

*i*(Figure 3), if two CNTs are in contact or penetrate each other, we

*intentionally*separate them with the shortest distance of 0.47 nm to avoid further penetration and then add tunneling resistance

*R*

_{tunnel}between them. This operation is very time-consuming for high CNT loading cases since there may be multiple contact points for one CNT with other neighboring CNTs. However, for low CNT loading cases discussed here, this operation is comparatively simple. The shortest distance (0.47 nm) is determined by matching the numerical predicted conductivity and piezoresistivity with the experimental ones. It is reasonable to set up this shortest distance (0.47 nm) since, in general, the equilibrium distance between two independent carbon atomistic structures under Lennard-Jones potential or van der Waals force ranges approximately from 0.3 to 0.5 nm. Furthermore, it is interesting to find that the shortest distance is much lower than the average diameter of the CNTs in the numerical simulations (i.e., 50 nm). Therefore, the range of distance

*d*between two central lines of two cylindrical CNTs to introduce the tunneling effect should be

*D*+

*d*

_{min}

*D*+

*d*

_{max}, where

*d*

_{min}and

*d*

_{max}are 0.47 nm and 1.0 nm, respectively.

where *J* is tunneling current density (see Figure 3), *V* is the electrical potential difference, *e* is the quantum of electricity, *m* is the mass of electron, *h* is Planck’s constant, *d* is the distance between CNTs, *λ* is the height of the barrier (for epoxy, 0.5 to 2.5 eV), and *A* is the cross-sectional area of the tunnel (the cross-sectional area of CNT is approximately used here).

Based on the well-known matrix representation for a resistor network [15] and Kirchhoff’s current law, the total electrical current *I* under an applied voltage can be estimated (see Figure 2). This is a large-scale linear system because the number of CNTs involved in the numerical model is very large, which ranges from several thousands, possibly up to several tens of thousands, depending on the aspect ratio of the CNTs. An iterative equation solver (i.e., the incomplete Cholesky conjugate gradient method) has been used to solve these linear equations for obtaining the total electrical current *I*. Then, the macroscopic electrical conductivity of the nanocomposites can be evaluated by Ohm’s law, which is used for predicting the resistance of the nanocomposites.

The reorientation of CNTs under the applied strain is simplified as a rigid-body movement due to a much higher Young’s modulus of CNTs compared with a polymer matrix and the very weak interface between CNTs and the matrix [6]. Therefore, the change of CNT position and orientation caused by the mechanical strain can be evaluated using the 3D fiber reorientation model [16] based on an affine transformation (see Figure 3). Corresponding to an updated CNT distribution under a prescribed mechanical strain, a new CNT network can be formed by re-calculating the possible intersections between CNTs and tunneling resistances between CNTs within the cutoff distance.

The switch of the CNT intersections to a possible tunneling effect due to the breakup of CNT contacts and the update distance of pre-existing tunneling effects were modeled. Then, the resistance of a nanocomposite at a fixed strain level can be re-evaluated using the 3D resistor network in an iterative way to finally calculate the piezoresistivity of the nanocomposite.

### Extension of the above model with considering the piezoresistivity of CNT

*D*

_{i}to the external diameter

*D*

_{o}is around 1:10. The length of the MWCNT is an average value provided by the maker (Nano Carbon Technologies Co., Ltd., Tokyo, Japan), i.e., 5 μm. The cap of the MWCNT was neglected in this RVE model. The sizes of the RVE are shown in Figure 7, where

*R*is used as a parameter to adjust the CNT loading in nanocomposites. Moreover, the parameters used in the RVE are summarized in Table 1. The previously reported mechanical properties of CNTs and epoxy [6, 7] were directly adopted.

**Parameters of the CNT and polymer**

Parameter | Value | |
---|---|---|

CNT | Length (μm) | 5 |

Diameter (nm) | 50 | |

Loadings (wt.%) | 2, 3, 4, 5 | |

Young’s modulus (GPa) | 1,000 | |

Poisson’s ratio | 0.1 | |

Polymer | Length (μm) | 10 |

Young’s modulus (GPa) | 2.73 | |

Poisson’s ratio | 0.34 |

*ε*

_{Poly}ranging from 0.0 % to 0.6 % on the top surface of the polymer matrix, the strain of the CNT (i.e.,

*ε*

_{CNT}) at various CNT loadings can be calculated from the average axial displacements measured at the top surface of the CNT in the FEA. As plotted in Figure 8, the strain of the CNT is linearly proportional to that applied on the polymer, which can be expressed as:

*α = ε*

_{CNT/}

*ε*

_{Poly}) at various CNT loadings are summarized in Table 2. It can be found that with the increase of CNT loading,

*α*decreases significantly as the deformation of nanocomposites is transferred to and distributed on more CNTs.

**Strain ratio**
α
**of CNT to polymer at various CNT loadings**

CNT loadingW | Strain ratioα in Equation3 |
---|---|

2 | 0.183 |

3 | 0.134 |

4 | 0.106 |

5 | 0.088 |

*θ*= 0 in Figure 9. For an oblique CNT with the strain of

*ε*

_{ θ }, we have

*ε*

_{ θ }with only several percent errors by comparing with the FEA results stated above.

*θ*can be determined from the coordinates of the two intersection points as follows:

where *l*_{ij} is the distance of the two intersection points calculated from (*x*_{i}, *y*_{i}, *z*_{i}) and (*x*_{j}, *y*_{j}, *z*_{j}).

where Δ*R* is the resistance change of a SWCNT due to a strain *ε*_{
θ
}, and *R*_{0} is the initial resistance of this SWCNT.

The samples of SWCNTs used to obtain the average piezoresistivity response in Equation 6 were (8,0), (8,1), (8,2), (8,8), etc., and their diameters were very small [14]. For CNTs with large diameters, e.g., 50 nm in this work, the influence of diameter on Δ*R*/*R*_{0} can be neglected since Equation 6 only describes the resistance change caused by quantum effects due to CNT lattice deformation based on the Landauer formula, in which the size effect on Δ*R*/*R*_{0} is removed. Moreover, the influences of the changes of CNT’s length and cross-sectional area in classical electrical equations are neglected based on the following reasons: (1) they are very small under small strains, and (2) quantum phenomena are dominant at the atomistic level.

Furthermore, in Table 1, the Young’s modulus of the CNTs was taken as 1,000 GPa, which is slightly higher than most reported values for MWCNTs, e.g., 500 to 800 GPa. By using a lower Young’s modulus for a CNT, its strain should increase. However, by considering the weak piezoresistivity of CNTs in Equation 6 and low CNT loading, this strain increase cannot lead to remarkable piezoresistivity change of nanocomposite sensors.

where *S*_{CNT} is the cross-sectional area of a CNT, and *l* is the CNT length. Note that in this work, the initial electrical conductivity *σ*_{0} is taken as 1 × 10^{4} S/m.

*R*under a strain

*ε*

_{ θ }, its electrical conductivity can be described as:

*ε*

_{Poly}is applied on a nanocomposite, the coordinates of the two intersection points i and j in the arbitrary CNT segment as shown in Figure 9 can be updated as follows by using the 3D fiber reorientation model [16] based on an affine transformation:

*ν*is the Poisson’s ratio of the nanocomposite. Note that although we consider the resistance change of a strained CNT here, the changes of geometries of the CNT (e.g., cross-sectional area of CNT, i.e.,

*S*

_{CNT}, and CNT length, i.e.,

*l*) are neglected for simplicity in Equation 9 since basically, after the axial deformation with a strain

*ε*, the changes of the length and cross-sectional area of a CNT can be described as

*l*+

*εl*, and

*S*

_{CNT}-ν

^{2}

*ε*

^{2}

*S*

_{CNT}, which are very small for the maximum strain 0.6 % considered in this work. Therefore, in the above relationship for updating the coordinates, we also neglect the coordinate changes caused by the strain of the CNT. After obtaining the new coordinates of the intersections, we can calculate the new distance between i and j as:

By substituting Equations 10 and 11 into Equations 5 and 9, we can update the orientation angle of the CNT and the electrical conductance *g*_{ij} between i and j.

Note that besides update of the electrical conductivity of the CNT segment on the internal conductive network, we also update the internal CNT network by judging the loss or formation of contacting points, and update the tunneling effect by changing the distance among those neighboring CNTs, as done in our previous work [12].

## Results and discussion

*λ*in Equation 2 was approximately determined by matching the zero-strain numerical electrical conductivities with the experimental ones at a reasonable range. To reduce the computational cost, the dimension of the unit cell (length × width × thickness) was set as 25 × 25 × 15 μm, which has been estimated to be large enough for an isotropic behavior and numerical convergence. The geometrical parameters of practical MWCNTs in our previous experiments [6, 7] were directly adopted with length of

*l*= 5 μm and diameter of

*D*= 50 nm. For those CNTs with a mutual distance between 0.47 and 1 nm, the tunneling effect was incorporated between them, as stated previously.

**Computational conditions for present numerical simulation**

Parameter | Value | |
---|---|---|

CNT | Length (μm) | 5 |

Diameter (nm) | 50 | |

Electrical conductivity (S/m) | 1 × 10 | |

CNT loading | 2, 3, 4, 5 | |

Barrier height | 1.5 | |

Tunneling distance range (nm) | Minimum value | 0.47 |

Maximum value | 1 | |

Unit cell size (μm | 25 × 25 × 15 |

*K*= 2). In this figure, it can be seen that the numerical simulations can qualitatively catch the main trend of the experimental results under tensile strains although the numerical results are higher. The nonlinear behaviors of numerical and experimental piezoresistivities can be observed clearly. It implies that the tunneling effect plays a very important role in determining the overall performance of the nanocomposites at low CNT loading. According to Equation 2, a 0.1nm increase of the distance between two CNTs (i.e.,

*d*) can lead to ten times lower tunneling current (

*λ*= 0.5 eV,

*A*= π(

*D*/2)

^{2}, and

*D*= 50 nm). From Equation 2, the tunneling resistance increases exponentially with the average distance

*d*. As the average distance

*d*is approximately assumed to change proportionally with an applied strain, the nonlinear relationship between the resistance change ratio and the applied strain is expected. Especially for the cases of low CNT loadings and high strains, this nonlinear behavior becomes more obvious, which indicates the significance of tunneling effect. Moreover, as similar to some experimental studies [6, 7, 9], low CNT loading cases show large resistance change ratios, which indicates the increase of the sensor sensitivity. The reason can be explained as follows. Generally, the resistance change is mainly caused by the breakup of CNT conductive network or tunneling effects. For a sparse conductive network, the breakup of a few conductive paths can lead to a huge increase of sensor resistance, which results in the higher sensor sensitivity. Inversely, for an intensive CNT conductive network, the breakup of a few conductive paths cannot cause a significant change of total sensor resistance.

Note that the CNTs in the present numerical model are distributed uniformly at random without considering their possible aggregations. From our recent experimental results (NH et al., unpublished data), some moderate aggregations can lead to a sparse conductive network of CNTs, and then result in a higher sensor sensitivity. However, in this case, the stability and repeatability of nanocomposite sensors decrease especially for the case of low CNT loading. Moreover, we deal with the polymer matrix as a completely insulating one. Some polymers may be electrically conductive to a certain extent. In this case, this point may also influence the piezoresistivity of nanocomposite sensors. Although it is quite difficult to precisely estimate this influence, it can be speculated from the following two aspects. Firstly, by considering Ohm’s law, the resistance *R*_{p} of a polymer matrix only is proportional to *L*/*A*, where *L* and *A* are the length and cross-sectional area of a nanocomposite sensor. Then, *R*_{p} will change by Δ*R*_{p} since both *L* and *A* vary when the sensor is subjected to a strain. Therefore, the total piezoresistivity of the nanocomposite sensor should increase when considering the contribution of Δ*R*_{p}. Secondly, the working mechanism, i.e., the change of internal network of the CNTs due to strain, might be weakened since the electric current can still pass through the matrix at the breakup points of the CNT network. However, when the electrical conductivity of the conductive polymer is much lower than that of the CNTs, this influence might be small since electric current always smartly chooses the components of much higher conductivity, i.e., CNTs, to pass through.

## Conclusions

In this work, we extend our previous 3D resistor network numerical model for predicting the piezoresistivity behaviors of strain sensors made from CNT/polymer nanocomposites. Besides the change of the internal conductive network formed by the conductive CNTs and the tunneling effect among neighboring CNTs, the contribution of the CNTs’ piezoresistivity was incorporated simultaneously. This modeling was realized by using a multi-scale technique in which we firstly compute the strain of a CNT in a RVE, then transform it into the strain of a randomly oriented CNT in the polymer matrix under a prescribed strain, and finally build up the relationship between the electrical conductivity change of the CNT using the above strain information and the CNTs’ piezoresistivity obtained from the first-principle computations [14]. An iterative model, which combines the 3D resistor network model and the fiber reorientation model, was constructed to predict the total piezoresistivity of the nanocomposite sensors from the updated information including CNT position, the internal conductive network, the tunneling effect among those neighboring CNTs, and the electrical conductivity of CNTs. The computational results qualitatively agree with our previous experimental results very well. Moreover, the contribution of the piezoresistivity of CNTs on the total piezoresistivity of the nanocomposite sensors is comparatively small, compared with those from the change of the internal conductive network and tunneling effects. The present numerical results can provide some valuable information for designing the highly sensitive resistance-type strain sensors made from the nanocomposites containing CNTs or other nanofillers.

## Declarations

### Acknowledgements

This work was partly supported by the Grand-in-Aid for Scientific Research (no. 22360044) from the Japanese Ministry of Education, Culture, Sports, Science and Technology to NH.

## Authors’ Affiliations

## References

- Ajayan PM, Zhou OZ: Carbon Nanotubes. In
*Carbon nanotubes: synthesis, structure, properties and applications*. Edited by: Dresselhaus MS, Dresselhausand G, Avouris PH. Springer, Berlin; 2001:391–425.View Article - Tombler TW, Zhou XC, Alexseyev L, Kong J, Dai H, Liu L, Jayanthi CS, Tang M, Wu SY: Reversible electromechanical characteristics of carbon nanotubes under local-probe manipulation.
*Nature*2000, 405: 769–772. 10.1038/35015519View Article - Frogley MD, Zhao Q, Wagner HD: Polarized resonance Raman spectroscopy of single-wall carbon nanotubes within a polymer under strain.
*Physical Review B*2002, 65: 113413.View Article - Dharap P, Li Z, Nagarajaiah S, Barrera EV: Nanotube film based on single-wall carbon nanotubes for strain sensing.
*Nanotechnology*2004, 15: 379–382. 10.1088/0957-4484/15/3/026View Article - Kang I, Schulz MJ, Kim JH, Shanov V, Shi D: A carbon nanotube strain sensor for structural health monitoring.
*Smart Materials & Structures*2006, 15: 737–748. 10.1088/0964-1726/15/3/009View Article - Hu N, Karube Y, Yan C, Masuda Z, Fukunaga H: Tunneling effect in a polymer/carbon nanotube nanocomposite strain sensor.
*Acta Mater*2008, 56: 2929–2936. 10.1016/j.actamat.2008.02.030View Article - Hu N, Karube Y, Arai M, Watanabe T, Yan C, Li Y, Liu YL, Fukunaga H: Investigation on sensitivity of a polymer/carbon nanotube composite strain sensor.
*Carbon*2010, 48: 680–687. 10.1016/j.carbon.2009.10.012View Article - Yin G, Hu N, Karube Y, Liu Y, Li Y, Fukunaga H: A carbon nanotube/polymer strain sensor with linear and anti-symmetric piezoresistivity.
*J Compos Mater*2011, 45: 1315–1323. 10.1177/0021998310393296View Article - Pham GT, Park YB, Liang Z, Zhang C, Wang B: Processing and modeling of conductive thermoplastic/carbon nanotube films for strain sensing.
*Composites Part B*2008, 39: 209–216. 10.1016/j.compositesb.2007.02.024View Article - Alamusi , Hu N, Fukunaga H, Atobe S, Liu Y, Li J: Piezoresistive strain sensors made from carbon nanotubes based polymer nanocomposites.
*Sensors*2011, 11: 10691–10723. 10.3390/s111110691View Article - Hu N, Masuda Z, Yan C, Yamamoto G, Fukunaga H, Hashida T: Electrical properties of polymer nanocomposites with carbon nanotube fillers.
*Nanotechnology*2008, 19: 215701. 10.1088/0957-4484/19/21/215701View Article - Liu Y, Hu N: Numerical simulations on piezoresistivity of CNT/polymer based nanocomposites.
*Computers Materials & Continua*2010, 20: 101–117. - Chen YR, Weng CI: Electronic properties of zigzag carbon nanotubes under uniaxial strain.
*Carbon*2007, 45: 1636–1644. 10.1016/j.carbon.2007.04.010View Article - Theodosiou TC, Saravanos DA: Numerical investigation of mechanisms affecting the piezoresistive properties of CNT-doped polymers using multi-scale models.
*Compos Sci Technol*2010, 70: 1312–1320. 10.1016/j.compscitech.2010.04.003View Article - Balberg I, Binenbaum N: Cluster structure and conductivity of three-dimensional continuum systems.
*Phys Rev A*1985, 31: 1222–1225. 10.1103/PhysRevA.31.1222View Article - Taya M, Kim WJ, Ono K: Piezoresistivity of a short fiber/elastomer matrix composite.
*Mech Mater*1998, 28: 53–59. 10.1016/S0167-6636(97)00064-1View Article - Simmons JG: Generalized formula for the electric tunnel effect between similar electrodes separated by a thin insulating film.
*J Appl Phys*1963, 34: 1793–1803. 10.1063/1.1702682View Article - Hu N, Fukunaga H, Lu C, Kameyama M, Yan B: Prediction of elastic properties of carbon nanotube-reinforced composites.
*Proceedings of the Royal Society A*2005, 461: 1685–1710. 10.1098/rspa.2004.1422View Article

## Copyright

This article is published under license to BioMed Central Ltd. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/2.0), which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.