You are on page 1of 16

Meccanica (2023) 58:1717–1732

https://doi.org/10.1007/s11012-023-01689-8

Design and analysis of a carbon nanotube‑based strain


gauge via multiscale modeling
G. Arana · A. Mora · I. Pérez · F. Avilés

Received: 30 September 2022 / Accepted: 30 June 2023 / Published online: 24 July 2023
© Springer Nature B.V. 2023

Abstract The gauge factor of strain gauges depends as electron tunneling. The gauge factors predicted by
on several factors such as the geometry of the gauge this model were moderately higher than the measured
and the piezoresistive properties of the materials used. values, given some idealizations of the model.
Herein, the piezoresistive response of a strain gauge
comprising a network of carbon nanotubes (CNTs) Keywords Carbon nanotube deposit · Strain
deposited over a polyimide (Kapton) substrate is the- gauge · Finite element · Multiscale simulation ·
oretically and experimentally investigated. Two com- Piezoresistivity · Carbon nanotubes
putational approaches are used, namely, a continuous
finite element model and a multiscale model com-
prising two dimensional scales. The finite element 1 Introduction
model is a simplified model which simulates the CNT
grid as a continuous solid. The multiscale approach Carbon nanotubes (CNTs) are nanomaterials with
utilizes a resistor network model to simulate the extraordinary mechanical and electrical properties.
change in electrical resistance of the CNT grid with To utilize these properties, CNTs have been used
strain and uses such effective electrical resistance as as fillers of polymeric nanocomposites as well as in
an input of the finite element model. It is found that CNT arrays or buckypapers, either free-standing or
the gauge factor increases with increased number of deposited over different substrates [1–3]. An impor-
end loops, decreasing grid width, and decreasing the tant characteristic of these materials is their piezore-
end loop height. Both modeling approaches agree sistivity, which is defined as the reversible change in
with the observed experimental tendencies. The mul- electrical resistance when the material is strained. In
tiscale approach is more realistic and accounts for CNT-based materials, piezoresistivity is mainly gov-
effects occurring at the micro and nanoscales, such erned by three parameters, viz. changes of the perco-
lated CNT network, changes in the number of contact
points between individual CNTs, and changes in the
Supplementary Information The online version distances between individual CNTs, which in turn
contains supplementary material available at https://​doi.​
org/​10.​1007/​s11012-​023-​01689-8.
affect the tunneling resistance [4, 5]. The piezore-
sistivity of CNT-based materials is exploited for the
G. Arana · A. Mora · I. Pérez · F. Avilés (*) development of strain sensors. Strain sensors are used
Centro de Investigación Científica de Yucatán, Unidad de in a wide variety of fields such as biomedicine, struc-
Materiales, Calle 43 No. 130 x 32 y 34, Col. Chuburná de
tural health monitoring, robotics, experimental stress
Hidalgo, 97205 Mérida, Yucatán, Mexico
e-mail: faviles@cicy.mx analysis, and transducer manufacturing, among many

Vol.: (0123456789)
13
1718 Meccanica (2023) 58:1717–1732

others [6–10]. The most commonly used strain sensor used finite element (FE) simulations to calculate the
are metallic strain gauges [6, 9–11]. The popularity of electrical resistance of a three-dimensionally printed
this type of sensor stems from its relatively low cost, graphenic sheets/polylactic acid strain sensor. They
ease of use, accuracy, and versatility [3, 6, 10, 11]. simplified their model by considering the grid made
However, a major downside of these sensors is that of graphenic sheets as a continuous solid (i.e., indi-
they present low gauge factors (GFs), between 2 and vidual graphenic sheets were not modeled), consid-
2.5, which limits their sensitivity and applicability [6, ering only geometrical effects. Their FE model was
12, 13]. This is because, in the case of metals, the pie- able to estimate the electrostatic properties of the
zoresistivity depends mostly on dimensional changes strain gauges, although the predicted electrical resist-
(geometrical effect), while changes in the electrical ance differed from the measured one. Han et al. [18]
resistivity of the metal may be negligible [4, 6–9]. fabricated a strain sensor with carbon black/polydi-
They also show low electrical resistances (from 30 methylsiloxane and used FE simulations to study the
Ω to 3 KΩ), temperature dependence of their electri- effects of grid geometry on the GF. Their approach
cal resistance [3], and are only suitable for measuring also considered the grid as a continuous material with
strains below 5% [14]. To overcome the downsides of constant resistivity, so only dimensional changes of
metallic gauges, CNT arrays (either free-standing or the sensor were accounted for. Chen et al. [28] devel-
deposited over a substrate) and CNT/polymer nano- oped a coupled electromechanical model of CNT/pol-
composites have been studied as strain sensors [2, ymer nanocomposites to simulate their piezoresistive
3, 14–19]. They can render higher sensitivity than response. To achieve this goal, the electrical resist-
metallic strain gauges [5, 18, 20]. Moreover, they ance of a CNT network was calculated using a Monte
often render good processability, flexibility, enhanced Carlo algorithm. Then, a FE model was implemented
thermal stability, and relative low cost [6]. to simulate the changes of the CNT network under
In spite of the existent research on CNT arrays and strain and the results were fed back to the Monte
CNT/polymer nanocomposites, their piezoresistive Carlo algorithm to obtain the electrical resistance of
behavior is still not fully understood, and there are the strained composite.
no accurate models which allow a full understanding Most computational studies on the piezoresistiv-
or prediction of their response [6]. Multiscale mod- ity of CNT-based materials focus only on the material
eling shows great potential addressing these issues, response rather than on the response of a geometri-
shedding light on the mechanisms, saving time and cally complex strain-sensing device, such as a strain
resources, and reducing the number of experiments gauge. Multiscale modeling simulations may greatly
required [21]. They also allow to investigate the com- assist in the design of such strain-sensing devices,
plicated multiscale and multiphysics phenomena that and there are only a few works exploiting a combined
affects piezoresistivity [22]. Many researchers have modeling-experimental approach [5, 23]. Further-
carried out computational simulations and experi- more, models with an analytical basis allow to read-
ments to investigate CNT-based materials [7, 10, 16, ily conduct parametric analysis and clearly investigate
17, 20, 23–27]. Among them, Lee at al [1]. proposed cause-effect relationships [29, 30].
a two-dimensional percolation model of a CNT film Therefore, in this study, multiscale computational
to obtain its piezoresistive properties, leading to a GF simulations of strain gauges based on CNTs deposited
of around 1.3. They validated their model experimen- on a thin polyimide substrate are developed to study
tally by characterizing a CNT-Pluronic buckypaper their piezoresistive response and the effect of varying
specimen. Huang et al. [2] fabricated a flexible strain the grid geometry on the GF. The initial FE model
sensor consisting on a CNT film between two layers considers the CNT grid on top of the polyimide sub-
of parylene-C, obtaining GFs two times higher than strate as a continuous solid with effective properties,
those of commercial gauges. Lee et al. [3] designed bypassing the detailed simulation of individual CNTs.
and fabricated gauges based on a CNT film spray- This allows to simulate the global piezoresistive
deposited on a polyimide substrate; they tested dif- response due solely to geometric effects, i.e., initially
ferent grid geometries and found that the sensitiv- ignoring the changes in the electrical resistivity of
ity increased with the number of end loops and film the CNT deposit. Then, a resistor network submodel
thickness, reaching GFs up to 16. Gooding et al. [7] is used at the microscale to update the electrical

Vol:. (1234567890)
13
Meccanica (2023) 58:1717–1732 1719

Fig. 1  Schematic of the


modeling process. On
the left, an unstrained
gauge with a certain CNT
arrangement. On the right,
a strained gauge with CNTs
rearranged. Dimensions
in mm

resistivity of the continuous grid in the FE model at account for the change in CNT-to-CNT electrical
each strain step, thus obtaining a multiscale model. resistance due to strain. This second model is hierar-
These simulations assist in designing the geometry chical and contains as a first step an “electrical resis-
and analyzing the response of a strain gauge at the tors network submodel”, which comprises a repre-
nano- (junction resistor), micro- (CNT resistor), and sentative volume element (RVE) of the CNT deposit
macroscales (strain gauge). Simulation results are that makes up the gauge subjected to strain, see insets
compared with selected experiments. of Fig. 1. In this second model, at each strain step the
resistivity of the CNT deposit is obtained from the
microscale submodel comprising individual CNTs, by
2 Multiscale modeling considering a deformed RVE and updating the CNT
volume fraction. Then the resistivity of the continu-
2.1 Modeling approach ous (CNT) grid in the FE model is updated using the
one obtained from the submodel, and gauge factors
A finite element model of the gauge was con- are computed from the FE model. A schematic of the
structed to simulate strain application and current process can be seen in Fig. 1, being ρ0 the unstrained
flow through the conductive path of the gauge. The resistivity and ρ1 the strained resistivity, where the
FE model considers the grid of CNTs as a continu- strained state implies a change in the CNT positions.
ous homogeneous solid, i.e., individual CNTs are not CNT rearrangement leads to a change in the effective
simulated in the FE approach. In a first (geometric, material’s resistivity, as depicted in Fig. 1. Finally,
macroscopic) FE model, the resistivity of the CNT the changes of electrical resistance as a function of
deposit was deemed constant and therefore only geo- strain are used to calculate the gauge factor (GF).
metrical changes of the solid conductive grid influ- Dimensions shown in Fig. 1 are nominal dimen-
ence the change in electrical resistance. Therefore, sions and were taken as the baseline ones for paramet-
in this initial model changes in the gauge factor are ric sensitivity analyses further conducted. The electri-
solely due to geometrical effects of the grid. In order cal resistance of the gauge was calculated at specific
to simulate the internal changes in resistivity of the axial strain (εx) values. For the FE simulations that
CNT deposit comprising the grid of the strain gauge, only considered geometric changes (FE model, first
the resistivity of this first model was updated at each model), the effective electrical resistance of the gauge
strain level using a multiscale approach. For this was calculated at εx = 0.075%, 0.15%, 0.3% and 2%,
purpose, a second (multiscale) model of the electri- given the linear behavior of the change in electrical
cal behavior of the CNT deposit was developed to resistance with strain. The strains under study for the
Vol.: (0123456789)
13
1720 Meccanica (2023) 58:1717–1732

a CNT grid width (GW) of 0.3 mm, end loop height


(LH) of 1.5 mm and 3 end loops (see Fig. 1). Sev-
eral designs (FE models, Fig. 2) of strain gauges were
investigated based on the nominal design and vary-
ing GW, LH and EL. These geometric variables were
studied by varying only one of the nominal dimen-
sions at the time and keeping the others constant. GW
varied from 0.1 to 0.5 mm in steps of 0.1 mm; LH
was considered as 0.3 mm, 0.6 mm, 0.9 and 1.5 mm;
the investigated EL were 1, 3, 5, and 7. Thus, the par-
ametric analysis of GW, LH and EL comprised eleven
strain gauges, namely, the nominal design and ten
variations.
A plane stress two-dimensional electro-mechan-
ical FE model was constructed using the software
ANSYS APDL 17.0 (Ansys Inc., Canonsburg, USA).
Eight-node quadrilateral elements “PLANE223”
(linear interpolation) with an average of 0.05 mm in
Fig. 2  Finite element model of a strain gauge (nominal dimen- size were used. The element provides structural (lin-
sions shown in Fig. 1). An equivalent electric circuit where the
CNT deposit is represented by the electrical resistance (R) is ear elastic) and electrostatic capabilities. An image
shown to the right of the meshed gauge is shown in Fig. 2. For illustra-
tive purposes, the elements shown in Fig. 2 are three
times larger than those actually used in the final
simulations with updated electrical resistivity ranged (converged) model. A convergence analysis (see sec-
from εx = 0–2%, in 0.5% intervals. In this second tion S1 of supplementary material) was performed
model (multiscale model), for each strained state a to ensure that the results are independent of the ele-
resistivity was calculated from the resistor network ment size. In order to simulate the deformation of the
submodel. The resistivity of the elements comprising gauges, a vertical displacement was applied to the top
the electrically insulating substrate was constant, and and bottom edges of the model in Fig. 2, according
they do not contribute to the GF results. to the desired axial strain (εx). In order to avoid free
body motion, the vertical center line was constrained
2.2 Finite element model of the strain gauge in the y-direction. A voltage (V) of 10 V was applied
to one end of the grid and 0 V to the other end, to
The geometry of the conductive grid (i.e., the CNT generate a current flow (I) through the conductive
deposit in Fig. 2) used to actively measure strain in grid (CNT deposit). The gauge consists of two mate-
a strain gauge, is commonly of a serpentine geome- rials, the Kapton substrate (backing material) and the
try, but its exact geometrical details and dimensions CNT deposit (blue serpentine grid in Fig. 2, consid-
may vary depending on the application. Commercial ered as a continuous solid) bonded to the electrically
metallic strain gauges have long, straight, and narrow insulating substrate (see Fig. 2.). The material proper-
segments to reduce the undesired influence of lateral ties used in the simulations were directly measured,
strains and typically have multiple turns to increase calibrated from dedicated measurements, or obtained
the effective length of the grid. These turns are the from the literature, as will be discussed in Sect. 2.4
sections that connect the straight segments and are (see Table 1). It was assumed that the thin (micromet-
called end loops (EL), see Fig. 2. Herein, the geom- ric) layer of CNT deposit was perfectly bonded to the
etry of the simulated gauges is inspired in that of a thicker Kapton substrate and that it does not modify
standard unidirectional metallic gauge. The nominal the stiffness of the substrate. Therefore, in the FE sim-
design shown in Figs. 1 and 2 has a commercial poly- ulations, the section of the model corresponding to
imide (Kapton) as backing material with an overall the grid was given the Kapton’s elastic modulus and
height and width of 8 mm and 4 mm, respectively, in-plane Poisson’s ratio (𝜈xy), but its own electrical

Vol:. (1234567890)
13
Meccanica (2023) 58:1717–1732 1721

properties (see Sect. 2.4). Two models were con- (4)


( )( )
A = t0 − 𝜖x 𝜈xz t0 w0 − 𝜖x 𝜈xy w0
ducted, one with constant resistivity (FE model) for
the continuous CNT grid and the second one (multi- where t0 is the initial thickness of the CNT deposit
scale model) with an electrical resistivity that changes (t0 = 1.36 μm, according to measurements of gauges
with strain (obtained from the submodel). For the FE to be discussed), w0 is the initial width of the straight
model, all elements had a constant resistivity, i.e., section (which varies for each gauge design, see
only geometric changes due to strain affect the GF Fig. 2), νxz is the transverse Poisson’s ratio of the
of the solid and continuous grid. In other words, the CNT deposit, and νxy is the in-plane Poisson’s ratio
change in resistivity of the bundles of CNTs making of the Kapton substrate. It should be noted that the
the deposit due to strain was not calculated directly in-plane Poisson’s ratio of the CNT deposit is con-
from the FE model. To simulate the changes in the sidered the same as that of its Kapton substrate
resistivity of the array of CNTs due to strain, a second (νxy = 0.34) because of the perfect bonding assump-
model was performed in which the electrical resistiv- tion. On the other hand, at the transverse (out-of-
ity of the conductive grid was updated at each strain plane, z) direction the CNT deposit has a free bound-
step with values obtained from the electrical resistors ary condition and such an effective Poisson’s ratio
network submodel (see Sect. 2.3). Thus, in this multi- is deemed to be that of the CNT deposit, νxz = −0.2
scale model changes in both, geometry and grid resis- [32]. The fractional change of electrical resistance
tivity, affect the computation of the GF. (ΔR/R0 = (R−R0)/R0, where R0 is the electrical resist-
For the simulations with variable resistivity, the ance at zero strain) was calculated for every strain
grid’s electrical resistance (R) was calculated at step and used to define a gauge factor such as [9],
strains ranging from 0 to 2% in 0.5% intervals. For
𝛥R∕R0
the simulations that only considered geometric GF = (5)
changes, the resistances were calculated at 0.075%, 𝜖x
0.15%, 0.3% and 2% strain. The electrical resistance
GFs for the investigated gauges were obtained
(R) was calculated via Ohm’s law as [31],
from the slope of the ΔR/R0 vs. εx plots.
V
R= (1)
I 2.3 Electrical resistors network submodel
where V is the applied voltage (10 V) and I is the
resulting effective electrical current flowing through A submodel based on a network of electrical resistors
the grid. The effective electrical current was obtained was used to calculate the effective electrical resistance
from the definition of current density in a conductor of an RVE of the CNT deposit at each level of applied
of uniform cross-sectional area, i.e. [31], strain. For this, three-dimensional cylindrical repre-
sentations of CNTs were generated for the submodel.
I The centerline of the cylindrical CNTs is discretized
J= (2)
A into n + 1 points defining n segments of length lS each
where A is the cross-sectional area of one straight [33]. These segments may have different orientations
segment in the grid and J is the effective current den- to obtain CNTs with different degrees of curliness. An
sity of the same straight segment. The effective (vol- example of a discretized CNT as seen over the xy plane
ume average) current density was calculated as, is shown in Fig. 3a. CNTs are generated inside an RVE
with the shape of a cuboid of size lRVEx y z
∑n × lRVE × lRVE ,
ψi Ji where lRVE is the length of the cuboid along direction
i
J = ∑i=1 (3)
n
ψ i, i ∈ {x, y, z}. The face lying on the xy plane is con-
i=1 i
sidered as the bottom face of the RVE and represents
where n is the number of elements that make up one the polymer substrate where CNTs are deposited. To
straight segment of the gauge, 𝜓 i and Ji are the vol- generate a CNT, the first point of the discretization,
ume and current density of element i, respectively. P0, is randomly generated on the xy plane within the
The cross-sectional area was calculated considering RVE. Then, the z-coordinate of P0 is calculated so
Poisson’s contraction as [9], that the corresponding CNT is either at the bottom of

Vol.: (0123456789)
13
1722 Meccanica (2023) 58:1717–1732

Fig. 3  Generation of CNT


network for the submodel.
a Representation of a CNT
on the xy plane, b first point
of a CNT placed on the xy
plane, c first point of a CNT
placed above an existing
CNT (point with label S),
d point Pi of a CNT placed
over point S that belongs
to another CNT, e segment
Pi−1 Pi of a CNT needs
to be rotated smoothly as
there are no other points
below point Pi. Points S and
T belong to two different
CNTs

the RVE (lying on the xy plane, Fig. 3b) or lying over CNTs). Then, segment Pi−1 Pi is rotated an angle θ/2
an existing CNT (Fig. 3c, where point S belongs to – Δ. Here, the term Δ is used to ensure segment Pi−1 Pi
another CNT). Thus, points are placed as if they were rotates even in the case θ is zero or close to zero. Oth-
laying on the xy-plane or on another CNT, to simulate erwise, if θ is zero (or close to zero), segment Pi−1 Pi
CNTs being deposited on a solid substrate. To gener- would remain in (or close to) its initial position parallel
ate any subsequent point Pi, an angle 𝜔i is randomly to the xy plane. Rotation of segment Pi−1 Pi is carried
generated. When i = 1, 𝜔1 follows a uniform distribu- out using a small angle, i.e., θ/2 – Δ, to obtain a smooth
tion in [0, 2π]. When i ≥ 2, 𝜔i follows a normal distri- bending of CNTs. A value of Δ = π/10 is used here.
bution in the interval [-α, α], where α ≤ π/2. The value Finally, points are generated until the required volume
of α determines the extent that each CNT segment may (or weight) fraction of CNTs is reached.
bend. Then, each new Pi is generated such that segment The CNT volume fraction in the RVE used in the
Pi−1 Pi is initially parallel to the xy plane and makes an submodel was 0.34 (34% v/v), which is the maximum
angle 𝜔i with the extension of the previous segment in volume fraction allowed by this approach. This volume
their projection on the xy plane, as shown in Fig. 3a. fraction targets the experiments discussed herein. In
When i = 1 there is no previous segment, so in this case order to show the similarity between the CNT network
the segment P0 P1 makes an angle 𝜔1 with the x-axis, as generated by the electrical resistors network submodel
in Fig. 3a. Angle 𝜔1 has a different distribution than 𝜔i and the experiments, Fig. 4 shows an example of a net-
for i ≥ 2 because there is no previous segment to limit work generated with this submodel (left), compared to
its value and, thus, may have any orientation. Next, an actual scanning electron microscopy image of the
segment Pi−1 Pi is rotated perpendicular to the plane CNT grid (right). The left image in Fig. 4 was gener-
xy such that the following conditions are met: (a) Pi ated using the data from the submodel and the open
lies over an existing CNT or the xy plane (see Fig. 3d, source visualization software ParaView (Kitware Inc.,
where point S belongs to another CNT); (b) segment Sandia National Laboratories, New Mexico). The net-
Pi−1 Pi and the extension of the previous segment make work generated by the submodel in Fig. 4 consists of
an angle smaller than π/2. This is because, following a CNTs of 3 μm in length and radius of 15 nm, accord-
previous work [33], π/2 is the largest angle that CNTs ing to the actual multiwall CNTs used in the experi-
are allowed to bend. In case neither of these two condi- ments. The RVE has a side length of 4.5 μm and height
tions is met, the procedure is as follows. Let θ be the of 1.36 μm (same height as the actual CNT deposit).
angle that segment Pi−2 Pi−1 forms with the xy-plane, as Similarities in the morphology of both networks are
shown in Fig. 3e (points S and T belong to two different

Vol:. (1234567890)
13
Meccanica (2023) 58:1717–1732 1723

Fig. 4  Comparison of
a CNT deposit obtained
from the submodel (left)
and a scanning electron
microscopy image of the
experimental CNT deposit
(right)

observed, showing the strong capabilities of the sub- junction resistance is typically much higher than the
model introduced herein to simulate curved CNTs and CNT electrical resistance, which further diminishes
capture the densely packed architecture of the CNT the influence of the assumption of solid or hollow
deposit. In general, the optimum CNT concentration for cross-sectional area. Other authors have compared
piezoresistive sensing applications is expected to be just the assumption of solid and hollow cylinders for the
above electrical percolation [4, 30]. The investigation of cross section of CNTs, finding negligible differences
such an optimal concentration for the densely packed in electrical conductivity [34]. Thus, we have consist-
CNT arrays discussed herein falls beyond the scope of ently assumed a solid cylinder for the cross section of
this work. the CNTs for both, the calibrated resistivity and the
The electrical resistivity of the RVE is assumed calculated electrical resistance.
to be the same as that of the CNT network, and is The boundary conditions used for the submodel
subsequently used as input effective property of the were the application of a voltage difference of 10 V to
continuous grid in the FE model. Thus, the electri- each pair of opposite faces of the submodel along the x-
cal resistance of the CNT network is calculated in the and y-directions. That is, 10 V were applied to all CNTs
submodel to obtain its electrical resistivity. For this connected to one (x or y) edge face of the RVE and all
purpose, the network of CNTs is turned into a net- CNTs connected to the opposite edge face were con-
work of electrical resistors. Two types of electrical nected to a ground node, i.e., 0 V.
resistors are considered, viz. CNT resistors and junc- A junction resistor comes from the transfer of elec-
tion resistors. trons between two contiguous CNTs, either by direct
A CNT resistor, RCNT, comes from the intrinsic contact or by tunneling of electrons. A junction resistor
electrical resistance of a CNT and its electrical resist- is added when two contiguous CNTs are separated less
ance is calculated as [9]. than a cutoff distance (dt), which is taken as dt = 1.8 nm
[35]. Figure 5 depicts a CNT (labeled C ­ NT2) with junc-
lCNT 𝜌CNT
RCNT = (6) tions with two contiguous CNTs (labeled ­CNT1 and
2
𝜋rCNT ­CNT3) and the equivalent resistor network. The value
of the junction resistance is calculated by the exponen-
where 𝜌CNT is the CNT resistivity, lCNT is the CNT tial equation presented by Hu et al. [20] as,
length and rCNT its radius. Notice that the choice of
a solid or hollow cross section for the CNT has no h2 d

4𝜋d √

RJ = exp 2m𝜆 (7)
significant influence in the numerical calculation of √
AT e2 2m𝜆 h
the GF. This is partly because the inner radius of the
CNT is just a few nanometers, which yields only a where e and m are the charge (1.602176565 × ­10−19 C)
small difference in the cross-sectional areas assum- and mass (9.10938291 × ­10−31 kg) of the electron, h is
ing solid or hollow cylinders. Furthermore, the the Planck’s constant (6.62606957 × ­10−34 ­kgm2/s), d

Vol.: (0123456789)
13
1724 Meccanica (2023) 58:1717–1732

Table 1  Input values for Parameter Value Reference


the simulations
Initial resistivity of CNT deposit (𝜌M,Ωm) 3.341 × 10−4 Measured
Kapton elastic modulus (GPa) 2.71 Measured
Kapton resistivity (Ωm) 1.0 × 10 17 [37]
Kapton Poisson’s ratio (νxy) 0.34 [37]
Transverse Poisson’s ratio of the CNT deposit (νxz) − 0.20 [32]
CNT resistivity (𝜌CNT , Ωm) 4.10 × 10−5 Calibrated
Height barrier (λ, eV) 0.1 Calibrated
CNT length (µm) 3 [38]
CNT radius (nm) 15 [38]

is the distance separating the CNTs, AT is the junc- Thus, 𝜌z is not considered here and i in Eq. (8) refers
tion’s cross-sectional area (approximated as the area to either x or y.
of two CNT crossed at a 90° angle [35]), and λ is the Simulations were performed in RVEs that were
height of the potential barrier. For cases were the found to be large enough to ensure convergence
CNTs are in contact, the van der Waals equilibrium and isotropy in the calculated resistivities (see Sec-
distance is used in Eq. (7), i.e. d = 0.34 nm. tion S2 for the convergence study of the submodel).
After the CNT network is turned into a network The RVEs chosen for the simulations have the same
of electrical resistors, the equivalent resistance height as the actual CNT deposit of the experiments
of the resistor network, Req, is obtained using the (i.e., 1.36 μm) and a square base with side-length
direct electrifying algorithm [36]. Then, the electri- lRVE = 2.5lCNT , i.e., lRVE = lRVE = lRVE . Note that
xy x y xy

cal resistivity of the RVE is calculated as [9]. these RVE dimensions are used for zero strain. In
addition, CNTs are generated until the RVE is fully
AReq
packed, which in our simulations corresponded to a
𝜌i = (8)
volume fraction of 0.34.
i
lRVE
For each value of strain applied, the RVE dimen-
where 𝜌i is the electrical resistivity of the RVE, lRVE
i
sions are updated considering linear elastic defor-
is the RVE’s side-length along the direction in which mation and Poisson’s effect (see Table 1 for elastic
𝜌i is measured, and A is the RVE’s cross sectional constants). The amount of CNTs is kept constant
area. It is noted that the equivalent electrical resist- as the RVE is deformed. Thus, the CNT volume
ance may be anisotropic along the x- and y-directions. fraction is updated at each value of applied strain
Thus, equivalent electrical resistances and resistivi- and the effective resistivity of the RVE at each
ties are calculated for the x- and y-directions. Flow stain level is computed. Ten simulations were per-
of electricity along the z-direction is not considered formed for each strain step (from 0 to 2% in 0.5%
as the RVE length along that direction is too small. intervals). Resistivities along x- and y-directions
of the ten simulations were averaged to obtain the
x- and y-resistivities of the RVE, 𝜌x and 𝜌y , respec-
tively. Given that both were numerically )similar, the
average RVE resistivity, 𝜌avg = 𝜌x + 𝜌y ∕2 , is used
(

as the updated resistivity in the FE model for each


level of strain.

2.4 Materials data input

Experiments were conducted to define key input


Fig. 5  Transformation of a CNT network into a resistor net- parameters of the model and to calibrate input val-
work ues such as the barrier height, which are difficult to

Vol:. (1234567890)
13
Meccanica (2023) 58:1717–1732 1725

measure directly. Input values used in the models


are summarized in Table 1. The initial resistance
of the CNT deposit was measured from a rectan-
gular deposit which is 0.3 mm wide, 3.2 mm long
and with a mean height of 1.36 μm (measured with
atomic force microscopy) over a Kapton substrate.
With these parameters, the mean initial resistivity
(𝜌M ) of the CNT deposit was measured. The elastic
modulus of the Kapton film was measured from uni-
axial tensile tests. The Poisson’s ratio and resistivity
of the Kapton film was obtained from the supplier’s
datasheet. The CNT resistivity (𝜌CNT ) and the height
barrier (λ) were calibrated as described in Section S3.

3 Experimental procedures

The fabrication of the strain gauges started by dis- Fig. 6  FE-predicted current density for constant grid resistiv-
persing acid-oxidized CNTs in chloroform via 15 min ity corresponding to the nominal design of the strain gauge
mechanical stirring and 30 min ultrasonic dispersion
in a conventional bath (70 W, 42 kHz). The dispersed
CNTs were then spray-deposited over 0.127 mm thick 4 Results and discussion
Kapton polyimide substrates and dried in an oven at
120 °C. An automated engraving machine was used 4.1 Finite element model using constant grid
to remove unwanted areas of the deposited CNTs, resistivity
obtaining the desired serpentine-shape of the strain
gauges. A parametric study of the grid geometry was under-
The nominal design (used as reference) comprises taken simulating the eleven gauge designs described in
3 loops, 1.5 mm end loop height and 0.3 mm width Sect. 2.2. The resistivities of the Kapton substrate and
of the straight sections, as shown in Fig. 1. For the CNT deposit is included in Table 1. Figure 6 shows
experimental investigation of the variation of geo- the resulting current density map predicted by the FE
metrical parameters, one parameter was varied at simulations corresponding to the nominal design of
the time. The number of loops tested were 3 and 5 the gauge (Fig. 1), where the CNT grid is deemed a
and the width of the straight section of the grid con- solid continuous material with a constant electrical
sidered were 0.3 and 0.5 mm. The end loop widths resistance. It is observed that the current density (J )
tested were 0.3 and 1.5 mm. is uniform at the straight segments of the gauge, and
The gauge factor was obtained from tensile testing gradients are present at the corners of the end loops.
of the gauges at εx ≤ 1% with simultaneous strain and The uniformity of J at the straight sections allows to
electrical resistance measurements. The tests were reliably obtain the electric current (I) flowing through
conducted in a Shimadzu® AGS-X universal testing the gauge using Eq. (2). Notice that the current density
machine (Kyoto, Japan) with a crosshead machine flowing through the farthest section of the end loops is
displacement of 2 mm/min (8.3%/min). The electri- two orders of magnitude smaller than the one circulat-
cal resistance was measured using a multifunction ing through the straight sections of the grid.
switch/measure unit model 34980A from Keysight Considering constant grid electrical resistivity, the
(Santa Rosa CA, USA). Five replicates were tested fractional change of electrical resistance (ΔR/R0) as a
for each configuration. The GF was obtained from function of strain (εx) showed a linear response for all
the linear slope of the piezoresistive response using eleven gauge designs, and the GF was computed from
strains between 0.1 and 0.9%. the slope of (ΔR/R0) vs. εx. Depending on the geome-
try of the gauge, the GFs ranged from 0.817 to 1.088,

Vol.: (0123456789)
13
1726 Meccanica (2023) 58:1717–1732

Fig. 7  Parametric analysis of gauge factor (GF) and initial electrical resistance (R0) obtained from the FE model using constant grid
electrical resistance. a Number of end loops, b grid width, c end loop height

with initial electrical resistances ranging from 7.17 kΩ gauges, although constant grid resistivity was assumed
to 37.1 kΩ, see Fig. 7. The changes in GF caused by in both works. In contrast, increasing LH significantly
the parameters GW and EL were small. Increasing the increased the GF, considerably more than when varying
number of end loops slightly increased R0 and the GF either GW or EL. Increasing LH form 0.3 to 1.5 mm
(Fig. 7a), the former being due to a longer path for elec- increases the GF in 23.8%. Increasing LH also slightly
trons. The increase in GF is likely due to the increased decreased R0, see Fig. 7c. The reduction in R0 comes
number of straight sections, where current flows par- from an increase in the cross-sectional area of the end
allel to strain. Decreasing the grid width caused an loops. However, since most of the electric current con-
increase in GF as well as in R0, Fig. 7b. This is due to centrates near the inner part of the end loop (see Fig. 6),
the narrowing of the cross-sectional area. Similarly, the increase in cross-sectional area has a small impact
using continuous FE simulations, Gooding et al. [7] on R0. Therefore, increasing the end loop height up to
predicted a small increase in GF with a decrease in 1.5 mm is a sound strategy to increase the GF.
grid width for three-dimensionally printed graphene/ When the dimension of GW is similar to that of
polylactic acid strain gauges. Han et al. [18] predicted LH, the sensitivity to transversal strains increases
an increase in GF with an increase of the number of [39], which explains the decrease in GF with increas-
end loops for carbon black/polydimethylsiloxane strain ing GW and decreasing LH. Under tensile axial strain,

Vol:. (1234567890)
13
Meccanica (2023) 58:1717–1732 1727

Fig. 8  Results of the RVE 3.40 8.54


resistor network submodel

Number of junction points (x10 5)


3.39
as a function of applied 8.52
strain. a Average resistivity 3.38

ρavg (10- 4 Ωm)


(𝜌avg), b number of junction 8.50
points 3.37

3.36 8.48

3.35
8.46
3.34

0.0 0.5 1.0 1.5 2.0 8.44


0.0 0.5 1.0 1.5 2.0
εx (%) εx (%)
(a) (b)

there is a positive change of electrical resistance in the axial strain applied. The RVE resistivities along
sections of the gauge where the current flows paral- the x- and y-directions (𝜌x and 𝜌y, respectively) were
lel to the axial (x) strain. Concomitantly, due to the numerically similar (see Figure S2.1), and thus the
Poisson effect of the substrate, a small compressive average is shown. The resistivity shown at zero strain
strain in the transverse direction (y) should develop. is the measured resistivity of the CNT deposit, 𝜌M
In sections of the gauge where the current flows in the (see Table 1). It is observed that the average RVE
transverse direction, the changes of electrical resist- resistivity increases nearly linearly as strain increases.
ance are expected to be negative. Because of these However, this increase is small (~ 0.4% for every
opposite effects, when measuring the resistance of the 0.5% increase in εx), as changes in resistivity are
gauge the change of electrical resistance is smaller observed in the third significant digit. This increase
(i.e., GF is lower) than if the end loops were wider in resistivity is attributed to a reduction in the num-
than the grid width. This matches findings by Fu et al. ber of conductive paths formed inside the RVE as
[39] and Chung et al. [40], who observed that having its volume increases with strain. That is, when the
similar LH and GW hinders the important discrimina- RVE is strained, its volume increases but the amount
tion between axial and transversal strains. Therefore, of CNTs remains the same, leading to an effective
it is recommended to use a narrow grid width in com- reduction of the CNT volume fraction. This reduction
bination with significantly longer end loops, to obtain in the volume fraction results in an effective reduced
higher GFs. number of conductive paths, which in turn results in
The changes in grid geometry also influence R0, higher resistivity of the RVE.
which in turn may also influence the GF. In particu- Figure 8b plots the number of CNT junction
lar, for carbon nanostructured materials, it is known points (Fig. 5) as a function of the applied strain, as
that increasing R0 tends to increase the GF [2]. For predicted by the submodel. Each CNT-to-CNT junc-
EL and GW, there is a tendency of increasing the GF tion is deemed to contribute with two junction points
with increasing R0, see Fig. 7a and b. However, for (see Fig. 5). The results plotted are the average and
LH the GF decreases as R0 (slightly) increases, but standard deviation of 10 simulations. The scattering
the change in R0 for LH = 0.3 mm and LH = 1.5 mm bars are somewhat large, because for each of those
is small (1 kΩ, i.e. 6.8% decrease). This suggests that 10 simulations the CNT network is randomly gener-
for some parameters (such as for LH) the geometry of ated. In spite of the dispersion, it is seen that the ini-
the gauge can have an impact on the GF that is domi- tial average number of junction points (~ 8.492 × ­105)
nant over the influence of R0. tends to lower values as the strain is applied, i.e. as
the RVE is enlarged in the x-direction. A reduction
4.2 Multiscale simulations using grid resistivity in the number of contact points means less number
calculated by the submodel of conductive paths, which influences the increase
in resistivity with strain observed in Fig. 8a. In addi-
Figure 8a shows the average RVE resistivity (𝜌avg, tion, upon strain application, the CNT-to-CNT dis-
obtained as described in Sect. 2.3) as a function of tances change within the grid where CNTs tend to

Vol.: (0123456789)
13
1728 Meccanica (2023) 58:1717–1732

4.3 Comparison to experimental results

The GFs obtained from the FE (constant grid resis-


tivity) and multiscale models are compared to experi-
mental results in Fig. 10. Given the versatility of the
modeling approaches, the modeling predictions cover
a wider range of results than the selected experiments
conducted. The experiments confirm the modeling
predictions regarding the trends of the three geo-
metrical parameters investigated. The GF increases
as the number of end loops increases, the loop height
increases, and the grid width decreases.
The experimental results show higher GFs than
Fig. 9  Fractional change of electrical resistance as a function the FE simulations assuming constant grid resistiv-
of strain for the nominal gauge design (shown in Fig. 1)
ity. This is because the simpler FE model does not
consider any contribution from the changes of elec-
be separated further apart. This is reflected in Fig. 8a trical resistivity of the CNT grid with strain. That is,
in the increase of resistivity of the RVE and also in the results of the FE model in Fig. 10 consider the
Fig. 8b as some CNTs become separated beyond the grid as a continuous solid of constant resistivity and
tunneling cutoff distance and thus the number of con- the piezoresistivity predicted is only due to geometri-
tact points is reduced. cal changes of such a continuous grid upon straining.
In the literature, it has been argued that the sensi- Therefore, the difference between the FE predictions
tivity of a CNT-based material cannot be explained and measurements of GF can be attributed to micro
from geometrical changes only [5, 41]. To gain more and nanoscale phenomena not accounted for in the FE
insight on this issue, results obtained from the model model, such as CNT-to-CNT contact resistance and
that only considers geometrical changes (FE model) tunneling [17, 20]. For example, for the nominal con-
are compared with the results of the model that con- figuration depicted in Fig. 1 (EL = 3, GW = 0.3 mm,
siders both geometrical and resistivity changes (mul- LH = 1.5 mm) the FE predicted GF is 1.01 and the
tiscale model). Figure 9 shows the fractional change measured one is 1.26. Assuming that the FE model
of electrical resistance (ΔR/R0) as a function of strain properly captures the geometric effect of the piezore-
(𝜖x ) predicted by the FE model with constant grid sistive tests, then the geometric contribution of the
resistivity (shown as “FE”) and by the multiscale GF is 80.1%, and the intrinsic material properties var-
model (i.e., using the submodel to account for varia- iation accounts for the remaining 19.9%.
ble resistivity and FE), for the nominal gauge design. As GW becomes smaller (0.2 mm), the geomet-
The GF in the multiscale model depends not only ric contribution to the GF decreases (76.9%). This
on the geometric changes of the grid, but also on the behavior is attributed to the increased contribution of
change in its electrical resistivity. It is observed that the straight segments to the total electrical resistance.
the increase in resistivity with strain predicted by the A similar trend is seen when adding more end loops;
multiscale model exceeds the increase in electrical for EL = 5 the geometric contribution predicted by
resistance caused by geometrical changes. Because of FE is 74.7% of the measured GF. Similarly, reducing
this, the sensitivity of the gauge (GF) increases while LH to 0.3 mm yields a geometric contribution of only
maintaining a nearly linear response with strain. A 67.5%. Furthermore, such a small LH is not recom-
similar piezoresistive response was observed for all mended because it could increase the undesired trans-
gauge designs. The GFs of all gauge designs pre- verse sensitivity [39, 42]. Notice that the configura-
dicted by the FE model were between 0.81 and 1.09 tions that lower the geometric contribution implicitly
depending on the design. For the multiscale model, increase the inherent material contribution (change
the GFs ranged from 1.60 to 1.87, which corresponds in resistivity at the nano- and microscales), and thus
to an average increase of around 79%. tend to increase the GF. This means that, at a given
strain, the resistance change due to changes in the

Vol:. (1234567890)
13
Meccanica (2023) 58:1717–1732 1729

Fig. 10  GFs predicted by


the FE model, multiscale
model, and measured, as a
function of the geometric
parameters shown in Fig. 2.
a Number of end loops,
b grid width, c end loop
height

CNT-to-CNT resistivity could be as important as the gauges. For example, the definition of thickness and
one caused by geometrical changes of the grid. edge of the CNT deposit could have slight variations
The multiscale simulations predict the high- throughout the gauge at the microscale, which could
est GFs, meaning that the CNT-to-CNT resistivity hinder the sensitivity to some extent [46]. In contrast,
changes add to the change in electrical resistance the FE model assumes a perfectly homogeneous grid
caused by the geometrical changes. The overpredic- thickness and perfectly straight edges of the CNT
tions of the multiscale modeling may be due to sev- grid. It also experimentally challenging to manufac-
eral factors connected with the model idealization. ture gauges with very narrow GW, although the mod-
The model assumes perfect stress/strain transfer, but eling predictions suggest that this practice increases
in practice it is known that the strain transfer from the the GF.
Kapton substrate to the CNT deposit is not perfect,
see e.g. [43, 44]. Strain transfer from CNT to CNT
within the grid may also present some shear lag, see 5 Conclusions
e.g. [45], and that is also not accounted for in the
multiscale model. The actual variability in the length A multiscale modeling strategy was implemented to
of the CNTs may also be a contributing factor [30], simulate the piezoresistive response of a strain gauge
as well as differences with the exact volume fraction made of a CNT deposit, investigating the main geo-
of the deposit. The length of the CNTs employed in metrical parameters that affect its response. Serpen-
the experiments comply with a lognormal distribution tine-shaped gauge designs with different number of
between 0.88 and 4.3 μm [38]. However, in the mul- end loops, grid width, and end loop height were simu-
tiscale simulations CNTs of 3 μm were used. There lated and tested. The combined modeling and testing
are also some limitations and idealizations from the strategy allowed to study the influence of such geo-
viewpoint of the fabrication of the CNT-based strain metrical parameters on the piezoresistive sensitivity

Vol.: (0123456789)
13
1730 Meccanica (2023) 58:1717–1732

(gauge factor, GF) of the gauges. The most influential Declarations


variable is the end loop height; changing it from 0.3
Conflict of interest Authors declare no known conflict of in-
to 1.5 mm increased the GF by 12.3% (as obtained terest.
by the multiscale model). The GF increases as the
number of end loops increases, the end loop height
increases, and the grid width decreases. The gauge
dimensions that yielded the highest GFs correspond References
to a grid width of 0.1 mm, a loop height of 1.5 mm
and 7 end loops. Thus, when designing a strain sens- 1. Lee BM, Loh KJ (2017) Carbon nanotube thin film
ing device with serpentine geometry, to obtain higher strain sensors: comparison between experimental tests
GFs it is recommended to have a small grid width and and numerical simulations. Nanotechnology 28:155502.
https://​doi.​org/​10.​1088/​1361-​6528/​aa6382
sufficiently long loop height, which also diminishes 2. Huang Y-T, Huang S-C, Hsu C-C et al (2012) Design
the transverse sensitivity of the gauge. and fabrication of single-walled Carbon Nanonet flexible
Comparison between experiments and two mod- strain sensors. Sensors 12:3269–3280. https://​doi.​org/​10.​
eling strategies showed that the piezoresistive 3390/​s1203​03269
3. Lee D, Hong HP, Lee CJ et al (2011) Microfabrication
response of the strain gauge is dominated by changes and characterization of spray-coated single-wall carbon
in geometric parameters of the gauge’s grid upon nanotube film strain gauges. Nanotechnology 22:455301.
strain application (around 65–80%), but the inherent https://​doi.​org/​10.​1088/​0957-​4484/​22/​45/​455301
changes of resistivity of the CNT deposit also play 4. Avilés F, Oliva-Avilés AI, Cen-Puc M (2018) Piezoresist-
ivity, strain, and damage self-sensing of polymer compos-
an important role. As the grid width reduces and the ites filled with Carbon Nanostructures. Adv Eng Mater
number of loops increases, the geometric contribution 20:1701159. https://​doi.​org/​10.​1002/​adem.​20170​1159
to the GF decreases, which increases the piezoresist- 5. Hu B, Hu N, Li Y et al (2012) Multi-scale numerical sim-
ive sensitivity. The multiscale model predicted higher ulations on piezoresistivity of CNT/polymer nanocompos-
ites. Nanoscale Res Lett 7:402. https://​doi.​org/​10.​1186/​
GFs than the constant resistivity FE model, and the 1556-​276X-7-​402
experimental results fell in between. This is attrib- 6. Garcia JR, O’Suilleabhain D, Kaur H, Coleman JN
uted mainly to the fact that the simplified FE model (2021) A simple model relating gauge factor to Filler
does not account for changes in the CNT grid resis- Loading in Nanocomposite strain sensors. ACS Appl
Nano Mater 4:2876–2886. https://​doi.​org/​10.​1021/​
tivity, and the more detailed multiscale model does acsanm.​1c000​40
not account for imperfect load transfer (CNT-to-CNT 7. Gooding J, Mahoney J, Fields T (2017) 3D printed strain
and substrate-CNT). In spite of those limitations, the gauge geometry and orientation for embedded sensing.
experiments confirmed the trends of the modeling In: 58th AIAA/ASCE/AHS/ASC structures, structural
dynamics, and materials conference. American Institute of
predictions and provided confidence for offering use- Aeronautics and Astronautics, Reston, Virginia
ful gauge design guidelines based on modeling. The 8. Hoffmann K (1989) An introduction to measurement
geometry of the grid in carbon nanostructure-based using strain gages. Hottinger Baldwin Messtechnik
strain gauges plays an important role that would allow GmbH, Darmstadt
9. Kobayashi AS (1993) Handbook on experimental
tailoring strain sensing devices to increase their sen- mechanics. VCH, Weinheim
sitivity without significantly changing their manufac- 10. Gouldstone C, Wu Y, Gouldstone A (2007) The geomet-
turing process. ric contribution to gauge factor of patterned lines on sub-
strates. Strain 43:306–310. https://​doi.​org/​10.​1111/j.​1475-​
Acknowledgements Special thanks go to Miguel Ángel Riv- 1305.​2007.​00353.x
ero Ayala for his help in fabricating the gauges. Authors would 11. Younis NT, Kang B (2011) Averaging effects of a strain
like to thank MBA Leonidas Andrade of DuPont for kindly pro- gage. J Mech Sci Technol 25:163–169. https://​doi.​org/​10.​
viding the Kapton HN sheets for this research free of charge. 1007/​s12206-​010-​1020-1
12. Lee BM, Loh KJ (2015) A 2D percolation-based model
Funding The research leading to these results received fund- for characterizing the piezoresistivity of carbon nanotube-
ing from the Office of Naval Research Global (ONRG), under based films. J Mater Sci 50:2973–2983. https://​doi.​org/​10.​
Award Number N62909-19-1-2119. CONACYT (Mexico) 1007/​s10853-​015-​8862-y
awarded the postdoctoral fellowship of Angel Mora (CVU 13. Pang C, Lee G-Y, Kim T et al (2012) A flexible and highly
#1047921), the research assistantship of Ivan Perez (CVU sensitive strain-gauge sensor using reversible interlocking
#1008770), and the MS scholarship granted to Gabriel Arana of nanofibres. Nat Mater 11:795–801. https://​doi.​org/​10.​
(CVU #951153). 1038/​nmat3​380

Vol:. (1234567890)
13
Meccanica (2023) 58:1717–1732 1731

14. Amjadi M, Yoon YJ, Park I (2015) Ultra-stretchable and 28. Chen X, Alian AR, Meguid SA (2020) Coupled electro-
skin-mountable strain sensors using carbon nanotubes– mechanical modeling of piezoresistive behavior of CNT-
Ecoflex nanocomposites. Nanotechnology 26:375501. reinforced nanocomposites with varied morphology and
https://​doi.​org/​10.​1088/​0957-​4484/​26/​37/​375501 concentration. Eur J Mech A/Solids 84:104053. https://​
15. Kang D-K, Kim C-W, Yang H-I (2017) Thermal effects doi.​org/​10.​1016/j.​eurom​echsol.​2020.​104053
on nonlinear vibration of a carbon nanotube-based mass 29. Haghgoo M, Ansari R, Hassanzadeh-Aghdam MK, Nan-
sensor using finite element analysis. Phys E Low-Dimens kali M (2022) A novel temperature-dependent percola-
Syst Nanostruct 85:125–136. https://​doi.​org/​10.​1016/j.​ tion model for the electrical conductivity and piezoresis-
physe.​2016.​08.​019 tive sensitivity of carbon nanotube-filled nanocomposites.
16. Ren X, Chaurasia AK, Oliva-Avilés AI et al (2015) Mod- Acta Mater 230:117870. https://​doi.​org/​10.​1016/j.​actam​
eling of mesoscale dispersion effect on the piezoresistivity at.​2022.​117870
of carbon nanotube-polymer nanocomposites via 3D com- 30. Haghgoo M, Ansari R, Kazem Hassanzadeh-Aghdam M
putational multiscale micromechanics methods. Smart et al (2022) Analytical formulation of the piezoresistive
Mater Struct 24:065031. https://​doi.​org/​10.​1088/​0964-​ behavior of carbon nanotube polymer nanocomposites:
1726/​24/6/​065031 the effect of temperature on strain sensing performance.
17. Xu S, Rezvanian O, Zikry MA (2013) Electro-mechanical Compos Part A Appl Sci Manuf 163:107244. https://​doi.​
modeling of the piezoresistive response of carbon nano- org/​10.​1016/j.​compo​sitesa.​2022.​107244
tube polymer composites. Smart Mater Struct 22:055032. 31. Ellingson S (2018) Electromagnetics. VT Publishing,
https://​doi.​org/​10.​1088/​0964-​1726/​22/5/​055032 Blacksburg
18. Han C-J, Chiang H-P, Cheng Y-C (2018) Using micro- 32. Hall LJ, Coluci VR, Galvão DS et al (2008) Sign change
molding and stamping to fabricate conductive polydi- of Poisson’s ratio for Carbon Nanotube Sheets. Science
methylsiloxane-based flexible high-sensitivity strain (80-) 320:504–507. https://​doi.​org/​10.​1126/​scien​ce.​11498​
gauges. Sensors 18:618. https://​doi.​org/​10.​3390/​s1802​ 15
0618 33. Mora A, Han F, Lubineau G (2018) Estimating and under-
19. Wang X, Li B, Zhang D et al (2021) Strain monitoring standing the efficiency of nanoparticles in enhancing the
using carbon nanotube buckypaper sensor on composite conductivity of carbon nanotube/polymer composites.
repaired structure. Appl Phys A 127:935. https://​doi.​org/​ Results Phys 10:81–90. https://​doi.​org/​10.​1016/j.​rinp.​
10.​1007/​s00339-​021-​05099-z 2018.​05.​019
20. Hu N, Karube Y, Yan C et al (2008) Tunneling effect in 34. Seidel GD, Puydupin-Jamin A-S (2011) Analysis of clus-
a polymer/carbon nanotube nanocomposite strain sen- tering, interphase region, and orientation effects on the
sor. Acta Mater 56:2929–2936. https://​doi.​org/​10.​1016/j.​ electrical conductivity of carbon nanotube–polymer nano-
actam​at.​2008.​02.​030 composites via computational micromechanics. Mech
21. Odegard GM (2018) Computational Multiscale mode- Mater 43:755–774. https://​doi.​org/​10.​1016/j.​mechm​at.​
ling—nanoscale to macroscale. Comprehensive composite 2011.​08.​010
materials II. Elsevier, pp 28–51 35. Li C, Thostenson ET, Chou T-W (2007) Dominant role
22. Zeng Q, Qin Y (2018) Multiscale modeling of hybrid of tunneling resistance in the electrical conductivity
machining processes. Hybrid machining. Elsevier, pp of carbon nanotube–based composites. Appl Phys Lett
269–298 91:223114. https://​doi.​org/​10.​1063/1.​28196​90
23. Bao WS, Meguid SA, Zhu ZH, Meguid MJ (2011) Mod- 36. Li C, Chou T-W (2007) A direct electrifying algorithm for
eling electrical conductivities of nanocomposites with backbone identification. J Phys A Math Theor 40:14679–
aligned carbon nanotubes. Nanotechnology 22:485704. 14686. https://​doi.​org/​10.​1088/​1751-​8113/​40/​49/​004
https://​doi.​org/​10.​1088/​0957-​4484/​22/​48/​485704 37. DuPont de Nemours I (2019) DuPont™ Kapton® HN. In:
24. Xia X, Du Z, Zhang J et al (2022) Modeling the impact of Wilmington, USA. https://​www.​dupont.​com/​conte​nt/​dam/​
glass transition on the frequency-dependent complex con- dupont/​amer/​us/​en/​ei-​trans​forma​tion/​public/​docum​ents/​
ductivity of CNT-polymer nanocomposites. Mech Mater en/​EI-​10206-​Kapton-​HN-​Data-​Sheet.​pdf. Accessed 19
165:104195. https://​doi.​org/​10.​1016/j.​mechm​at.​2021.​ Apr 2021
104195 38. Avilés F, May-Pat A, Canché-Escamilla G et al (2016)
25. Luo J, Peng L-M, Xue ZQ, Wu JL (2005) Total energy of Influence of carbon nanotube on the piezoresistive behav-
charged carbon nanotubes and single-electron tunneling. ior of multiwall carbon nanotube/polymer composites.
Phys E Low-Dimens Syst Nanostruct 27:26–31. https://​ J Intell Mater Syst Struct 27:92–103. https://​doi.​org/​10.​
doi.​org/​10.​1016/j.​physe.​2004.​09.​019 1177/​10453​89X14​560367
26. Ansari R, Hassanzadeh-Aghdam MK (2017) Microme- 39. Fu X, Al-Jumaily AM, Ramos M et al (2019) Stretchable
chanical characterizing elastic, thermoelastic and viscoe- and sensitive sensor based on carbon nanotubes/polymer
lastic properties of functionally graded carbon nanotube composite with serpentine shapes via molding technique.
reinforced polymer nanocomposites. Meccanica 52:1625– J Biomater Sci Polym Ed 30:1227–1241. https://​doi.​org/​
1640. https://​doi.​org/​10.​1007/​s11012-​016-​0512-1 10.​1080/​09205​063.​2019.​16276​49
27. Oliva-Avilés AI, Alonzo-García A, Zozulya VV et al 40. Chun S, Choi Y, Park W (2017) All-graphene strain sen-
(2018) A dielectrophoretic study of the carbon nanotube sor on soft substrate. Carbon 116:753–759. https://​doi.​org/​
chaining process and its dependence on the local electric 10.​1016/j.​carbon.​2017.​02.​058
fields. Meccanica 53:2773–2791. https://​doi.​org/​10.​1007/​ 41. Oliva-Avilés AI, Avilés F, Seidel GD, Sosa V (2013)
s11012-​018-​0869-4 On the contribution of carbon nanotube deformation to

Vol.: (0123456789)
13
1732 Meccanica (2023) 58:1717–1732

piezoresistivity of carbon nanotube/polymer compos- Mater Struct 15:737–748. https://​doi.​org/​10.​1088/​0964-​


ites. Compos Part B Eng 47:200–206. https://​doi.​org/​10.​ 1726/​15/3/​009
1016/j.​compo​sitesb.​2012.​09.​091 46. Bu L, Steitz J, Kanoun O (2010) Influence of processing
42. Xu W, Allen MG (2013) Deformable strain sensors based parameters on properties of strain sensors based on carbon
on patterned MWCNTs/polydimethylsiloxane composites. nanotube films. In: 2010 7th international multi-confer-
J Polym Sci Part B Polym Phys 51:1505–1512. https://​doi.​ ence on systems, signals and devices. IEEE, pp 1–6
org/​10.​1002/​polb.​23361
43. Yang T, Han E, Wang X, Wu D (2017) Surface decoration Publisher’s Note Springer Nature remains neutral with regard
of polyimide fiber with carbon nanotubes and its appli- to jurisdictional claims in published maps and institutional
cation for mechanical enhancement of phosphoric acid- affiliations.
based geopolymers. Appl Surf Sci 416:200–212. https://​
doi.​org/​10.​1016/j.​apsusc.​2017.​04.​166
Springer Nature or its licensor (e.g. a society or other partner)
44. Arana G, González-Díaz MO, Castillo-Atoche A, Avilés
holds exclusive rights to this article under a publishing
F (2023) Improving adhesion between polyimide surface
agreement with the author(s) or other rightsholder(s); author
and carbon nanotube arrays for strain-sensing devices.
self-archiving of the accepted manuscript version of this article
Mater Today Commun 34:105008. https://​doi.​org/​10.​
is solely governed by the terms of such publishing agreement
1016/j.​mtcomm.​2022.​105008
and applicable law.
45. Kang I, Schulz MJ, Kim JH et al (2006) A carbon nano-
tube strain sensor for structural health monitoring. Smart

Vol:. (1234567890)
13

You might also like