1 Introduction

Piezoresistive material-based strain sensors have been intensively used in, such as wearable electronics [1,2,3], displacement detection [4], physiological monitoring [5], and dosage surveillance [6]. Among the reported piezoresistive materials, carbon nanotubes (CNTs) that were first reported in the 1990s [7,8,9] have attracted growing interest in recent years. They possess superior properties, such as electrical, mechanical, optical, and chemical properties; their strength is over tenfold larger compared to the other industrial fibers. The characteristics of some piezoresistive materials against CNTs are summarized in Table 1 [10]. CNTs are seamless cylindrical structures of single or multiple layers of graphene, denoted as single-wall CNTs (SWNTs) or multi-wall CNTs (MWNTs), respectively. The structures of SWNTs and MWNTs are depicted in Fig. 1. SWNCTs are cylinder-like structured which are formed through a single sheet of graphene rolled seamlessly; it has a diameter of 1–2 nm and length of up to centimeters. MWCNTs consist of multiple concentric SWNCTs which are separated by about 0.35 nm. SWNTs usually have higher electrical conductivity (106 S/cm) and thermal conductivity (∼3500 W/mK) than MWNTs (3 × 104 S/cm, 3000 W/mK), been acting as a metallic role. As a common piezoresistive strain sensor type, CNTs–polymer composites that change in electrical resistance upon an external force (namely piezoresistive effects) have gained popularity [11].

Table 1 Characteristics comparison between the CNTs and other materials
Full size table
Fig. 1
figure 1

A Structure of SWNTs. B Structure of MWNTs

Full size image

The piezoresistive characteristics of CNTs-filled polymer composites were demonstrated in [19, 20], and they were introduced comprehensively in [21, 22] as strain sensors. The works aforementioned mainly focused on the relationship between the applied pressure and resistance change of the composites. On the other hand, the piezoresistive mechanism of the CNTs–polymer composites should be more comprehensively studied from the view of the changes in the micro-scale. A 3D statistical resistor network model was introduced to simulate the resistance change when the strain was applied to MWNTs-filled polymer composites [23]. Different piezoresistive properties can be found in CNTs–polymer composites with different polymer matrixes [21]. The piezoresistivity of conductive polymer composite that used CNT and silicon rubber was analyzed in terms of the change in the conductive network [24]. CNTs can be more easily embedded into polymer composite structures for in situ measurement than the other traditional strain sensor materials. The CNTs have a high aspect ratio, their diameter usually does not exceed 100 nm, and the length can achieve up to hundreds of millimeters, which exhibits a 1D nanostructure. Consequently, the CNTs are usually filled in polymers to form an effective conductive network in a composite with a 3D structure. The piezoresistivity and conductivity of CNTs–polymer composites were investigated in [25,26,27,28,29]; in such composites, CNTs were used in their conductive phase, while the polymer performed as insulation. The optimal ratio of the CNTs is about 2 vol% (the volume fraction) [24], and the high ratio will cause the deterioration of the mechanical property of the entire composites, while decreasing the ratio of CNTs will lead to increased resistance, making the integration of the other circuit systems difficult.

This paper aims to survey fabrication processes, working principles, and sensor applications of the CNTs–polymer composites. The article is organized as follows: the second section summarizes novel approaches and functional nano-materials for the fabrication of flexible CNTs–polymer composites in detail; in the third section, the mechanisms involved in the strain-responsive behavior of piezoresistive-type sensors are explained; the fourth section introduces elastic and sensitive strain sensors that commonly serve as E-skin; the next section discusses possible challenges for the skin-mountable and wearable sensor applications; the last section gives the conclusion points.

2 Fabrication processes

The CNTs–polymer composites are in general fabricated by either direct mixing of the CNT and polymer in a suitable solvent [30,31,32], melt processing of bulk composite [33, 34], melt processing of composite fibers [35, 36], polymerization [37,38,39], or coagulation spinning of composites [40,41,42]. However, it is challenging that pristine CNTs cannot be well dispersed in most solvents; thus, the performance of the CNTs–polymer composites is greatly degraded following the traditional fabrication procedures. Recently, new approaches for uniform distribution of the CNTs in the polymer have been investigated; the spinning approaches, dispersion conditions, and characterizations of various CNTs–polymer composites are summarized in Table 2. A layer-by-layer approach that involves building up a layered composite film by alternate dipping of a substrate into the dispersion of CNTs and polyelectrolyte solution was introduced in [43, 44]. The major merits of this approach include controllable thickness and polymer–nanotube ratio and high nanotube loading level. The swelling technique is another promising post-processing approach that enables the incorporation of CNTs into insoluble or temperature-sensitive polymers. For example, MWNTs-Kevlar and MWNTs-polyethylene (PE) composites can be produced by swelling Kevlar fibers and PE film in a suspension of MWNTs in N-methylpyrrolidone (NMP) [45] and tetrahydrofuran (THF) [46] under ultra-sonication. The resulting composites exhibit higher electrical conductivity and mechanical ability. The bucky-paper-based approach has been extensively reported [47,48,49] to produce a thin porous assembly of CNTs. The fabricated composites normally have a laminar structure with a random orientation of the bundles of tubes in the plane of the sheet [50].

Table 2 Summary of CNT–polymer composites fabrication approaches and their key characteristics
Full size table

The other functional nano-particles that can be added into the CNTs–polymer to improve the sensing performance include the following: in [70], the silver nano-particles (AgNPs) were added in the polymer composites to prevent the intact interaction between neighboring CNTs as well as tailor the conductive pathways in the composite; Fe3O4 nano-particles were mixed with CNTs-Poly (vinylidene fluoride) (PVDF) to improve the thermal conductivity of the composites, which in turn converted microwave energy into Joule heating systems [71]. In this review, the composites that are formed by mixing the CNTs and the polymer and their corresponding fundamental structure are emphasized. Regarding support materials for the CNTs, silicone-based elastomers [e.g., polydimethylsiloxane (PDMS)] [72,73,74,75,76] and rubbers [77,78,79] are widely used. Figure 2 illustrates an overall fabrication process of a popular CNTs–polymer reported in [80], and it is composed of MWNTs, PDMS, and reverse micelle solution (RMS). The components and their weight ratio are given in Fig. 2a. RMS solution is formed after ultrasonic vibration of 30 min and the mechanical stirring for another 30 min until it becomes a homogeneous black gel-like solution (Fig. 2b), so that MWNTs can be able to be evenly distributed to form a uniform RMS-MWNTs solution. The mixture of PDMS and RMS-MWNTs then is poured into a flat and enclosed substrate to form a film (Fig. 2c).

Nozzle-jet printing is used on the target substrate to design various line patterns (Fig. 2d). The final step is to cure the PDMS and solidify the mixture. The temperature for the process follows the heating-treatment requirement as listed at the bottom of Fig. 2(a). The reverse micelles had evaporated, producing a porous structure during the heating process. The more MWNTs were mixed, and the smaller pores were formed (Fig. 2e). Computer-aided design (CAD) can also be used in the nozzle-jet-printing for line pattern, as depicted in Fig. 3 [80]. MWNTs mixed-rubber solution was patterned by moderating the flow rates and the nozzle movement speeds after patterning is done. Silicone oil was used to cover the rubber solution, because it was immiscible with RMS and does not damage the pressure-sensitive rubber pattern. The rubber solution adhered to the flexible PET substrate, so the final strain sensor can suffer from numerous deformations, as shown in Fig. 3b. The magnified view of uniformly distributed pores is shown in Fig. 3c.

Fig. 2
figure 2

a The flowchart of the fabrication process of flexible WMNTs-based strain sensor. b Illustration of mixing procedure. c The flat thin strain sensor film formation method. d The line pattern formed using the nozzle-jet printing method. e RMS behaviors during the heat treatment procedure [80]

Full size image
Fig. 3
figure 3

Illustration of the nozzle-jet printing process for the patterned PPSR [80]

Full size image

2.1 Strain-responsive principle

The structure of the piezoresistive sensor plays a critical role in the strain-responsive principle. The sensor was usually designed as a sandwich structure [81], double-percolation structure [82], porous structure [83], segregation structure [84], sponge structure [85], etc. [86]. In [87], a flexible piezoresistive sensor with an interlocked structure was explored as bionic human skin, to detect three-dimensional force; A piezoresistive membrane with a four-petal structure, was designed as a Wheatstone bridge, to be applied in low-pressure measurement applications [88]; The MWCNTs were assembled onto a polydimethylsiloxane (PDMS) film with a pyramidal microarray structure, to output tactile force signal [89]. In this review, the structure design of the sensor is not emphasized; only the working principle of the CNTs–polymer of which deformation results in electrical resistance change is analyzed.

An SEM image of the CNTs distribution in the polymer is shown in Fig. 4 [90]. It is observed that the conductive particles are randomly distributed and the adjacent CNTs are insulated by thin polymer films. However, the tunneling transport of electrons can make the separated CNTs conductively connected through junction gaps, forming a conductive network [32]. The so-called “tunnel effect” demonstrates a phenomenon where the current can flow over adjacent conductive particles as long as their gaps are very small [91, 92]. Therefore, the tunnel effect is commonly presented to explain the conductivity of the CNTs–polymer [23, 93].

Fig. 4
figure 4

SEM image of CNTs dispersion [90]

Full size image

As shown in Fig. 5a, the whole resistance of the random CNTs network includes RN of all the CNT segments, RJ of all the tunneling resistances due to the junctions. When the external force is applied on the CNTs-filled composite, the resistance of the composite will change, which refers to three possible reasons (Fig. 5b): (1) variation of RJ occurs due to the change of junction gap; (2) geometry deformation induced-CNT resistance variation ∆ρCNT; (3) junction length along with CNTs changes ∆lCNT due to slippage. Reasons 2 and 3 contribute to the variation of RN. Actually, for the CNTs–polymer composites, RN does not dominate the resistance of the composites network, because RN is usually far less as compared to RJ, and the effect of ∆RN can nearly be ignored for the resistance variation of the composites. Therefore, the average junction gap variation (AJGV) is a quantitative parameter to mainly determine the piezoresistive performance of the network when the composite suffers from external strain or compression.

Fig. 5
figure 5

Schematic illustration of a an electrical conductive CNTs in a polymer matrix, and b three possible reasons for the piezoresistive effect [94]

Full size image

Because the conductivity of the composite is mainly based on the CNTs-formed conductive links, the entire resistance is mainly determined by the AJGV between the adjacent CNTs. Based on the tunneling effect theory, the tunneling current density J over the insulation gap can be given by [95]

$$J = \frac{{3e^{2} U\sqrt {2m\varphi } }}{{2h^{2} de^{{4\pi h^{ - 1} d\sqrt {2m\varphi } }} }},$$

where U is applied voltage, d is the thickness of the insulation, φ means the height of the potential barrier, m is electron mass, e is electron charge, and h is Planck’s constant. It can be observed that an increasing gap d between the neighbor CNTs caused by the stretching of the sensor leads to an increasing resistance change, while the decreased junction gap caused by compression will cause a decreasing resistance change.

Moreover, disconnection and reconnection of the CNTs-formed conductive links or the occurrence of cracks in the composites can also cause the resistance change during deforming the sensor [96]. Figure 6 illustrates a cracked surface of the conductive polymer composites containing CNTs and the concept of the effective conductive pathways. It is assumed that the initial region can be divided into N0 identical conductive pathways [97] with a width d, the initial electrical resistance R0 of the composite specimen, and its whole width W can be expressed by

$$\left\{\begin{array}{c}{R}_{0}={R}^{E}/{N}_{0}\\ W=\sum_{i=1}^{{N}_{0}}d\end{array},\right.$$

where RE denotes an effective resistance of a single conductive pathway. When the cracks occur and its propagation breaks effective conductive pathways, the crack extension ∆W can be obtained as follows:

$$\Delta W=W-\sum_{i=1}^{{N}^{^{\prime}}}d,$$

where N′ is the number of effective conductive pathways after crack propagation. Then, the resistance of the composite specimen becomes

Fig. 6
figure 6

Illustration of the effective conductive pathways and cracked surface of the polymer composites

Full size image
$${R}^{^{\prime}}={R}^{E}/{N}^{^{\prime}}.$$

Thus, the change in the resistance of the composite specimen, ∆R = R′-R0, can be approximately regarded in terms of ∆W and W as

$$\Delta R=\frac{\Delta W}{W-\Delta W}{\cdot R}_{0}.$$

Therefore, the crack extension ∆W caused by the external force or deformation of the polymer composites causes an increasing resistance change ∆R, explaining the principle of the sensors which are based on the change in contact resistance between the cracked fracture structures.

To clarify the relationship between the strain and resistance, Fig. 7a shows a basic setup for measuring the electrical resistance of the sensing element under different levels of pressure. The material testing machine can apply pressure on the testing material by controlling the movement of the movable head. The resistance is recorded by a digital multimeter with the applied pressure. When the external force is applied to the CNTs–polymer composites, the distance between separated CNTs will change accordingly. If the gap is reduced to a threshold range where the tunneling effect can be able to happen, a new conductive path can be formed, so that the current can be able to penetrate the insulating matrix. Accordingly, the formation or destruction of the effective conductive paths may change the actual number of the conductive links, so the resistance of the entire composites changes based on the AJGV between the adjacent CNTs in the conductive network. Specifically, external pressure may make the CNTs move far away from each other, decreasing the number of effective conductive paths; this, in turn, increases the resistance. Figure 7b shows that the resistance increases with the increase of pressure for four different tested elements. This result convicts that the compression enlarges the gap between carbon nanotubes and reduces the number of effective conductive paths, resulting in destruction effects at conductive links.

Fig. 7
figure 7

a The measurement setup for sensing the electrical resistance under different applied pressures [24]; b resistance–pressure curves of the four sensing elements [24]

Full size image

2.2 Applications

Flexible pressure (or strain)-sensitive CNTs–polymer composites can be used in a variety of engineering fields. The most promising applications include wearable electronics, robotic skins, and medical health monitoring [98]. In this section, the sensors based on CNTs–polymer composites that serve as skin-mountable strain gauges are emphasized, for example, the electronic-skin (E-skin). In general, the resistance or capacitance of the sensor that deploys at the skin surface or joints changes when it deforms. Human physiological signals and real-time wrist pulse detection can be monitored according to the output signal of the sensor. Owing to the excellent sensing performance of the sensor in terms of rapid response time, great stability, and repeatability, the E-skin can work in conjunction with the wearable device for the prevention and prediction of illnesses. As shown in Fig. 8, a flexible pressure-sensitive polymer film based on the MWNTs adhered to the wrist, the geometry of the film changed correspondingly for the deformation of the skin surface; consequently, the film was stretched and bent. The two insets on the right of Fig. 8 show the resistance variations of a wearable sensor according to different bending frequencies. Therefore, the repetitive bending motions of the human wrist can be monitored by doing so, and the cyclic tests did not induce any unwanted signal and material degradation, exhibiting a stable working performance.

Fig. 8
figure 8

Monitoring the resistance change according to the stretching and bending of the skin due to wrist movements; g–h) magnified views of measured signals for high-frequency and low-frequency movements [80]

Full size image

As well known, speech can cause the deformation of the epidermis and muscles around the throat; therefore, E-skin can be able to provide a novel and achievable application for voice recognition. Figure 9a shows a prototype of a strain sensor-based E-skin; it can be used for monitoring pressure difference at a person’s neck during voicing, because the resistance would change corresponding to the muscle movement induced by speech. Figure 9b depicts different I–t profiles when the tester spoke different words. By doing so, different voices and speech can be recognized through fast and sensitive strain sensing. To investigate the point that the similar I–t curves can be repetitively obtained for speaking the same thing, saying “hello” and “one world one dream” three times and Fig. 9c, d shows the corresponding I–t curves, respectively. It is observed that the three I–t curves have nearly the same characteristic peaks and valleys, which convict the point aforementioned. This application can help those patients who damage their vocal cords recover their voicing ability via controlling their throat muscle movement.

Fig. 9
figure 9

a Prototype of E-skin directly adheres to a tester’s neck for monitoring the muscle movement. b–d Real-time muscle movement monitor via I–V curves due to resistance change of flexible pressure sensor [99]

Full size image

A flexible pressure sensor can also provide great potential for modern biomedical applications. Figure 10 depicts a prototype of an MWNTs-based pressure sensor for detecting the wrist pulse. Wrist pulse can be able to indicate arterial blood pressure, heart rate, and other physiological information which helps the non-invasive medical diagnosis. For example, atherosclerosis can cause arterial pulse pathologic and affect arterial pressure, while wrist pulse is a key indicator to monitor arterial blood pressure, which can perform a rapid, easily operated, and effective method for diagnosing such cardiovascular diseases [100,101,102]. As shown in Fig. 10a, the pressure sensor that was placed on the wrist can differentiate the subtle pressure difference between those who are indifferent body conditions. Figure 10b provides two continuous I–t curves monitoring the wrist pulse for healthy people and a pregnant woman (21 weeks). It is seen that people with different body conditions owned different wrist pulse shapes and frequencies. The healthy person showed a regular and repeatable pulse shape, and the pulse frequency was 75 beats/min within a normal range. While, the pregnant woman showed an irregular pulse shape and intensity, and pulse frequency was 91 beats/min (out of the normal range).

As well known, speech can cause the deformation of the epidermis and muscles around the throat; therefore, E-skin can be able to provide a novel and achievable application for voice recognition. Figure 9a shows a prototype of a strain sensor-based E-skin; it can be used for monitoring pressure difference at a person’s neck during voicing, because the resistance would change corresponding to the muscle movement induced by speech. Figure 9b depicts different I–t profiles when the tester spoke different words. By doing so, different voices and speech can be recognized through fast and sensitive strain sensing. To investigate the point that the similar I–t curves can be repetitively obtained for speaking the same thing, saying “hello” and “one world one dream” three times and Fig. 9c, d shows the corresponding I–t curves, respectively. It is observed that the three I–t curves have nearly the same characteristic peaks and valleys, which convict the point aforementioned. This application can help those patients who damage their vocal cords recover their voicing ability via controlling their throat muscle movement.

Fig. 10
figure 10

a Prototype of electronic skin device for detecting wrist pulses. b I–V curves demonstrate the wrist pulse of a healthy person and a pregnant woman by using the pressure sensor [99]

Full size image

Figure 11 gives an example where a tank-like robot (MINDSTORMS NXT 2.0, LEGO, USA) was remotely controlled by a wearable sensor array. The instruction signals (moving forward, acceleration, deceleration, etc.) can be transmitted via the output voltage variation over each circuit. Figure 11a, b shows the images of piezoresistive sensors using MWNTs-filled polymer composites for detecting pressure and strain, respectively. The pressure sensor array consists of four channels, each of which functions different action command for controlling the motion of the robot. The strain sensors that were mounted on the middle and index fingers consist of two channels, triggering acceleration and deceleration, respectively. Each skin-mounted sensor was connected in series to a resistor of 100 kΩ (used as reference), and the six resistor-sensor pairs were connected in parallel. When a DC bias of 5 V was applied to the circuit, the output signals were measured from the point between the resistor and the sensor (Fig. 11j right top inset). The custom-made LabVIEW program analyzed the output signal changes of the six skin-mounted sensors induced by mechanical motions (e.g., pressing or stretching) and activated a specific keypad. The blue tooth embedded robot can receive the wireless signals and operate the corresponding instructions. Figure 11 provides an example for the human–machine interface (HMI) and convinces that the MWNTs-based sensor indeed exhibits a promising potential for HMI due to its high linearity and sensitivity. Actually, through this example, It can also be observed that CNTs used in tactile sensors own wider working ranges, low commercial cost, and less power consumption than optical, magnetic, ultrasonic materials. Their structures associated with read-out electronics are also simply deployed. For future applications, CNTs-based piezoresistive sensors can be widely used for humanoid robots or other healthcare control devices.

Fig. 11
figure 11

Wearable mechanical sensor for remotely controlled applications. A–I images of flexible MWNTs-based strain sensors with different circuit structures. The respective functions of each sensor are displayed. J Output voltage signals from the sensor array while controlling the robot wirelessly [80]

Full size image

2.3 Discussion and challenges

Though CNTs–polymer composites-based strain sensor owns lots of advantages and offers great potential for a variety of applications, several challenges can still not be ignored. At least two significant issues need to be addressed here, including repeatability and stability.

  • Repeatability: it refers to repeatable resistance change under cyclically applied strain. According to some latest studies, the resistance of CNTs–polymer composites shifts over time even an applied mechanical load is removed [103, 104]. This phenomenon will bring lots of serious problems in practical applications. Resistance drifting may attribute to breakdown and exfoliation of the surface of CNTs, as well as heat-induced by current passing through the conductive network.

  • Stability: it denotes a stable resistance behavior over a long operation period. As for many piezoresistive strain sensors including CNTs, the I–V pulses are not similar to each other even the same strain is applied, inducing some precision issues. Moreover, owing to irreversible deterioration at interfaces of the CNTs and polymer composite, CNTs-based flexible strain sensors usually perform resistance hysteresis during cyclically operated strain loadings.

The selection of polymer is also an important factor. The weak polymer encapsulation is hard to sustain repeatable and permanent deformations and a high strain force (larger than 0.04%). Therefore, the fragility of the sensors is a critical issue that needs to be overcome.

3 Conclusion

The carbon nanotube is a popular material nowadays and exhibits a wide range of applications. Carbon nanotube-based pressure sensor with a flexible polymer composite shows promising trends in the field of tactile sensing, electronic-skin, and other human–machine interfaces. The working principle is mainly based on the tunneling theory which indicates that the resistance of the insulating phase is varying according to the geometry change. The sensing examples aforementioned demonstrate the feasibilities and high efficiency of the CNTs’ conductive phase, paving the way to future healthcare monitoring, intelligent controlling devices.