1. Introduction
Researchers have paid much attention to the strain sensors due to their key contribution to the flexibility and stretchability in foldable electronics, soft robotics, personal/rehabilitation health monitoring[1-4]. So far, various new methods and techniques have been introduced and developed for efficient stretchable and flexible strain sensors for detecting deformation in body motion at large scale[5, 6]. New materials whose properties are extremely high in terms of conductivity, sensitivity and better electrical stability under high strain have been paid more attention for these types of the devices. Carbon nanotubes (CNTs) are considered as one of the main materials in the field of nanotechnology possessing more tensile strength than steel, better electrical and thermal conductivity, and, owing to these properties, it has been intensively demanded by many researchers in the field of highperformance artificial muscles, battery electrodes and flexible super capacitors[7-10]. It has opened various fronts for research to fabricate CNTbased devices due to its tiny size and fibril structure[11]. It has been observed from many research findings that CNT-based sensors have low sensitivity, nonlinearity, poor restorability, narrow strain window in linear response having large hysteresis upon stretching and releasing process[12-14]. This may be the effect of applying strain: the CNT network could be damaged easily. However, it is crucial to resolve these problems before implementation of CNT-based strain sensors in real life. Tremendous work has been done in the field of actuators and sensors to explore this versatile material. Many researchers have investigated piezoresistive properties of individual CNTs by using micro-electro-mechanical system (MEMS)[15, 16]. An individual CNT possesses solid properties for sensor applications, but its poor process ability & integration complexity into bulk structures narrowed their widespread use. It has been reported by many researchers that dispersing CNTs into polymers or liquids by introducing some complex methods could improve the performance of sensors by forming 3D CNT network to strengthen the conductivity[17, 18]. It has been investigated from recent studies that CNT-based strain sensors have produced excellent results versus conventional sensors. However, it has been observed that without applying mechanical load[4, 19], CNT/polymer film sensors' resistance have been changed with time, which revealed an alarming situation for practical applications. This drift was thought to be the defects and breakdown of CNT shell structure[20], and could be from the current effect when current passed through the CNT sensor[19]. Moreover, it has been investigated that strain sensor based on CNT composite shows hysteresis resistance under repeated strain loads[21, 22], which reveals progressively worse CNT/polymer interfaces. In addition, temperature effect on sensitivity remained unsolved. For fabricating strain gauges, CNT Bucky papers have also been used[23, 24]. However, due to their low strength and deformation (approximately 0.5%), they are not convenient to be used for multi-functionality and reproducibility applications. In another method, CNTs were used as sensor & actuators by dipping them into electrolyte with high voltage (EDP method)[25]. From the investigations, it has been concluded that these CNT-based sensors are unsuitable to be applied in commercialized applications.
Herein, we report MWCNTs-based strain sensors fabricated by three different methods. We investigated and presented strain and resistance of the three strain sensors fabricated with different techniques. Our findings revealed after going through various experiments and measurements that the spray coated sample showed 60% strain better linearity and with small hysteresis curve.
2. Experiment & fabrication methodology
2.1 Synthesis of the MWCNTs
In this paper, we used two kinds of MWCNTs. One is the commercial MWCNTs purchased from Beijing Dk Nano technology Co., Ltd, Beijing, China with the diameter of 8-15 nm. The other kind was MWCNT arrays grown on SiO2 wafer through floating catalyst chemical vapor deposition (CVD)[26, 27]. A schematic diagram of floating catalyst CVD process has been illustrated in Fig. 1. A few samples of 300-nm thick SiO2 silicon substrate were placed inside the quartz tube on a high temperature at the middle of the furnace tube. 10% H2 and 90% Ar gas as a carrier gas for growth of CNTs was introduced to quartz tube and precursor solution was injected by an injector at 80 ℃. In our experiment 2.106 g of ferrocene in 120 ml xylene were mixed together to prepare precursor solution followed by sonication and starrier. This solution was pre-heated at 200 ℃ with an injector's speed of 8 mL/h. By controlling the growth time, we can get different length of CNT arrays. Fig. 2 illustrates the SEM images of the two kinds of MWCNTs, respectively. The diameter of the CNTs has some difference, but the performance of strain sensors based on CNTs is mainly confirmed by the structure of the sensors not the diameter of CNTs.
2.2 The preparation of PDMS substrate
The PDMS is commonly used by many researchers with the same following process. PDMS (Dow Corning Sylgard 184; proportion of base to cross-linker, 10:1 by mass) was blended, then it was degassed in vacuum for half an hour and poured against the cleaned surface of a silicon wafer with a 300 nm silicon oxide layer. After that, it was put in an oven or in room temperature for 24 h to cure.
2.3 Spray coating of CNTs solution
For spray coating sample, a solution of MWCNTs was prepared from 10 mg of MWCNTs powder with the 10-20 ml of N-methylpyrolidone followed by 30 min sonication. MWCNTs sample was fabricated by spray coating of 40 times with 9 s delay at the temperature of 200 ℃ on PDMS substrate.
2.4 Direct deposition of MWCNTs
At 100 ℃ dry powder of MWCNTs was deposited on PDMS substrate by applying pressure to spread MWCNTs on PDMS substrate. The sketch of MWCNTs powder-based strain sensor is shown in Fig. 3.
2.5 CNT array-based sensors
To fabricate CNT array-based strain sensors, the PDMS solution was dropped on the surface of CNT arrays at about 25 ℃ for 24 h to solidify the PDMS. Then the CNT arrays were immersed in PDMS and stuck onto the PDMS, which were peeled off from 300-nm thick SiO2 silicon substrate for strain sensor measurement. The structure for MWCNT array-based strain sensor is illustrated in Fig. 3.
3. Results and discussions
The output resistance change was observed by applying strain on the PDMS layer of MWCNTs sheet to obtain the sensing properties. Fig. 4 reveals that the resistance response with strain has a wider linear range than strain sensor-based CNTs and graphene (2%-5%) reported by other groups[11, 12, 28, 29], the hysteresis curve reveals the wider range and can be noticed in Fig. 5. CNTs low density could be the consequence on the substrates. Linear equation can be established for linear region between strain and resistance for calculating direct values of strain applied on the device corresponding detected resistance while noticing the change in curves. It might be convenient to detect the change in strain accurately. Wider linear range reveals that strain can work in higher strain region and the strain can be calculated accurately with linear equation.
Hysteresis curves in respect to wider range are presented in Fig. 5. The change in corresponding resistance can be observed linearly by applying strain in Fig. 5(a). The sample was recorded with excellent performance and maximum strain of 60% which is better than the generally available strain sensors. However, low sensitivity and nonlinear response of the dry powder MWCNTs sample and different lengths CVD-grown array CNTs can be observed in Figs. 5(b), 5(c), and 5(d). We discovered that spray coating sensor on PDMS substrate showed higher sensitivity in Fig. 5(a) than the other sensors reported Figs. 5(b), 5(c), and 5(d). The high sensitivity reveals that each individual CNT is connected with neighboring CNTs on MWCNTs sheet along its length as shown in SEM images of Fig. 3. When the strain is increased, it causes disconnection of CNTs thereby causing increase in resistance change.
Moreover, the sensitivity of samples prepared by dry MWCNTs powder showed better results in terms of low hysteresis and good linearity than that of MWCNT arrays in two cycles of the stress and strain measurement. All in all, MWCNT arrays grown by CVD process on SiO2 showed nonlinearity, high hysteresis and small stretchability compared to spray coated and dry CNT sample. It could be because the high density of CNT arrays can prevent the change of contact resistance between CNTs when the strain is applied to the sample.
In addition, the long-term stability and repeatability of the spray coating sample on PDMS substrate was carried out by using standard test machine. Periodically, strain was applied while investigating the performance of the sample. The change in resistance result can be observed in Fig. 6. We can see clearly, the variations are within 10%, which indicates that the fabricated strain sensor possesses better stability and can be reliable for motion capturing. In brief, the strain sensors produced by spray coating have almost the same excellent performance as in the earlier report[30, 31], but the preparation method is much simpler and cheaper.
To get a better understanding about the influence of MWCNT structure on the performance of flexible strain sensors, SEM images were utilized to check the microstructure of MWCNT-based sensors. The SEM images for MWCNT strain sensors fabricated by spray coating, direct deposition and MWCNT arrays were shown in Fig. 7. The surface structures of sensors fabricated by spray coating and direct deposition, in Figs. 7(a) and 7(b), presented that the MWCNT sheet prepared by directly depositing on PDMS substrate was not uniform and has more holes. For MWCNT arrays, Fig. 7(c) showed the sample had high density of MWCNTs. The cross-sectional SEM images indicated that some MWCNTs were immersed in the PDMS substrates fabricated by direct deposition. For the sensors fabricated by spray coating shown in Fig. 7(d), the CNTs had close contact with the PDMS substrate, but the CNTs had not penetrated into PDMS as in Fig. 7(e). When PDMS immerged into the MWCNTs arrays, shown in Fig. 7(f), the structure of MWCNTs arrays kept the same. It is well known that the connection properties between MWCNTs would dominate the stretchability and sensitivity of MWCNT strain sensors. When more CNTs were immersed in the PDMS substrates, it could decrease the sensitivity of these strain sensors because CNT-polymer composite has lower sensitivity[31]. MWCNT arrays have orientation, which is beneficial to produce strain sensors with directional selectivity. However, in our experiment, the density of MWCNTs limited the sensitivity of this kind of sensor. High density of MWCNTs means more connections between MWCNTs, which implied strain can hardly change the connection resistance. Controlling the density of aligned CNTs may improve the performance.

4. Conclusions
In summary, different flexible strain sensors with different structures based on MWCNTs were fabricated by different techniques. One of the sensors prepared by spray coating MWCNTs solution on PDMS substrate exhibits higher stretchability and sensitivity compared with other sensors, while the CNT arrays-based strain sensors have the lowest response to the strain. The microstructure of these different CNT-based strain sensors from SEM images elucidated that controlling the distribution, density and connection of CNT network is the key factor in fabricating CNT-based strain sensors. This sensor produced by spray coating MWCNTs with uniform CNT network and better performance in sensitivity than the generally available sensors reveals that spray coating technique is a more economic method to develop CNT-based strain sensors in detecting human body motion and other applications.