The overall design of the piezoresistance sensor is illustrated in Fig. 2a, which consist of the top and bottom electrodes and middle porous elastomer. The electrodes as output signal layer were fabricated by etching the copper-clad PI film. The middle part is CNTs-SEBS porous elastomer as sensitive layer, which can sense compression. Conductive silver gel is utilized to bond the porous elastomer and the electrodes together for forming a reliable contact. Such contact resistance determines the initial resistance of the device. The elastomer was designed as a cylinder with radius of 3mm and height of 5mm. Accordingly, the size of electrodes was designed to be 8mm×8mm, so that the elastomer can be completely covered. Due to the adjunction of liquid paraffin in the preparation process of the elastomer, vaporized gas would escape during the melt-molding process, which plays an important role in forming the pores in the interior of the elastomer. In the meantime, the aggregation effect of CNTs also contributes to the formation of porous structure. Figure 2b was the SEM characterization image of the cross section of the elastomer. There were many small pores in a size of hundreds of micrometers distributed inside the elastomer. The enlarged SEM image of a single small hole was shown in Fig. 2b. The introduction MWCNTs not only contribute to form the pores, but also form the conductive path for mechanical sensing. Figure 2b was the enlarged SEM picture of the CNTs inside the elastomer. Multiple MWCNTs with a length of 8–14 µm are densely and randomly distributed in the material.
The presence of micropores in the elastomer provides free space for the flexible materials to expand under pressure, so that less pressure is required to deform the elastomer layer. As shown in Fig. 2d, when the elastomer is under pressure, the internal pores are squeezed to shrink to disappear and the conductive material around the pores contact, which would increase the conductive path inside the elastomer. As the pressure continues to increase, the interior of the elastomer is compressed, and the spacing between the MWCNTs is reduced. Due to the tunneling effect, the resistance inside the material is decreased, so that the pressure response can be detected in a larger range.
As the sensitive material in the piezoresistive sensor, CNTs plays a key role in the performance of the sensor. In order to systematically investigate the effect of CNTs on the sensor performance, we herein fixed the total mass of the materials (CNTs and SEBS), while tuning the mass ratio of CNTs and SEBS for the following study. We set the mass ratio of CNTs/SEBS at 2:10, 2.5:10, 3:10, to evaluate the compression response of the sensor in different mass ratio of CNT/SEBS. The sensors with different mass proportion of CNTs show different sensitivity (gauge factor, GF) as shown in Fig. 3a. Here, the GF is defined as\(GF=\frac{\frac{\varDelta R}{{R}_{0}}}{\epsilon }=\left(\frac{{R}_{0}-R}{{R}_{0}}\right)/\epsilon\)
where \({R}_{0 }\)is the initial resistance without applied strain, ε is the corresponding linear strain, and R is the resultant resistance under external strain. The figure indicate that the sensor shows lower sensitivity in lower mass proportion of CNTs. Because the lower proportion of CNTs (0.2gCNT/1gSEBS) results in the weak aggregation effect which reduces pores in the structure. However, when the mass ratio of CNTs/SEBS is increased to 3:10, the sensitivity decrease, which might be attributed by the stiffness of the material. The Fig. 3c presents the stress-strain response of elastomers with different mass ratios of CNTs/SEBS. It can be seen from the figure that the elastomer exhibited minimum Young's modulus when the mass ratios of CNTs/SEBS is 2.5:10. The lower Young's modulus makes the elastomer easier to be compressed and the conductive path inside the elastomer easier to be formed which increase sensitivity of the sensor. Considering comprehensively, the CNTs/SEBS of 2.5:10 was adopted as the optimized parameter for the sensor.
The Fig. 3b present the pressure-resistance response of the sensor. Owing to the porous microstructure inside the elastomer, the sensor exhibited greater sensitivity under low pressure. Its sensitivity was 0.848\({N}^{-1}\) the range of 0 ~ 1 N. When subjected to greater pressure, the spacing between the CNTs in the sensing material continues to shrink, which caused conductivity increased to detect pressure. The sensitivity reaches 0.041\({N}^{-1}\)within 1-4N. Because each elastomer is cut from a large piece of elastomer, it exhibited a high degree of consistency between the sensors in Fig. 3b. As shown in Fig. 3d, the response time and recover time of the piezoresistive sensor under high pressure were measured to be 278 ms and 273 ms, respectively. The stability of the device was also tested (Fig. 3g). We first monitored the response of the device to different cyclic stress of 0.5N, 1N, 2N (Fig. 3e). The stepped waves under different stress and the stable waveforms corresponded to each stress, indicating that the proposed sensor has excellent stability over the compression range. To examine the potential for long-term use, we monitored the relative resistance variation of the sensor under a periodic compressing/releasing (0.5N) for 10000 cycles, as shown in Fig. 3f. The sensor exhibited a constant resistance variation over the whole duration, which indicates the excellent stability of the device for potential long-term use.
Because the device is mainly used on mechanical palm, the sensor needs to be highly robust to ensure that the sensor can work normally on the mechanical palm in case of mechanical collision in industrial production activities. A running car was used for testing the robustness of our sensor under extreme mechanical conditions, which can generate a high pressure and a large shear stress at its tire tread. We fixed the sensor on the ground and the resistance signal was collected using the digital source meter which was connected with the sensor through data cable. When the test vehicle is running, the tire tread bears a normal pressure of ~ 350 kPa and a shear stress of about 7 kPa, which is a great challenge for a flexible pressure sensor to survive under such complicated mechanical conditions. Figure 3i shows the resistance response of the sensor when the front and rear tires of the test vehicle press over the sensor respectively. The sensor exhibited a constant resistance variation in the twice press and recovered to the initial resistance when the stress released. To further illustrate the robustness of the sensor, we compared the pressure-resistance response of the sensor before and after the test in Fig. 3j. The twice responses showed high consistency which indicates the strong robustness of the device.
Similar to the design of the piezoresistive sensor above, the capacitive sensor consists of three layers. The top and bottom were electrode layer and the mid layer was dielectric layer (Fig. 4a). The dielectric layer is an elastomer with conductive filler added. To avoid mechanical mismatch, we utilize the same flexible material with the piezoresistive sensor to made the dielectric elastomer. In addition, compared with other flexible materials such as PDMS and ecoFlex, SEBS shows a lower Young's modulus, which makes the elastomer easier to be compressed, thereby improving the sensitivity of the sensor. The composite material system is heterogeneous, which will produce interfacial polarization effect, which has a significant impact on the dielectric properties of the composite material. The size of conductive filler has a great influence on interface polarization. When the size of the filler is smaller, the specific surface area of the filler is larger, and the interfacial polarization is more significant. Carbon black is a kind of spherical particle with a size of 30-50nm, which can produce a larger specific surface area than other fillers (Fig. 4c). Therefore, we choose carbon black as the conductive filler.
Figure 4d exhibit the surface and the cross section of the dielectric elastomer. There were some small pores in a size of tens of micrometers distributed inside the elastomer. Compared with the piezoresistive elastomer above, the count of pores is less and the size is smaller, which might be attributed by the small particle sizes of CB nanoparticles and thus a weak aggregation effect. As shown in the Fig. 4e, when pressing the sensor, the thickness of the dielectric elastomer decreases and the cross section increases, resulting in a decrease in capacitance.
The dielectric properties of dielectric elastomers are affected by the mass of conductive fillers and the internal small pores [31]. When the volume fraction of conductive filler reaches near the percolation threshold of the material, the dielectric elastomer will achieve the best dielectric properties. To systematically investigate the effect of the mass proportion of the filler on the dielectric performance, we herein fixed the total mass of the materials (CB and SEBS), while tuning the mass ratio of CB and SEBS to test the pressure-capacitance response of the capacity sensor. As shown in Fig. 4f, the sensitivity of the capacity sensor increased with the mass ratio of CB/SEBS, and then decreased when the ratio exceeds 0.25wt%. The sensitivity of a capacitive-type sensor is defined as \(S=(\varDelta C/{C}_{0})/\varDelta P\), where the \({C}_{0}\) is the initial capacitance before loading and \(\varDelta C\) is the change in capacitance with the change in pressure, \(\varDelta P\). It was found that, the CB-SEBS blend elastomer showed good capacitance sensitivity at the 0.25% mass proportion. In addition, the dielectric properties of the elastomer are further improved due to the presence of air in the internal pores.
To evaluate the stability of the device, we investigated the real-time response of the device by applying different pressures for cycles(Fig. 4g). The device exhibits stable and reproducible response signals under varied pressures. Owing to each elastomer is cut from a large piece of elastomer, it exhibited a high degree of consistency between the sensors. We applied varied pressures on three capacitive sensors. The pressure sensing performances are shown in Fig. 4h. It can be clearly seen that all of the sensing curves exhibit similar incremental changes.
In addition to the perception of pressure stimuli, the perception of proximity stimuli is also highly expected in the device. Proximity sensing capability is the capability to perceive an approaching object and thus provide a timely prediction to transfer the feedback signal. Distinguishing from pressure sensing, proximity sensing in our device can be attributed to the disturbance of the electric field. As shown in Fig. 4i, an approaching finger could disturb the electric field of the sensor, resulting in a reduction in capacitance. The proximity sensing properties are evaluated in Fig. 4j, which shows the real-time capacitance response to the approaching finger. The finger was brought closer to the sensor from a distance of 20cm. The capacitance variation increased when the distance less than 9cm, and increased prominently when the distance less than 5cm. These results indicate that our capacitive sensor can perceive pressure/proximity with discriminable signal, good sensibility, and excellent stability. We also evaluated four sensors’ proximity sensing properties in Fig. 4k. It can be clearly seen that all of the sensing curves exhibit similar incremental changes.
We assemble the capacitive sensors and piezoresistive sensors into a sensing array, which was mounted on five fingers and palm of the robot hand as shown in Fig. 6a. The four sensing units in the middle of the palm are capacitive sensing units, which are mainly to perceive pressure, proximity and direction. The other parts are piezoresistive sensing unites, which are mainly to detect pressure and object shape.
As shown in Fig. 5a, a hand approached a robot hand equipped with the sensor array from four directions and the responses of four capacitive sensing units are recorded. It was clearly found that the responses changed significantly with the direction, which indicate that the sensor array can help the robot hand detect the direction of approaching objects. When the robotic hand tries to catch an object, it will approach the object firstly, and then grasp it. To monitor the capacitance variation of the capacitance sensing unit in this process, we used a metal cup to approach the robotic hand from a distance of 10cm and recorded the capacitive response in real-time. As shown in Fig. 5b, the capacitance increased when the cup approached the robotic hand, while when the robotic hand grasped the cup, the capacitance decreased significantly because of the pressure stimuli. The test indicated that the variation of capacitance can reflect the process of the robotic hand approaching and grasping the object, whether it is grasping or approaching.
To examine the ability of shape recognition, we manipulated the robot hand equipped with sensor array to grasp diverse fruits and then compare the resistance and capacitance responses of the sensor array. Three kinds of fruit (Mango, banana and peer), which have obvious differences in shape, were used in the experiment of object recognition.
As shown in Fig. 6d, the output signal can clearly represent the loaded regions, indicating excellent shape recognition ability.
We also developed a neural network model to increase the sensing accuracy for the various shape of object. The neural network model was composed of a fully connected (FC) (18,10) layer, rectified linear unit (ReLU) function, FC (10,6) layer, ReLU function, FC (6,3) layer, and Softmax, as shown in Fig. 6e. We choose three kinds of fruits with great shape differences (banana, pear, mango) to carry out the test, and each kind of fruit was chosen with three sizes. The resistance and capacitance responses of the sensor array on the robot hand was used as the input signals, and the outputs were the predicted kind of the grasped fruit. Sixteen signals of the sensor array on the five fingers and palm of the robotic hand were defined as a set of data. Each fruit was grasped from different orientations to collect the data sets. 270 sets of data (90 sets for each kind of fruit) were used to train and optimize the model. Afterward, we used another test dataset (containing 180 sets) to verify the trained model. Figure 6f illustrates the test verification result of the test dataset by comparing the predicted results with the actual objects. The verification results indicate that the total classification accuracy reaches about 90% in recognizing the objects with different sizes and shapes.