Human motion recognition using SWCNT textile sensor and fuzzy inference system based smart wearable

Human motion recognition using SWCNT textile sensor and fuzzy inference system based smart wearable

Accepted Manuscript Title: Human motion recognition using SWCNT textile sensor and fuzzy inference system based smart wearable Authors: Chi Cuong Vu, ...

2MB Sizes 0 Downloads 47 Views

Accepted Manuscript Title: Human motion recognition using SWCNT textile sensor and fuzzy inference system based smart wearable Authors: Chi Cuong Vu, Jooyong Kim PII: DOI: Reference:

S0924-4247(18)30859-8 https://doi.org/10.1016/j.sna.2018.10.005 SNA 11051

To appear in:

Sensors and Actuators A

Received date: Revised date: Accepted date:

20-5-2018 20-9-2018 4-10-2018

Please cite this article as: Vu CC, Kim J, Human motion recognition using SWCNT textile sensor and fuzzy inference system based smart wearable, Sensors and amp; Actuators: A. Physical (2018), https://doi.org/10.1016/j.sna.2018.10.005 This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Human motion recognition using SWCNT textile sensor and fuzzy inference system based smart wearable

IP T

Chi Cuong Vu, and Jooyong Kim*

Republic of Korea

author. E-mail: [email protected] (J. Kim)

U

*Corresponding

A

M

EP

TE

 

- A way of recognizing human motion by combining textile stretch sensors in a realistic applications is proposed. - The method relies on combination of the SWCNT - PET/SP sensor and fuzzy inference system (FIS). - The fabricated sensors are extremely thin, lightweight, sensitive. - In addition, it has been revealed that this method is highly preferable for mass production mainly due to simple process as well as highly accurate with high comfort.

D



N

Highlights: 

SC R

Department of Organic Materials and Fiber Engineering, Soongsil University, Seoul 156-743,

CC

ABSTRACT:

A

Wearable technology through proper combination of e-textiles and wearable devices is one of the good ways to accomplish the functions required to monitor and recognize human motions in daily life. However, most of the current research was only done in just simple trials without any significant analytical algorithms. This study aims to introduce a complete combination of e-textile stretch sensor based on single-walled carbon nanotube (SWCNTs) and spandex fabric

(PET/SP) with fuzzy inference system (FIS) in order to recognize human motions in a realistic applications. The wearing experiments were conducted for different types of human motions in order to examine the feasibility of the FIS model embedded in the sensing system developed. In the research, the performance of the method developed would be evaluated both by

IP T

characteristics of the fabrication sensor and accuracy in the application.

SC R

Keywords: Textile sensor; SWCNT; Spandex fabric; FIS; Human motion monitoring

N

U

1. Introduction

A

By focusing on revealing the multi-dimensional aspects of human life, the way of moving and

M

interacting with the surrounding environment, the wearable technology can be prevailly applied in medical and healthcare [1-3], monitoring devices [4-7], safety instrument and planar

D

waveguides [8], capacitive energy storage [9], etc. Especially, smart clothes fabricated by

EP

recognition [10-13].

TE

embedding smart sensors in normal fabric would be a promising technology in human motion

Most of the operating mechanism of the sensors are based on the relationship between physical

CC

quantity or chemical quantity such as temperature, pressure, stretch, vibration, humidity, and

A

electrical properties such as resistance, electromagnetic, capacitance of constituent conductive materials. According to this principle, the popular approach of designing wearable sensors [1415,27-28] is to integrate electronic devices including temperature, stretch gage, proximity, accelerometer, and pulse-oximeter sensors into a small hard packet added on clothes, jewelry [1,4,16-17] or directly on the skin [18,19].

The research fabricated a textile stretch sensor based on single-walled carbon nanotubes (SWCNT) [20-22] and spandex fabric (PET/Spandex - PET/SP). This textile sensor was embedded in smart pants in order to monitor human movements. Those motions would be detected by analyzing a set of receiving data from the sensor. This data would continuously flow into the memory of microcontroller chip and processed in order to get important factors

IP T

like as input variables of the classification model.

However, the real world applications have to face high levels of uncertainties because the data

SC R

acquired is complex and highly variable. Besides, such factors are nonlinear and have mutual

interactions among them; so it is not easy to create an exact correlation between the input

U

variables and the final decision by using normal mathematical or statistical methods. So, a

N

fuzzy inference system (FIS) [23-25] is proposed as a signal processing method in order to

A

upgrade accuracy of the model. The parameters chosen for developing the FIS system are the

M

average of amplitude (AMP), the standard deviation of amplitude (STD), and the average cycle (CYC). The final decision of the human motions can be stored or transmitted to monitoring

D

devices [26]. This way will provide a realistic motion sensing wearable products without

TE

unnecessary heavy and uncomfortable electronic devices.

EP

As expected, the SWCNT - PET/SP textile sensor is extremely thin, lightweight, sensitive, and thus high-level flexibility. The fabricated sensor is no offence, no irritation or allergies to the

CC

skin. In addition, the production process is simple, easy to implement. The results of the experiment also indicated that the developed FIS can use as a smart simulator in order to

A

classify motions with robust and accurate recognition rate. So this method really is acceptable for mass production. 2. Experimental method 2.1. Fabrication of textile stretch sensor

This experimentation was used the PET/SP fabric with a polyethylene terephthalate/spandex ratio = 76/24, item 16043 A, 341 g/YD, 262 g/SQM (SNT Co. Ltd, Seoul, South Korea). The PET/SP fabric is inherently composed of conventional PET/SP multifilament yarns with high elasticity. The PET/SP fibers can be converted into carbon fibers via coating and surface

and treated by acid solution (laboratory grade, HNO3:H2SO4 = 3:1).

IP T

treatment. The SWCNT raw powder (KH Chemical Co. Ltd, Seoul, South Korea) was prepared

Summary of fabrication process steps is shown in Fig. 1. In order to fabricate the textile sensor,

SC R

carbon powder ink was applied by water-based single-walled carbon nanotube solution (SWCNT dispersed in H2O - SWCNT ink 12-15) with nanotubes at 1.0 ~ 1.3 nm diameter and

U

0.1 wt% concentration. The PET/SP fabric was prepared and immersed in SWCNT ink for 3 ~

N

5 seconds, and then, mixed with the stirring machine at 60 ~ 80 ℃, 1000 rpm, 24 hr for 1 – 2

A

mins and treated with ultra-sonification (2 hr, 19.990 Hz). The impregnating process would

M

keep the conditions that allow the SWCNT particles to penetrate well (pressure roll speed: 1.0 m/min, air cylinder pressure: 3 bar (0.3 MPa) over). This process would make the SWCNT

D

particles adhere the stretchable fabric after dipping and squeezing. And then, the two-way

TE

drying machine was used in order to get rid the excess water in the stretchable fabric. The drying conditions were optimized at the time of drying: 1 – 3 mins, the range of temperature:

EP

180 – 200 0C, and the speed of circulation fan: 1500 rpm. Finally, it was maintained for 3 ~ 5

CC

hours under normal room temperature condition. The textile fabric sensor was cut to form smaller specimens for experimental. The ‘U’ line shape has been known to be better in

A

sensitivity and resolution compared to the single line design especially under small deformation such as human muscle movement. Sensor area was laminated by thermal pressing with a stretchable thin film on both sides in order to cover sensor surface and protect human skin from SWCNTs. The analysis of the muscle activity reveals that the largest stretch occurs in the sagittal axis direction. According

to this, the stretch sensor was designed at the middle of thigh. In this position, the deformation of the sensors and the output signals are maximized. Two ends of sensor were fixed on the pants by thread. And then, two conductive snap buttons were used to connecting processing circuit.

IP T

As shown in Fig. 3, the method of recognizing specific motion is strongly based on the relationship between mechanical and electrical properties of the fabric sensors. The resistance would change according to stretching or releasing by responsive mechanism of crack

SC R

propagation. The cracks originate and propagate in the thin conductive layers coated on the

PET/SP fibers during continuous mechanical stretching. And then, they would recover to their

U

initial states after releasing the stretch force imposed on the sensors. Edges of the cracks would

N

reconnect at this point, ensuring complete recovery of the electrical resistance.

A

2.2. Signal processing method

M

Fig. 4 showed block diagram of system and flowchart of signal processing. The whole data

D

acquisition system uses A/D conversion channel with 10-bit A/D converter with internal. For

TE

resolution reason, mathematical mapping of voltage values between 0 to 3.7 volts into digital values between 0 to 1023 (3.7/1023 = 0.0037 V or 3.7 mV per unit) has been made by pre-

EP

calculating the actual data. It is calculated to take about 0.01 s (10 ms) to read a signal input, and maximum reading is about 100 times per second. These data are written into EEPROM

CC

memory (512 bytes) of the microcontroller chip (ATtiny85) and processed every 2 minutes in

A

order to analyze important characteristics. Three parameters were suggested in order to create inputs for the FIS module such as the average of amplitude (AMP), the standard deviation of amplitude (STD), and the average cycle (CYC).

The AMP is the average of the magnitude of all instantaneous values in the duration of the signal. Considering a signal, A1, A2, A3, etc. are magnitudes of the signal at instants of 1, 2, 3, etc., respectively. The AMP for this duration can be determined as follows:

AMP =

𝐴1 + 𝐴2 + 𝐴3 + ⋯ + 𝐴𝑛 n

(1)

IP T

The AMP does not show all aspects of a signal. The signal can be very uniform with the data all bunched around the average or spread out a long way from the average. So, the STD is a

∑|𝐴 − 𝐴̅|2 𝑛−1

(2)

U

STD = √

SC R

measure of how the numbers are spread out and it can be calculated as follows:

N

where A represents an individual value, 𝐴̅ represents the mean value, and n represents the total

A

number of values.

M

The CYC is the shortest period in which action is repeated. Average cycle included process time, during which a unit is acted upon to bring it closer to an output, and delay time, during

D

which a unit of work is spent waiting to take the next motion. This is the most important value

TE

for analyzing data in order to give a final decision. As shown in Fig. 5, each cycle motion was

EP

calculated through a threshold. And the average cycle (CYC) could be calculated following

CC

Equation (3):

CYC =

∑𝑛 𝑖=1 𝐶𝑖 𝑛

(3)

A

where 𝐶𝑖 represents a cycle time and n represents the total number of cycles. 2.3.Fuzzy inference system model Based on natural language with uncertain or incomplete data, fuzzy inference system (FIS) [2324] reflects human reasoning in order to obtain a better classification in real usage environment.

The FIS is a form of many-values logic in which the true values of variables may be any real number between 0 and 1. It is used to handle the concept of partial truth, where the truth value may range between completely true and completely false. The main advantages and steps of FIS are shown below.

IP T

According to the proposed of this paper, the inputs are pre-determined, based on practical considerations of available data and motions that can be sensed. When creating the inputs and

states, the MFs are created by using the trapezoidal distribution and defined by intuitive

SC R

parameters. GUI for the MFs [23] creation is shown in Fig. 6. This FIS system is consisted (MF of input, output and rules [23]):

U

• Three membership functions as input variables (feature space representation as MF0, MF1

N

and MF2). Here MF0 is represented the average of amplitude (AMP), MF1 is represented the

A

standard deviation of amplitude (STD) and MF2 is represented the average cycle of the original

M

signal (CYC).

D

• A membership function as output variable. It is represented the decision.

TE

• A set of rules are evaluated in parallel by using fuzzy reasoning.

EP

All MFs were built on the trapezoidal (trapmf) distribution curves. It has a flat top and really is just a truncated triangle curve. The trapezoidal curve can be a function of vector, x, and

A

CC

depend on four scalar parameters a, b, c and d as given by:

y = trapmf(x, [𝑎 𝑏 𝑐 𝑑 ]) 0, 𝑥−𝑎 , 𝑏−𝑎 1, f(x; a, b, c, d) = 𝑑−𝑥 , 𝑑−𝑐 { 0,

𝑥 ≤𝑎 𝑎 ≤𝑥 ≤𝑏 𝑏 ≤𝑥 ≤𝑐

(4)

𝑐 ≤𝑥 ≤𝑑 𝑑 ≤𝑥

}

The parameters a and d are located at the “feet” of the trapezoid and the parameters b and c are located at the “shoulders”. Fig. 7, 8 showed membership functions of input variables and output variable. In order to simplify processing of the implemented rules, present fuzzy set categories were defined in form letters (L1, L2, etc.). Based on the results of experiments, the scalar parameters were chosen and shown in Table 1. For instance, according to the ranges and codes

IP T

given (Table 1), an experimental set of "AMP = [0 0 25 45], STD = [0 0 10 30], and CYC = [0 0 1 4]" was coded as "L1", respectively.

SC R

Fuzzy rules are the convenient way to represent knowledge by through a series of IF-THEN statements. The results of the rules are a combined model of the MFs with the historical

U

understanding of the study. Based on both developed fuzzy set categories and ranges of the

N

existing measured data, a total of 22 rules were established in the IF-THEN format for best-fit

A

model structure. Table 2 shows rule sets of the developed system. For instance, IF "AMP is

M

L1, STD is L1, and CYC is L8" THEN "Decision is Standing", respectively. Fuzzy rules define fuzzy patches, which is also the key idea in fuzzy logic. This paper used Mamdani type [23] to

TE

D

deal data in order to recognize the motions. 3. Results and Discussion

EP

3.1. Textile sensor structure

CC

Scanning electron microscopy (SEM) was used to show the structure of the fabricated sensor. Fig. 9 showed SEM image of the standard PET/SP fabric with the magnified view showing no

A

coating on the fiber and the coated PET/SP fabric with SWCNT, respectively. The filaments are an about 10µm diameter, loosely twisted and ample of free space between the microfiber bundles. The electric particles can be observed in the form of the thin coatings and stuck randomly onto PET/SP fiber surface with 80% coated rate. Following that structure, the PET/SP textile sensor will work based on the change in electrical resistance in sensor layer

along with stretch. That means the sensor resistance will increase while stretching (disconnection between the particles) and decrease while releasing (reconnection between the particles). 3.2. Stretchability and sensitivity (Gauge factor)

IP T

Tensile tests were performed to evaluate the mechanical behaviour of the sensors, and the typical strain-stress curves are presented in Fig. 10a. That curves showed linearity at strain of

SC R

< 40% with Young's modulus of 0.0003 MPa. The relatively Young's modulus indicates that the stretching of this sensor is sensitive to stress, which is an advantage for the fabrication of

the applications. The stretch-ability of a material will depend on the micro/nanostructures and

U

the fabrication process. As show on Fig. 10b, the structure of PET/SP fabric is the main reason

N

for the high stretchable (ε > 50 %). If the tension is too high (ε > 50 %), this structure will be

A

destroyed. Following that, the stretch-ability performed the best working range of this sensor

M

(ε = 0 % - 30 %) would be enough for realistic applications. The ratio of a relative change in resistance (∆R/R) and stretch (ε) is defined as the sensitivity or gauge factor (GF). It is clear

D

that the resistance increases along with the stretch increases, and vice versa. For the stretch

TE

textile sensor, the GF depends mainly on the nanostructures of SWCNTs. As calculated in

EP

Equation (5), the value of the GF ranges from 3.5 to 8.5 (Fig. 10c). This value is sensitive and

∆R ∆R R GF = = R △L ε L

A

CC

fits applications in the experimental.

(5)

ε = stretch(%)

3.3. Current – Voltage curves (I - V) The I – V curve is an important characteristic for usability of any sensor. Fig. 10d showed graphical curves which are used to define the operation of the developed sensor under different

static stretchs from 0 – 40 % within the experimental. The slope of I – V curves reduces with an increase of the tension, at 0 % – 12.5 % – 25 % – 37.5 %, indicating that an increase in applied stretch leaded to an increase in the sensor’s resistance. 3.4. Hysteresis

IP T

The hysteresis is defined as a behavior whose output does not only depend on the current input but also on the history of the input. Typical causes for the hysteresis are friction and structural

SC R

changes in the fabric [29]. The hysteresis increases with the decrease of the fabric density,

which determines the number of contact points within a given length of fabric. This characteristic becomes important when the stretch textile sensor is used in the dynamic

U

applications such as human motion monitoring, ECG monitoring, etc. The hysteresis behaviors

N

of 3 samples are shown in Fig. 11a, indicating a linear rise in resistance when applying stretch

A

and only a small hysteresis. A different resistance of the sensor appeared when the stretch was

M

applied and released, which correspond to the same input stimulus. This hysteresis error may lead to the inaccuracy in applications. In order to weaken the hysteresis, we suggest the coating

EP

3.5. Durability

TE

sensor with a low hysteresis.

D

silicon layer in the sensor fabrication [30]. This process can fabric a composited silicon-textile

CC

The dynamic durability is the stable electrical functionality and mechanical integrity of the stretch sensor in the stretching/releasing cycles. This characteristic depends on the fatigue and

A

plastic deformation of PET/SP fibers under high stretch which causes damage to fibers (PET) and the sensing nanomaterials (SWCNTs). In order to prepare for experimental, the textile stretch sensor was carefully stitched on the inner fabrics of the muscle pants. The durability performed under Lab customized UTM (Fig. 11b) clearly revealed that the durability was enough for realistic applications. Resulting fabric surface was kept intact after 30,000 abrasion

cycles. Fig. 11b showed the resistance of the sensor in the tension test (30 %). The resistance was measured every 5,000 cycles, and all results of each sample were uniform. All samples showed resistance changes of less than 10 % after 30,000 cycles of 30 % tension. For washability, the sensor area was laminated by thermal pressing with a stretchable thin film on both

thin film with 100% polyurethane (PU) and thickness is 20 µm. 3.6. Response and recovery time

IP T

sides in order to prevent SWCNT fall-off. The material used for laminating sensors is elastic

SC R

Response time (RST) and recovery time (RCT) are two of the important parameters for

evaluating performance in the dynamic application. Fig. 11c showed the response time of 200

U

ms and recovery time of 220 ms at ε = 30 %. The RST exists in the sensors is mainly affected

N

by the viscoelastic nature of PET/SP fabric. The RCT is affected by the friction force and the

A

reconnected ability between the SWCNT coatings with PET/SP fibers. Self-recovery process

M

of SWCNTs ensures recovery of the electrical property of the stretch sensor and avoids the degradation of the device performance during the large deformation. The high RST causes the

D

bias in the final results. To address this limitation, we propose the vacuum drying in the sensor

TE

fabrication. This process can create a strong connection between SWCNTs and PET/SP fibers.

EP

Reducing the response time and the hysteresis error are also the future directions of the project. 3.7. Human motion recognition

CC

The application capability of the textile stretch sensor would be evaluated by the experimental on eight human motions such as sitting, standing, walking, running, limping, sprinting,

A

jumping, and climbing. The result of the experimental is shown in the comparison between the output of the system and the actual motion through accuracy. The accuracy is determined by Equation (6) where N here represents the total number of motions and P represents the correct classified values.

A=

P N

x 100 %

(6)

During the experiment, the participant was asked to wear the smart muscle pants while moving in the straight corridor, and then to climb flights of a stairway. Fig. 12 showed one sample data shape of different motions in 2 minutes of the experimental. The developed model was

IP T

evaluated through 800 motion samples (100 samples for each motion) to verify the robustness of the system.

SC R

Fig. 13 showed final recognition process of the fabrication system. Data acquisition of motions will be analyzed to get three important parameters (AMP, STD, and CYC). Based on that, the FIS system uses set of fuzzy rules and database to make a final decision of the motion. Fig. 14

U

showed the class-based performance for eight motion types with the number of testing. This

N

proposed model obtained a mean performance accuracy of 100 % while sitting and standing,

A

86.3 % while limping, 87.5 % while jumping, 88.8 % while climbing, 82.5 % while walking,

M

83.8 % while running, and 82.5 % while sprinting, respectively. Average value of the accuracy

D

(A > 80 %) showed that there is an agreement between the measured and classified values.

TE

Accordingly, it is clear that the sitting and standing motions are easy to classification (A = 100 %). However, the walking, running and sprinting motions are easy to confusion (A = 82 % -

EP

83 %). Especially, the precision of the running and sprinting motions are lowest (A = 82.5 %). The accuracy of the limping, jumping and climbing motions are acceptable (A > 85 %). In

CC

brief, through the statistical indicator, the accuracy was demonstrated that this system can be

A

applied as an intelligent device for recognizing human motions in a real application. 4. Conclusions This study developed a complete combination of the wearable application based on SWCNT PET/SP textile sensor and fuzzy inference system to analysis sensing signals on a real product. The first promising highlight of the research is the fabrication process of the stretch sensor is

simple, stretchable and flexible. And the second, this research has improved the accuracy of the human motion recognition when developing classification model from three characteristic parameters (AMP - STD - CYC). These also are unique to this system. This paper also showed an ability to bring the product from experimental to daily life with a high economic efficiency.

IP T

Acknowledgements This study is sponsored by Soongsil University Research Fund.

SC R

References

U

[1] S. Majumder, T. Mondal, M. J. Deen, Wearable Sensors for Remote Health Monitoring,

N

Sensors 17 (2017) 130-175.

A

[2] V. Kaushik, J. Lee, J. Hong, S. E. Lee, S. A. Lee, J. Seo, C. Mahata, T. Lee, Textile-

M

Based Electronic Components for Energy Applications: Principles, Problems, and

D

Perspective, Nanomaterials 5 (2015) 1493-1531.

TE

[3] D. Son, J. Lee, S. Qiao, R. Ghaffari, J. Kim, J. E. Lee, C. Song, S. J. Kim, D. J. Lee, S. W. Jun, S. Yang, M. Park, J. Shin, K. Do, M. Lee, K. Kang, C. S. Hwang, N. Lu, T.

EP

Hyeon, D. H. Kim, Multifunctional wearable devices for diagnosis and therapy of

CC

movement disorders, Nat. Nanotechnol. 9 (2014) 397–404.

A

[4] A. Servati, L. Zou, Z. J. Wang, F. Ko, P. Servati, Novel Flexible Wearable Sensor Materials and Signal Processing for Vital Sign and Human Activity Monitoring, Sensors 17 (2017) 1622-1642.

[5] Y. Enokibori, K. Mase, Human joint angle estimation with an e-textile sensor, IEEE 18th ISWC 9 (2014) 129-130.

[6] C. Kallmayer, E. Simon, Large area sensor integration in textiles, Int. Multi-Conf. on Systems, Signals and Devices 9 (2012) 1-5. [7] N. H. I. Lopez, M. A. Munoz, Wearable Inertial Sensors for Human Motion Analysis: A Review IEEE Sens. J. 16 (2016) 7821-7834.

IP T

[8] B. J. Munro, T. E. Campbell, G. G. Wallace, J. R. Steele, The intelligent knee sleeve: A wearable biofeedback device, Sens. Actuator B-Chem. 131 (2008) 541-547.

SC R

[9] D. Yu, K. Goh, H. Wang, L. Wei, W. Jiang, Q. Zhang, L. Dai, Y. Chen, Scalable synthesis

of hierarchically structured carbon nanotube–graphene fibres for capacitive energy

U

storage, Nat. Nanotechnol. 9 (2014) 555–562.

N

[10] C. Mao, H. Zhang, Z. Lu, Flexible and wearable electronic silk fabrics for human

A

physiological monitoring, Smart Mater. Struct. 26 (2017) 095033-095042.

M

[11] Y. Wang, L. Wang, T. Yang, X. Li, X. Zang, M. Zhu, K.Wang, D. Wu, H. Zhu, Wearable and Highly Sensitive Graphene Strain Sensors for Human Motion Monitoring, Adv.

TE

D

Funct. Mater. 24 (2014) 4666–4670. [12] Y. Jiao, C. W. Young, S. Yang, S. Oren, H. Ceylan, S. Kim, K. Gopalakrishnan, P. C.

EP

Taylor, L. Dong, Wearable Graphene Sensors With Microfluidic Liquid Metal Wiring for Structural Health Monitoring and Human Body Motion Sensing, IEEE Sens. J. 16

CC

(2016) 7870-7875.

A

[13] D. O. Lara, A. M. Labrador, A Survey on Human Activity Recognition using Wearable Sensors, IEEE Commun. Surv. Tutor. 15 (2012) 1192-1209.

[14] M. Amjadi, K. Kyung, I. Park, M. Sitti, Stretchable, Skin‐ Mountable, and Wearable Strain Sensors and Their Potential Applications: A Review, Adv. Funct. Mater. 26 (2016) 1678–1698.

[15] F. M. Guo, X. Cui, K. L. Wang, J. Q. Wei, Stretchable and compressible strain sensors based on carbon nanotube meshes, Nanoscale 8 (2016) 19352–19358. [16] M. Stoppa, A. Chiolerio, Wearable Electronics and Smart Textiles: A Critical Review, Sensors 14 (2014) 11957–11992.

IP T

[17] M. C. Lina, B. F. Alison, Smart fabric sensors and e-textile technologies: a review, Smart Mater. Struct. 23 (2014) 053001-053028.

SC R

[18] T. Someya, Z. Bao, G. G. Malliaras, The rise of plastic bioelectronics, Nature 540 (2016) 379–385.

U

[19] X. Wang, Y. Gu, Z. Xiong, Z. Cui, T. Zhang, Silk-Molded Flexible, Ultrasensitive, and

N

Highly Stable Electronic Skin for Monitoring Human Physiological Signals, Adv.

A

Mater. 26 (2014) 1336–1342.

M

[20] B. Bhushan, Springer Handbook of Nanotechnology, third ed., Springer, Berlin, 2010.

D

[21] X. Guo, Y. Huang, Y. Zhao, L. Mao, L. Gao, W. Pan, Y. Zhang, P. Liu, Highly

TE

stretchable strain sensor based on SWCNTs/CB synergistic conductive network for wearable human-activity monitoring and recognition, Smart Mater. Struct. 26 (2017)

EP

095017-095026.

CC

[22] E. Roh, B. -U. Hwang, D. Kim, B. -Y. Kim, N. -E. Lee, Stretchable, Transparent, Ultrasensitive, and Patchable Strain Sensor for Human–Machine Interfaces Comprising

A

a Nanohybrid of Carbon Nanotubes and Conductive Elastomers, ACS Nano 9 (2015) 6252-6261.

[23] J. -S. R. Jang, N. Gulley, Fuzzy Logic Toolbox User's Guide, first ed., The MathWorks, Natick, 1995.

[24] C. H. Lim, E. Vats, C. S. Chan, Fuzzy human motion analysis: A review, Pattern Recognit. 48 (2015) 1773-1796. [25] B. Bruno, F. Mastrogiovanni, A. Saffiotti, A. Sgorbissa, Using Fuzzy Logic to Enhance Classification of Human Motion Primitives, in: A. Laurent, O. Strauss, B. Bouchon-

in Knowledge-Based Systems, Springer, Berlin, 2014, pp 596-605.

IP T

Meunier, R. R. Yager (Eds.), Information Processing and Management of Uncertainty

SC R

[26] K. V. S. S. S. S. Sairam, N. Gunasekaran, S. R. Redd, Bluetooth in wireless communication, IEEE Commun. Mag. 40 (2002) 90-96.

[27] J. Zhong, Q. Zhong, Q. Hu, N. Wu, W. Li, B. Wang, B. Hu, J. Zhou, Stretchable Self‐

N

U

Powered Fiber‐ Based Strain Sensor, Adv. Funct. Mater. 25 (2015) 1798–1803.

A

[28] Q. Zhong, J. Zhong, X. Cheng, X. Yao, B. Wang, W. Li, N. Wu, K. Liu, B. Hu, J. Zhou,

M

Paper‐ Based Active Tactile Sensor Array, Adv. Mater. 27 (2015) 7130–7136. [29] J. Fraden, Handbook of Modern Sensors - Physics, Designs, and Applications, fourth

TE

D

ed., Springer, Berlin, 2010.

[30] A. Atalay, V. Sanchez, O. Atalay, D. M. Vogt, F. Haufe, R. J. Wood, C. J. Walsh, Batch

EP

Fabrication of Customizable Silicone‐ Textile Composite Capacitive Strain Sensors for

A

CC

Human Motion Tracking, Adv. Mater. Technol. 2 (2017) 1700136.

Author Biography:

Chi Cuong Vu received his B.S. in Electronic &

program

in

Organic

IP T

Telecommunication in 2014. He finished M.S. Materials

and

Fiber

SC R

Engineering at Soongsil University, Seoul, Korea.

He is currently a Ph.D. student in Organic Materials

U

and Fiber Engineering at Fashionoid R&D Center,

N

Soongsil University, Seoul, Korea. His research interests include flexible-stretchable electronic

A

materials and their application in wearable human-activity monitoring and personal

Jooyong Kim received the B.S. and M.S. degree in fiber and polymer science & engineering from Seoul National University, Seoul, Korea, in 1990 and 1992, respectively. He is currently working as a professor at department of organic materials and fiber engineering of Soongsil University, Seoul,

A

CC

EP

TE

D

M

healthcare.

Korea since 1999 after Ph.D. at North Carolina State University, Raleigh, USA. From 1998 to 1999, he was a post-doctoral researcher at department of mechanical & aerospace engineering of UCLA, California, USA. His research interest includes the development of smart fashion products based on electronic textiles.

IP T SC R U N A

A

CC

EP

TE

D

M

Fig. 1. Summary of fabrication process steps

SC R

IP T

Fig. 2. Shape parameters of the sensors

A

CC

EP

TE

D

M

A

N

U

Fig. 3. The working mechanism of the sensor

(b)

Fig. 4. Block diagram of system (a) and flowchart of signal processing

IP T SC R

A

CC

EP

TE

D

M

A

N

U

Fig. 5. The parameters of the signal

Fig. 6. The advantages (a) and the steps of FIS (b)

IP T SC R U N A M D TE EP CC A

Fig. 7. Membership functions of (a) average amplitude, (b) standard deviation, and (c) average cycle

IP T SC R

A

CC

EP

TE

D

M

A

N

U

Fig. 8. Membership functions of decision

Fig. 9. Surfaces of the fabricated sensors under different magnifications: (a) untreated 100µm, (b) treated 100µm, (c) untreated 15µm, and (d) treated 15µm

IP T SC R U N A M D TE

EP

Fig. 10. The electro-machenical properties of the sensor (I): (a) typical tensile stress-strain

A

CC

curve, (b) resistance–stretch characteristics, (c) gauge factor, and (d) current–voltage curves

IP T SC R U N A M D

TE

Fig. 11. The electro-mechanical properties of the sensor (II): (a) hysteresis, (b) results of

A

CC

EP

30,000 cycles of dynamic tension test (30%), and (c) response and recovery time

IP T SC R U N A M D

A

CC

EP

TE

Fig. 12. Data shape in one sample of each motion

IP T SC R U N A M

A

CC

EP

TE

D

Fig. 13. Final recognition process of the system

IP T SC R U

A

CC

EP

TE

D

M

A

N

Fig. 14. Accuracy of human motion recognition based on the FIS model

Table 1. Number of trapezoidal membership functions (trapmf) and their ranks for each of the input and output variables considered in the present fuzzy sets Output

Level of Input variables

variable

membership STD

CYC

L1

[0 0 25 45]

[0 0 10 30]

[0 0 1 4]

L2

[30 70 110 150]

[20 50 70 100]

[2 7 14 18]

[130 180 230

[80 110 130

280]

160]

[260 310 370

[140 170 180

420]

210]

[400 430 470

[200 240 260

500]

300]

29]

TE

[550 600 640

EP

690]

[660 710 720

[33 37 43 47]

U N

D

570]

M

[480 520 540 L6

[27 30 34 37]

A

L5

[280 310 340

[35 39 43 [43 47 52 57] 47] [45 50 55 [53 57 63 67]

370]

65]

[360 385 395

[57 65 77 [63 67 73 77]

420]

85]

[400 440 550

[80 90 200

L8

770]

CC

[13 17 23 27]

[23 27 33 37]

L4

[73 77 83 87] 550]

200]

[730 780 1023

[83 90 100 —

L9

A

[0 0 10 17]

[16 20 25

L3

L7

Decision

IP T

AMP

SC R

functions (trapmf)

1023]

— 100]

Table 2. The fuzzy rule sets Input variables

Number of

Output variable

AMP

STD

CYC

Decision

1

L9

L2

L8

Sitting

2

L8

L2

L8

Sitting

3

L2

L4

L6

Walking

4

L2

L4

L7

Walking

5

L2

L4

L8

6

L2

L3

L6

7

L2

L3

L7

8

L2

L3

L5

9

L2

L4

10

L2

L3

11

L2

12

L4

13

L3

Walking Walking Walking Limping Limping

L3

Limping

L3

L4

Limping

L5

L2

Running

L4

L2

Running

L4

L7

Sprinting

L6

L4

L4

Sprinting

L6

L4

L3

Sprinting

17

L5

L8

L3

Jumping

18

L5

L6

L3

Jumping

19

L7

L4

L8

Climbing

20

L7

L3

L8

Climbing

21

L8

L4

L8

Climbing

15

A

16

M D

TE L7

EP

14

A

L5

CC

N

U

SC R

IP T

fuzzy rules

D

TE

EP

CC

A

IP T

SC R

U

N

A

M

22 L1 L1 L8 Standing