Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Design and Performance Evaluation of a Wearable Sensing System for Lower-Limb Exoskeleton

Design and Performance Evaluation of a Wearable Sensing System for Lower-Limb Exoskeleton Hindawi Applied Bionics and Biomechanics Volume 2018, Article ID 8610458, 9 pages https://doi.org/10.1155/2018/8610458 Research Article Design and Performance Evaluation of a Wearable Sensing System for Lower-Limb Exoskeleton 1,2 1,2 1 1,2 1,2 Chunfeng Yue, Xichuan Lin, Ximing Zhang, Jing Qiu , and Hong Cheng The School of Automation Engineering, University of Electronic Science and Technology of China, China Center for Robotics, University of Electronic Science and Technology of China, China Correspondence should be addressed to Hong Cheng; hcheng@uestc.edu.cn Received 12 February 2018; Revised 27 April 2018; Accepted 13 June 2018; Published 18 September 2018 Academic Editor: Dongming Gan Copyright © 2018 Chunfeng Yue et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Because the target users of the assistive-type lower extremity exoskeletons (ASLEEs) are those who suffer from lower limb disabilities, customized gait is adopted for the control of ASLEEs. However, the customized gait is unable to provide stable motion for variable terrain, for example, flat, uphill, downhill, and soft ground. The purpose of this paper is to realize gait detection and environment feature recognition for AIDER by developing a novel wearable sensing system. The wearable sensing system employs 7 force sensors as a sensing matrix to achieve high accuracy of ground reaction force detection. There is one more IMU sensor that is integrated into the structure to detect the angular velocity. By fusing force and angular velocity data, four typical terrain features can be recognized successfully, and the recognition rate can reach up to 93%. 1. Introduction and standing, but they also rebuild their confidence in daily life. There are three typical commercial products that have Lower limb exoskeleton (LLE) robotic technology has been been developed: ReWalk 6.0 [7], Ekso GT [8], and HAL-5 [9]. They each have a weight of about 20 kg. Besides these, developed rapidly for the past 20 years. Three implementa- lightweight ASLLEs have also been developed to assist spinal tion fields, that is, augmentation, rehabilitation, and living cord injury patients, such as Phoenix 3.0 [10] and INDEGO support are explored. First, the human augmentation type [11]. The researchers are also trying to prove their benefit LLE (AULLE) was developed for the military and aims at to patients who have used the ASLLEs [12, 13]. improving a soldier’s weight loading ability. The fields of In our previous research, we focused on the human aug- application extend to disaster relief and industrial transport mentation exoskeleton [14–16] and an ASLLE named the assistance. Kazerooni et al. developed the typical first gener- ation AULLE which is named BLEEX [1, 2]. Then, the third AIDER [17, 18] for individuals. An illustration of a prototype of AIDER with a SCI patient is shown in Figure 1. Based on generation AULLE which is named HULC was developed AIDER and two crutches, the patient can form a stable area for carrying a load of about 90 kg. Second, the rehabilitation to keep balance. Four DC motors are installed on the hip type LLE (RELLE) was developed for patients whose lower and knee joints respectively. The patient controls the AIDER limbs are inconvenienced [3–5]. Typical patients include those with foot drop, spinal cord injuries, and strokes. The by two controllers which are installed on the grab handles of the crutches. A battery and control circuit placed in the back- most famous RELLE is Lokomat, which is developed by the pack are the core part for controlling the DC motors. This Hocoma Company [6]. The assistive-type LLE (ASLLE) is paper is an extension for my previous work [19]. used to assist patients with lower limb disabilities but whose Avoiding harm to the patient when piloting an exoskele- upper limbs are normal. The ASLLEs assist patients to return to their normal life. They do not only assist in walking ton is a critical issue. Table 1 shows the influence of three kinds of LLEs on the interaction between human, machine, motion on flat ground, climbing on stairs, or sitting down 2 Applied Bionics and Biomechanics Company. Based on this product, the relationship between the gait characteristics and foot pressure for overweight chil- Pilot dren was analyzed [20]. Besides, researchers also designed a number of wearable sensing systems to meet their research requirements. In [21], the researchers designed sensing shoes Bag to estimate the CoM displacement continuously using an ambulatory measurement system which contains 2 IMUs and 2 6DOF force/moment sensors. Although the precision of the sensor is really high, the large size influences gait anal- Control button Hip joint ysis resulting from normal walking. Even high precision data can be collected easily. The large size of the IMU and pressure sensor gives the pilot poor wearing experience. Liu et al. have worked on gait analysis for serval years. They developed Knee joint smart shoes for human gait analysis [22, 23]. The novel point of the smart shoes is that they contain three 6-axis force sen- sors with 1 cm thickness. The sensors are mounted on the heel, arch, and forefoot respectively to subdivide the phase of human walking and get high accuracy gait data. Bamberg et al. developed a multiple sensing system and integrated it with shoes. The sensing system contains 6 kinds of sensors Crutch including an accelerometer, gyroscope, force sensitive resis- tor, bend sensor, polyvinylidene fluoride strip, and electric field sensor [24]. The multiple sensors provide redundancy Shoes gait data to ensure stability. Due to the complex environment as shown in Table 1, we intend to develop a wearable sensing Figure 1: The prototype of AIDER with a SCI patient. system for AIDER which can not only detect human body motion but also detect the state of the environment, such as the features of the ground. Table 1: The influence for three kinds of exoskeletons on the human-machine-environment. 2. Motivation Factors AULLE RELLE ASLLE 2.1. User and Application Environment for AIDER. AIDER is Human Health Weak Weak intended for users with SCI with injury levels from T9 to T12 Machine Low High High caused by traumatic injuries (e.g., vehicular crashing or fall- Environment Daily life/unsafe Hospital/safe Daily life/unsafe ing from buildings) or disease (e.g., myelitis) [25, 26]. The AIDER is aimed at extending the range of activities to advance their rehabilitation programs for SCIs. Besides, and the environment. For an AULLE, the mechanical parts walking upright makes the patients feel more confident follow the pilot’s motion and support the weight of the load. because they can make a conversation with friends at eye The pilot takes the initiative in the human-machine- level and walk like normal persons. environment system. For a RELLE, the machine controls As illustrated in the Introduction, the target environ- the human motion because of the lower-limb paralysis of ments for the application of AIDER are daily life and clinic SCI patients. The working environment of a RELLE is very rehabilitation. Compared with the clinical environment, the safe because the pilot is always protected by fixed support. daily life environment is more complex. Therefore, in this For an ASLLE, although the potential pilots are the same as research, we pay more attention in analyzing the main fea- those of a RELLE, the working environment for an ASLLE tures of the daily life environment, especially the ground. is complex daily life. The safety for the pilot of an ASLLE Generally, two main features of the ground influence the gait relies on the exoskeleton’s stability. Besides, patients with for normal walking, that is, hardness and terrain. Figure 2 complete injury have lost their sensory ability. They cannot shows the relationship between these two features and the keep their balance when wearing the exoskeleton. Therefore, common implementation environment. The typical mate- the ASLLE should be able to sense the human-machine state rials in daily life have two relative features: for example, the and adjust the gait trajectory for the pilot to avoid a danger- typical features of marble ground are flat and hard. ous situation. Based on the analysis for the safety of a human-machine 2.2. Gait Analysis and Environment Detection for the AIDER. system, a sensing system is necessary to improve the stability For healthy people, the gait is changed adaptively when they of ASLLEs. The shoes are suitable components for installing cross from one terrain to another. For instance, the gait will sensors. In related works, researchers have used foot- change when someone crosses from hard ground to sand. sensing systems to detect gait information. Footscan is a typ- However, the AIDER works on a customized gait to realize ical commercial product which was developed by the RSscan the walking motion for SCI patients [15]. It would cause a Applied Bionics and Biomechanics 3 Hardness Terrain Typical material (d) To avoid the force from exceeding the acceptable range, the force measurement range of the wearable Cobbled sensing system should be from 0 to at least 120 kg. road (e) To cut the cost, the hardware cost of the wearable Flat sensing system should be less than ¥2000. Marble ground (f) To provide enough data for control strategies, the Hard wearable sensing system should be designed to be Rugged able to detect and recognize ground features. Ramp (g) Finally, the wearable sensing system needs to realize Soft attitude measurement and gait analysis. Lawn Slope 3.2. Design of the Wearable Sensing System. As proposed in Carpet Section 3.1, 7 design requirements should be met. The mechanical design of the wearable sensing system is shown Figure 2: The common application environment for users of in Figure 3. In Figure 3(a), there are 3 layers that form the AIDER. The solid line and dotted line stand for hard and soft sensing part for force detection. The bottom layer is con- features, respectively; red, blue, and black denote flat, rugged, and structed of wear-resistant rubber which is used to ensure that slope features, respectively. the pilot’s foot does not slip. A hook and loop tape is used to fasten the foot. The middle layer is a holder for the force sen- sor. The total thickness of the 3 layers is 18 mm which can potential safety hazard because it cannot adapt to the change meet the conditions of requirement (a). Seven strain gauge of terrains adaptively in a social environment. force sensors are employed to sense the center of force in Gait analysis mainly focuses on two parameters which the z-axis. The top layer is used to install the 7 force sensors are ground reaction force (GRF) and body posture. These which are made of aluminum alloy. It is necessary to recog- two parameters can be utilized to confirm whether the sys- nize the force for the heel and forefoot [27, 28]. Therefore, tem state is suitable for the next motion. For the stability the top layer is made of two separated aluminum alloys. control of biped robots, findings from [27] indicated that, Three force sensors on the forefoot form a stable plane. The comparing with a hard ground, step height tends to other 4 force sensors form a trapezoid to keep stable. The increase for avoiding collision between a robot’s feet and accuracy of each force sensor is about 0.1%. All the seven soft ground. Besides, terrain features are also the essential force sensors together are capable of high accuracy measure- factors for gait adjustment. For AIDER, the control strat- ment. Because the range of a force sensor is about 25 kgf, the egy of stability not only depends on the system controller measured range of the wearable sensing system is about but also the environmental features. Typical environments 175 kgf. The IMU sensors and control circuit are used to col- for AIDER are shown in Figure 2. Therefore, we intend to lect attitude data which is installed in the circuit box as design a wearable sensing system for AIDER which can be shown in Figure 3(a). The connection rod is designed to link used to detect and recognize environmental features and the wearable sensing system and the shanks of AIDER. To CoP of the feet in this paper. meet requirement (e), the cost of the wearable sensing system is listed in Table 2. 3. Method and Materials 3.3. Force Measurement Experiment. To test the performance 3.1. Design Requirement of a Wearable Sensing System. of our force sensing system in terms of accuracy and dynamic Section 1 shows the benefit of a wearable sensing system for stability, a force measurement experiment was conducted. In gait analysis and the safety of a human-machine system. this experiment, we used the force platforms to verify and Therefore, a wearable sensing system for the feet is proposed calibrate the accuracy of the wearable sensing system. to realize gait analysis and environment detection. The Figure 4 describes the setup of the force measurement exper- following design requirements are proposed according to iment, where the pilot stands astraddle on two force plat- the features of a SCI patient: forms. During the experiment, the pilot shifts the support foot at the center of his body weight from the left to the right (a) To make the user comfortable, the thickness of sole and then shifts back again to the left. The motion frequency should be less than 20 mm. is about 2 seconds. Finally, we used a wireless module to translate the data of the wearable sensing system to a PC (b) To ensure convenience, people should be able to put for sensing system analysis. The experimental results are on the shoe using one hand. indicated in Figure 5. The red line indicates the output of (c) To ensure the accuracy for gait analysis, the magni- the force plates and the blue line indicates the output of tude of output force from the sensors should be the wearable sensing system. This figure shows that the data from the shoes follow the data from the force plate with high obtained. 4 Applied Bionics and Biomechanics Bandage Top layer criteria. CoP coincides with ZMP when the system is under a quasistatic state. Di et al. developed a cane robot to realize Connection human fall detection by estimating the CoP [29]. In our Circuit box research, CoP is also involved in AIDER for stability estima- tion of the human-machine system. To estimate the CoP of Top layer the human-machine system, the first step is to calculate the Middle layer ground reaction force (GRF) and the CoP of the feet. Accord- ing to [30], the CoP can be estimated by Bottom layer (a) x ⋅ Fx dx = , CoP Middle layer Fx dx y ⋅ Fy dy Y = CoP Fy dy Based on the mechanical design of the force sensory system, after being dispersed (1), the CoP of the foot is obtained by (b) ∑x ⋅ f i ni X = , CoP ∑f ni ∑y ⋅ f i ni Y = , CoP ∑f ni where P x , y ,(i =1,2,… ,7) denotes the coordinate for i i i each force sensor. f denotes the force that is obtained ni by each force sensor. n is the mark for recognizing the left and right foot. Based on mechanical design, the coordinate of each sensor can be obtained by Figure 3(c). A verification experiment is designed to prove the perfor- mance of a wearable sensing system for CoP detection. The experimental setup is similar to that in Figure 4. The differ- ence is that the pilot is walking in a daily life state but on a (c) force platform. Figure 6 showed the experimental results from the point when the heel touches the ground up to the Figure 3: The mechanical structure of the wearable sensing system point when the toe lifts from the ground. In this experiment, for AIDER. (a) The structure of the wearable sensing system. (b) The prototype of the wearable sensing system. (c) The coordinate the trajectory transforms from the heel to the big toe as system for the left foot. shown in Figure 6(a). Figure 6(b) shows the magnitude of the total force in the z-axis. The trajectory of CoP in the X- Y plane is shown in Figure 6(c). Based on [31], the trajectory of CoP agrees with human habit because of a similar curve. Table 2: The cost of wearable sensing system. Name Unit cost Unit quantity Total price/¥ 4. Ground Characteristic Analysis IMU 25 piece 2 50 and Recognition Force sensor 15 piece 14 210 4.1. Ground Characteristic Analysis. Based on Section 2, the Mechanical parts 500 Set 2 1000 main features of the application environment contain hard- Circuit board 100 Piece 4 400 ness and terrain. More specifically, soft/hard and flat/slope Hook and loop tape 10 piece 2 20 are two pairs of critical factors for a control strategy. After Sum 1680 considering the environment in daily life, carpet, ramp, and marble ground are selected as the recognized subjects. For the ramp, uphill and downhill is the difference. For a flat ground, soft and hard is the main difference feature. There- precision. With a shaking motion, the wearable system fore, the main purpose of the wearable sensing system is to detected the body shaking accurately. recognize the following combined features which are flat/ 3.4. Center of Pressure (CoP). For biped locomotion control, hard (F/H), flat/soft (F/S), uphill/hard (U/H), and down- ZMP (zero moment point) and CoP are two important hill/hard (D/H). Applied Bionics and Biomechanics 5 Pilot Wearable sensing sytem data Force platform Wearable sensing sytem Force platform data Figure 4: The setup of the stability test for a force measurement experiment. 4.2. Principal Component Analysis (PCA) for the Four Ground Feature Extraction. PCA is a data analysis method −200 that uses an orthogonal transformation to obtain principal −400 components which are used to present the original data fea- −600 ture by a low-dimensional variable. As a popular pattern rec- 01 23 45 6 ognition method, PCA is widely used in face recognition Time (1 휇s/Div) ×10 [33]. This method is aimed at reducing the dimension for Platform output the eigenvector. In our research, the dimension of the eigen- Shoes output vector for an environmental feature extraction is 30 which contains the sum, mean, and variance of the 7 normalized Figure 5: Results for the accuracy verification of a wearable sensing force sensor output and 3-axis motion acceleration. After system. analyzing the data by PCA, the variance that explains princi- pal components are obtained. According to the result of Figures 8–11, the variance that To recognize these features, the data from the IMU and explained the first three principal components is more than force sensor are necessary. Generally, a walking motion can 85%. Therefore, the first three principal components are be divided into 8 phases, that is, initial contact (IC), loading enough to distinguish the four ground features. Figures 8 response (LR), midstance (MS), preswing (PS), initial swing and 9 particularly show that the uphill and downhill motions (IS), midswing (MS), and terminal swing (TS) [24]. For a are easy to describe using the first 2 principal components. In ramp, 3 force sensors are used in [32] to recognize the slope Figures 9 and 10, the variance explained on the third princi- by adjusting the sequence of the force sensor output. The IC pal component is more than 10%. The results indicate that and LR phases contain impact information caused by the the soft and hard features are relatively complex. Finally, hardness of the ground. Therefore, the sensor data from the ground features are described by the first three and two the IC and LR phases are collected by the data window for principal components in Figures 12 and 13, respectively. feature recognition. Due to the four ground features, it is easy to classify the A total of 7 force sensors and one IMU is used to sense minimum-distance classifier [34] which is employed to clas- the ground features. The force of each sensor is f ni sify the four features. (i =1,2,… ,7). The output of the IMU sensor is angular velocity ω = ω ω ω and acceleration a = x y z 4.3. Experiments. To verify the performance of the recogni- a a a . To get a credible result, the gravitational x y z tion method, 5 experimental subjects wore the wearable sens- acceleration is removed and the resultant force of the 7 sen- ing system and walked on carpet, ramp (uphill and sors is normalized. To keep the data in the same magnitude, downhill), and flat ground surface in a normal gait, and about the force is multiplied with a scale factor. The drastic vibra- 2000 steps were obtained. After preprocessing, half of data tion makes the IC and LR phases easy to detect and the force are used as training data. The left part is used for testing output also increases. Therefore, according to the output of the training model, and the experimental results are obtained the force sensors, the data window for feature recognition is as shown in Figure 14. The black circles indicate the points obtained as shown in Figure 7. that are not classified in features on the right. Force (N) X (mm) 6 Applied Bionics and Biomechanics −50 20 0 −100 −50 −150 −50 0 −200 50 −250 −100 (a) −150 −200 −250 −50 050 −250 −200 −150 −100 −50 0 X (mm) Y (mm) (b) (c) Figure 6: The CoP for the left foot; the black circle denotes the pressure-bearing point. (a) The trajectory of the CoP. (b) The magnitude of the total force during contact of the left foot to the ground. (c) The trajectory of CoP when the heel touches the ground up to point when the toe lifts from the ground. Data window Downhill/hard 100 100 90 90 80 80 70 70 60 60 50 50 40 40 −20 20 20 −40 10 10 0 0 −60 12 3 Principal component 1.185 1.19 1.195 1.2 1.205 1.21 1.215 1.22 Time (0.004 ms/Div) ×10 Figure 8: The relationship between the principal component and the variance explained for D/H. Data window AccZ AccX Force 20 AccY Finally, the recognition rate for 4 ground features is listed Figure 7: Data window for the feature extraction. The force in Table 3. The recognition rate of hard/flat ground and multiplied by 20 is the aim for analyzing in the same magnitude. downhill/hard ground is more than 95%. This result shows that the eigenvector which is extracted by PCA and the minimum-distance classifier is suitable for the ground fea- ture recognition. Y (mm) Force (N/9.8) Force (N/9.8) Variance explained (%) Y (mm) (%) Component 2 Applied Bionics and Biomechanics 7 Uphill/hard Soft/flat 90 90 90 90 80 80 80 80 70 70 70 70 60 60 60 60 50 50 50 50 40 40 40 40 30 30 30 20 20 20 10 10 10 0 0 0 0 1 12345 Principal component Principal component Figure 9: The relationship between the principal component and Figure 11: The relationship between the principal component and the variance explained for U/H. the variance explained for S/F. Hard/flat 100 100 70 70 0.5 60 60 50 50 −0.5 40 40 −1 −1.5 30 30 1 20 2 −1 −2 −3 −1 0 0 −4 Blue: downhill/hard Principal component Red: uphill/hard Green: hard/flat Figure 10: The relationship between the principal component and Black: soft/flat the variance explained for H/F. Figure 12: The four ground features are described by the first three principal components. 5. Discussion A mechanical design has been proposed in Section 3.2 that terrains are involved in this paper. However, the daily life meets design requirements (a), (b), (d), and (e). To meet environment is more complex than an experiment. The data design requirement (c), 7 force sensors were used to form a in Table 3 indicates the accuracy of the ground feature recog- force measurement plate. The force sensor can bear the nition. The maximum error is about 5.3% which occurred on weight of a pilot. The main contribution of this research is soft and hard ground feature detection. An error of about that ground feature detection and recognition were realized 4.4% occurred in the up and down features. which is mentioned in requirements (f) and (g). Attitude and force data were combined to get the data window which is used to analyze the ground features. Besides, the main work 6. Conclusions and Future Work of the IMU is to obtain the attitude data for the shoes. The detection result is also the effect of the properties of the mate- In this work, we introduced the application environment for rial used in the bottom layer. The rubber layer can absorb the AULLE, RELLE, and ASLLE respectively. As an ASLEE, noise from the motion of touching the ground. AIDER is used to help SCI patients return to a normal life. PCA and the minimum-distance classifier are involved We proposed a wearable sensing system that is able to in realizing the ground feature recognition. Four classical improve the flexibility and safety of the pilot by detecting gait Component 1 Variance explained (%) Variance explained (%) (%) (%) Component 3 Variance explained (%) (%) Component 2 8 Applied Bionics and Biomechanics results indicated that the wearable sensing system is able to realize human gait trajectory detection, and the trajectory 1.5 trend of CoP agrees with the normal human trajectory. PCA is involved in ground feature recognition because of the large dimension of the eigenvector. The analysis result 0.5 showed that the first three principal components are enough for the uphill/hard, downhill/hard, hard/flat, and soft/flat ground feature extraction. Finally, a test was carried out to −0.5 verify the recognition performance, and the results showed −1 that the recognition rate is more than 93%. −1.5 In the future, more environmental situations should be considered into the recognition experiment to verify the per- −2 formance of the wearable sensing system, for example, a road −4 −3 −2 −10 1 made of sand and cobblestones. Until now, the recognition Component 1 algorithm is still executed on a PC, which is not convenient Blue: downhill/hard for real time work. Red: uphill/hard Green: hard/flat Data Availability Black: soft/flat Figure 13: The four ground features are described by the first two The data used to support the findings of this study are principal components. available from the corresponding author upon request. Test result Conflicts of Interest The authors declare that there is no conflict of interests regarding the publication of this paper. Acknowledgments This research project is supported by the National Key −2 Research and Development Plan (2017YFB1302300) and the National Natural Science Foundation of China (nos. U1613223 and 61503060). 0 0 References −2 −1 −4 [1] A. B. Zoss, H. Kazerooni, and A. Chu, “Biomechanical design −2 −6 of the Berkeley lower extremity exoskeleton (BLEEX),” IEEE/ ASME Transactions on Mechatronics, vol. 11, no. 2, pp. 128– Figure 14: The recognition results for the 4 ground features. 138, 2006. [2] R. Steger, S. H. Kim, and H. Kazerooni, “Control scheme and Table 3: The recognition rate for the 4 features. networked control architecture for the Berkeley lower extrem- ity exoskeleton (BLEEX),” in Proceedings 2006 IEEE Interna- D/H U/H H/F S/F tional Conference on Robotics and Automation, 2006. ICRA D/H 95.620% 4.380% 0 0 2006, pp. 3469–3476, Orlando, FL, USA, May 2006. U/H 0.680% 93.878% 2.721% 2.721% [3] C. Teng, Z. Wong, W. Ten, and Y. Chong, “Design and development of inexpensive pneumatically-powered assisted H/F 0 3.448% 96.552% 0 knee-ankle-foot orthosis for gait rehabilitation-preliminary S/F 0 0 5.333% 94.667% finding,” in 2012 International Conference on Biomedical Engineering (ICoBE), Penang, Malaysia, February 2012. [4] Y. Eguchi, H. Kadone, and K. Suzuki, “Standing mobility and GRF. The mechanical design of the wearable sensing device with passive lower limb exoskeleton for upright loco- shoes is proposed to fulfill the design requirements. Seven motion,” IEEE/ASME Transactions on Mechatronics, vol. 23, force sensors are used to form two rigid planes to detect the no. 4, pp. 1608–1618, 2018. GRF that showed a good performance on force and CoP [5] T. Vouga, K. Z. Zhuang, J. Olivier et al., “EXiO—A brain- detection. The IMU sensor was installed on the wearable controlled lower limb exoskeleton for rhesus macaques,” IEEE sensing system to sense the attitude and acceleration data Transactions on Neural Systems and Rehabilitation Engineer- which were calculated for ground feature recognition. A ing, vol. 25, no. 2, pp. 131–141, 2017. verification experiment of the wearable sensing system was [6] M. Munera, A. Marroquin, L. Jimenez et al., “Lokomat therapy executed to test the performance for CoP detection. The in Colombia: current state and cognitive aspects,” in 2017 Component 1 Component 3 Component 2 Applied Bionics and Biomechanics 9 International Conference on Biomedical Engineering and Infor- International Conference on Rehabilitation Robotics (ICORR), pp. 394–399, London, UK, July 2017. matics, pp. 184–188, Dalian, China, October 2014. [7] M. Talaty, A. Esquenazi, and J. E. Briceño, “Differentiating [21] H. M. Schepers, E. van Asseldonk, J. H. Buurke, and P. H. ability in users of the ReWalk™ powered exoskeleton,” in Veltink, “Ambulatory estimation of center of mass displace- 2013 IEEE 13th International Conference on Rehabilitation ment during walking,” IEEE Transactions on Biomedical Robotics (ICORR), pp. 1–5, Seattle, WA, USA, June 2013. Engineering, vol. 56, no. 4, pp. 1189–1195, 2009. [8] P. N. Gad, Y. P. Gerasimenko, S. Zdunowski et al., “Iron ‘Elec- [22] G. Li, T. Liu, J. Yi, H. Wang, J. Li, and Y. Inoue, “The lower triRx’ man: overground stepping in an exoskeleton combined limbs kinematics analysis by wearable sensor shoes,” IEEE Sen- with noninvasive spinal cord stimulation after paralysis,” in sors Journal, vol. 16, no. 8, pp. 2627–2638, 2016. 2015 37th Annual International Conference of the IEEE Engi- [23] T. Liu, Y. Inoue, K. Shibata, and K. Shiojima, “Three-dimen- neering in Medicine and Biology Society (EMBC), pp. 1124– sional lower limb kinematic and kinetic analysis based on a 1127, Milan, Italy, August 2015. wireless sensor system,” in 2011 IEEE International Conference [9] Y. Sankai, “Hal: hybrid assistive limb based on cybernics,” in on Robotics and Automation, pp. 842–847, Shanghai, China, Robotics Research. Springer Tracts in Advanced Robotics, vol May 2011. 66, M. Kaneko and Y. Nakamura, Eds., pp. 25–34, Springer, [24] S. J. M. Bamberg, A. Y. Benbasat, D. M. Scarborough, D. E. Berlin, Heidelberg, 2010. Krebs, and J. A. Paradiso, “Gait analysis using a shoe- [10] http://www.suitx.com. integrated wireless sensor system,” IE EE Transactions on Infor- mation Technology in Biomedicine,vol. 12, no.4,pp.413–423, [11] A. D. Gardner, J. Potgieter, and F. K. Noble, “A review of com- mercially available exoskeletons’ capabilities,” in 2017 24th International Conference on Mechatronics and Machine Vision [25] M. Pazzaglia and M. Molinari, “The embodiment of assistive in Practice (M2VIP), Auckland, New Zealand, November devices-from wheelchair to exoskeleton,” Physics of Life Reviews, vol. 16, pp. 163–175, 2016. [12] P. Asselin, S. Knezevic, S. Kornfeld et al., “Heart rate and [26] P. J. O’Connor, “Trends in spinal cord injury,” Accident Anal- oxygen demand of powered exoskeleton-assisted walking in ysis & Prevention, vol. 38, no. 1, pp. 71–77, 2006. persons with paraplegia,” Journal of Rehabilitation Research [27] K. Hashimoto, H. Kang, M. Nakamura et al., “Realization of and Development, vol. 52, no. 2, pp. 147–158, 2015. biped walking on soft ground with stabilization control based [13] L. A. Rapp, V. Huang, S. Firpi, L. Tovar, J. M. Zanca, and on gait analysis,” in 2012 IEEE/RSJ International Conference J. Baeza-Dager, “The age of the bionic man has arrived: the on Intelligent Robots and Systems, pp. 2064–2069, Vilamoura, use of Ekso exoskeleton in acute paraplegia,” The Journal of Portugal, October 2012. Spinal Cord Medicine, vol. 36, no. 5, pp. 504–523, 2013. [28] M. K. Shepherd and E. J. Rouse, “The VSPA foot: a quasi- [14] M. K. Duong, H. Cheng, H. T. Tran, and Q. Jing, “Minimizing passive ankle-foot prosthesis with continuously variable stiff- human-exoskeleton interaction force using compensation ness,” IEEE Transactions on Neural Systems and Rehabilitation for dynamic uncertainty error with adaptive RBF network,” Engineering, vol. 25, no. 12, pp. 2375–2386, 2017. Journal of Intelligent & Robotic Systems, vol. 82, no. 3-4, [29] P. Di, J. Huang, S. Nakagawa, K. Sekiyama, and T. Fukuda, pp. 413–433, 2016. “Fall detection for elderly by using an intelligent cane robot [15] R. Huang, H. Cheng, Q. Chen, H.-T. Tran, and X. Lin, based on center of pressure (COP) stability theory,” in 2014 “Interactive learning for sensitivity factors of a human- International Symposium on Micro-NanoMechatronics and powered augmentation lower exoskeleton,” in 2015 IEEE/ Human Science (MHS), pp. 1–4, Nagoya, Japan, November RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 6409–6415, Hamburg, Germany, September- [30] T. T. Pham, S. T. Moore, S. J. G. Lewis et al., “Freezing of gait October 2015. detection in Parkinson's disease: a subject-independent detec- [16] R. Huang, H. Cheng, H. Guo, Q. Chen, and X. Lin, “Hierarchi- tor using anomaly scores,” IEEE Transactions on Biomedical cal interactive learning for a human-powered augmentation Engineering, vol. 64, no. 11, pp. 2719–2728, 2017. lower exoskeleton,” in 2016 IEEE International Conference on [31] A. De Cock, J. Vanrenterghem, T. Willems, E. Witvrouw, and Robotics and Automation (ICRA), pp. 257–263, Stockholm, D. De Clercq, “The trajectory of the centre of pressure during Sweden, May 2016. barefoot running as a potential measure for foot function,” [17] X. Shu, H. Cheng, and X. Lin, “Adaptive stair-ascending and Gait & Posture, vol. 27, no. 4, pp. 669–675, 2008. stair-descending strategies for powered lower limb exoskele- [32] K. Suwanratchatamanee, M. Matsumoto, and S. Hashimoto, ton,” in 2017 IEEE International Conference on Mechatronics “Haptic sensing foot system for humanoid robot and ground and Automation (ICMA), pp. 1579–1584, Takamatsu, Japan, recognition with one-leg balance,” IEEE Transactions on August 2017. Industrial Electronics, vol. 58, no. 8, pp. 3174–3186, 2011. [18] Q. Chen, H. Cheng, C. Zou, R. Huang, and H. Guo, “Intention [33] W. Xueyun and Z. Wei, “Application of kernel PCA for foetal recognition for exoskeleton,” in 2017 2nd International Con- ECG estimation,” Electronics Letters, vol. 54, no. 6, pp. 340– ference on Advanced Robotics and Mechatronics (ICARM), 342, 2018. pp. 535–539, Hefei, China, August 2017. [34] S. Bag and G. Sanyal, “An efficient face recognition approach [19] C. Yue, H. Cheng, Y. Chen, Q. Deng, and X. Lin, “Design of a using PCA and minimum distance classifier,” in 2011 Interna- wearable sensing system for a lower extremity exoskeleton,” in tional Conference on Image Information Processing, pp. 1–6, 2017 IEEE International Conference on Mechatronics and Auto- Shimla, India, November 2011. mation (ICMA), pp. 464–468, Takamatsu, Japan, August 2017. [20] S. Yan, W. Ren, X. Liang, and K. Zhang, “Gait characteristics of overweight and obese children with different ages,” in 2014 7th International Journal of Advances in Rotating Machinery Multimedia Journal of The Scientific Journal of Engineering World Journal Sensors Hindawi Hindawi Publishing Corporation Hindawi Hindawi Hindawi Hindawi www.hindawi.com Volume 2018 http://www www.hindawi.com .hindawi.com V Volume 2018 olume 2013 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 Journal of Control Science and Engineering Advances in Civil Engineering Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 Submit your manuscripts at www.hindawi.com Journal of Journal of Electrical and Computer Robotics Engineering Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 VLSI Design Advances in OptoElectronics International Journal of Modelling & Aerospace International Journal of Simulation Navigation and in Engineering Engineering Observation Hindawi Hindawi Hindawi Hindawi Volume 2018 Volume 2018 Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com www.hindawi.com www.hindawi.com Volume 2018 International Journal of Active and Passive International Journal of Antennas and Advances in Chemical Engineering Propagation Electronic Components Shock and Vibration Acoustics and Vibration Hindawi Hindawi Hindawi Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Applied Bionics and Biomechanics Hindawi Publishing Corporation

Design and Performance Evaluation of a Wearable Sensing System for Lower-Limb Exoskeleton

Download PDF
10 pages

Loading next page...
 
/lp/hindawi-publishing-corporation/design-and-performance-evaluation-of-a-wearable-sensing-system-for-ErrRWSl0t2

References (35)

Publisher
Hindawi Publishing Corporation
Copyright
Copyright © 2018 Chunfeng Yue et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
ISSN
1176-2322
eISSN
1754-2103
DOI
10.1155/2018/8610458
Publisher site
See Article on Publisher Site

Abstract

Hindawi Applied Bionics and Biomechanics Volume 2018, Article ID 8610458, 9 pages https://doi.org/10.1155/2018/8610458 Research Article Design and Performance Evaluation of a Wearable Sensing System for Lower-Limb Exoskeleton 1,2 1,2 1 1,2 1,2 Chunfeng Yue, Xichuan Lin, Ximing Zhang, Jing Qiu , and Hong Cheng The School of Automation Engineering, University of Electronic Science and Technology of China, China Center for Robotics, University of Electronic Science and Technology of China, China Correspondence should be addressed to Hong Cheng; hcheng@uestc.edu.cn Received 12 February 2018; Revised 27 April 2018; Accepted 13 June 2018; Published 18 September 2018 Academic Editor: Dongming Gan Copyright © 2018 Chunfeng Yue et al. This is an open access article distributed under the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited. Because the target users of the assistive-type lower extremity exoskeletons (ASLEEs) are those who suffer from lower limb disabilities, customized gait is adopted for the control of ASLEEs. However, the customized gait is unable to provide stable motion for variable terrain, for example, flat, uphill, downhill, and soft ground. The purpose of this paper is to realize gait detection and environment feature recognition for AIDER by developing a novel wearable sensing system. The wearable sensing system employs 7 force sensors as a sensing matrix to achieve high accuracy of ground reaction force detection. There is one more IMU sensor that is integrated into the structure to detect the angular velocity. By fusing force and angular velocity data, four typical terrain features can be recognized successfully, and the recognition rate can reach up to 93%. 1. Introduction and standing, but they also rebuild their confidence in daily life. There are three typical commercial products that have Lower limb exoskeleton (LLE) robotic technology has been been developed: ReWalk 6.0 [7], Ekso GT [8], and HAL-5 [9]. They each have a weight of about 20 kg. Besides these, developed rapidly for the past 20 years. Three implementa- lightweight ASLLEs have also been developed to assist spinal tion fields, that is, augmentation, rehabilitation, and living cord injury patients, such as Phoenix 3.0 [10] and INDEGO support are explored. First, the human augmentation type [11]. The researchers are also trying to prove their benefit LLE (AULLE) was developed for the military and aims at to patients who have used the ASLLEs [12, 13]. improving a soldier’s weight loading ability. The fields of In our previous research, we focused on the human aug- application extend to disaster relief and industrial transport mentation exoskeleton [14–16] and an ASLLE named the assistance. Kazerooni et al. developed the typical first gener- ation AULLE which is named BLEEX [1, 2]. Then, the third AIDER [17, 18] for individuals. An illustration of a prototype of AIDER with a SCI patient is shown in Figure 1. Based on generation AULLE which is named HULC was developed AIDER and two crutches, the patient can form a stable area for carrying a load of about 90 kg. Second, the rehabilitation to keep balance. Four DC motors are installed on the hip type LLE (RELLE) was developed for patients whose lower and knee joints respectively. The patient controls the AIDER limbs are inconvenienced [3–5]. Typical patients include those with foot drop, spinal cord injuries, and strokes. The by two controllers which are installed on the grab handles of the crutches. A battery and control circuit placed in the back- most famous RELLE is Lokomat, which is developed by the pack are the core part for controlling the DC motors. This Hocoma Company [6]. The assistive-type LLE (ASLLE) is paper is an extension for my previous work [19]. used to assist patients with lower limb disabilities but whose Avoiding harm to the patient when piloting an exoskele- upper limbs are normal. The ASLLEs assist patients to return to their normal life. They do not only assist in walking ton is a critical issue. Table 1 shows the influence of three kinds of LLEs on the interaction between human, machine, motion on flat ground, climbing on stairs, or sitting down 2 Applied Bionics and Biomechanics Company. Based on this product, the relationship between the gait characteristics and foot pressure for overweight chil- Pilot dren was analyzed [20]. Besides, researchers also designed a number of wearable sensing systems to meet their research requirements. In [21], the researchers designed sensing shoes Bag to estimate the CoM displacement continuously using an ambulatory measurement system which contains 2 IMUs and 2 6DOF force/moment sensors. Although the precision of the sensor is really high, the large size influences gait anal- Control button Hip joint ysis resulting from normal walking. Even high precision data can be collected easily. The large size of the IMU and pressure sensor gives the pilot poor wearing experience. Liu et al. have worked on gait analysis for serval years. They developed Knee joint smart shoes for human gait analysis [22, 23]. The novel point of the smart shoes is that they contain three 6-axis force sen- sors with 1 cm thickness. The sensors are mounted on the heel, arch, and forefoot respectively to subdivide the phase of human walking and get high accuracy gait data. Bamberg et al. developed a multiple sensing system and integrated it with shoes. The sensing system contains 6 kinds of sensors Crutch including an accelerometer, gyroscope, force sensitive resis- tor, bend sensor, polyvinylidene fluoride strip, and electric field sensor [24]. The multiple sensors provide redundancy Shoes gait data to ensure stability. Due to the complex environment as shown in Table 1, we intend to develop a wearable sensing Figure 1: The prototype of AIDER with a SCI patient. system for AIDER which can not only detect human body motion but also detect the state of the environment, such as the features of the ground. Table 1: The influence for three kinds of exoskeletons on the human-machine-environment. 2. Motivation Factors AULLE RELLE ASLLE 2.1. User and Application Environment for AIDER. AIDER is Human Health Weak Weak intended for users with SCI with injury levels from T9 to T12 Machine Low High High caused by traumatic injuries (e.g., vehicular crashing or fall- Environment Daily life/unsafe Hospital/safe Daily life/unsafe ing from buildings) or disease (e.g., myelitis) [25, 26]. The AIDER is aimed at extending the range of activities to advance their rehabilitation programs for SCIs. Besides, and the environment. For an AULLE, the mechanical parts walking upright makes the patients feel more confident follow the pilot’s motion and support the weight of the load. because they can make a conversation with friends at eye The pilot takes the initiative in the human-machine- level and walk like normal persons. environment system. For a RELLE, the machine controls As illustrated in the Introduction, the target environ- the human motion because of the lower-limb paralysis of ments for the application of AIDER are daily life and clinic SCI patients. The working environment of a RELLE is very rehabilitation. Compared with the clinical environment, the safe because the pilot is always protected by fixed support. daily life environment is more complex. Therefore, in this For an ASLLE, although the potential pilots are the same as research, we pay more attention in analyzing the main fea- those of a RELLE, the working environment for an ASLLE tures of the daily life environment, especially the ground. is complex daily life. The safety for the pilot of an ASLLE Generally, two main features of the ground influence the gait relies on the exoskeleton’s stability. Besides, patients with for normal walking, that is, hardness and terrain. Figure 2 complete injury have lost their sensory ability. They cannot shows the relationship between these two features and the keep their balance when wearing the exoskeleton. Therefore, common implementation environment. The typical mate- the ASLLE should be able to sense the human-machine state rials in daily life have two relative features: for example, the and adjust the gait trajectory for the pilot to avoid a danger- typical features of marble ground are flat and hard. ous situation. Based on the analysis for the safety of a human-machine 2.2. Gait Analysis and Environment Detection for the AIDER. system, a sensing system is necessary to improve the stability For healthy people, the gait is changed adaptively when they of ASLLEs. The shoes are suitable components for installing cross from one terrain to another. For instance, the gait will sensors. In related works, researchers have used foot- change when someone crosses from hard ground to sand. sensing systems to detect gait information. Footscan is a typ- However, the AIDER works on a customized gait to realize ical commercial product which was developed by the RSscan the walking motion for SCI patients [15]. It would cause a Applied Bionics and Biomechanics 3 Hardness Terrain Typical material (d) To avoid the force from exceeding the acceptable range, the force measurement range of the wearable Cobbled sensing system should be from 0 to at least 120 kg. road (e) To cut the cost, the hardware cost of the wearable Flat sensing system should be less than ¥2000. Marble ground (f) To provide enough data for control strategies, the Hard wearable sensing system should be designed to be Rugged able to detect and recognize ground features. Ramp (g) Finally, the wearable sensing system needs to realize Soft attitude measurement and gait analysis. Lawn Slope 3.2. Design of the Wearable Sensing System. As proposed in Carpet Section 3.1, 7 design requirements should be met. The mechanical design of the wearable sensing system is shown Figure 2: The common application environment for users of in Figure 3. In Figure 3(a), there are 3 layers that form the AIDER. The solid line and dotted line stand for hard and soft sensing part for force detection. The bottom layer is con- features, respectively; red, blue, and black denote flat, rugged, and structed of wear-resistant rubber which is used to ensure that slope features, respectively. the pilot’s foot does not slip. A hook and loop tape is used to fasten the foot. The middle layer is a holder for the force sen- sor. The total thickness of the 3 layers is 18 mm which can potential safety hazard because it cannot adapt to the change meet the conditions of requirement (a). Seven strain gauge of terrains adaptively in a social environment. force sensors are employed to sense the center of force in Gait analysis mainly focuses on two parameters which the z-axis. The top layer is used to install the 7 force sensors are ground reaction force (GRF) and body posture. These which are made of aluminum alloy. It is necessary to recog- two parameters can be utilized to confirm whether the sys- nize the force for the heel and forefoot [27, 28]. Therefore, tem state is suitable for the next motion. For the stability the top layer is made of two separated aluminum alloys. control of biped robots, findings from [27] indicated that, Three force sensors on the forefoot form a stable plane. The comparing with a hard ground, step height tends to other 4 force sensors form a trapezoid to keep stable. The increase for avoiding collision between a robot’s feet and accuracy of each force sensor is about 0.1%. All the seven soft ground. Besides, terrain features are also the essential force sensors together are capable of high accuracy measure- factors for gait adjustment. For AIDER, the control strat- ment. Because the range of a force sensor is about 25 kgf, the egy of stability not only depends on the system controller measured range of the wearable sensing system is about but also the environmental features. Typical environments 175 kgf. The IMU sensors and control circuit are used to col- for AIDER are shown in Figure 2. Therefore, we intend to lect attitude data which is installed in the circuit box as design a wearable sensing system for AIDER which can be shown in Figure 3(a). The connection rod is designed to link used to detect and recognize environmental features and the wearable sensing system and the shanks of AIDER. To CoP of the feet in this paper. meet requirement (e), the cost of the wearable sensing system is listed in Table 2. 3. Method and Materials 3.3. Force Measurement Experiment. To test the performance 3.1. Design Requirement of a Wearable Sensing System. of our force sensing system in terms of accuracy and dynamic Section 1 shows the benefit of a wearable sensing system for stability, a force measurement experiment was conducted. In gait analysis and the safety of a human-machine system. this experiment, we used the force platforms to verify and Therefore, a wearable sensing system for the feet is proposed calibrate the accuracy of the wearable sensing system. to realize gait analysis and environment detection. The Figure 4 describes the setup of the force measurement exper- following design requirements are proposed according to iment, where the pilot stands astraddle on two force plat- the features of a SCI patient: forms. During the experiment, the pilot shifts the support foot at the center of his body weight from the left to the right (a) To make the user comfortable, the thickness of sole and then shifts back again to the left. The motion frequency should be less than 20 mm. is about 2 seconds. Finally, we used a wireless module to translate the data of the wearable sensing system to a PC (b) To ensure convenience, people should be able to put for sensing system analysis. The experimental results are on the shoe using one hand. indicated in Figure 5. The red line indicates the output of (c) To ensure the accuracy for gait analysis, the magni- the force plates and the blue line indicates the output of tude of output force from the sensors should be the wearable sensing system. This figure shows that the data from the shoes follow the data from the force plate with high obtained. 4 Applied Bionics and Biomechanics Bandage Top layer criteria. CoP coincides with ZMP when the system is under a quasistatic state. Di et al. developed a cane robot to realize Connection human fall detection by estimating the CoP [29]. In our Circuit box research, CoP is also involved in AIDER for stability estima- tion of the human-machine system. To estimate the CoP of Top layer the human-machine system, the first step is to calculate the Middle layer ground reaction force (GRF) and the CoP of the feet. Accord- ing to [30], the CoP can be estimated by Bottom layer (a) x ⋅ Fx dx = , CoP Middle layer Fx dx y ⋅ Fy dy Y = CoP Fy dy Based on the mechanical design of the force sensory system, after being dispersed (1), the CoP of the foot is obtained by (b) ∑x ⋅ f i ni X = , CoP ∑f ni ∑y ⋅ f i ni Y = , CoP ∑f ni where P x , y ,(i =1,2,… ,7) denotes the coordinate for i i i each force sensor. f denotes the force that is obtained ni by each force sensor. n is the mark for recognizing the left and right foot. Based on mechanical design, the coordinate of each sensor can be obtained by Figure 3(c). A verification experiment is designed to prove the perfor- mance of a wearable sensing system for CoP detection. The experimental setup is similar to that in Figure 4. The differ- ence is that the pilot is walking in a daily life state but on a (c) force platform. Figure 6 showed the experimental results from the point when the heel touches the ground up to the Figure 3: The mechanical structure of the wearable sensing system point when the toe lifts from the ground. In this experiment, for AIDER. (a) The structure of the wearable sensing system. (b) The prototype of the wearable sensing system. (c) The coordinate the trajectory transforms from the heel to the big toe as system for the left foot. shown in Figure 6(a). Figure 6(b) shows the magnitude of the total force in the z-axis. The trajectory of CoP in the X- Y plane is shown in Figure 6(c). Based on [31], the trajectory of CoP agrees with human habit because of a similar curve. Table 2: The cost of wearable sensing system. Name Unit cost Unit quantity Total price/¥ 4. Ground Characteristic Analysis IMU 25 piece 2 50 and Recognition Force sensor 15 piece 14 210 4.1. Ground Characteristic Analysis. Based on Section 2, the Mechanical parts 500 Set 2 1000 main features of the application environment contain hard- Circuit board 100 Piece 4 400 ness and terrain. More specifically, soft/hard and flat/slope Hook and loop tape 10 piece 2 20 are two pairs of critical factors for a control strategy. After Sum 1680 considering the environment in daily life, carpet, ramp, and marble ground are selected as the recognized subjects. For the ramp, uphill and downhill is the difference. For a flat ground, soft and hard is the main difference feature. There- precision. With a shaking motion, the wearable system fore, the main purpose of the wearable sensing system is to detected the body shaking accurately. recognize the following combined features which are flat/ 3.4. Center of Pressure (CoP). For biped locomotion control, hard (F/H), flat/soft (F/S), uphill/hard (U/H), and down- ZMP (zero moment point) and CoP are two important hill/hard (D/H). Applied Bionics and Biomechanics 5 Pilot Wearable sensing sytem data Force platform Wearable sensing sytem Force platform data Figure 4: The setup of the stability test for a force measurement experiment. 4.2. Principal Component Analysis (PCA) for the Four Ground Feature Extraction. PCA is a data analysis method −200 that uses an orthogonal transformation to obtain principal −400 components which are used to present the original data fea- −600 ture by a low-dimensional variable. As a popular pattern rec- 01 23 45 6 ognition method, PCA is widely used in face recognition Time (1 휇s/Div) ×10 [33]. This method is aimed at reducing the dimension for Platform output the eigenvector. In our research, the dimension of the eigen- Shoes output vector for an environmental feature extraction is 30 which contains the sum, mean, and variance of the 7 normalized Figure 5: Results for the accuracy verification of a wearable sensing force sensor output and 3-axis motion acceleration. After system. analyzing the data by PCA, the variance that explains princi- pal components are obtained. According to the result of Figures 8–11, the variance that To recognize these features, the data from the IMU and explained the first three principal components is more than force sensor are necessary. Generally, a walking motion can 85%. Therefore, the first three principal components are be divided into 8 phases, that is, initial contact (IC), loading enough to distinguish the four ground features. Figures 8 response (LR), midstance (MS), preswing (PS), initial swing and 9 particularly show that the uphill and downhill motions (IS), midswing (MS), and terminal swing (TS) [24]. For a are easy to describe using the first 2 principal components. In ramp, 3 force sensors are used in [32] to recognize the slope Figures 9 and 10, the variance explained on the third princi- by adjusting the sequence of the force sensor output. The IC pal component is more than 10%. The results indicate that and LR phases contain impact information caused by the the soft and hard features are relatively complex. Finally, hardness of the ground. Therefore, the sensor data from the ground features are described by the first three and two the IC and LR phases are collected by the data window for principal components in Figures 12 and 13, respectively. feature recognition. Due to the four ground features, it is easy to classify the A total of 7 force sensors and one IMU is used to sense minimum-distance classifier [34] which is employed to clas- the ground features. The force of each sensor is f ni sify the four features. (i =1,2,… ,7). The output of the IMU sensor is angular velocity ω = ω ω ω and acceleration a = x y z 4.3. Experiments. To verify the performance of the recogni- a a a . To get a credible result, the gravitational x y z tion method, 5 experimental subjects wore the wearable sens- acceleration is removed and the resultant force of the 7 sen- ing system and walked on carpet, ramp (uphill and sors is normalized. To keep the data in the same magnitude, downhill), and flat ground surface in a normal gait, and about the force is multiplied with a scale factor. The drastic vibra- 2000 steps were obtained. After preprocessing, half of data tion makes the IC and LR phases easy to detect and the force are used as training data. The left part is used for testing output also increases. Therefore, according to the output of the training model, and the experimental results are obtained the force sensors, the data window for feature recognition is as shown in Figure 14. The black circles indicate the points obtained as shown in Figure 7. that are not classified in features on the right. Force (N) X (mm) 6 Applied Bionics and Biomechanics −50 20 0 −100 −50 −150 −50 0 −200 50 −250 −100 (a) −150 −200 −250 −50 050 −250 −200 −150 −100 −50 0 X (mm) Y (mm) (b) (c) Figure 6: The CoP for the left foot; the black circle denotes the pressure-bearing point. (a) The trajectory of the CoP. (b) The magnitude of the total force during contact of the left foot to the ground. (c) The trajectory of CoP when the heel touches the ground up to point when the toe lifts from the ground. Data window Downhill/hard 100 100 90 90 80 80 70 70 60 60 50 50 40 40 −20 20 20 −40 10 10 0 0 −60 12 3 Principal component 1.185 1.19 1.195 1.2 1.205 1.21 1.215 1.22 Time (0.004 ms/Div) ×10 Figure 8: The relationship between the principal component and the variance explained for D/H. Data window AccZ AccX Force 20 AccY Finally, the recognition rate for 4 ground features is listed Figure 7: Data window for the feature extraction. The force in Table 3. The recognition rate of hard/flat ground and multiplied by 20 is the aim for analyzing in the same magnitude. downhill/hard ground is more than 95%. This result shows that the eigenvector which is extracted by PCA and the minimum-distance classifier is suitable for the ground fea- ture recognition. Y (mm) Force (N/9.8) Force (N/9.8) Variance explained (%) Y (mm) (%) Component 2 Applied Bionics and Biomechanics 7 Uphill/hard Soft/flat 90 90 90 90 80 80 80 80 70 70 70 70 60 60 60 60 50 50 50 50 40 40 40 40 30 30 30 20 20 20 10 10 10 0 0 0 0 1 12345 Principal component Principal component Figure 9: The relationship between the principal component and Figure 11: The relationship between the principal component and the variance explained for U/H. the variance explained for S/F. Hard/flat 100 100 70 70 0.5 60 60 50 50 −0.5 40 40 −1 −1.5 30 30 1 20 2 −1 −2 −3 −1 0 0 −4 Blue: downhill/hard Principal component Red: uphill/hard Green: hard/flat Figure 10: The relationship between the principal component and Black: soft/flat the variance explained for H/F. Figure 12: The four ground features are described by the first three principal components. 5. Discussion A mechanical design has been proposed in Section 3.2 that terrains are involved in this paper. However, the daily life meets design requirements (a), (b), (d), and (e). To meet environment is more complex than an experiment. The data design requirement (c), 7 force sensors were used to form a in Table 3 indicates the accuracy of the ground feature recog- force measurement plate. The force sensor can bear the nition. The maximum error is about 5.3% which occurred on weight of a pilot. The main contribution of this research is soft and hard ground feature detection. An error of about that ground feature detection and recognition were realized 4.4% occurred in the up and down features. which is mentioned in requirements (f) and (g). Attitude and force data were combined to get the data window which is used to analyze the ground features. Besides, the main work 6. Conclusions and Future Work of the IMU is to obtain the attitude data for the shoes. The detection result is also the effect of the properties of the mate- In this work, we introduced the application environment for rial used in the bottom layer. The rubber layer can absorb the AULLE, RELLE, and ASLLE respectively. As an ASLEE, noise from the motion of touching the ground. AIDER is used to help SCI patients return to a normal life. PCA and the minimum-distance classifier are involved We proposed a wearable sensing system that is able to in realizing the ground feature recognition. Four classical improve the flexibility and safety of the pilot by detecting gait Component 1 Variance explained (%) Variance explained (%) (%) (%) Component 3 Variance explained (%) (%) Component 2 8 Applied Bionics and Biomechanics results indicated that the wearable sensing system is able to realize human gait trajectory detection, and the trajectory 1.5 trend of CoP agrees with the normal human trajectory. PCA is involved in ground feature recognition because of the large dimension of the eigenvector. The analysis result 0.5 showed that the first three principal components are enough for the uphill/hard, downhill/hard, hard/flat, and soft/flat ground feature extraction. Finally, a test was carried out to −0.5 verify the recognition performance, and the results showed −1 that the recognition rate is more than 93%. −1.5 In the future, more environmental situations should be considered into the recognition experiment to verify the per- −2 formance of the wearable sensing system, for example, a road −4 −3 −2 −10 1 made of sand and cobblestones. Until now, the recognition Component 1 algorithm is still executed on a PC, which is not convenient Blue: downhill/hard for real time work. Red: uphill/hard Green: hard/flat Data Availability Black: soft/flat Figure 13: The four ground features are described by the first two The data used to support the findings of this study are principal components. available from the corresponding author upon request. Test result Conflicts of Interest The authors declare that there is no conflict of interests regarding the publication of this paper. Acknowledgments This research project is supported by the National Key −2 Research and Development Plan (2017YFB1302300) and the National Natural Science Foundation of China (nos. U1613223 and 61503060). 0 0 References −2 −1 −4 [1] A. B. Zoss, H. Kazerooni, and A. Chu, “Biomechanical design −2 −6 of the Berkeley lower extremity exoskeleton (BLEEX),” IEEE/ ASME Transactions on Mechatronics, vol. 11, no. 2, pp. 128– Figure 14: The recognition results for the 4 ground features. 138, 2006. [2] R. Steger, S. H. Kim, and H. Kazerooni, “Control scheme and Table 3: The recognition rate for the 4 features. networked control architecture for the Berkeley lower extrem- ity exoskeleton (BLEEX),” in Proceedings 2006 IEEE Interna- D/H U/H H/F S/F tional Conference on Robotics and Automation, 2006. ICRA D/H 95.620% 4.380% 0 0 2006, pp. 3469–3476, Orlando, FL, USA, May 2006. U/H 0.680% 93.878% 2.721% 2.721% [3] C. Teng, Z. Wong, W. Ten, and Y. Chong, “Design and development of inexpensive pneumatically-powered assisted H/F 0 3.448% 96.552% 0 knee-ankle-foot orthosis for gait rehabilitation-preliminary S/F 0 0 5.333% 94.667% finding,” in 2012 International Conference on Biomedical Engineering (ICoBE), Penang, Malaysia, February 2012. [4] Y. Eguchi, H. Kadone, and K. Suzuki, “Standing mobility and GRF. The mechanical design of the wearable sensing device with passive lower limb exoskeleton for upright loco- shoes is proposed to fulfill the design requirements. Seven motion,” IEEE/ASME Transactions on Mechatronics, vol. 23, force sensors are used to form two rigid planes to detect the no. 4, pp. 1608–1618, 2018. GRF that showed a good performance on force and CoP [5] T. Vouga, K. Z. Zhuang, J. Olivier et al., “EXiO—A brain- detection. The IMU sensor was installed on the wearable controlled lower limb exoskeleton for rhesus macaques,” IEEE sensing system to sense the attitude and acceleration data Transactions on Neural Systems and Rehabilitation Engineer- which were calculated for ground feature recognition. A ing, vol. 25, no. 2, pp. 131–141, 2017. verification experiment of the wearable sensing system was [6] M. Munera, A. Marroquin, L. Jimenez et al., “Lokomat therapy executed to test the performance for CoP detection. The in Colombia: current state and cognitive aspects,” in 2017 Component 1 Component 3 Component 2 Applied Bionics and Biomechanics 9 International Conference on Biomedical Engineering and Infor- International Conference on Rehabilitation Robotics (ICORR), pp. 394–399, London, UK, July 2017. matics, pp. 184–188, Dalian, China, October 2014. [7] M. Talaty, A. Esquenazi, and J. E. Briceño, “Differentiating [21] H. M. Schepers, E. van Asseldonk, J. H. Buurke, and P. H. ability in users of the ReWalk™ powered exoskeleton,” in Veltink, “Ambulatory estimation of center of mass displace- 2013 IEEE 13th International Conference on Rehabilitation ment during walking,” IEEE Transactions on Biomedical Robotics (ICORR), pp. 1–5, Seattle, WA, USA, June 2013. Engineering, vol. 56, no. 4, pp. 1189–1195, 2009. [8] P. N. Gad, Y. P. Gerasimenko, S. Zdunowski et al., “Iron ‘Elec- [22] G. Li, T. Liu, J. Yi, H. Wang, J. Li, and Y. Inoue, “The lower triRx’ man: overground stepping in an exoskeleton combined limbs kinematics analysis by wearable sensor shoes,” IEEE Sen- with noninvasive spinal cord stimulation after paralysis,” in sors Journal, vol. 16, no. 8, pp. 2627–2638, 2016. 2015 37th Annual International Conference of the IEEE Engi- [23] T. Liu, Y. Inoue, K. Shibata, and K. Shiojima, “Three-dimen- neering in Medicine and Biology Society (EMBC), pp. 1124– sional lower limb kinematic and kinetic analysis based on a 1127, Milan, Italy, August 2015. wireless sensor system,” in 2011 IEEE International Conference [9] Y. Sankai, “Hal: hybrid assistive limb based on cybernics,” in on Robotics and Automation, pp. 842–847, Shanghai, China, Robotics Research. Springer Tracts in Advanced Robotics, vol May 2011. 66, M. Kaneko and Y. Nakamura, Eds., pp. 25–34, Springer, [24] S. J. M. Bamberg, A. Y. Benbasat, D. M. Scarborough, D. E. Berlin, Heidelberg, 2010. Krebs, and J. A. Paradiso, “Gait analysis using a shoe- [10] http://www.suitx.com. integrated wireless sensor system,” IE EE Transactions on Infor- mation Technology in Biomedicine,vol. 12, no.4,pp.413–423, [11] A. D. Gardner, J. Potgieter, and F. K. Noble, “A review of com- mercially available exoskeletons’ capabilities,” in 2017 24th International Conference on Mechatronics and Machine Vision [25] M. Pazzaglia and M. Molinari, “The embodiment of assistive in Practice (M2VIP), Auckland, New Zealand, November devices-from wheelchair to exoskeleton,” Physics of Life Reviews, vol. 16, pp. 163–175, 2016. [12] P. Asselin, S. Knezevic, S. Kornfeld et al., “Heart rate and [26] P. J. O’Connor, “Trends in spinal cord injury,” Accident Anal- oxygen demand of powered exoskeleton-assisted walking in ysis & Prevention, vol. 38, no. 1, pp. 71–77, 2006. persons with paraplegia,” Journal of Rehabilitation Research [27] K. Hashimoto, H. Kang, M. Nakamura et al., “Realization of and Development, vol. 52, no. 2, pp. 147–158, 2015. biped walking on soft ground with stabilization control based [13] L. A. Rapp, V. Huang, S. Firpi, L. Tovar, J. M. Zanca, and on gait analysis,” in 2012 IEEE/RSJ International Conference J. Baeza-Dager, “The age of the bionic man has arrived: the on Intelligent Robots and Systems, pp. 2064–2069, Vilamoura, use of Ekso exoskeleton in acute paraplegia,” The Journal of Portugal, October 2012. Spinal Cord Medicine, vol. 36, no. 5, pp. 504–523, 2013. [28] M. K. Shepherd and E. J. Rouse, “The VSPA foot: a quasi- [14] M. K. Duong, H. Cheng, H. T. Tran, and Q. Jing, “Minimizing passive ankle-foot prosthesis with continuously variable stiff- human-exoskeleton interaction force using compensation ness,” IEEE Transactions on Neural Systems and Rehabilitation for dynamic uncertainty error with adaptive RBF network,” Engineering, vol. 25, no. 12, pp. 2375–2386, 2017. Journal of Intelligent & Robotic Systems, vol. 82, no. 3-4, [29] P. Di, J. Huang, S. Nakagawa, K. Sekiyama, and T. Fukuda, pp. 413–433, 2016. “Fall detection for elderly by using an intelligent cane robot [15] R. Huang, H. Cheng, Q. Chen, H.-T. Tran, and X. Lin, based on center of pressure (COP) stability theory,” in 2014 “Interactive learning for sensitivity factors of a human- International Symposium on Micro-NanoMechatronics and powered augmentation lower exoskeleton,” in 2015 IEEE/ Human Science (MHS), pp. 1–4, Nagoya, Japan, November RSJ International Conference on Intelligent Robots and Systems (IROS), pp. 6409–6415, Hamburg, Germany, September- [30] T. T. Pham, S. T. Moore, S. J. G. Lewis et al., “Freezing of gait October 2015. detection in Parkinson's disease: a subject-independent detec- [16] R. Huang, H. Cheng, H. Guo, Q. Chen, and X. Lin, “Hierarchi- tor using anomaly scores,” IEEE Transactions on Biomedical cal interactive learning for a human-powered augmentation Engineering, vol. 64, no. 11, pp. 2719–2728, 2017. lower exoskeleton,” in 2016 IEEE International Conference on [31] A. De Cock, J. Vanrenterghem, T. Willems, E. Witvrouw, and Robotics and Automation (ICRA), pp. 257–263, Stockholm, D. De Clercq, “The trajectory of the centre of pressure during Sweden, May 2016. barefoot running as a potential measure for foot function,” [17] X. Shu, H. Cheng, and X. Lin, “Adaptive stair-ascending and Gait & Posture, vol. 27, no. 4, pp. 669–675, 2008. stair-descending strategies for powered lower limb exoskele- [32] K. Suwanratchatamanee, M. Matsumoto, and S. Hashimoto, ton,” in 2017 IEEE International Conference on Mechatronics “Haptic sensing foot system for humanoid robot and ground and Automation (ICMA), pp. 1579–1584, Takamatsu, Japan, recognition with one-leg balance,” IEEE Transactions on August 2017. Industrial Electronics, vol. 58, no. 8, pp. 3174–3186, 2011. [18] Q. Chen, H. Cheng, C. Zou, R. Huang, and H. Guo, “Intention [33] W. Xueyun and Z. Wei, “Application of kernel PCA for foetal recognition for exoskeleton,” in 2017 2nd International Con- ECG estimation,” Electronics Letters, vol. 54, no. 6, pp. 340– ference on Advanced Robotics and Mechatronics (ICARM), 342, 2018. pp. 535–539, Hefei, China, August 2017. [34] S. Bag and G. Sanyal, “An efficient face recognition approach [19] C. Yue, H. Cheng, Y. Chen, Q. Deng, and X. Lin, “Design of a using PCA and minimum distance classifier,” in 2011 Interna- wearable sensing system for a lower extremity exoskeleton,” in tional Conference on Image Information Processing, pp. 1–6, 2017 IEEE International Conference on Mechatronics and Auto- Shimla, India, November 2011. mation (ICMA), pp. 464–468, Takamatsu, Japan, August 2017. [20] S. Yan, W. Ren, X. Liang, and K. Zhang, “Gait characteristics of overweight and obese children with different ages,” in 2014 7th International Journal of Advances in Rotating Machinery Multimedia Journal of The Scientific Journal of Engineering World Journal Sensors Hindawi Hindawi Publishing Corporation Hindawi Hindawi Hindawi Hindawi www.hindawi.com Volume 2018 http://www www.hindawi.com .hindawi.com V Volume 2018 olume 2013 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 Journal of Control Science and Engineering Advances in Civil Engineering Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 Submit your manuscripts at www.hindawi.com Journal of Journal of Electrical and Computer Robotics Engineering Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 VLSI Design Advances in OptoElectronics International Journal of Modelling & Aerospace International Journal of Simulation Navigation and in Engineering Engineering Observation Hindawi Hindawi Hindawi Hindawi Volume 2018 Volume 2018 Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com www.hindawi.com www.hindawi.com Volume 2018 International Journal of Active and Passive International Journal of Antennas and Advances in Chemical Engineering Propagation Electronic Components Shock and Vibration Acoustics and Vibration Hindawi Hindawi Hindawi Hindawi Hindawi www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018 www.hindawi.com Volume 2018

Journal

Applied Bionics and Biomechanics – Hindawi Publishing Corporation

Published: Sep 18, 2018

There are no references for this article.