Garment-Integrated Wearable Sensing for Knee Joint Monitoring

Transcription

1 Garment-Integrated Wearable Sensing for Knee Joint Monitoring Guido Gioberto University of Minnesota Ph.D. Candidate Computer Science and Engineering Wearable Technology Lab 340 McNeal Hall 1985 Buford Avenue St. Paul, MN Abstract Body monitoring is one of the most intuitive and direct applications for technologies that are wearable. Wearable devices are capable of detecting body movements using wearable sensors, and using signals to capture anomalies as well as good patterns in our daily activities. Clothes provide the most accessible platform for embedding sensors and electronic components, preserving imperceptibility and user Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. Copyrights for third-party components of this work must be honored. For all other uses, contact the Owner/Author. Copyright is held by the owner/author(s). ISWC'14 Adjunct, September 13 17, 2014, Seattle, WA, USA ACM /14/09. comfort, especially for long term body monitoring applications. Both perceptibility and comfort variables are associated with the willingness of the user to wear the device and with the quality of the data captured (that should reflect the unbiased wearer activity as if the user would not be wearing any device). Author Keywords E-textile; wearable technology; smart clothing; wearable sensor; stitched sensor. ACM Classification Keywords H.1.2. User/Machine Systems: Human factors; I.2.9. Robotics: Sensors. Introduction The Theory and Fundamentals of the sensing approach, as well as the specifics of the garment-integrated sensors used to build my body sensing application have already been presented in [1]. A novel garment movement approach allowed characterization of the garment noise distribution on a pair of pants for waist, hip, thigh, calf, and ankle [2]; and quantification of errors directly related to garment wearability and body movement [3]. Truly garment-integrated sensors made out of apparel material (i.e., a conductive yarn) feeling and looking like normal stitches have also been presented [4] and sensitivity analyzed [5]. 113

2 The work presented here is the application of the novel garment-related variables and sensing approach to find the best correlation between garment-integrated sensor signals and joint movements, specifically expressed in terms of the knee bend angle error. Background Traditional inertial systems to measure joint angle typically rely on accelerometers, to measure inclination; gyroscopes, to measure rotation speed; and magnetometers, to measure azimuth respect to horizontal reference. Errors were found to be less than 3deg for flexion/extension and abduction/adduction, and up to 9deg for internal/external rotation [8]. To translate this approach into a more wearable form factor, an array of equidistant conductive Fiber Sensors is integrated into flexible, skin-tight fabrics surrounding the knee joint. Resistance changes across these conductive fibers are used to measure angles through the use of a non-linear predictor after an initial, onetime calibration. During motion, sensors can move and shift over the body surface, and calibration coefficients are dynamically updated by minimizing the average squared error between each sensor s estimate and the average estimate with respect to the sensors offset (self-calibration). The comparison between a reference goniometer and the angle estimated from the sensor response showed an error of 5.4deg for the linear predictor and of 3.2deg for the quadratic predictor [7]. Two flex sensors have been integrated on the dorsal side of a knee brace like skin-tight textile to develop a wearable system to monitor joint flexion. The change in resistivity of the flex sensors with the knee bend showed a correlation of 0.98, and compared to a goniometer an average bend prediction error of 6.92deg, after fusing the two sensors with Extended Kalman Filter (EKF) [9]. Gemperle et al. discussed how a truly wearable device should always be wearable during body motion [6]. Keeping minimal the bulkiness of a wearable device, maximizing comfort, ease of use and application of the wearable device are all variables addressed and optimized in my work with the final outcome to increase willingness and quality of use. Stitched Stretch Sensors Stretch sensing through stitched stretch sensors is our novel method. Three types of stitches commonly used in apparel have been studied: Bottom Thread Coverstitched sensor, with an almost linear inversely proportional analog output response; Top Thread Coverstitched sensor, which output is a proportional linear response that saturates at ~15% stretch [4]; and Overlock Stitched sensor, different than the previous two because it must be integrated on textile edges, which response also saturates like the Top Thread at similar stretch percentage but with smaller resistance amplitude than the Top Thread Coverstitched sensor depending on the inner needle tension on the upper looper [10]. Figure 1 shows the three stitched stretch sensors responses of 3 long sensors when repeatedly stretched and relaxed for about 25% of the sensor length with an INSTRON tensile machine. 114

3 SESSION: DOCTORAL SCHOOL equidistant 6 long sensors spaced about.5 apart were integrated on the front side of the pants across knee numbering the sensors from inner side to outer side of the knee as indicated in figure 2, to absorb the pants stretch. Figure 1. Stitched Stretch Sensor responses of 3 long sensors for 25% elongation (.75 ) Both Top Thread and Overlock Stitched sensors exhibit almost immediately saturation, while the Bottom Thread sensor keeps decreasing its response until eventually the stitch and/or fabric breaks due to excessive stretch. This makes the Bottom Thread Coverstitched sensor more suited to sense more and larger stretches, for the same stitch length. Knee Bend Sensing Consistent Knee Bends were evaluated using an animatronic mannequin (Cyberquins, UK) wearing a pair of skin-tight stretchable pants. The running mannequin repeats over time the same basic running gait cycle, without completely extending his leg with maximum extinction and flexion of ~ deg, measured with a goniometer. At maximum extension we were able to locate the maximum pant stretch on the very front of the knee equal to about 2. Thus, five Figure 2. Bottom Thread Coverstitched sensors stitched on the frontal side of the left knee of a cyber mannequin. Using a DMM for consecutive gait cycles and recoding each sensor separately, we found that sensors located far from the region of maximum stretch on the very front of the knee showed poor output responses, as showed in figure

4 sensor in the middle is used to build the model by solving the curve fitting least square problem to map sensor resistance output to true bend angle estimated with the function angle between three points of a VICON system. The model is then used to predict the other two sensors on the sides. A more practical application would probably use a printed board capable to capture the three sensors at the same time, scarifying however the calibration precision of the DMM. Here we make the assumption that the pants are not moving much so that between consecutive trials pants are located in the same position, as we were also able to validate experimentally. Figure 3. Bottom Thread Coverstitched Stretch sensor integrated on the frontal side of the mannequin knee Therefore we refer our knee bend angle prediction analysis to the sensor located on the very front of the knee, in the region of maximum extension. To approximately capture the full maximum stretch region, two additional sensors are integrated ¼ apart on the left and right side of the sensor in the very middle. This is not just a simplification of the problem. Apparel manufacturing techniques used to integrate our garment-integrated stitched stretch sensors are affected by many imprecision variables such as sensor placement, stitch length or stitch structure, as we discuss in [11]. Given that the pants tested are skintight and do not move significantly during the mannequin run, considering the sensors located on the very front is a meaningful choice. Results The three front sensors responses are recorded for 11 consecutive cycles. Removing from each response the offset given by the sensor output value for full knee extension we obtain the sensors response is showed in figure 4. The three sensors spaced ¼ apart on the every front of the knee were capture by 3 consecutive trials. The 116

5 SESSION: DOCTORAL SCHOOL Figure 4. Model and maximum stretch region sensors responses Using the raw data of Figure 4 for the model to predict the knee bend angle from lateral sensors responses, we evaluated the errors in Table 1. Trials no. ¼ Left Sensor ¼ Right Sensor Table 1. Knee bend prediction error Angle Estimation Error Flexion = ; Extension = 8.68 Flexion = 7.04 ; Extension = 6.31 Angle Prediction Error Analysis The angle prediction errors all combined results in an average prediction error of ~8.39deg compared to the true reference angle calculated with VICON for each trial. The error appears larger to the sensor on the left, while the other two sensors were occupying a more central position. The less practical case of normalized resistance, where the sensors outputs are normalized by the maximum resistance value recorded over the 11 gait cycles, brings down the average prediction error to 5.95deg. A smaller value of the prediction error is also evaluated is the normalization is done on a single gait cycle basis. Open Questions The work the still needs to be done prior completion of my PhD work can be summarized in the following bullet points: Refinement of Stretchable Pants for knee joint sensing, with three identical sensors on the region of maximum stretch across the knee that will be captured at the same time with a printed board Test of refined stretchable pants for different pants sizes, and knee bend prediction error comparison between the different sizes Human Stretch Testing Analysis Comparison between Human and Mannequin Stretch Sensing Tests As soon as the refined stretchable pants are available, I will test them with a 3-channels/3-amplifiers board already build and available in the lab to test the pants on the mannequin, using algorithms that I have already developed and tested to sense and correlate the three sensors responses but for pants affected by sensors and fabrication imperfections [11]. I will then compare the correlation results for the different sizes. To complete my work, I will re-run the same tests but on a human, and finally compare human and mannequin testing results. Aim and Objective Focus of this thesis work is to provide an alternative approach for body monitoring through truly wearable sensors that preserve not only perceptibility of others but also perceptibility of the user him/herself, together with comfort related user variables. Specifically, the goal is here to develop a novel garment-integrated body sensing method to characterize knee angle, to measure the accuracy of the garment-integrated sensing method, and to contextualize and validate this accuracy by comparison to the accuracy requirements of benchmark wearable sensing applications. 117

6 Acknowledgements This work was supported by the National Science Foundation under grant number IIS A very special thank to the University of Minnesota, Dr. Lucy Dunne, and all Wearable Technology Lab collaborators. References [1] Gioberto, G. Measuring Joint Movement Through Garment-Integrated Wearable Sensing. Doctoral School, UbiComp'13 Adjunct, 2013, Zurich, Switzerland [2] Dunne, L.E., Gioberto, G., Koo H. A Method of Measuring Garment Movement Error for Wearable Sensing. Proc. of the 13th IEEE International Symposium on Wearable Computers, 2011 San Francisco, CA [3] Gioberto, G., Dunne, L.E. Garment Positioning and Drift in Garment-Integrated Wearable Sensing. Proc. of the 16th IEEE International Symposium on Wearable Computers, 2012, Newcastle, UK [4] Gioberto, G., Dunne, L.E. Theory and Characterization of a Top-Thread Coverstitched Stretch Sensor. Proc. of 2012 IEEE International Conference on System, Man, and Cybernetic, 2012 Seoul, Korea [5] Gioberto, G., Coughlin, J., Bibeau, K, Dunne, L.E., Detecting Bends and Fabric Folds using Stitched Sensors. Proc. of the 17th IEEE International Symposium on Wearable Computers, 2013, Zurich, CH [6] Gemperle, F., Kasabach, C., Stivoric, J., Bauer, M., Martin, R. Design for Wearability. Wearable Computers, Digest of Papers. Second International Symposium on, Oct. 1998, Pittsburgh, PA, USA [7] Gibbs, P.T., Asada, H.H. Wearable Conductive Fiber Sensor for Multi-Axis Human Joint Angle Measurements. Journal of NeuroEngineering and Rehabilitation, March 2005, 2:7 [8] Favre, J., Jolles, B.M., Aissaoui, R., Aminian, K. Ambulatory measurement of 3D knee joint angle. Journal of Biomechanics, 41, 2008, [9] Bakhshi, S., Mahoor, M.H. Development of a Wearable Sensor System for Measuring Body Joint Flexion. Body Sensor Networks, 2011 International Conference on, May 2011, 35-4-, Dalla, TX, USA [10] Gioberto, G., Dunne, L.E. Overlock-Stitched Stretch Sensors: Characterization and Effect of Fabric Property. Journal of Textiles and Apparel Technology and Management (JTATM), 2013, Vol 8 No 3 [11] Gioberto, G., Min, C.H., Compton, C., Dunne, L.E. Garment Positioning and Drift in Garment-Integrated Wearable Sensing. Proc. of the 18th IEEE International Symposium on Wearable Computers, 2014, Seattle, WA 118