Downloaded from http://journals.lww.com/jpojournal by BhDMf5ePHKbH4TTImqenVIhz0dzqyKWSaAVAhdpJ64OO96NUjl2ApMegP2AeZRsbDicKmzyfL8Q= on 10/09/2020 ORIGINAL RESEARCH ARTICLE Application of Novel Graphite Flex Sensors in Closed-Loop Angle Feedback on a Soft Robotic Glove for Stroke Rehabilitation Aaron Jing-Yuan Goh, BEng, Hong-Kai Yap, PhD, Gokula Krishnan Ramachandran, PhD, Chen-Hua Yeow, PhD ABSTRACT Introduction: Stroke survivors require physiotherapy and rehabilitation programs to restore their hand function so that they can carry out activities of daily living (ADLs) independently. Soft robotic gloves are designed to assist in these rehabilitation pro- grams and reduce manpower costs, but their pressure-activated actuation mechanisms require closed-loop position feedback to allow for finer motor coordination for the hands, thereby improving hand rehabilitation for patients who had stroke. We present a novel design of graphite-based flex sensors that we implemented in a soft robotic glove to evaluate its performance in closed-loop metacarpophalangeal (MCP) joint angle feedback. Materials and Methods: The graphite-based flex sensors are embedded into a sensor glove and characterized in terms of baseline stability and drift over 20 continuous loading cycles per trial for five times. Curve-fitting using both linear and nonlinear equa- tions was done to determine the relationship between resistance and MCP joint angle, using Vicon MX motion capture system to obtain 3D coordinates and joint angles, as well as separate Arduino circuitry to obtain signal voltage samples. Pneumatic pressures are regulated using proportional-integral-derivative (PID) control, with a safety factor (SF) of 1.2. Two con- trol algorithms were developed to make use of angular feedback to control set point pressures: 1) Intent Recognition Mode makes use of a single MCP angle threshold at 50° to activate a maximum output pressure was set at 100 kPa (83.33 kPa after SF); and 2) Fixed Interval Assist Mode makes use of different MCP joint angle values (30°, 45°, 60°, and 90°) to derive correspond- ing set point pressures set at 25, 50, 75, and 100 kPa (20.83, 41.67, 62.50, and 83.33 kPa after SF). Results: Nonlinear equations consistently provided a reasonably better fit as compared with the linear equation fit. However, in this work, the linear MCP joint angle models are preferred as a calibration method, because nonlinear equations are hard to im- plement in control algorithms in practice. PID control for Intent Recognition activates and deactivates at approximately 18% and 95% of each full flexion-extension exercise cycle progression, respectively. For Fixed Interval Assist Mode, thumb MCP joint angle feedback is less repeatable compared with that of the other fingers in the same experiment, possibly because of the diffi- culty in placement of the sensor at the thumb MCP joint, close proximity to other sensors, and physiological crosstalk between the fingers. Conclusions: This work has presented a novel integration and implementation of graphite-based flex sensors with a soft robotic glove for stroke rehabilitation. The relationship between the signal voltage and the MCP joint angle varies greatly with anatom- ical differences between each individual, and with sensor placement. However, based on the experimental results, a linear map- ping calibration algorithm for the graphite-based flex sensors was implemented, which also complements its robustness for the potential application on stroke rehabilitation. The effectiveness of the calibration algorithm is also thus demonstrated via the Intent Recognition and Fixed Angle Assist control algorithms. (J Prosthet Orthot. 2020;32:272–285) KEY INDEXING TERMS: sensors, soft robotics, flex sensors, rehabilitation, negative feedback, PID S troke is a highly debilitating disease that occurs when there is neural tissue damage in the brain. 1,2 This can happen in two ways—ischemic and hemorrhagic stroke. In either of these cases, the disease usually results in patient death or hemiplegia, where the patient is only able to move one side of the body and the other side is left much weaker and having much difficulty in carrying out various activities of daily living (ADLs). The affected activities include normal gait and movement of the hands to grasp a cup, for instance, among other basic needs such as maintaining personal hygiene and work activities such as writing and typing. 1,2 This in turn re- duces the stroke survivor's quality of life (QOL) 3 significantly, and rehabilitation is usually required to restore his or her phys- ical independence in performing these ADLs. AARON JING-YUAN GOH, BEng; HONG-KAI YAP, PhD; GOKULA KRISHNAN RAMACHANDRAN, PhD; and CHEN-HUA YEOW, PhD, De- partment of Biomedical Engineering, National University of Singapore, Singapore. Disclosure: The authors declare no conflict of interest. This work has been done with NUS MOE ARC Tier 2 Funding, and the healthy human subject trials are conducted with the approval of the National University of Singapore Institutional Review Board (NUS-IRB reference number B-16-050). Copyright © 2020 American Academy of Orthotists and Prosthetists. Correspondence to: Chen-Hua Yeow, PhD, Department of Biomedical Engineering, National University of Singapore, Blk E6, 5 Engineering Dr 1, Level 7, Singapore 117608; email: bieych@nus.edu.sg 272 Volume 32 • Number 4 • 2020 Copyright © 2020 by the American Academy of Orthotists and Prosthetists. Unauthorized reproduction of this article is prohibited.