Fabrication and Characterization of a Soft and Stretchable Capacitive Strain Sensor for Hand Gesture Recognition

• Published in: IEEE sensors journal Vol. 25; no. 1; pp. 601 - 612
• Main Authors: Tchantchane, Rayane, Zhou, Hao, Zhang, Shen, Alici, Gursel
• Format: Journal Article
• Language: English
• Published: New York IEEE 01.01.2025; The Institute of Electrical and Electronics Engineers, Inc. (IEEE)
• Subjects: Automation; Bending; Capacitive sensors; Capacitive strain sensor (CSS); Cyclic loads; Deformation wear; Discriminant analysis; Fingers; Gesture recognition; Hand (anatomy); hand gesture recognition (HGR); Hands; human-machine Interface (HMI); Hysteresis; Industrial robots; Machine learning; machine learning (ML); Man-machine interfaces; Multi wall carbon nanotubes; Real time; Robot sensing systems; Robotics; Sensitivity; Sensor phenomena and characterization; Sensors; Soft robotics; Strain; Stretchability; Wearable computers; Wearable sensors; Wearable technology
• ISSN: 1530-437X; 1558-1748
• DOI: 10.1109/JSEN.2024.3493160

In line with recent progress in soft robotics, human-machine interfaces (HMIs), and wearable sensors, there has been an increasing need for flexible and stretchable strain sensors, especially high-performance and low-cost capacitive strain-based sensors. Our sensor, based on a multiwalled carbon nanotube (MWCNT)/Ecoflex composite, conforms to curved and irregular surfaces to detect and respond to mechanical deformations, including tensile and bending modes while maintaining exceptional flexibility, stretchability (230%), and comfort without interfering with hand movements. It exhibits a low hysteresis (maximum hysteresis error of <inline-formula> <tex-math notation="LaTeX">\le 2.5 </tex-math></inline-formula>%), high sensitivity characterized by a gauge factor (GF) of 0.80 at 100% elongation, and 0.147 at 90° bending. The sensor also demonstrates durability under cyclic loads, enduring over 1000 bending cycles. By employing different machine learning (ML) classifiers, including random forest (RF), linear discriminant analysis (LDA), and logistic regression (LR), the strain sensor can recognize various finger angles from five subjects with accuracies of 99%, 98%, and 97%, respectively, demonstrating its promising applications in enhancing human-machine interactions and wearable technology, paving the way for future research on flexible sensing systems capable of real-time, precise gesture recognition.