Abstract
This paper presents a novel lightweight fabric strain sensor array specifically designed for comprehensive knee joint monitoring. The sensor system features a unique two-layer design incorporating eight strategically positioned sensing elements, enabling effective spatial mapping of strain distribution across the knee during movement. This configuration offers advantages in capturing complex multi-axis kinematics (flexion/extension, rotation) and localized tissue deformation when compared to simpler sensor layouts. To evaluate the system, ten subjects performed three distinct activities (seated leg raise, standing, walking), generating resistance data from the sensors. A hybrid deep learning model (CNN + BiLSTM + Attention) processed the data and significantly improved performance to 95%. This enhanced accuracy is attributed to the model's ability to extract spatial-temporal features and leverage long-term dependencies within the time-series sensor data. Furthermore, channel attention analysis within the deep learning model identified sensors 2, 4, and 6 as major contributors to classification performance. The results demonstrate the feasibility of the proposed fabric sensor array for accurately recognizing fundamental knee movements. Despite limitations in the diversity of postures, this system holds significant promise for future applications in rehabilitation monitoring, sports science analytics, and personalized healthcare within the medical and athletic domains.