A rigid-flexible coupled elbow rehabilitation monitoring system based on neural network fusion
Year: 2026
Type: article
Abstract: Low-cost stretchable sensors offer promising solutions for wearable joint angle monitoring in home rehabilitation but suffer from viscoelastic hysteresis and relaxation drift. This paper proposes a wearable elbow angle monitoring system achieving high accuracy using only resistive stretch sensors. The system employs a dual-sensor antagonistic configuration and an LSTM network for hysteresis compensation. The antagonistic layout includes a pre-strained inner sensor to eliminate signal dead zones near full extension. A temporary resistive bend sensor provides dynamic ground truth during a one-time calibration phase and is subsequently removed for daily deployment. The neural network learns the history-dependent mapping from corrupted signals to compensated angle estimates. Experimental results demonstrate that the proposed system reduces the hysteresis loop width by 63.8\% and stabilizes estimates despite raw resistance drift. Evaluating random rehabilitation movements yields an RMSE of 6.2$^\circ$ and an $R^2$ of 0.95, representing a 42.0\% accuracy improvement over baseline linear regression. The achieved accuracy satisfies clinical requirements for home rehabilitation and validates the system as a comfortable and cost-effective alternative to rigid motion capture technologies.
Authors: Sufeng Hu, Aiguo Song
Language: English
Subfield: Biomedical Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: bronze
Funder: Government of Jiangsu Province
OpenAlex ID: W7234196536
