A smart neuromuscular stimulator for autonomous upper-limb rehabilitation

Neuromuscular Electrical Stimulation (NMES) is widely applied in rehabilitation therapy to restore muscle function and prevent atrophy. However, conventional NMES systems typically operate in open-loop configurations with fixed stimulation parameters, often leading to rapid muscle fatigue and reduced therapeutic efficiency. This study presents the design and prototype-level validation of a fatigue-aware closed-loop NMES system integrating an MPU6050 inertial measurement unit (IMU) and embedded machine learning for adaptive stimulation control.
The proposed system utilizes real-time tri-axial acceleration data to monitor contraction dynamics during electrically induced muscle activation. Time-domain features extracted from IMU signals were processed using a Random Forest regression model deployed on an ESP32 microcontroller to estimate fatigue progression. Based on predicted fatigue levels, the system dynamically adjusted pulse width through a closed-loop control algorithm to maintain contraction stability.
Experimental validation demonstrated that IMU-derived kinematic signals exhibited progressive amplitude reduction and increased variability during sustained stimulation, consistent with fatigue development. The embedded machine learning model achieved stable real-time inference performance, enabling adaptive pulse-width modulation without computational instability. Compared to open-loop operation, the closed-loop configuration maintained contraction amplitude within a narrower functional range and reduced rapid degradation during prolonged stimulation.
The results confirm the feasibility of using motion-based sensing as a non-invasive alternative to electromyography for fatigue-aware NMES control at prototype level. The integration of low-cost IMU sensing and embedded machine learning demonstrates the potential for intelligent, portable, and adaptive rehabilitation devices.