A Real-Time Control Method for Upper Limb Exoskeleton Based on Active Torque Prediction Model

• Published in: Bioengineering (Basel) Vol. 10; no. 12; p. 1441
• Main Authors: Li, Sujiao, Zhang, Lei, Meng, Qiaoling, Yu, Hongliu
• Format: Journal Article
• Language: English
• Published: Switzerland MDPI AG 01.12.2023
• Subjects: Active control; active torque prediction model; Activities of daily living; Adaptability; Analysis; Back propagation networks; Bioengineering; Care and treatment; Control methods; Design; Diagnosis; Elbow; Electromyography; Embedded systems; Exoskeleton; Exoskeletons; Fitness training programs; Health aspects; Hemiplegia; Limbs; Medical research; Model accuracy; Muscles; Neural networks; Patient outcomes; Patients; Physical therapy; Prediction models; Prevention; Real time; real-time control; Rehabilitation; rehabilitation exoskeleton; Rehabilitation robots; Risk factors; Robotics; Robots; sEMG; Stroke; Stroke (Disease); Therapeutics, Physiological; Torque; China; active torque prediction model; sEMG; real-time control; rehabilitation exoskeleton
• ISSN: 2306-5354; 2306-5354
• DOI: 10.3390/bioengineering10121441

Exoskeleton rehabilitation robots have been widely used in the rehabilitation treatment of stroke patients. Clinical studies confirmed that rehabilitation training with active movement intentions could improve the effectiveness of rehabilitation treatment significantly. This research proposes a real-time control method for an upper limb exoskeleton based on the active torque prediction model. To fulfill the goal of individualized and precise rehabilitation, this method has an adjustable parameter assist ratio that can change the strength of the assist torque under the same conditions. In this study, upper limb muscles' EMG signals and elbow angle were chosen as the sources of control signals. The active torque prediction model was then trained using a BP neural network after appropriately extracting features. The model exhibited good accuracy on PC and embedded systems, according to the experimental results. In the embedded system, the of this model was 0.1956 N·m and 94.98%. In addition, the proposed real-time control system also had an extremely low delay of only 40 ms, which would significantly increase the adaptability of human-computer interactions.