Control Method of Lower Limb Rehabilitation Robot Based on Nonlinear Time Series Prediction Model and Sensor Technology

• Published in: IEEE access 2024, Vol.12, p.152532-152543
• Main Authors: Xing, Yanli, Jiao, Weiwei
• Format: Article
• Language: English
• Subjects: Adaptation models; Assistive robots; Brain modeling; Compensation; Computational modeling; Control methods; Gait; Hip; Hip joint; Inertial sensing devices; Joints (anatomy); Knee; lower limb rehabilitation; Neural networks; Nonlinear control; Nonlinear time series prediction; Patient rehabilitation; Prediction models; Predictive control; Predictive models; Radial basis function; radial basis function neural network; Rehabilitation; Rehabilitation robots; Robot control; Robot dynamics; Robot sensing systems; Robots; sensor technology; Solid modeling; Time series analysis; Tracking control; Tracking errors; Training; Trajectories; Trajectory
• ISSN: 2169-3536; 2169-3536
• DOI: 10.1109/ACCESS.2024.3480252

With the application of intelligent technology in the medical field, patient rehabilitation training is gradually developing towards intelligence. Aiming at the lower limb rehabilitation robots, a lower limb gait curve prediction model based on nonlinear auto-regressive neural network and a compensation control model for rehabilitation robots based on radial basis function neural network are constructed. Firstly, inertial sensors are used to collect upper limb motion data. A nonlinear auto-regressive neural network is used to build a lower limb gait curve prediction model. Then, to design trajectory tracking control for the rehabilitation robot, a radial basis function neural network is applied to build a compensation control model for the rehabilitation robot. The results indicated that the prediction results of the lower limb gait curve prediction model were relatively close to the actual situation. The error between the predicted curve of the hip joint and the actual curve was within 7°. The prediction curve error of the knee joint was within 10°. After adding radial basis function neural network compensation, the motion trajectory and velocity tracking error of the hip joint were significantly reduced. The tracking error remained stable at around 0.007rad and 0.003rad/s, respectively. The motion trajectory and velocity tracking error of the knee joint were stable at around 0.011rad and 0.008rad/s, respectively. According to the research results, the constructed model has good application effects, which is beneficial for improving the rehabilitation effect of lower limb rehabilitation robots on patients.