Neural network-based adaptive command filtering control for pneumatic artificial muscle robots with input uncertainties. (January 2022)

• Record Type: Journal Article
• Authors: Liu, Gendi; Sun, Ning; Liang, Dingkun; Chen, Yiheng; Yang, Tong; Fang, Yongchun
• Is Part Of: Control engineering practice. Volume 118(2022)
• Publication Date: 2022-01
• Subjects: Pneumatic artificial muscle (PAM) -- Neural network (NN) -- Adaptive control -- Command filtering -- Dead zone -- Saturation; Automatic control -- Periodicals; 629.89
• DOI: 10.1016/j.conengprac.2021.104960
• Languages: English
• ISSNs: 0967-0661

Abstract

Abstract: Due to such advantages as large output forces, variable stiffness, and strong safety, flexible pneumatic artificial muscles (PAMs) have been widely used in important fields, such as medical rehabilitation training and military exoskeleton assist. However, their complex hysteresis, creep, input uncertainties (caused by air pressure thresholds and unidirectional saturations), and sensitivity to external noises, etc., lead to difficulties in accurate modeling, parameter identification, and nonlinear control of PAM robots. Aiming at these problems, this paper designs an adaptive command filtering control method based on neural networks, which realizes satisfactory tracking control of a dual-PAM arm robot. Specifically, by introducing a barrier term, the designed feedback controller actuates tracking errors to converge to the neighborhoods of zero, and always limits the tracking errors within the desired bounds. Furthermore, a Lyapunov function is chosen to prove the stability of the closed-loop system. Compared with most of existing methods, this paper gives the first continuous controller to simultaneously deal with unmodeled dynamics, parametric uncertainties, and multiple input constraints of PAM robots with only measurable outputs being required. In the case, differential noises can be effectively suppressed, which is pretty beneficial to the control of PAM robots that are driven by highly compressed air. Finally, the feasibility and robustness of the proposed method are validated by a series of hardware experiments on a self-built PAM humanoid arm robot testbed. Highlights: Adaptive neural network control for nonlinear PAM robots with input uncertainties. Command filtering control to handle differential noises. Ensured convergence without violating the desired constraints of tracking errors. Rigorous proof of stability based on Lyapunov techniques. Validity demonstration of hardware experimental results of designed controllers.