Study On The Motion Trajectory Tracking Control Of The Pneumatic Muscle Actuated Servo System

Pneumatic artificial muscle(PAM)is a humanoid pneumatic actuator to provide contraction force,which inherits the advantages of high power to weight ratio,flexibility,and cleanliness.It has widespread prospects in the fields of medical rehabilitation,aerospace as well as emergency relief.However,owing to characteristics of high nonlinearity,time-varying,hysteresis,and creep,the PAM actuated servo system's tracking performance cannot meet the demands of complicated tasks.In this dissertation,the most popular single PAM actuated servo system is considered.Aiming at the actual tracking performance and uncertainty compensation,the PAM nonlinear model,adaptive control,sliding mode/robust control,hysteresis compensation,and neural network adaptive control are investigated in turn.This dissertation is organized as follows:In chapter 1,the objective and significance of this project are illustrated,and the literature associated with PAM model as well as the control schemes of PAM actuated servo systems are reviewed.In chapter 2,regarding the PAM contraction force,the force-displacement hysteresis is found to be asymmetry,with nonlocal memory,weak correlation of high internal pressure,and quasi rate-independent.Then,six Prandtl-Ishlinskii(PI)and Bouc-Wen(BW)hysteresis models are developed,among the rest,the modified PI+Dead-zone model can predict the hysteresis best with the absolute mean error less than 1N and the mean variance less than 1.5N under each contraction condition.Moreover,with the development of the pressure dynamic,the PAM nonlinear model with hysteresis force is developed.In chapter 3,a direct adaptive robust controller(DARC)and an integrated direct indirect adaptive robust controller(DIARC)are designed respectively based on the state-space model of the single PAM actuated servo system.The parametric uncertainty is compensated by the adaptive part,and the remaining uncertainty is attenuated by the robust feedback further,then,prescribed transient and steady-state performance can be guaranteed.Moreover,inherence effects of the tuning function based adaptive law,least square based adaptive law,and fast dynamic compensation are discussed along with the preliminaries of linear system theory.Comparative experiments demonstrate the well transient and steady performance of these two controllers,especially the steady-state root-mean-square error of DIARC is less than 0.21 mm during tracking sine signal under the short stroke.In chapter 4,an adaptive robust controller with nonlocal memory hysteresis compensation(DIARC+Hys)is developed for the single PAM actuated servo system.The classical PI model is employed in the development of the robust control law with hysteresis feedback linearization compensation.Then,model parameters including operator weights are updated online by the recursive least square estimation,hence,the parametric uncertainty and the hysteresis non-local memory characteristic are attenuated.Moreover,the problem of unbounded uncertain nonlinearities introduced by the PI hysteresis term is addressed by applying online monitoring so that the tracking error is guaranteed to converge to a small residual set.Comparative experimental results demonstrate the significance of the hysteresis compensation,especially,the steady-state root-mean-square tracking error of DIARC+Hys is less than 0.82 mm during tracking sine signal under the large stroke,which is decreased by 45% compared to the similar controller without hysteresis compensation.Furthermore,the non-local memory characteristic is proved to be attenuated by means of the operator weights' adaption.In chapter 5,the neural network is introduced to the previous adaptive robust controller for the purpose of attenuating the unmodeled nonlinearity.These unmodeled errors are divided into the inherent and repeatable unknown nonlinearity and the nonrepeatable nonlinearity.Then,RBF neural network and BP neural network based adaptive robust controller are developed respectively by means of the neural network's function approximation for the unknown nonlinearity,the former is implemented with hidden-output layer weights adaption,and the latter is implemented with input-hidden and hidden-output weights adaption.Comparative experiments demonstrate the steadystate root-mean-square tracking error of the BP neural network based adaptive robust controller is less than 0.14mm(short stroke)and 0.74mm(large stroke)during tracking sine signal,which is decreased by 33% and 50% respectively compared to the DIARC,which reveals the compensation effects of the neural network estimation for the unknow nonlinearity.In chapter 6,the main conclusions of this dissertation and the future work for the PAM actuated servo system are described.