【作者】 张颖；

【导师】 郝丽娜；

【作者基本信息】 东北大学 ， 机械电子工程， 2020， 博士

【摘要】 随着计算机、材料科学、控制等学科的发展进步,机器人的应用范围不断拓宽,具有高度柔顺性和连续变形能力的软体机器人不仅可以适应非结构化环境需求,而且具备更安全的人机交互特性,成为目前机器人方向的研究热点之一。在康复领域,软体机器人可辅助残疾患者完成日常活动;在野外探险、救灾领域,软体机器人可适应各种繁杂环境,承当勘探、侦察等工作;在医疗手术领域,软体机器人能够与生物组织安全兼容,有利于外科微创手术的实现。作为一种柔性驱动器,气动人工肌肉(pneumatic artificial muscle,PAM)因制作简单、功率密度比高、输出力大、可产生大范围变形等优点在众多软体机器人驱动器中脱颖而出。然而,PAM驱动器具有复杂的静态迟滞行为和输入频率相关的动态迟滞行为,不仅会降低驱动系统运动精度,甚至造成系统振荡。同时,由于PAM的非线性大变形,使得由其驱动的软体机器人存在空间大挠度形变耦合,以及气体动力耦合。上述因素对由PAM驱动的软体机器人的大变形致动分析及精准运动控制造成了很大的挑战。因此,本论文深入研究了 PAM驱动器及基于PAM驱动的软体机械臂的建模及控制理论与方法,从系统建模方法、控制器的设计和实验验证三个层面展开,主要研究内容如下:(1)针对现有PAM静态迟滞唯象模型泛化性不足的问题,建立了幅值相关的非对称静态迟滞模型和基于此模型的直接逆迟滞补偿器,并通过和PID控制器复合,研究了复合PID控制方法,从而有效提升PAM驱动器的控制精度。将非线性输入函数引入到迟滞算子的同时,利用卷积神经网络(CNN)良好的特征提取能力和学习能力,构造一种基于迟滞算子和CNN的深度学习网络结构,即增强型扩展非平行 PI(extended unparallel PI model,EUPI)模型(即 D-EUPI 模型),解决了传统算子迟滞模型泛化能力不足的问题,并验证了所提模型的泛化性和准确性。在此基础上,直接应用该模型描述PAM驱动器的逆迟滞曲线,构建直接逆静态迟滞补偿器(I-DC),将其和PID控制器并联,对PAM驱动器进行迟滞补偿控制。实验结果证明了 I-DC迟滞补偿控制器的可行性。(2)针对静态迟滞模型难以捕捉PAM输入频率和负载变化信息,造成无法准确描述PAM动态迟滞行为的问题。提出了完整表征PAM驱动器动态输出的负载相关的综合动态迟滞模型(E-RFNARMAX模型)和基于E-RFNARMAX逆模型的动态迟滞补偿控制方法。从分析输入频率、外部负载和PAM迟滞因素对PAM驱动特性的综合影响入手,通过迟滞子模型和负载相关的动态子模型串联,构成E-RFNARMAX模型。同时设计了动态迟滞补偿器(I-RC),对PAM进行开环补偿控制理论与实验研究。实验结果证明了所提的综合动态迟滞模型可以精确地表征PAM的动态驱动特性。同时,该建模方法的提出有利于提升PAM负载相关的动态迟滞开环补偿控制精度,无需复杂的反馈控制器。此外,本文第一次提出了基于E-RFNARMAX逆模型的负载预测方法,能够有效预测驱动过程中PAM末端的负载变化,为柔性驱动器的驱动感知一体化设计提供了新的思路。(3)介绍了基于伸长型PAM和收缩型PAM并联混合驱动的气动软体机械臂系统,该机械臂系统具有变刚度特点,可以实现空间位置和刚度的解耦。由于几何结构的复杂性和材料超弹性,软体机械臂存在大挠度变形耦合,决定其位姿变化。针对上述问题,本文对PAM驱动软体机械臂进行了大变形机理分析,计算了静平衡条件下软体机械臂的弯曲位姿。首先通过分析PAM的材料弹性及几何非线性,提出了改进的PAM输出力模型。基于PAM改进输出力模型和连续介质力学理论,完整地分析了 PAM的几何非线性、材料非线性、充气方式、负载与重力对软体机械臂不同方向大挠度变形影响,以及变形与机械臂位姿关系模型。实验结果表明,本文所提出的模型能够准确预测软体机械臂在不同工作条件下的弯曲位姿,同时,系统地、定量地研究了该机械臂的运动特性和刚度特性。(4)针对软体机械臂系统的多输入多输出耦合问题,提出了 PAM驱动软体机械臂的动态迟滞耦合模型,并设计了迟滞补偿解耦控制器,完成机械臂的变刚度位置控制。通过实验分析软体机械臂的不同收缩型PAM之间的迟滞耦合行为,及气压输入频率和伸长型PAM气压对该行为的影响,将PAM驱动器迟滞模型与动态耦合子模型串联,描述软体机械臂迟滞耦合效应。第一次完整地描述了该软体机械臂的耦合运动特点。为进一步提高软体机械臂的位置控制精度,基于神经网络和动态迟滞耦合模型,设计了多维系统的迟滞补偿解耦控制器。通过对比迟滞补偿解耦控制器和未解耦控制器在不同工况条件下对机械臂的控制效果,证明了本文提出的控制器对提升软体机械臂变刚度位置控制精度的必要性。更多还原

【Abstract】 Along with the rapid development of control,material science,computer and other disciplines,the application fields of robots are expanding.The soft robots with continuous deformation ability and high flexibility can not only adapt the unstructured environment,but also have safer human-computer interaction characteristics,and they have become one of the hot topics in human research recently.In the field of rehabilitation,soft robots can assist the disabled to complete their daily activities;in the field of field exploration and disaster relief assistance,soft robots can adapt to different complex environments and undertake exploration and reconnaissance;in the field of medical surgery,soft robots can be safely compatible with biological tissues,which is conducive to the realization of minimally invasive surgery.As a kind of flexible actuator,the pneumatic artificial muscle(PAM)stands outamong many kinds of actuators for soft robots because of its simple fabrication,high power density ratio,large output force and large deformation.However,the PAM actuator has complex static hysteresis behavior and dynamic hysteresis behavior dependent on input frequency,which can easily reduce the motion accuracy of the system,and even cause system oscillation.At the same time,due to the flexible large deformation of PAM,the soft robot driven by PAMs has large deflection deformation coupling and aerodynamiccoupling.These factors pose a great challenge to the large deformation analysis and precise positioning control of soft robots driven by PAMs.This dissertation thoroughly studied the theory and method of hysteresis nonlinear modeling and compensation control of PAM actuator and the soft manipulator driven by PAMs.This paper is carried out from three aspects:system modeling,controller design and experimental verification,and the main research contents are as follows:(1)In view of the poor generalization ability of the existing PAM static hysteresis model,the amplitude dependent asymmetric static hysteresis model and the direct inverse hysteresis compensator are proposed.By compounding with PID controller,the compound PID controller is studied,so as to effectively improve the control accuracy of PAM actuator.By introducing nonlinear input function into hysteresis operator and taking advantage of excellent feature extraction and learning ability of convolutional neural network,a deep learning network structure based on hysteresis operator and convolutional neural network is constructed,namely enhanced extended unparallel PI model(D-EUPI model),which effectively solve the problem of poor generalization ability of the EUPI model.And the precision and generalization ability of the D-EUPI model are verified by experimental data.On this basis,the model is directly applied to describe the inverse hysteresis curve of PAM actuator,and a direct inverse hysteresis compensator(I-D C compensator)is constructed.Then,the compensator is paralleled with PID controller to compensate the hysteresis of PAM actuator.The experimental results verify the effectiveness of the I-D C compensator.(2)The static hysteresis model is difficult to capture the input frequency and load change information of PAM,which leads to the problem that the dynamic hysteresis behavior of PAM cannot be accurately described.A load dependent dynamic hysteresis model(E-RFNARMAX model)and a dynamic hysteresis compensation controller based on the inverse E-RFNARMAX model are proposed.Starting from the analysis of the comprehensive effects of input frequency,external load and PAM hysteresis on the driving characteristics of PAM,the E-RFNARMAX model is proposed by connecting the hysteresis sub model with the load related dynamic sub model in series.In order to facilitate the parameter identification of the model,the nonlinear hysteresis sub part and the dynamic sub part are separated and identified.At the same time,the dynamic hysteresis compensator(I-R C compensator)is designed to study the open-loop compensation control for the PAM.Experimental results show that the proposed comprehensive dynamic hysteresis model can well characterize the dynamic driving characteristics of PAM,and the E-RFNARMAX model can improve the open-loop control accuracy of PAM with load-dependent dynamic hysteresis,avoiding complex feedback controller.In addition,A load forecasting method based on the inverse ERFNARMAX model applied to the end of PAM is first proposed,which provides a new idea for the driving and sensing integration design of flexible actuator.(3)A pneumatic soft manipulator based on extensor and contractile PAMs is introduced.The manipulator system possesses variable stiffness,which can achieve decouple the spatial position and stifness.Due to the complexity of geometric structure and flexibility of materials,large deflection deformation coupling exists in the soft manipulator.In view of the above problems,the mechanism of large deformation of the pneumatic soft manipulator is analyzed,and the bending posture of the manipulator under static balance condition is calculated.Firstly,by analyzing the material hyperelasticity and geometric nonlinearity of the PAM,an improved PAM output force model is proposed.Based on the improved PAM output force model and continuum mechanics theory,the effects of geometric nonlinearity,material nonlinearity,inflation mode,load and gravity on large deflection deformation of flexible manipulator in different directions are analyzed.The experimental results show that the proposed model can accurately predict the bending posture of the soft manipulator under different working conditions.At the same time,based on the model,this paper systematically and quantitatively analyzes the motion and stiffness characteristics of the soft manipulator.(4)Aiming at the multi input and multi output coupling problem of the soft manipulator system,a dynamic hysteresis coupling model of the soft manipulator is proposed,and the hysteresis compensation decoupling controller is designed to realize the variable stiffness position control of the manipulator.The hysteresis coupling behavior between different contraction type PAM and the influence of air pressure input frequency and extended PAM pressure on the behavior are analyzed experimentally.The hysteresis model of the PAM actuator and the dynamic coupling sub model are connected in series to describe the coupling hysteresis effect of soft manipulator.For the first time,the coupled motion characteristics of the soft manipulator are described completely.In order to further improve the position control accuracy of the soft manipulator,a hysteresis compensation decoupling controller based on the dynamic hysteresis coupling model and neural network for the multi-dimensional system is designed.By comparing the trajectory tracking results of the hysteresis compensation decoupling controller and the uncoupling controller under different working conditions,it is proved that the controller proposed in this paper is necessary to improve the position control accuracy of the flexible manipulator with variable stiffness.更多还原