【作者】 陈海波；

【导师】 向馗；

【作者基本信息】 武汉理工大学 ， 控制科学与工程， 2020， 硕士

【摘要】 自上世纪90年代起,串联弹性作动器(Series Elastic Actuators,SEA)被广泛研究并用于仿生机器人关节中。SEA常用金属扭簧作为弹性体,其力矩估计准确,但是参数调整和加工安装都不方便,难以实现高集成度,实际应用受到限制。如果用高弹性聚合物(如橡胶)代替金属扭簧,集成方便,结构紧凑,但聚合物具有粘弹性,若将其简单假设成理想弹簧,会导致力矩估计精度大幅降低。10年前,国外大学研究发现,高弹性聚合物的粘性阻尼并非完全有害,它对速度敏感,有助于提高作动器的输出力矩带宽。为此提出串联粘弹性作动器(Series Viscoelastic Actuators,SVA)。最近,SVA的研究逐步得到重视,尤其是其力矩估计是当下的研究热点。目前,针对SVA粘弹性体的力矩估计,主要是根据聚合物材料的刚度和阻尼估算力矩值,方法简单,但精度有限。以常用的聚合物材料橡胶为例,其力学特性复杂,并非刚度和阻尼项的简单叠加。而且橡胶在温度等因素作用下会老化变性,其力学参数时变特性明显。本文采用橡胶作为粘弹性体自主设计了SVA,并系统研究了其负载力矩估计方法。分别采用粘弹性本构方程和电机电流估计了SVA力矩,然后用双扩展卡尔曼滤波算法(DEKF)融合两种方法的优点,进一步提高了力矩估计精度。主要研究内容如下:(1)采用橡胶作为粘弹性体,设计SVA模块化关节。根据设计指标对驱动源、传动机构、橡胶材料、编码器等进行合理选择、模块化设计和装配;设计并调试基于STM32的系统控制电路、无刷电机矢量控制电路和信号采集电路。(2)基于Zenner本构方程建立橡胶体力矩估计模型。测试发现,橡胶力学特性不仅与刚度、阻尼有关,还与应力松弛、蠕变等粘弹性特性有关。依据测试结果,选择Zenner本构方程来建立橡胶体的力矩模型,并用最小二乘法辨识模型参数,力矩拟合误差小于0.03Nm。但是在长期使用时,因温度变化和材料老化,力学参数发生改变,误差会逐渐增大。(3)基于电机电流估计负载力矩。分析和简化SVA中各运动部件的动力学特性,以Stribeck静态摩擦和角位置相关动态摩擦来描述谐波减速机摩擦力矩,建立了基于电机电流的力矩估计模型。实验发现,在电机运转平稳时,该方法的力矩估计误差小于0.02Nm。但是,当电机处于低速、加速和换向等不平稳阶段,误差大于0.5Nm。(4)基于双扩展卡尔曼滤波的力矩估计。基于双扩展卡尔曼滤波理论,以橡胶模型为基础设计力矩状态估计器,以电流模型为基础设计橡胶参数估计器。两种估计器级联运行,以同时估计力矩和参数。经过实验验证,该方法融合了方法(2)、(3)的优点,在所有工况下力矩估计误差均小于0.03Nm。本文基于橡胶粘弹性体自主设计了新的SVA,研究了基于橡胶和电流估计力矩的方法。考虑到两种方法存在各自缺陷,提出了基于双扩展卡尔曼滤波算法,融合两者的优点,提高力矩估计精度,为SVA输出力矩的精确控制奠定了基础。更多还原

【Abstract】 Since the 1990s,Series Elastic Actuators(SEA)have been widely studied and used in bionic robot joints.SEA commonly uses metal torsion springs as the elastic body,and its torque estimation is accurate.But its parameter adjustment,processing and installation are not convenient,and it is difficult to achieve high integration.So its practical application is limited.If a highly elastic polymer(such as rubber)is used to replace the metal torsion spring,the integration is convenient and the structure is compact.But the polymer has viscoelasticity.If it is simply assumed to be an ideal spring,the accuracy of torque estimation will be greatly reduced.Ten years ago,a foreign university study found that the viscous damping of highly elastic polymers is not completely harmful.It is sensitive to speed and helps to increase the output torque bandwidth of the actuator.For this purpose,Series Viscoelastic Actuators(SVA)are proposed.Recently,the research of SVA has been gradually paid attention,especially the torque estimation is the current research hotspot.At present,the SVA viscoelastic bodies is mainly based on the stiffness and damping of polymer materials to estimate the torque.The method is simple,but the accuracy is limited.Taking the commonly used polymer material,rubber,as an example,its mechanical properties are complex,and it is not a simple superposition of stiffness and damping terms.Moreover,rubber will age and degrade under the influence of temperature and other factors,and its mechanical parameters have obvious timevarying characteristics.This paper uses rubber as a viscoelastic body to design SVA independently,and systematically studies its load torque estimation method.The viscoelastic constitutive equation and the motor current were used to estimate the SVA torque respectively,and then the Dual Extended Kalman Filter algorithm(DEKF)was used to fuse the advantages of the two methods to further improve the torque estimation accuracy.The main research contents are as follows:(1)Using rubber as a viscoelastic body to design SVA modular joints.According to the design indicators,the drive source,transmission mechanism,rubber material,encoder,etc.are reasonably selected,modularized and assembled.Designed and debugged the STM32-based system control circuit,brushless motor vector control circuit and signal acquisition circuit.(2)The rubber body torque estimation model is established based on the Zenner constitutive equation.The test found that the mechanical properties of rubber are not only related to stiffness and damping,but also related to viscoelastic properties such as stress relaxation and creep.Based on the test results,the Zenner constitutive equation is selected to establish the torque model of the rubber body,and the model parameters are identified by the least square method.The torque fitting error is less than 0.03 Nm.However,during long-term use,due to temperature changes and material aging,the mechanical parameters change,and the error will gradually increase.(3)Estimate the load torque based on the motor current.The dynamic characteristics of each moving part in SVA are analyzed and simplified.Stribeck static friction and angular position-dependent dynamic friction are used to describe the harmonic torque of the harmonic reducer.A torque estimation model based on motor current is established.Experiments have found that when the motor is running smoothly,the torque estimation error of this method is less than 0.02 Nm.However,when the motor is in an unstable stage such as low speed,acceleration and commutation,the error is greater than 0.5Nm.(4)Torque estimation based on Dual Extended Kalman Filtering.Based on the Dual Extended Kalman Filtering theory,the torque state estimator is designed based on the rubber model,and the rubber parameter estimator is designed based on the current model.Two estimators operate in cascade to simultaneously estimate torque and parameters.After experimental verification,this method combines the advantages of methods(2)and(3),and the torque estimation error is less than 0.03 Nm under all working conditions.In this thesis,a new SVA is independently designed based on rubber viscoelastic body,and the method of estimating torque based on rubber and current is studied.Considering that the two methods have their own shortcomings,a Dual Extended Kalman Filtering algorithm is proposed,which combines the advantages of the two methods and improves the accuracy of torque estimation.更多还原