【作者】 张飞；

【导师】 金明河；

【作者基本信息】 哈尔滨工业大学 ， 机械电子工程， 2012， 硕士

【摘要】 随着我国空间站建设步伐的加快，大型空间机械臂的研制变得更加迫切。关节是大型空间机械臂的关键组件，将工作在大温差、高真空、强辐射的环境下，其性能好坏将直接影响到整个在轨任务的完成。为此，本文结合实验室自主课题“空间大型机械臂的研制”开展了大型空间机械臂关节性能测试平台的研制工作，并基于该平台开展了关节动力学参数辨识的研究。大型空间机械臂关节的定位精度和负载能力是关节最主要的两个性能指标，因此使用相应的测试设备检测其性能指标是否满足设计指标要求就显得尤为重要。针对这种需求，本文研制了HIT大型空间机械臂关节性能测试平台，该测试平台集成了力矩传感器、位置传感器、齿轮增速器和用于加载的磁粉制动器等部件，从而达到对关节性能进行测试的目的。由于关节内部使用了谐波减速器作为核心传动部件，而谐波减速器的摩擦和刚度非线性将直接造成关节刚度和摩擦的非线性特性，为此本文基于考虑非线性因素的谐波减速器动力学模型建立了HIT大型空间机械臂关节的动力学模型。关节内部传感器的精度将直接影响关节动力学参数辨识准确性，高精度的传感器能大大提高参数辨识的效率，为此本文标定了关节内部传感器，包括力矩传感器、磁编码器、旋转变压器和电流传感器，并且针对旋转变压器精度较低问题提出了基于BP神经网络的误差补偿方法，并完成多次补偿实验。实验结果表明该方法能有效提高旋转变压器的测角精度。准确的动力学参数能有效改善关节的控制精度，因此本文开展了基于电流和位置双闭环关节控制模式下的HIT大型空间机械臂关节动力学参数辨识研究。采用粘滞+库伦摩擦模型、三次多项式摩擦模型和Stribeck摩擦模型进行关节摩擦辨识研究，结果表明Stribeck摩擦模型更能反映实际的关节阻尼。刚度辨识研究结果表明由于力矩传感器和一些关节连接件的变形造成关节的刚度略小于谐波减速器刚度。采用最小二乘法进行基于关节动力学模型的离线参数辨识，结果表明考虑库伦+粘滞摩擦时辨识得到的刚度更加接近实际关节刚度值。在离线辨识的基础上采用改进的递推最小二乘法并考虑基于误差的遗忘因子模型进行在线参数辨识，实验结果表明各个参数能够收敛到相对稳定值，该稳定值与离线参数辨识结果接近。更多还原

【Abstract】 With the accelerated pace of the building of the China Space Station, developmentof the large-scale space manipulators should be put into practice desirably. Jointsworking in large temperature, high vacuum, hard radiation environment are keycomponents of the large-scale space manipulator, whose performance will directlyaffect the entire implementation of the orbital mission. Therefore, with the support ofthe laboratory autonomous program “Development on the large-scale spacemanipulators” the HIT large-scale space manipulator joint performance test platform aredeveloped and the joint dynamic parameters identification is thoroughly studied in thetest platform.The positioning accuracy and the load capacity are the two most important indicesof performance. So it is very important to check if indices of performance satisfy designindices. For satisfying this requirement, the HIT large-scale space manipulator jointperformance test platform is designed, which integrates a torque sensor, a positionsensor, a gear increaser and a magnetic particle brake for loading. So the indices of thejoint are completely measured.Harmonic drive is the core transmission part of the joint, whose stiffness andfriction nonlinearity will cause the nonlinearity of the joint. So the dynamic model ofthe HIT large-scale space manipulator joint is established, according to the nonlinearityof the dynamic model for the harmonic drive.The accuracy of the sensors in the joints will directly influence the accuracy ofparameters identification. High accuracy sensors can greatly increase efficiency ofparameters identification. For the reason the sensors in the joint are calibrated, includingthe resolver, the torque sensor, the current sensor and the encoder. The problem of thelow accuracy of the resolver will be solved by an error compensation method based onBP neural network. For many compensation experiments it is proved that this methodcan improve the accuracy of the resolver greatly.The exact dynamic parameters can greatly improve the joint control accuracy. Forthis reason in the control mode of the current close-loop and the position close-loopcontrol parameters identification is researched on the HIT large-scale space manipulatorjoint. The research on the coulomb-hydraulic friction model, the cubic polynomialfriction model and Stribeck friction model are done, experimental results show that theStribeck model can reflect the actual joint damping better. Experimental results of thestiffness identification show that the stiffness is less than the harmonic drive’s samplevalues, which is caused by deformation of the torque sensor and connectors of the joint.Then the off-line parameters identification of the overall dynamic model using theleast-squares method is obtained, and experimental results show that the identified stiffness is closer to the actual joint stiffness, considering the damping torque. Finally,on-line parameters identification of the joint is carried out using the modified recursiveleast-squares algorithm, considering the error model based on a forgetting factor. Theexperimental results show that the parameters can converge to the relative stable valueswhich are near to the off-line identification parameters.更多还原