【作者】 冯涛；

【导师】 杨向东；

【作者基本信息】 清华大学 ， 机械工程， 2009， 硕士

【摘要】 随着科技的不断进步,目前机器人在各种自动化生产系统中扮演着越来越重要的角色。而关节型机器人以其较小的占用空间、灵活的工作范围等特点,在医疗领域占有相当大的比重。但是与直线关节型机器人相比,其关节空间到末端执行器位形空间的映射关系比较复杂,误差影响因素比较多,这对提高医疗机器人的绝对定位精度是相当不利的。本文以超声引导微波消融肝癌治疗机器人系统原型机为载体,对机器人正运动学、误差理论、关节参数标定、以及机器人系统定位精度分析和补偿等方面的内容进行了研究。主要内容如下:1.分析了影响穿刺机器人系统定位精度的各种因素,根据手术流程和相互作用关系,将众多因素归为三大类:分割、重构引起的图像误差,靶点映射误差及穿刺机器人机械本体误差。根据经典误差理论,建立了系统总体误差与各误差分量之间的关系。2.根据穿刺机器人构型特点,根据旋量理论建立了机器人本体的正运动学模型和误差模型,从根本上避免了使用D-H方法建模带来的奇异性问题。导出了只使用机器人末端位置数据对各关节参数进行标定的理论公式,为使用关节式测量臂对穿刺机器人进行快速标定奠定了基础。设计了简便有效的标定实验步骤和算法,对机器人实际关节参数进行辨识和提取。3.根据穿刺机器人结构,设计了一类特殊结构的人工神经网络。其不仅能够作为机器人控制模型中的运动补偿器;同时还能够从神经网络的权重、偏置参数中提取关节参数。这使得人工神经网络不再单纯地作为完成非线性映射的“黑箱子”。同时,这种设计思想也为确定神经网络的层数和节点数提供了参考。4.对新型定位装置——磁定位器在手术室条件下的特性进行了初步研究,改进了其标定配准方法,有效提高了磁定位器及机器人系统整体的定位精度。通过对上述内容的研究,大幅度提高了穿刺机器人的定位精度,为临床手术实验的顺利进行奠定了良好的基础。同时本文中的理论推导和实验方法具有一定的普遍性,能够为今后的机器人结构设计和标定工作提供指导和参考。更多还原

【Abstract】 With the development of modern industrial technology, robotic systems are playing a more important role in automation systems. Meanwhile, articulated robots occupy a great market in the field of medical instrument due to their non-linear characteristics and differences with regards to Cartesian coordinate robots. But this kind of mechanism configuration also causes a complex mapping between joint-space and task-space, which leads to the position inaccuracy of end-effectors. Based on the development project of ultrasound-guided medical robotic system for liver cancer coagulation therapy, this thesis studies on forward kinematic, calibration, accuracy analysis and compensation. The primary matters are described as follows:First, the factors that cause the position error of the robotic system are analyzed. According to surgery process and data flow, these factors can be classified into three groups. The first describes the graphic error, which includes errors caused by segmentation, three-dimension ultrasound reconstruction and registration. The second relates to the electromagnetic track device. The last includes those affecting the robot position inaccuracy, such as the error caused by joint parameters. The system error propagation is established according to classical error theory.Second, according to the feature of mechanism configuration, the method of kinematic modeling based on Product-of-Exponentials (POE) is adopted to avoid the singularity caused by Denavit-Hartenberg representation. Kinematic calibration formulas are introduced, with which the actual parameters of joints can be derived from the position data of end-effector. This method significantly simplifies the process of calibration experiment. Experiment results show that the average position accuracy of medical robot increases significantly after calibration.Third, a special artificial neural network (ANN) is designed, which can not only be used as a position compensator in robot control system, but also be able to derive joint parameters from weights and biases. Thus the ANN is no longer a black box just for nonlinear fitting and mapping, but grayer or even whiter. This kind of design ideas also provides a method to estimate the number of layers and nodes of ANN for a practical robot.Last, the performance of electromagnetic trackers, a new kind sensor of position and orientation, is studied under the environment of surgical room. New registration method is developed to improve the position accuracy of robotic system.On summary, this thesis provides various methods to improve the performance of the medical robotic system, which can also be considered as a reference for mechanical design and kinematic calibration in the future works.更多还原