【作者】 所丽娜；

【导师】 周欣荣；

【作者基本信息】 哈尔滨理工大学 ， 测试计量技术及仪器， 2003， 硕士

【摘要】 水轮机转轮叶片的维修与维护一直是各国水电部门的一项重要工作，每年都要对其耗费巨大的人力、物力、财力，尤其是我国河流泥沙含量大，此项工作就更为艰巨。为了减少人工修复的劳动强度，提高维护、维修的质量和效率，我们引进并自行开发了适合于我国国情的水轮机叶片修复专用机器人。由于机器人测控系统是机器人的核心，为了提高机器人的整体性能与自主能力以及智能化水平，本文设计了开放式通用机器人测控系统的硬、软件结构。 本课题是省科技攻关项目“机坑内修复小型焊补机的研制”的重要组成部分，本文的研究工作是针对水轮机修复专用机器人进行的，应用D-H方法建立了此机器人的各关节之间的坐标关系及其运动学正问题方程，为机器人系统的运动控制提供了可靠依据。 由于此种机器人的工作方式是示教再现型的，为了实现要求的空间轨迹(直线或圆弧)，必须对机器人运动学逆问题求解，但传统解法对于具有冗余自由度的水轮机修复专用机器人来讲，不但解法复杂，而且解难以确定。为此，我们提出应用基于改进的遗传算法的神经网络来建立机器人运动学逆问题模型，针对该模型，详细讨论了改进后的遗传算法的各个算子以及编码方案，并针对该机器人第一关节是平动关节，建立了该机器人各关节与相应驱动电机转角之间的关系模型，将各关节驱动电机的转角做为神经网络训练样本对的输出样本，为统一到适应度函数的公式中做了必要的准备。本文对用软件实现该神经网络进行了程序设计，并用两杆平面机械手对应用遗传算法学习神经网络的权系数的可行性和有效性进行了仿真研究，仿真结果表明，该方法可极大地提高机械手逆运动学解的精度，确保快速达到全局收敛。更多还原

【Abstract】 The repair and maintenance work of the turbine blades is the international problem for all the hydro power plant. A lot of manpower and resources is costed every year, especially in China, because the rivers contain plentiful sand, the problem is more serious. In order to lessen labor intensity and increase the welding and grinding precision and efficiency of the robot, introducing and exploring a repair robot used for welding and grinding on the turbine blades is very useful for those plants. Due to the measurement-control system is the core of robot, in order to improve whole capability and automation level of the robot, hardware and software structure of the system is designed.The theme is an important part of province project for tackling key problem concerning the development of minitype welding and repairing instrument suiting for working in the turbine pit, the theme is worked in the welding and grinding robot. With D-H method the kinetic coordinate and equations are built which is the base of controlling robot.Due to the robot depends on demonstration to realize the required track, we must get the inverse kinematic solutions for the special purpose robot repairing hydraulic turbines with redundant degree of freedom, but the traditional approach is complicated. A kind of multilayer forward neural network based on improved genetic algorithm is applied to build the inverse kinematics model of manipulator, aim at the model, each operator and coding plan of the improved algorithm are discussed in detail, because the first joint of the robot is not rotational, for to unify to the fitness function, the relation model between every joint of the robot and corresponding electromotor rotational angle which is regarded as output sample neural network is build. The software in which the neural network is realized is designed in the paper, and with two-link planar manipulator, the feasible and effective of using GA to learn the weights of NN is studied, simulations show that the proposed method improves considerably the inverse kinematic solutions for robot manipulator and guarantees a rapid global convergence.更多还原