【作者】 孙一兰；

【导师】 柳洪义；

【作者基本信息】 东北大学 ， 机械电子工程， 2009， 博士

【摘要】 固体火箭发动机是航天工程中重要的动力装备,粘贴在燃烧室(即发动机壳体)内壁的绝热层,对燃烧室壳体起到隔热、防止燃烧产物冲刷等作用。目前,我国在绝热层的打磨中仍采用手工操作方式。手工打磨,不仅效率低、打磨时周围充满粉尘,工作环境极其恶劣,并且难以保证打磨的均匀性。而打磨厚度过多或过少都会使绝热层以至整个发动机壳体的报废,造成极大的经济损失,这种制造方式已不能适应航天工业当今和未来发展的需要。研制先进的“绝热层自动打磨装备”是航天工业的发展趋势,也是所在研究室与中国航天科工集团某厂合作的实际项目“固体火箭发动机壳体内壁绝热层打磨机器人”。基于以上实际需要,本文对绝热层打磨机器人研制中的关键技术问题进行了详细的分析与论述,并给出有效的解决方案与方法,主要研究内容与成果如下：(1)固体火箭发动机绝热层的制造,涉及军事、航天等敏感领域,国际上相关报道很少,而国内该领域的自动化生产水平低下,没有前例可以参照。同时由于绝热层结构尺寸与内壁精确打磨要求的特殊性,对打磨装置本身的形状尺寸及其在操作过程中的运动都有特殊地要求与限制,使其不同于传统的零部件外表面的打磨。因此,在对绝热层结构尺寸及其打磨技术要求进行详细地分析、并借鉴应用于其它领域的自动化打磨装置的设计的基础上,设计了一个安装有专用打磨装置的特种绝热层内壁打磨机器人,并从机械结构、打磨方式、功能模块、性能指标等方面给出了整体设计方案。(2)根据建立的打磨机器人机械本体综合优化模型的特点,采用基于改进编码方式和加入小范围竞争的遗传算法进行优化求解。优化模型中的优化变量皆为两位或三位的正整数,所以采用正整数十进制组合作为基因表对染色体进行编码,这样对基因的遗传操作实质上就是对优化变量的遗传操作,既利用了遗传算法的优化机制,又符合常规的对优化变量进行搜索的优化设计思想,同时也减少了遗传操作中染色体的个数,提高了优化的效率。而小范围竞争可以避免由于大量的同一父代产生的个体在下代中进行遗传操作而陷入局部极限。优化结果表明,在GA算法中采用正整数十进制组合编码方式和小生竞争机制,既可以明显地减少搜索代数,又避免了由于“近亲繁殖”而在某一数值附近重复搜索,增强了全局搜索能力。通过对打磨机器人机械本体的综合优化设计,得到最佳的设计方案,指导机器人机械本体的设计与制造。(3)针对传统的应用于自由操作空间的机器人冗余度干涉规避方法,无法解决“细长”形壳体内壁绝热层打磨机器人各关节杆件间、关节杆件与内壁间的干涉规避问题,本文提出一种在受限操作空间中无冗余干涉避障方法。该方法在不增加打磨机器人自由度的前提下,只改变末端执行器(即打磨头)的设计形状,然后通过数学模型判别在打磨轨迹上各关节杆件间、关节杆间与壳体内壁问是否发生干涉,动态地调整打磨角,从而改变机器人各杆件的位置与姿态,实现干涉和碰撞的规避。无冗余干涉避障方法在最低程度改变原有机械本体的前提下,通过对机器人各关节杆件在操作过程中的运动形态的优化控制实现了干涉规避,为机器人的干涉避障提供了新的方法与思路。(4)在机器人轨迹规划中引入数控S形加减速规划方法,并通过分析规划方法中各参数之间的关系,给出求解S形加减速算式的约束条件,简便了求解过程。同时,在对比了等切线长、等弦长和等弧长三种时间分割法的基础上,选用弓高误差最小的等弧长时间分割法对椭圆打磨轨迹段进行插补计算,把经由前S形加减速算规划每一运动控制周期的位移量,精确的、实时的转换为相应的打磨点坐标。仿真数据说明,打磨机器人的运动轨迹规划与插补方法,精度高、速度快、平稳性好,为实现壳体内壁绝热层的精确打磨提供了必要的前提条件。(5)由于绝热层在局部为几何参数未知的非规则平滑曲面,本文在理想运动轨迹规划和经典控制方法的基础上,采用基于神经网络预测参考轨迹的阻抗控制系统进行控制补偿,使打磨机器人具备柔顺控制的能力。通过推导,证明了以位置传感器和力传感器的采集信号作为输入、以非规则表面预测轨迹参数和环境预测刚度为输出的4层神经网络感知器预测参考轨迹模型的可行性。另外,针对阻抗控制系统中的实际检测接触力Pm很难直接获得,采用采集各伺服电机转矩值、经过换算得到末端执行器与外部环境的接触力的方法,来替代直接采用力传感器反馈末端执行器接触力信号；对期望接触力Pd,则设计了一种实验方法得到打磨力和在其作用下的被打磨厚度,预估出期望的打磨力范围,再根据神经网络预测参考轨迹的阻抗控制系统在线进行局部的微量调整,以满足打磨要求。实验数据表明,打磨厚度在打磨要求范围内,保证了打磨的质量与安全。(6)在打磨过程中,机械本体的大部分在壳体内部工作,有必要建立机械本体各杆件运动的三维图形在线监测系统。因此,本文利用Solid Edge三维建模软件及其二次开发技术,实现机器人参数化建模；在此基础上,再通过Solid Edge公开地装配体和零部件操作中的五个驱动函数,用在线位置反馈数据驱动模型中各关节杆件运动,简便地实现了打磨机器人各关节杆件运动形态的在线三维监测,以便快速准确地作出判断与决策。更多还原

【Abstract】 Solid rocket motor is the most important space power equipment. The insulation pasted in the combustion chamber (namely, the engine shell) inner-wall plays an important role in thermal insulation and avoiding washing to shell by combustion products. Presently, the domestic production is still by manual method. This manufacturing not only is inefficient, labor-intensie, extremely poor working conditions, but also is not able to ensure uniformity of grinding. The thickness of grinding too much or too little will lead to scrap the insulation as well as the whole engine. So this manufacturing has not meet today’s aerospace industry and future development needs.Development of advanced automatic grinding equipment insulation is the development trend of the space industry and also one practical co-operation projects between my research department and a factory of China Aerospace Science and Industry Group. Consequently, in this dissertation, the key technical issues of rocket engines inner-wall grinding robot have been made a detailed analysis and discussion. Moreover, according with the key problems, the effective solutions and their theoretical basis and technical support are given. The main research content and results are as follows:(1) The manufacturing of advanced automatic rocket engines inner-wall grinding equipment involves in military, aerospace and other sensitive areas, so there are few relevant reports international. And the automation level of domestic production is backward. Consequently, there is no precedent for reference. Moreover, due to the specificity requirements of grinding and the structure of insulation, they all limit the shape and size of robot mechanical structure and the movement patterns in grinding process. Hence, the manufacturing of advanced automatic grinding is different from the traditional outer surface. Hence, reference the design of grinding automation applied to other areas, the overall function and structure of grinding robot is designed from the aspects of the polished manner, functional modules and performance indicators.(2) Based on the characters of integrated optimal design of grinding robot mechanical structure, the improved encoding method and involing competition in small scope in GA is applied to optimize. Because the optimization model has six variables and all of them are positive integers, combination of ten numbers is used as chromosome table to encode genes. Consequently, the genetic manipulation of genes is essentially the genetic manipulation of optimal variables. This method not only uses the optimization mechanism of genetic algorithm, but also accords with the conventional optimization design idea. Moreover, combination encoding also reduces the numbers of chromosome so that it can also improve the efficiency of optimization search. Moreover, the competition in small scope is able to avoidng the individuals generated by the same parental type great carrid on genetic manipulation.(3) Traditional robot collision avoidance method applied to free operation space is carried out by adding robot DOF(Degree of Freedom). However, this method isn’t suitable to limited operation space such as a kind of slightness solid-propellant rocket engines inner-wall. Hence, this dissertation presents a new robot method with non-redundant DOF for limited operation space collision avoidance. Without adding robot DOF, this method judges collision situation in grinding trajectory by mathematical model, then adjusts dynamically the grinding interface of the robot end-effector. So robot bodies’poses are transformed to avoid collisions between the different robot bodies or between robot bodies and inner-wall. Moreover, kinematics optimization and simulation were carried on by setting the smoothness of joints’velocity variety as optimization objective and discriminant of collision as constraints. On the premises of minimum changing of original, this method use the optimal control way to collision avoidance.(4) Referring to the pre-interpolation S-shape acceleration/deceleration and smooth transfer in CNC, the grinding robot trajectory planning is carried on. The trajectory planning is able to avoid impact, overshoot and oscillation in the start and stop processes of robot basic. And through the analysis of parametric relationship, the constraints of S-shape acceleration/deceleration is presented to simplify the solution. Moreover, the heads of solid-propellant rocket engines body are ellipsoid thin-wall structure so the ellipse trajectory curve is not approximated by general linear or circular interpolation method. Otherwise, it will create a big bow high error. In view of the defects of above methods of interpolation, based on the contrast of three different ellipse interpolation algorithm, the equal arc length time-sharing interpolation algorithm is used to transform the displacement planned by S-shape acceleration/deceleration. The simulation results prove the inner-wall grinding robot’s trajectory planning and interpolation algorithm with high precision, speed and smoothness. All of these provide the necessary precondition for accuracy grinding of missile inner-wall grinding.(5) Because the insulation exist some local convex and cupped, this dissertation presents a impedance control system with multilayer perception NN-based predicting reference trajectory to compensate the ideal trajectory and control. Through formula deduction, the feasibility of NN-based predicting trajectory is proved with the NN sample inputs set as the actual contact trajectory points and contact forces, and the NN sample outputs set as the environment predictive trajectory points, stiffness and tangential angle.In addition, the actual force Pm and expected force Pd by the predictive trajectory based on neural network impedance reference control system are difficult to be directly access. Hence, Pm is substituted by the acquisition of the servo motor torque, which can be converted to the contact force of the end effector. And an experimental method has been designed to get Pd. In order to meet the requirements the method predict the expected grinding force referred to the grinding force that be indirectly measured and the thickness that be polished, and adjust grinding force slightly based on neural network reference trajectory of the impedance control system. Experimental data show that the thickness of griding can meet the requirements and ensure the quality and safety of grinding.(6) In the grinding process, the grinding robot mechanical structure is almost in the rocket engines, so it is necessary to build the three dimensions graph monitoring system on-line of grinding robot’s movement. In this dissertation, Solid Edge software and its secondary development technology are used to establish the parameters three dimensions model of grinding robot. Then, through five command functions provided openly by Solid Edge software, the grinding robot’s monitoring system online are build simply.更多还原