机械外文文献及其翻译.doc

《机械外文文献及其翻译.doc》由会员分享，可在线阅读，更多相关《机械外文文献及其翻译.doc（13页珍藏版）》请在淘文阁 - 分享文档赚钱的网站上搜索。

1、-/与机械相关的外文及翻译Multidisciplinary Design Optimization of Modular Industrial Robots by Utilizing High Level CAD Templates1、IntroductionIn the design of complex and tightly integrated engineering products, it is essential to be able to handle interactions between different subsystems of multidisciplinary

2、 nature 1. To achieve an optimal design, a product must be treated as a complete system instead of developing subsystems independently 2. MDO has been established as a convincing concurrent design optimization technique in development of such complex products 3,4.Furthermore, it has been pointed out

3、 that, regardless of discipline, basically all analyses require information that has to be extracted from a geometry model 5. Hence, according to Bow-cutt 1, in order to enable integrated design analysis and optimization it is of vital importance to be able to integrate an automated parametric geome

4、try generation system into the design framework. The automated geometry generation is a key enabler for so-called geometry-in-the-loop6 multidisciplinary design frameworks, where the CAD geometries can serve as framework integrators for other engineering tools.To eliminate noncreative work, methods

5、for creation and automatic generation of HLCt have been suggested by Tarkian 7.The principle of high HLCts is similar to high level primitives(HLP) suggested by La Rocca and van Tooren 8, with the exception that HLCts are created and utilized in a CAD environment.Otherwise, the basics of both HLP an

6、d HLCt can, as suggested byLa Rocca, be compared to parametric LEGOV Rblocks containing a set of design and analysis parameters. These are produced and stored in libraries, giving engineers or a computer agent the possibility to first topologically select the templates and then modify the morphology

7、, meaning theshape,of each template parametrically.2、Multidisciplinary Design FrameworkMDO is a “systematic approach to design space exploration”17, the implementation of which allows the designer to map the interdisciplinary relations that exist in a system. In this work, the MDO framework consists

8、 of a geometry model, a finite element(FE) model, a dynamic model and a basic cost model. The geometry model provides the analysis tools with geometric input. The dynamic model requires mass properties such as mass, center of gravity, and inertia. The FE model needs the meshed geometry of the robot

9、as well as the force and torque interactions based on results of dynamic simulations.High fidelity models require an extensive evaluation time which has be taken into account. This shortcoming is addressed by applying surrogate models for the FE and the CAD models. The models are briefly presented b

10、elow.2.1 High Level CAD TemplateGeometry ModelTraditionally, parametric CAD is mainly focused on morphological modifications of the geometry. However, there is a limit to morphological parameterization as follows:The geometries cannot be radically modified.Increased geometric complexity greatly incr

11、eases parameterization complexity.The geometry model of the robot is generated with presaved HLCts, created in CATIA V5. These are topologically instantiated with unique internal design variables. Topological parameterization allows deletion, modification, and addition of geometricelements which lea

12、ds to a much greater design space captured.Three types of HLCts are used to define the industrial robot topologically; Datum HLCt which includes wireframe references required for placement for the Actuator HLCTs and Structure HLCts, as seen Fig.2.Fig. 2 An industrial robot (left) and a modular indus

13、trial robot(right)The names of the references that must be provided for each HLCt instantiation are stored in the knowledge base (see Appen-dix A.4), which is searched through by the inference engine. In Appendix A, pseudocode examples describes how the references are retrieved and how they are stor

14、ed in the knowledge base.The process starts by the user defining the number of degrees of freedom (DOF) of the robot (see Fig. 3) and is repeated until the number of axis (i) is equal to the user defined DOF.In order to instantiate the first Structure HLCt, two Datum and two actuator instances are n

15、eeded. References from the two Datum instances help orienting the structure in space, while the geometries of the actuator instances, at both ends of the link, are used to construct the actuator attachments, as seen in Figs. 2 and 3. For the remaining links, only one new instance of both datum and a

16、ctuator HLCts are required, since the datum and actuator instances from adjacent links are already available.Appendix A.2 shows a pseudocode example of an instantiation function. The first instantiated datum HLCt is defined with reference to the absolute coordinate system. The remaining datum HLCt i

17、nstances are placed in a sequential order, where the coordinate system of previous instances is used as reference for defining the position in space according to user inputs (see also AppendixA.3). Furthermore, the type of each actuator and structure instance is user defined.Fig. 3 The high level CA

18、D template instantiation processSince it is possible to create new HLCts in the utilized CAD tool, the users are not forced to merely choose from the templates available. New HLCts can be created, placed in the database and parametrically inserted into the models.2.2 Dynamic Model The objective of p

19、erforming dynamic simulation of a robot is to evaluate system performance, such as predicting acceleration and time performance, but it also yields loads on each actuated axis, needed for actuator lifetime calculations and subsequent stress analysis based on FE calculations. The dynamic model in the

20、 outlined framework is developed in Modelica using Dymola, and it constitutes a seven-axis robot arm based on the Modelica Standard library 18. The dynamic model receives input from the geometry model,as well as providing output to the FE model, which is further described in Sec. 2.3. However, to be

21、tter understand the couplings between the models, the NewtonEuler formulation will be briefly discussed. In this formulation, the link velocities and acceleration are iteratively computed, forward recursively When the kinematic properties are computed, the force and torque interactions between the l

22、inks are computed backward recursively from the last to the first link2.3 FE Surrogate Model To compute the structural strength of the robot, FE models for each robot link is created utilizing CATIA V5, see Fig. 4. For each HLCt, mesh and boundary conditions are manually preprocessed in order to all

23、ow for subsequent automation for FE-model creation. The time spent on preprocessing each FE-model is thus extensive. Nonetheless, the obtained parametric FE-model paves way for automated evaluation of a wide span of concepts. Each robot link is evaluated separately with the load conditions extracted

24、 from the dynamic model. The force (fi-11and fi) and torque (i-1and ti) are applied on the surfaces where the actuators are attached.2.4 Geometric Surrogate Models. Surrogate models are numerically efficient models to determine the relation between inputs and outputs of a model 19. The input variabl

25、es for the proposed application are the morphological variables thickness and link height as well as a topological variable actuator type. The outputs of the surrogate models are mass m, Inertia I, and center of gravity ri,ci. To identify the most suitable type of surrogate model for the outlined pr

26、oblem, a range of surrogate models types are created and evaluated using 50 samples. The precision of each surrogate model is compared with the values of the original model with 20 new samples. The comparison is made using the relative average absolute error (RAAE) and relative maximum absolute erro

27、r (RMAE) as specified by Shan et al. 20, as well as the normalized root mean square error (NRMSE), calculated as seen in Eq. (3). All precision metrics are desired to be as low as possible, since low values mean that the surrogate model is accurate The resulting precision metrics can be seen in Appe

28、ndix B and the general conclusion is that anisotropic kriging 21, neural networks 22, and radial basis functions 23 are the most promising surrogate models. To investigate the impact of increasing number of samples, additional surrogate models of those three are fitted using 100 samples, and the res

29、ults compiled in Appendix B. The resulting NRMSEs for 50 and 100 samples for anistotropic kriging, neural networks, and radial basis functions can be seen in Fig.5. The figures inside the parentheses indicate the number of samples used to fit the surrogate models.Fig. 5 Graph of the NRMSEs for diffe

30、rent surrogate models, fitted using 50 and 100 samplesAccording to Fig. 5, anisotropic kriging outperforms the other surrogate models and the doubling of the number of samples usedfor fitting the surrogate model increases the precision dramatically.2.5 FE Surrogate Models For generating FE surrogate

31、 models, the anisotropic kriging was also proven to be the most accurate compared to the methods evaluated in Sec. 2.4. Here, one surrogate model is created for each link. Inputs are thickness, actuators, force (fi-11and fi) and torque (i-1and ti). The output for each surrogate model is maximum stre

32、ss (MS).A mean error of approximately 9% is reached when running 1400 samples for each link. The reason for the vast number of samples, compared to geometry surrogate models, has to do with a much larger design space.利用高水平CAD模板进行模块化工业机器人的多学科设计优化1 介绍指出,除了规则,基本上所有的分析都需要信息,而这些信息需要从一个几何模型中提取。因此,根据Bowcut

33、t1中,为了使综合设计分析和优化，最重要的是能够将在设计的复杂和紧密集成的工程产品的过程中,必须要有能力处理不同的子系统的多学科性质之间的相互作用。达到一个最优的设计,一个产品必须被视为一个完整的系统,而不是正在开发子的独立系统 。此外,已经一个自动化的参数化几何生成系统融入到设计框架中。自动化的几何生成对于所谓几何循环多学科设计框架是一个关键驱动因素，在这个框架中CAD几何图形可以作为框架连接者来连接其他工程工具。消除没有创新的工作，Tarkian已经提出了创造和生成HLCt的方法 。高HLCts的原理类似于La Rocca 和van Tooren提出的高水平的原始(HLP)，例外的是，HL

34、Cts是在CAD的环境中被创造和应用的。此外，HLP 和 HLCt的基础被比作参数LEGOVR块包含一个组设计和分析参数。这些都是生产和存储在库中,使工程师或计算机代理可能首先选择模板,然后修改形态、意义的形状,每个模板参数化。2 多学科设计框架 MDO“是一个用来设计空间探索的系统化的方法”，是允许设计师来映射存在于系统中的跨学科关系的执行。在这个过程中,MDO框架由一个几何模型、有限元(FE)模型、动态模型和一个基本的成本模型构成。这个几何模型提供了有几何输入的分析工具。这个动态模型需要诸如质量,中心的重力和惯性等大量的属性。有限元模型需要网状的几何机器人以及基于动态模型结果的力与力矩的相

35、互作用。高保真模型需要一个广泛的、已经被考虑的评估时间。这个缺点是通过在有限元与CAD模型中运用代理模型来解决的，下面简要介绍的模型。2.1高水平的CAD模板几何模型传统上,参数化CAD主要集中在几何形态的变化。然而,有限度的形态学参数如下: 这个几何图形不能彻底修改。 增加几何复杂性大大增加参数的复杂性。几何模型的机器人与presaved生成HLCts,创建在CATIA V5。这些是拓扑实例化与独特的内部设计变量。拓扑参数化允许删除,修改和添加导致获得一个更大的设计空间的几何元素。三种类型的HLCts用于定义工业机器人拓扑; 基准HLCt包括需要的执行机构HLCTs位置和结构HLCTs的线框

36、引用,见图。 图2致动器(左),基准(中心)和结构HLCt(右) 引用的名称,必须为每个存储在通过推理引擎才能搜索到的知识库的HLCt实例提供。在附录A,伪代码示例描述了引用如何被检索和它们是如何存储在知识库中的。 这个过程开始于由用户定义机器人的自由度（见图3)和重复直到数量的轴(I)等于用户定义的自由度。图3高水平CAD模块安装程序为了实例化第一HLCt结构,需要两个基准和两个致动器实例。引用这两个基准实例帮助定向结构在空间,而几何图形的致动器实例的两端连接,用于构建致动器附件,见图2和3。对于剩下的链接,只有一个新实例的两个基准和致动器HLCts是必需的,因为来自于相邻链接的基准面和致动

37、器实例已经可用。附录A.2显示了一个实例化的伪代码例子，第一个实例化基准HLCt是指绝对坐标系统的引用。剩下的基准面HLCt实例被放置在一个顺序中,在这个顺序中坐标系统先前的实例作为参考来定义空间的位置，而这个位置根据用户输入。此外,每个执行机构的类型和结构实例是由用户定义的。因此在利用CAD工具创建新的HLCts是可能的,用户不是被迫仅仅选择可用模板。可以创建新HLCts,放置在数据库和参数化地插入到模型中。2.2动态模型进行动态仿真机器人的目的是评估的一个机器人系统的性能,如预测加速度和时间性能,但它也有利于在每个驱动轴载荷, 计算所需的致动器寿命和随后的基于有限元应力分析计算。在大体框架

38、中的动态模型是在使用Dymola的Modelica中被发展的，它构成基于Modelica标准库的并联机器人臂。动态模型从几何模型中接受输入，以及提供输出到有限元模型,进一步在Sec。2.3描述。然而,为了更好地理解模型之间的耦合,牛顿欧拉公式将简要讨论。在这个过程中,速度和加速度的链接会被反复计算,递归地向前。 当计算运动特性的时候,力和力矩之间的相互作用被从最后一个到第一个链接递归计算。 2.3有限元模型为计算机器人的结构强度,有限元模型创建每个机器人链接利用CATIA V5,见图4。 图4.有限元模型的奇偶的链接 对于每个HLCt、网格和边界条件是手动预处理，其目的是考虑到有限元模型的创建

39、的后续自动化。花费在每一个有限元模型预处理上的时间是非常多的。然而,参数化的有限元模型为广泛概念的自动评价方法铺平道路。每个机器人链接被从动态模型中取得的负载条件单独地评估。2.4几何替代模型 代替模型是确定模型输入和输出之间的关系的数值地、有效的模型。提出的应用形态的输出变量有厚度和链接高度以及拓扑变量执行机构类型。代替模型的输出是质量m，惯性I和重心引力ri,ci。为概述问题确定最适合的代替模型，一系列代替模型被建立，并且使用50个样本来评估。每一个代替模型是与原始模型的数值和20个新样本作比较的。比较是通过使用相对平均绝对误差(RAAE)和相对最大绝对误差(RMAE)所指定的山et al

40、。20,以及归一化均方根误差(NRMSE)来确定的，,计算见Eq。(3) 所有的精度指标是想要尽可能低,因为低值意味着代理模型是准确的。 结果精度指标中可以看出附录B和总的结论是各向异性克里格21,神经网络22,径向基函数23是最有前途的替代模型。对于日益增加的样本的调查影响，那三个附加的代替模型使用100个样本是合适的，结果编译在附录B。由此产生的NRMSEs 50和100个样本作为各向异性克里格,神经网络和径向基函数，如图5。 图5不同代理模型的NRMSEs图，安装使用的50和100个样本 括号内的数字表示用于配合代理模型的样品的数量。根据图5,各向异性克里格优于现有的其他代理模型和两倍数量的用于配件代理模型的样本，其精度大大提高。2.5有限元替代模型 为了生成有限元替代模型，各向异性克里格与2.4的评估方法相比被证明是为准确的。在这里, 为每个链接的一个代理模型被创建。输入厚度、致动器、力(fi1和fi1)和转矩(si1和si1)。输出为每个代理模型是最大应力(MS)。在运行各个环节的1400样品时平均误差约9%就达到了。与代理模型的体积相比，大数量的样本的原因与较大设计空间有关。