外文翻译--MCB-工业的机械手论文

附录 A MCB Industrial Robot Feature Article The BarrettHand grasper programmable flexible part handling and assembly Abstract This paper details the design and operation of the BarrettHand BH8-250, an intelligent, highly flexible eight-axis gripper that reconfigures itself in real time to conform securely to a wide variety of part shapes without tool-change interruptions. The grasper brings enormous value to factory automation because it: reduces the required number and size of robotic work cells (which average US$90,000 each not including the high cost of footprint) while boosting factory throughput； consolidates the hodgepodge proliferation of customized gripper-jaw shapes onto a common programmable platform； and enables incremental process improvement and accommodates frequent new-product introductions, capabilities deployed instantly via software across international networks of factories. Introduction This paper introduces a new approach to material handling, part sorting, and component assembly called “grasping”, in which a single reconfigurable grasper with embedded intelligence replaces an entire bank of unique, fixed-shape grippers and tool changers. To appreciate the motivations that guided the design of Barretts grasper, we must explore what is wrong with robotics today, the enormous potential for robotics in the future, and the dead-end legacy of gripper solutions. For the benefits of a robotic solution to be realized, programmable flexibility is required along the entire length of the robot, from its base, all the way to the target work piece. A robot arm enables programmable flexibility from the base only up to the tool plate, a few centimeters short of the target work piece. But these last few centimeters of a robot must adapt to the complexities of securing a new object on each robot cycle, capabilities where embedded intelligence and software excel. Like the weakest link in a serial chain, an inflexible gripper limits the productivity of the entire robot work cell. Grippers have individually-customized, but fixed jaw shapes. The trial-and-error customization process is design intensive, generally drives cost and schedule, and is difficult to scope in advance. In general, each anticipated variation in shape, orientation, and robot approach angle requires another custom-but-fixed gripper, a place to store the additional gripper, and a mechanism to exchange grippers. An unanticipated variation or incremental improvement is simply not allowable. By contrast, the mechanical structure of Barretts patented grasper, illustrated in Figure 1, is automatically reconfigurable and highly programmable, matching the functionality of virtually any gripper shape or fixture function in less than a second without pausing the work cell throughput to exchange grippers. For tasks requiring a high degree of flexibility such as handling variably shaped payloads presented in multiple orientations, a grasper is more secure, quicker to install, and more cost effective than an entire bank of custom-machined grippers with tool changers and storage racks. For uninterrupted operation, just one or two spare graspers can serve as emergency backups for several work cells, whereas one or two spare grippers are required for each gripper variation potentially dozens per work cell. And, its catastrophic if both gripper backups fail in a gripper system, since it may be days before replacements can be identified, custom shaped from scratch, shipped, and physically replaced to bring the affected line back into operation. By contrast, since graspers are physically identical, they are always available in unlimited quantity, with all customization provided instantly in software. Gripper legacy Most of todays robotic part handling and assembling is done with grippers. If surface conditions allow, vacuum suction and electromagnets can also be used, for example in handling automobile windshields and body panels. As part sizes begin to exceed the order of 100gms, a grippers jaws are custom shaped to ensure a secure hold. As the durable mainstay of handling and assembly, these tools have changed little since the beginning of robotics three decades ago. Grippers, which act as simple pincers, have two or three unarticulated fingers, called “jaws”, which either pivot or remain parallel during open/close motions as illustrated in Figure 2. Well organized catalogs are available from manufacturers that guide the integrator or customer in matching various gripper components (except naturally for the custom jaw shape) to the task and part parameters. Payload sizes range from grams for tiny pneumatic grippers to 100+ kilograms for massive hydraulic grippers. The power source is typically pneumatic or hydraulic with simple on/off valve control switching between full-open and full-close states. The jaws typically move 1cm from full-open to full-close. These hands have two or three fingers, called “jaws”. The part of the jaw that contacts the target part is made of a removable and machine ably soft steel or aluminum, called a “soft jaw”. Based on the unique circumstances, an expert tool designer determines the custom shapes to be machined into the rectangular soft-jaw pieces. Once machined to shape, the soft-jaw sets are attached to their respective gripper bodies and tested. This process can take any number of iterations and adjustments until the system works properly. Tool designers repeat the entire process each time a new shape is introduced. As consumers demand a wider variety of product choices and ever more frequent product introductions, the need for flexible automation has never been greater. However, rather than make grippers more versatile, the robotics industry over the past few years has followed the example of the automatic tool exchange technique used to exchange CNC-mill cutting tools. But applying the tool-changer model to serial-link robots is proving expensive and ineffective. Unlike the standardized off-the-shelf cutting tools used by milling machines, a robot tool designer must customize the shape of every set of gripper jaws a time-consuming, expensive, and difficult-to-scope task. Although grippers may seem cheap at only US$500 each, the labor-intensive effort to shape the soft jaws may cost several times that. If you multiply that cost times a dozen grippers as in the example above and throw in a tool changer and tool-storage rack for an additional US$10,000, the real cost of the “few-hundred-dollar” gripper solution balloons to US$20,000 to US$60,000. To aggravate matters, unknowns in the customization process confound accurate cost projections. So the customer must commit a purchase order to the initial installation fee on a time and materials basis without guarantee of success or a cost ceiling. While priced at US$30,000, intelligent graspers are not cheap. However, one can “customize” and validate the process in software in a matter of hours at the factory in a single day. If the system does not meet performance targets, then only a days labor is wasted. If the system succeeds, then there are not any hidden expenses following the original purchase order. Beyond cost, the physical weight of tool changer mechanisms, located at the extreme outer end of a serial-link robotic arm, limits the useful payload and dynamic response of the entire system. The additional length of the tool changer increases the critical distance between the wrist center and payload center, degrading kinematic flexibility, dynamic response, and safety. Description of the BarrettHand Flexibility and durability in a compact package The flexibility of the BarrettHand is based on the articulation of the eight joint axes identified in Figure 3. Only four brushless DC servomotors, shown in Figure 4, are needed to control all eight joints, augmented by intelligent mechanical coupling. The resulting 1.18kg grasper is completely self-contained with only an 8mm diameter umbilical cable supplying DC power and establishing a two-way serial communication link to the main robot controller of the work cell. The graspers communications electronics, five microprocessors, sensors, signal processing electronics, electronic commutation, current amplifiers, and brushless servomotors are all packed neatly inside the palm body of the grasper. The BarrettHand has three articulated fingers and a palm as illustrated in Figure 5 which act in concert to trap the target object firmly and securely within a grasp consisting of seven coordinated contact vectors one from the palm plate and one from each link of each finger. Each of the BarrettHands three fingers is independently controlled by one of three servomotors as shown in Figure 6. Except for the spread action of fingers Fl and F2, which is driven by the fourth and last servomotor, the three fingers, Fl, F2, and F3, have inner and outer articulated links with identical mechanical structure. Each of the three finger motors must drive two joint axes. The torque is channeled to these joints through a patented, TorqueSwitch mechanism (Figure 7), whose function is optimized for maximum grasp security. When a fingertip, not the inner link, makes first contact with an object as illustrated in Figure 8, it simply reaches its required torque, locks both joints, switches off motor currents, and awaits further instructions from the microprocessors inside the hand or a command arriving across the communications link. But when the inner link, as illustrated in Figure 9, makes first contact with an object for a secure grasp, the TorqueSwitch, reaches a preset threshold torque, locks that joint against the object with a shallow-pitch worm, and redirects all torque to the fingertip to make a second, enclosing contact against the object within milliseconds of the first contact. The sequence of contacts is so rapid that you cannot visualize the process without the aid of high-speed photography. After the grasper releases the object, it sets the TorqueSwitch threshold torque for each finger in anticipation of the next grasp by opening each finger against its mechanical stop with a controlled torque. The higher the opening torque, the higher the subsequent threshold torque. In this way, the grasper can accommodate a wide range of objects from delicate, to compliant, to heavy. The finger articulations, not available on conventional grippers, allow each digit to conform uniquely and securely to the shape of the object surface with two independent contact points per finger. The position, velocity, acceleration, and even torque can all be processor controlled over the full range of 17,500 encoder positions. At maximum velocity and acceleration settings, each finger can travel full range in either direction in less than one second. The maximum force that can be actively produced is 2kg, measured at the tip of each finger. Once the grasp is secure, the links automatically lock in place allowing the motor currents to be switched off to conserve power until commanded to readjust or release their grasp. While the inner and outer finger-link motions curl anthropomorphically, the spread motion of Figure 10 is distinctly non-anthropomorphic. The spread motion is closest in function to a primates opposable (thumb) finger, but instead of one opposable finger, the BarrettHand has twin, symmetrically opposable fingers centered on parallel joint axes rotating 180 degrees around the entire palm to form a limitless variety of gripper-shapes and fixture functions. The spread can be controlled to any of 3,000 positions over its full range in either direction within 1/2 second. Unlike the mechanically lockable finger-curl motions, the spread motion is fully back drivable, allowing its servos to provide active stiffness control in addition to control over position, velocity, acceleration, and torque. By allowing the spread motion to be compliant while the fingers close around an object, the grasper seeks maximum grasp stability as the spread accommodates its position, permitting the fingers to find their lowest energy states in the most concave surface features. Electronic and mechanical optimization Intelligent, dexterous control is key to the success of any programmable robot, whether it is an arm, automatically guided vehicle, or dexterous hand. While robotic intelligence is usually associated with processor-driven motor control, many biological systems, including human hands, integrate some degree of specialized reflex control independent of explicit motor-control signals from the brain. In fact, the BarrettHand combines reflexive mechanical intelligence and programmable microprocessor intelligence for a high degree of practical dexterity in real-world applications. By strict mathematical definition, dexterity requires independent, intelligent motor control over each and every articulated joint axis. For a robot to be dexterous, at least n independent servomotors, and sometimes as many as n + 1 or 2n, are required to drive n joint axes. Unfortunately, servomotors constitute the bulkiest, costliest, and most complex components of any dexterous robotic hand. So, while the strict definition of dexterity may be mathematically elegant, it leads to impractical designs for any real application. According to the definition, neither your hand nor the BarrettHand is dexterous. Naturally, their superior versatility challenges the definition itself. If the BarrettHand followed the strict definition for dexterity, it would require between eight and 16 motors, making it far too bulky, complex, and unreliable for any practical application outside the mathematical analysis of hand dexterity. But, by exploiting four intelligent, joint-coupling mechanisms, the almost-dexterous BarrettHand requires only four servomotors. In some instances reflex control is even better than deliberate control. Two examples based on your own body illustrate this point. Suppose your hand accidentally touches a dangerously hot surface. It begins retracting itself instantly, relying on local reflex to override any ongoing cognitive commands. Without this reflex behavior, your hand would burn while waiting for the sensations of pain to travel from your hand to your brain via relatively slow nerve fibers and then for your brain, through the same slow nerve fibers, to command your arm, wrist, and finger muscles to retract. As the second example, try to move the outer joint of your index finger without moving the adjacent joint on the same finger. If you are like most people, you cannot move these joints independently because the design of your hand is optimized for grasping. Your muscles and tendons are as streamlined and lightweight as possible without forfeiting functionality. The design of the BarrettHand recognizes that intelligent control of functional dexterity requires the integration of microprocessor and mechanical intelligence. Control electronics Inside its compact palm, the BarrettHand contains its central supervisory microprocessor that coordinates four dedicated motion-control microprocessors and controls I/O via the RS232 line. The control electronics, partially visible in Figure 4 are built on a parallel 70-pin backplane bus. Associated with each motion-control microprocessor are the related sensor electronics, motor commutation electronics, and motor-power current-amplifier electronics for that finger or spread action. The supervisory microprocessor directs I/O communication via a high-speed, industry-standard RS232 serial communications link to the work cell PC or controller. RS232 allows compatibility with any robot controller while limiting umbilical cable diameter for all power and communications to only 8mm. The openly published grasper communications language (GSL) optimizes communications speed, exploiting the difference between bandwidth and time-of-flight latency for the special case of graspers. It is important to recognize that graspers generally remain inactive during most of the work cell cycle, while the arm is performing its gross motions, and are only active for short bursts at the ends of an arms trajectories. While the robotic arm requires high control bandwidth during the entire cycle, the grasper has plenty of time to receive a large amount of setup information as it approaches its target. Then, with precision timing, the work cell controller releases a “trigger”command, such as the ASCII character “C” for close, which begins grasp execution within a couple milliseconds. Grasper control language (GCL) The grasper can communicate and accept commands from any robot-work cell controller, PC, Mac, UNIX box, or even a Palm pilot via standard ASCII RS232-C serial communication the common denominator of communications protocols. Though robust, RS232 has a reputation for slow bandwidth compared to USB or FireWire standards, but its simplicity leads to small latencies for short bursts of data. By streamlining the GCL, we have achieved time of flight to execute and acknowledge a command (from the work cell controller to the grasper and then back again to the work cell controller) of the order of milliseconds. The initial effort to develop a highly optimized grasper language based on such a standard protocol means that the GCL is upwardly compliant with any future industry-standard protocol. The grasper has two control modes: supervisory and real time. Supervisory is the normal mode used to control the grasper. It is made up of a simple command structure, designed for optimal performance and minimized learning curve. Supervisory mode has the following grammatical structure: Object (prefix) Verb (command) Subject (parameters) Qualifiers (values) The prefix refers to motors 1 through 4 with the ASCII values for 1, 2, 3, and 4 corresponding to the fingers Fl, F2, F3, and the spread motion. Any number of prefixes may be used in any order. If the prefix is omitted, then the grasper applies the command to all available axes. As an example, the ASCII character “C” represents the command which drives the associated motor (s) at its individual default (or user defined) velocity and acceleration profile(s) until the motor(s) stops for the default (or user defined) number of milliseconds. As each motor reaches this state its position is locked mechanically in place. 1C closes finger Fl. 2C closes finger F2. 12C closes fingers Fl and F2. C is equivalent to 1234C and closes all three fingers and the spread motion. We also have defined “S” (derived from “spread”) as a shortcut for “4” and “G” (from “grasp”) as a short cut for “123”, so that: GC is equivalent to 123C SC is equivalent to 4C There are similar commands for opening fingers, moving any combination of the four axes to an array of positions, incremental opening and closing by default or user-defined distances, reading and setting user-defined parameter values, and reading the (optional) strain gages on the three fingers. The latest version of the BH8-250 firmware has 21 commands and 28 parameter settings, giving it almost unlimited flexibility. The real time mode is reserved for advanced uses such as real time teleportation control and is frequently accessed through Barretts user-friendly GUI for PCs running Windows95/98/NT. In real time mode, the user specifies a tailored packet-structure in supervisory mode. Barretts PC software gives the user a histogram of 20 successive time-of-flight tests so that the user can refine the packet structure by quantitatively balancing information content with latency. The GUI accelerates the prototyping of tasks and includes a pictorial of the grasper with sliders for position and rate control. The GUI also has a novel “Generate C+ Code” button which enables anyone to save and later recall successful algorithms without any knowledge of C or C+ programming. But, with C+ programming familiarity, you can also edit the code as desired. Once real time mode is initiated, packets are exchanged in full duplex until an ASCII control character is issued to break out of real time mode and return to supervisory mode. The system has proven effective and robust in a variety of customer applications. Conclusion Although the BarrettHand BH8-250 was only introduced commercially in 1999, 30 units have been put into service around the globe at a price of US$30,000 each. The largest concentration of graspers is among automotive manufacturers and suppliers in Japan, including Honda, Yamaha Motorcycles, and NGK (ceramic substrates for catalytic converters). At this time, these manufacturers are only beginning to explore the capabilities of this versatile device, while some customers, such as Fanuc Robotics and the US and Japanese space programs have become repeat customers. Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 MCB 工业的机械手论文 巴雷特机械手爪可编 程式可弯曲部分的搬运和组装 摘要 本文详细介绍了巴雷特机械手爪 BH8 250 型的设计和运行，一个智能的，灵活的八轴夹具 , 一个可以随时进行自我完善 ,改变或者中断各种危险行为的工具。机械手爪带来巨大的价值 -工厂自动化 ,因为它 :降低所需机器人工作单元的数量和尺寸 (平均每项 90,000 美元不包括高成本的占地面积 ),从而提高了工厂的生产能力 ,通过一个可编程平台控制整合了各种各样的机械手抓 ；渐渐的改进和推出新产品介绍，通过工厂里的软件进行国际联网。 介绍 本文介绍了一种新的方法来进行材料处理，零件分类和构件组装 ，我们称它为“抓”，即一个单一的嵌入式智能可重构机械手爪，取代了独特的，固定形状的夹子和整个换刀库。指导的目的是为了感谢巴雷特机械手爪的设计，今天我们必须探讨什么是错误的机器人技术，机器人在未来的巨大潜力，以及以前遗留下来的行不通的手爪的解决方案。 为了实现机器人的优秀解决方案，可编程的灵活性，需要沿着整个机器人的设计史，从它的诞生，到现在。机器人手臂能够从基础的编程升级到灵活性的钣金，让机器人的外壳越来越薄。但即使要让这些机器人的外壳变薄，也必须嵌入智能软件 Excel，以确保每个机器人能正常运转和适应新的 复杂的功能。就像在串行链中最薄弱环节，不灵活的爪子限制了整个工作单元机器人的生产力。 机械爪子已经进行了独特的设计，但是固定颚板的形状还没有确定。在设计的过程中，一般难以预计硬盘成本和进度的范围。一般来说，机器人的每个形状、方向和接近角的预期的变化，需要其他自定义，但是爪子固定的位置，存放爪子的地方和更换爪子的器械，是不容许擅自改变和增加的。 相比之下，巴雷特的专利机械手爪如图 1 所示，机械结构，自动重新配置和高度可编程性，不到一秒钟匹配，地工作单元不停顿的数据交换量，交换手爪的几乎任意形状的变换功能。 对于 需要处理的可变等多种有效载荷的方向，提出了高度灵活性的任务，一个能让机械手爪更安全，更快捷的安装，以及比定做加工夹具更低的成本和大容量的存储机架。 不间断运行时，工作单元只有一个或两个备用机械手爪可以作为应急备份，而一个或两个备用的手抓，是要求每个手爪都能变化 - 可能每个工作单元需要几十人。而且，悲剧的是如果两个手爪都系统备份，如果失败，因为它会存储前几天的可以识别的数据，很多自定义形状，装运和自身装配，所以会影响后面的操作。与此相反，由于机械手爪是数据相同，他们总是可以通过特定的软件及时提供无限量的数据 。 传统夹具 今天的机器人，装配零件的处理大部分是通过夹具。如果表面的条件允许，真空吸力和电磁铁也可以应用，例如：处理汽车挡风玻璃和车身。作为部分尺寸开始超过 100gms，颚板的自定义形状，以确保安全运行。由于处理和装配耐用的主体，这些工具没有什么变化，因为机器人是从三十年以前才开始的。 夹具，可以看做简单的类似钳子的动作，有两个或三个非铰链手指，被称为“钳口”，保持平行或者进行打开或者观关上的运动，如 2 所示。良好的组织目录可从制造商那里得到客户所匹配（除了自定义形成的素具）的各种组成手爪所需要的部分任务和 部分参数。 大型液压夹钳的有效载荷的尺寸范围从微克至 100+公斤。驱动是典型的气动或液动 ,用简单的开 /关阀控制切换全开或全闭状态。钳口通常移动 1cm就可以全开至全关。这两只手，两个或三个手指，被称为“钳口”。对钳口部分的制造目标是可移动的机械软钢或铝，被称为“软钳口”。 在特殊的情况下，工具专家设计师决定要对矩形软钳口件进行自定义加工。一旦加工形状完成，软钳口夹持器就要连接到各自的机构进行测试。这一过程可以采取任何数量的迭代和调整，直到系统正常工作。模具设计者需要时间来重复整个过程中每一个新的形态。 随着消 费者需求产品选择的多元化，更加频繁的产品介绍，对可调节自动化的需求前所未有地强烈。然而，并没有使机械手爪更全面，在过去几年中机器人产业一直遵循的工具自动来交换数控磨刀具技术的例子。 但是，应用换模型串行工具连接机器人证明是昂贵的和无效的。不同于使用标准规范外的铣床工具进行成品切割，机器人工具设计者必须定制每一套机械爪的形状 ， 一个耗时多、昂贵且难以确定范围的任务。虽然机械手爪只便宜了 500 美元，但是每个工人努力制造机械手爪会耗资数倍。通过上面的例子，如果你要增加 12 个爪子的换刀架和刀具库，那么花费就不止10,000 美元，而是剧增到 20,000 美元至 60,000 元。 更严重的是，在定制过程中准确的预测成本将是未知数。因此，客户必须提交一份包括初装费、所需时间和材料的成本在内的采购订单，这是是交易的基础。虽然在美国售价 30,000 美元，智能机械手爪并不便宜。然而，人们可以“定制”在某一天运行一个小时软件来验证的系统的性能。如果系统的性能不能达到目标要求，那么那一天的劳动是白费的。如果系统成功，那么按照原订单将没有任何隐藏的费用。 除了成本、转换机制的物理重量，在末端串行连接的机械臂外，限制了有效载荷和整个系统的 动态响应。该转换长度增加了额外的有效载荷中心，运动的灵活性，动态响应和关键的安全距离。 巴雷特机械手爪的说明 灵活性和耐久性的简洁介绍 巴雷特机械手爪在灵活性的基础上，按图 3 中确定的八种联合轴连接。只有四个直流伺服电动机，如图 4 所示，需要控制所有八个关节，由智能机械耦合增强。由此产生的机械手爪是自身总共只有一点一八公斤重，只有 8mm直径的连接电缆提供直流电源，建立双向串行通讯连接到机器人的主要工作单元控制器中。该机械手爪的通讯电子，五微处理器，传感器，信号处理电路，电子换相，电流放大器，和伺服电机都整齐的安 装在机械手的体内。 巴雷特机械手爪的手掌有三个手指关节，如图 5 所示，通过联系手掌及每个手指，协调他们的动作，能让他们在目标范围内，牢牢的抓住物体。 每个巴雷特机械手爪的三根手指都是由三个独立伺服电机控制，如图 6所示。除了 F1 手指和 F2 手指的伸展动作 , 都是由伺服电动机驱动， F1、 F2、F3 三个手指的内外部的机械结构都是用一样的铰链连接的。 三个手爪的每个马达都必须驱动两个关节轴。通过专用通道来控制扭矩，扭矩控制器如图 7 所示，这些关节的作用是能让爪子在抓东西时更安全。如图 8 所示，当手爪第一次接触物体时，需要 传递简单的扭矩，确定他们的接合处，关闭电机电流，并等待手抓内部的微处理器发出进一步的指示命令。 但是，当扭矩控制器如图 9 所示时，为使安全和控制，扭矩控制机制第一次接触物体时，用事先设置好的扭矩的上限值，再次确定物体螺纹的联系，并当手爪第二次接触时，在一毫秒之内更改第一次接触时的所有扭矩信息。接触顺序快速确定，没有高速摄像机的帮助下你将无法想象其中的过程。当机械手爪放开被抓对象时，它会让扭矩控制器对每个手爪的扭矩的上限值进行设定，而可控扭矩可以让机械的停止的爪子再一次打开。当扭矩达到最高，那就是它的上限。这样 的话，机械手爪就有一个可抓起物体重量的准确范围。 与传统的机械手爪不同，这种手抓里的连接，容许每个手指的两个一致的独立接触点来稳固的抓住物体的表面。不管位置，速度，加速度，还是扭矩，都可以通过所有 17,500 种型号的编码器进行控制处理。当速度和加速度设置为最大时，每个手指可以在不到一秒钟的时间内向任何一个的方向上进行运动。通过测量，每个手指都能够迅速的产生 2 公斤的力。一旦确定稳定的抓住物体时，则链接将自动锁定，并关掉电机，这样可以节省电源，直到需要重新调整或者放开物体时，电机会自动打开。 虽然手抓的张开与收 缩是拟人化，但如图 10 所示，清楚的显示了机器的非拟人化。机器伸展的动作非常近似于灵长类动物的手指（拇指）的动作，但是又不能替代，巴雷特机械手爪还有一种类似的手爪，整个手爪可以旋转180 度，手指中心对称且平行于连接轴，是一种可以抓住各种夹具形状的多功能手爪。 在 1/2 秒内可控制的机器在 3000位置中任意伸展。不同的是机械锁定了手爪伸缩的运动，使伸缩运动变回了可驱动状态，允许其控制手爪的位置，速度，加速度和扭矩。通过允许手爪的伸缩运动，使手指紧密的感应物体，机械手爪的主要目的是找到一个接触面积最大的同时耗 能又是最小的地方。 电子和机械的优化 可编程机器人智能灵巧的控制是成功的关键，不论是控制机械手臂，自动引导