After 25 years, the basic control architecture of the robot industry is changing. Here?s the network causing all the commotion.
By Joseph E. Campbell
Adept Technologies Inc.
Edited by Miles Budimir
High-speed digital servo networks are paving the way to new control architectures for industrial robots. And the IEEE-1394 FireWire protocol is emerging as a leading standard in high-performance motion control. It has the bandwidth to support network traffic, distributed processing, and heavy data rates from advanced sensors such as machine vision.
ALL FIRED UP
FireWire got its start in the consumer market, and is still most prevalent in digital video cameras, desktop publishing, and business applications such as video conferencing. The emergence of digital video and multimedia applications brought with it the need to quickly move large amounts of data between peripherals and computers.
FireWire offers guaranteed delivery of multiple data streams through isochronous data transport. Using special ICs, FireWire multiplexes different types of digital signals such as compressed video, digitized audio, and device-control commands on two twisted-pair conductors (a total of four wires). It uses what's called a fairness arbitration approach to assure that all nodes having information to transmit get a chance to use the bus. Standard Ethernet doesn't provide this type of arbitration.
Like Ethernet, FireWire is a layered transport system. The IEEE-1394 standard defines three layers as physical, link, and transaction. The physical layer provides signals required by the FireWire bus. The link layer takes raw data from the physical layer and formats it into recognizable 1394 packets. The transaction layer takes the packets from the link layer and presents them to the application.
The current transfer speed called out by the IEEE-1394 standard is 400 Mbits/sec, with a revision that delivers 800 Mbits/sec. The newest FireWire standard, IEEE-1394b, allows data rates up to 1.6 Gbits/sec at distances to 100 m. This high bandwidth delivers communication speeds equal, or more typically higher, than actual net backplane bus communication rates. For example, the typical net bandwidth from a VME backplane, after factoring in all latencies and overheads, is 128 Mbits/sec. And because it's a network, the FireWire bandwidth can be used for distributed processing and control.
In comparison, the newest Sercosbased systems run at up to 16 Mbits/sec. While this is adequate for remotely controlling multiple axes of motion, it is not enough bandwidth for advanced dataintensive functions such as trajectory planning under full kinematics coupled with handling machine vision data and distributed processing. Furthermore, FireWire is deterministic and offers multiple communication modes including an isochronous transfer mode that guarantees message handling.
Another advantage of FireWire is cost. The consumer market is driving down the cost of hardware and software drivers. The push toward digital image processing facilitates the development of FireWire-based network cameras. Coupled with high bandwidth and distributed-processing capabilities, this means a single FireWire cable can carry the digital I/O, servo commands, and vision images to and from the robot mechanism and end effector.
FireWire features a guaranteed and deterministic bandwidth, critical for distributed processing of time-critical motion-control functions. While Ethernet offers 100 Mbits/sec transmission rates, it has no means to ensure that transmissions occur at a specified time or regular intervals. FireWire also supports hot-pluggable devices, automatic ID assignment to network devices, and software-driven parameters. Each bus node supports up to 63 devices.
CONQUERING NEW TERRITORY
While it's still dominant in the consumer world, FireWire has slowly been moving into the industrial arena. In 1999, Ormec announced a general-purpose servo network based on IEEE1394, called ServoWire. It supported up to eight axes in traditional machinecontrol applications such as precise electronic gearing, electronic cams, and profiling. More recently, Yaskawa, NEC, and Mitsubishi have been collaborating to develop a FireWire-based digital servodrive interface. Meta Controls and Sony have developed FireWire-based industrial cameras, and Meta has introduced a line of FireWire network servoamplifiers called FireBlox. And NyQuist Industrial Control has introduced a FireWire interface module called the FireWire Connection, and has announced partnerships with Control Techniques and Kollmorgen.
Adaptec has created its own network based on FireWire called SmartLink, which reduces wiring requirements to a few twisted pairs. Traditional robot architectures use heavy multiconductor cables and connections between the controller's motion-control board and the power amps, and between the power amps and the robot mechanism. These cables and the traditional motion-control/interface board are now replaced by the Adept SmartLink network and its twisted-pair physical layer.
Large, multiconductor cables are also difficult to install, often requiring a dedicated wireway or large-diameter conduit. Also, cables have historically been a reliability problem, because each wire crimp and contact-connection point is a potential failure point. SmartLink replaces hundreds of contact points with a simple, standard network connection. For example, on a four-axis semiconductor tool machine, wire count dropped from 150 to 15. These wires carry power, encoder feedback, and data from sensors and cameras.
In addition, SmartLink is a distributed processing network architecture in which the controller's CPU runs the trajectory planner, while the servoloop is closed in Adept's SmartAmp amplifier. SmartLink eliminates the servoamp panel, motion interface board, and expansion slots in the primary controller, a 70% reduction in panel space. This processing scheme delivers the best performance and flexibility.
Also, SmartLink becomes a way to add controller features. In a traditional architecture, for instance, the controller includes expansion slots and attendant power-supply capacity, connectors, rack space, and increased footprint. Now, additional features such as digital I/O, general-purpose motion control, and additional SmartLink mechanisms are added by simply connecting the module directly to the network.
OH THE ROBOTS THEY ARE A CHANGIN'
Besides the rise of high-speed digital networks, fundamental changes are taking place in the robotics industry. Robots and their controllers are morphing into new form factors. They are becoming less expensive, easier to deploy, and more reliable. Due to smaller microprocessors, controllers are shrinking to a fraction of their current footprint. Meanwhile, power electronics and servomotor processing are being distributed into the body of the robot.
The first robots used hydraulic drives and mechanical timer-based controls. Since then, industrial robots have evolved quite a bit. Robots nowadays generally consist of three major subsystems: controller, power amplifiers, and mechanism. Controllers handle the data-intensive number crunching needed to calculate trajectories and movements. Power amplifiers take signals from the controller and regulate the power to the motors on the mechanism.
Controllers and power amplifiers traditionally are housed in a single enclosure sometimes as big as a washing machine. Now, they can be separate rackmounted components. However, there is considerable wiring between components. Signal cables connect the power amplifiers to the controller, and fat power and signal cables connect the power amps to the mechanism.
Furthermore, while vendors continue to work on incremental cost reductions, the core components of most robot systems such as encoders, motors, connectors, cables, bearings, and castings don't offer sizable cost reductions. The basic architectures include some expensive elements (such as fat multiconductor cables), to say nothing of the fact that performance and reliability barely get addressed in the process.
Then along came the network. It started out in the factory as a means to connect terminals, PCs, and intelligent devices back to a host computer. Early applications were limited to data being delivered up, and new programs delivered down. This spawned the rapid growth of PLC networks, including peer-to-peer communications. Finally, device networks evolved, essentially replacing hard wiring for I/O and signal interfaces, including data-intensive applications like simple motion control.
The next major evolution was the digital servo network for robotic motion control, the most demanding, time-critical, data-intensive application. Digital networks like Sercos replaced standard ?10 V analog signals. Sercos linked the controller to the drive. These were new, smart controllers and drives, with DSPs doing loop closure. Plus, it used a noise-immune fiber-optic cable.
CRS Robotics was a pioneer in this field, with its C500C controller, and a proprietary servo network to connect to their F3 mechanisms. Key benefits for robot manufacturers included an 80% reduction in wire count, reduced failure rates, simplified scalability of advanced features, and reduced hardware costs from eliminating dedicated motion-control interfaces. Another early pioneer, Robotics Research Corporation, Cincinnati, developed its I Type robot family, an extremely scalable and flexible multiaxis anthropomorphic arm with servoamps embedded in each arm segment.