Engineers who design industrial machines still have to contend with a hodgepodge of networking technologies — perhaps a fieldbus or two for all the I/O, a proprietary motion control network and Ethernet for everything else. Supporting all those networks, though, doesn’t come cheap. Think of all the wires to pull, hardware to buy and software to write. Wouldn’t it be nice if one network could just do everything?
Well, that’s exactly the appeal of industrial Ethernet. With its ability to reduce the installed cost of factory automation and play nice with corporate IT systems, industrial versions of Ethernet are poised for huge growth in the coming years. According to a recent study by Harry Forbes, a senior analyst with the ARC Advisory Group, the market for industrial Ethernet switches and related network infrastructure will grow to nearly $1 billion over the next five years, up from $260 million in 2006.
It’s not just IT guys who will drive that growth. “Engineers think if they can switch over to industrial Ethernet, they can connect everything in the factory with a length of CAT5 cable and networking devices that they buy at their local electronics store,” says Scott Hibbard, vice president of technology for Bosch Rexroth Drives and Controls. “It’s really a utopian view of automation.”
Until recently, motion control has been blocking the view. “To use Ethernet for motion control, it has to do things it was never designed to do,” Hibbard says. Ethernet, as it was originally conceived, simply didn’t provide the cycle time, determinism or quality of service needed to close high-performance servo loops in real time or update complex trajectories on the fly. “Ethernet TCP/IP wasn’t originally something you’d want to run motion over,” says Ed Nicolson, a Ph.D. electrical engineer and director of technology development for Yaskawa Electric.
Over the past seven years, however, engineers have developed clever ways to make Ethernet networks offer real-time performance suitable not just for less demanding I/O applications but also for complicated motion control tasks. And Forbes predicts a bright future for motion over Ethernet. “Many if not most of the motion control applications will eventually migrate to industrial Ethernet,” he says.
But first, engineers will have to figure out which of the competing approaches to industrial Ethernet best meets their motion control needs. It’s not an easy call given the number of network technologies that make some claim to real-time performance. Currently, the Real Time Ethernet website lists 22 of them.
Only a handful of them qualify as front runners, open standards that will likely gain a foothold in larger automation systems requiring significant amounts of motion control. They include PROFINET, SERCOS III, Ethernet/IP, EtherCAT, ETHERNET Powerlink and Modbus. Of these, some motion experts exempt Ethernet/IP and Modbus from the real-time club because they don’t currently have performance on par with the other four networks.
Ethernet/IP, one of the dominant industrial Ethernet variants in North America, is in the midst of beefing up its real-time capabilities with the addition of CIP Sync, a network clock synchronization protocol built on IEEE 1588, according to Matheus Bulho, motion product manager for Rockwell Automation. But actual motion control products that use CIP Sync have yet to hit the streets. Until they do, many motion applications will be beyond Ethernet/IP’s reach, according to Bill Seitz, president of IXXAT Inc., a developer of communication hardware and software for industrial and automotive applications. “Think of Ethernet/IP and Ethernet/IP with CIP Sync as two different networks for the purposes of motion control,” says Seitz, who has familiarity with Ethernet/IP through his company’s activities as a Rockwell value added design partner. Nicolson also thinks “CIP Sync could change the game” for Ethernet/IP. “Without it, I don’t believe Ethernet/IP is deterministic enough for anything more complex than sending simple end-point commands,” he says.
Making industrial Ethernet even more confusing is all the hype surrounding the various networking technologies. Proponents of each real-time Ethernet contender have been engaging in no small amount of “specsmanship” as they jockey to come out ahead of the pack. Much of the jockeying takes the form of “my cycle time is faster than your cycle time” or the even geekier “my network is more deterministic than your network.”
Cycle time and determinism obviously matter a great deal in motion control applications, but these specs alone won’t tell you which network is best for your application. Here’s why:
When it comes to each network’s cycle time — or the time it takes for the network’s nodes to communicate — it’s tempting to think faster is always better. As things stand now, though, none of the four highly deterministic networking technologies can claim a truly compelling cycle time advantage in the context of all but a few real-world motion requirements (see sidebar). “All of the deterministic Ethernet approaches available today offer enough performance for the vast majority of motion applications,” says Seitz.
With cycle times of 1 ms and below, all the fastest networks offer more than enough performance if you look at the response times required in typical machine applications. Zuri Evans, manager for Siemens’ SIMOTION products, says the hydraulic axis on injection molding machines typically requires a response time between 250µ and 1 ms, while the axes on printing machines normally need about 3 ms. He says converting machines range between 2 and 6 ms and packaging equipment between 2 and 4 ms. “PROFINET is more than capable of meeting these requirements,” he says.
And so are the other three fastest networks. Remember that only a few microseconds of cycle time separate the fastest networks. And time differences that small rarely come into play in real-world motion control applications — because the dynamics of mechanical components and the frequency response of many industrial machines impose the speed limit rather than network cycle times. “You’ll run up against issues with dynamics of your mechanical systems long before you notice microsecond differences in Ethernet cycle times,” Nicolson says. “Engineers need to ask themselves what kind of cycle times they really need,” he says.
He’s not alone in that assessment. “Remember that real time is a relative concept,” says Wayne Baron, president and former technical director at Galil Motion Control. “It all depends on what the application actually requires.”
Jitter, or the network’s ability to deliver communications at a precise time and an indication of its determinism, falls into the same boat as cycle time in the sense that the four key motion-capable networks all deliver adequate performance with jitter values less than 1 µs. “In most applications a bit of jitter won’t hurt you,” says Baron. “You can easily account for it in the software.”
What really differentiates the various real-time networks is how they deliver determinism. Bosch Rexroth’s Hibbard, who also serves on the board that oversees the SERCOS III standard, notes there’s a spectrum of technical responses to standard Ethernet’s inherent lack of determinism — which can mostly be traced to the original Ethernet’s collision detection mechanism and its tendency to occasionally and unpredictably delay the delivery of data packets.
One end of the spectrum consists of approaches that try to make standard, unmodified Ethernet and TCP/IP behave in a deterministic way using IEEE 1588 to synchronize clocks across the network. This is roughly the approach taken by Ethernet/IP with “CIP Sync” implemented.
Hibbard describes the other end of the spectrum as the “scorched earth” approach. “You rip apart the old Ethernet standard until all that’s left is bits of copper and connectors,” he jokes. He puts proprietary motion networks that only make use of Ethernet’s physical layer in this category.
Open deterministic networking standards like SERCOS III, PROFINET IRT, EtherCAT and ETHERNET Powerlink fall somewhere in between and use hardware, software stacks or both to prioritize and separate deterministic communications from the network traffic that’s less time sensitive. “All of them have a decent amount of determinism,” Hibbard says. “The real question is how they get it. That’s where the trade-offs come in.” Jeremy Bryant, a networking specialist for Siemens, agrees. “The different approaches to maintaining determinism have important consequences beyond motion control,” he says.
These consequences include the cost of the system, since different types of networks differ substantially in the amount of hardware they need to implement determinism. Depending on the type of network, this hardware can range from off-the-shelf or embedded Ethernet switches to custom switches to special ASICs or even FPGAs. Proponents of each type of industrial Ethernet are locked in heated debate as to which one is the most cost effective. For his part, though, IXXAT’s Seitz estimates the various industrial Ethernet networks cost somewhere between $30 and $50 per node depending on how much specialized hardware and application development they need to achieve determinism.
Another important consequence of how determinism is achieved relates to the Ethernet control network ability to handle non-motion tasks, which is one of the chief benefits of going to industrial Ethernet in the first place. “It is imperative to implement an Ethernet-based solution which is powerful enough to handle the motion requirements, while flexible enough for all of the other network requirements,” says Bryant. These requirements might include I/O for machine control, diagnostic data, OPC for data collection, HMI connectivity, safety and more.
To transmit all this data, all the real-time deterministic networks have access to the same 100 mbit/s of bandwidth (100BaseT), “but we all use that bandwidth more or less efficiently,” says Helmut Kirnstoetter, a vice president at B&R Industrial Automation, the company that devised ETHERNET Powerlink. This efficiency, which still sparks hot debates about the optimum size and handling of data packets, affects the number of nodes a system can handle and the amount of bandwidth left over for other tasks on the same network.
“At the end of the day, engineers have to ask themselves ‘what cycle time do I need, how many nodes do I have and what kind of hardware do I need to get the job done,’” Kirnstoetter says. Answers those three questions accurately and you’re well on your way to picking the right network.
In Molding, Cycle Time Matters
|While cycle time of communications buses do matter in motion applications, there is a “point of diminishing returns when you’re trying to control a mechanical system,” according to Chris Choi, technology director of Husky Injection Molding Systems Ltd. He should know. His company builds some of the world’s fastest injection- molding machines for applications that include high-speed production of bottle preforms and thin-wall packaging.
Like some of the other motion control experts cited in this article, Choi points out the small cycle time differences between the various real-time industrial Ethernet networks don’t make much of a difference when considered in the context of the dynamic response of most actuators and mechanical components used in industrial machines. “Factors such as inertia are usually more of limiting factors than fieldbus cycle time in mechanical systems,” he says.
Yet he’s quick to add that in the world of high-speed injection molding even a seemingly miniscule amount of cycle time can make a big difference. He says Husky’s goal with machine control is not simply to control the actuation of the machine’s mechanical elements but to control the injection-molding process. To do that, Husky’s control algorithms have to infer the state of the plastic melt — it’s temperature, pressure, shear conditions and more — from the machine’s behavior. “Every microsecond counts when you’re controlling a fast process like this,” he says.
So when Husky began to work on a new control architecture for its high-speed bottle and packing machine, Choi embraced EtherCAT as the new communications bus. “We’re in the process of implementing EtherCAT now,” he says. With EtherCAT, Husky is capable of seeing reaction times in its process control system as fast as 100 µs, Choi says. He says Husky picked EtherCAT not just for the speed of its network communications but also because it processes its unique data packet on the fly, without the use of switches or other hardware that can add to the overall reaction time.
Industrial Ethernet Hardware Developments
With industrial Ethernet poised to grow rapidly over the coming years, suppliers have already started to roll out switches and Ethernet-ready controllers tailor made for real-time control applications.
Rockwell Automation recently took a big step in that direction. At its Automation Fair, the company unveiled a line of managed and unmanaged industrial Ethernet switches, including models that integrate the Catalyst switch architecture from Cisco Systems Inc. Managed switches, in both modular and embedded versions, have features that should prove useful in high-performance motion applications over Ethernet/IP. These features include IEEE-1588 time synchronization and prioritization of network traffic for improved quality of service (QoS). The new switches will allow set-up and diagnostics from within Rockwell’s Integrated Architecture. Rockwell plans to make the new switches available starting mid-2008.
Another recent hardware development with real-time implications comes from Innovasic Semiconductor, which has developed 32-bit microcontrollers with built-in provisions for real-time applications. These fido microcontrollers — with “fido” standing for “flexible input, deterministic output” — feature an architecture that puts much of the real-time functionality into the chip rather than relying on RTOS software. “We call it an RTOS Kernel In A Chip,” says Dave Alsup, Innovasic’s R&D director. Constructed on a set of five “hardware contexts” that work like individual virtual CPUs, the RTOS Kernel performs real-time operations such as scheduling, priority control, memory protection and timer control.
The result is low jitter and determinism “even in the presence of low-priority traffic,” says Keith Prettyjohns, Innovasic’s CEO and a Ph.D. electrical engineer. He points to recent lab tests in which Schneider Electric engineers processed a high-priority Modbus/TCP application along with background Ethernet traffic on a fido 1100 microcontroller and an ARM9-based processor. According to the study, the fido 1100 microcontroller achieved a worst-case latency of 1.1 msec and worst-case jitter of 130 µsec, while the ARM9-based controller achieved a worst-case latency of 1.8 msec and worst case jitter of 760 µsec. The fido won out despite a slower clock speed of 66 MHz, versus 133 MHz for the ARM9.
Alsup says the fido microcontroller also includes other features that will prove useful in control applications. One is a deterministic cache. It provides a permanent storage space for pieces of code that need to be executed quickly, thereby eliminating the jitter associated with cache misses. The other is programmable I/O capabilities in the form of four Universal I/O Controllers (UICs). Each of these can be programmed to support a variety of I/O protocols.