When General Motors Powertrain upgraded its manufacturing facility in Toledo, Ohio, speed was a critical element. GM plans daily production of 2,200 GF6 transmissions for its Chevy Malibu and Cruze lines, so every machine has to operate quickly with a high level of synchronization and precision.
That’s a high volume for complex six-speed transmissions, which contain many components. Networked automation is a key to the process, as GM strives to cut costs and press defects to ever-lower levels. This automation program begins by delivering workpieces with autonomous guided vehicles. The workpieces are tracked by radio-frequency identification (RFID) tags placed on each pallet. Model number, serial number and build status information are all contained in the tags.
While speed is a key factor for automotive production, quick changes are also critical. To meet the demands of GM’s flexible automated assembly system, all hardware and software is designed with a deterministic functionality. After a detailed deterministic study, GM opted to employ a high-level Ethernet protocol, Profinet, in which all the control elements for every assembly operation and test station would be fully integrated.
Third-party software package provider Elite Engineering Inc., of Rochester Hills, Mich., teamed up with the Siemens Automotive Center of Competence in nearby Troy, Mich., to devise a networking and production program that would meet these demands. Their networking selection highlights the growing role of deterministic Ethernet in demanding applications.
Profinet gives GM a deterministic network with no special hardware required, cutting costs. Further cost cutting was achieved by eliminating a safety network. All safety devices are now networked over Profisafe protocol, a certified safety network, eliminating time-consuming and difficult-to-maintain traditional hardwired safety connections.
The combination of determinism and safety shine the spotlight more directly on the real-time variations of Ethernet. When these networks handle multiple critical data sets, the importance of their speed and timing is matched by the need for reliability.
Once engineers determine which version of Ethernet best meets their demands, their design challenges shift to the topology. This choice will determine performance traits, as well as reliability, as it is critical to ensure that devices remain connected when a cable is disconnected or cut.
“Most of the time, these networks also require fault tolerance. Developers want to ensure that data gets there even if there’s some sort of break,” says Martins Jansons, network consultant at Siemens Industry Inc., the Alpharetta, Ga.-based automation supplier.
The networking topology will play a key role in determining the level of fault tolerance. It will also be a factor in costs. Some architectures require duplicate cabling, while others require some additional hardware. Selecting a topology is not a simple decision.
“We spend a lot of time with customers trying to determine which topology they need,” says Nate Holmes, Product Manager for Motion and EtherCat at National Instruments Corp. (NI), a test and automation supplier headquartered in Austin, Texas. “We usually break out six to 10 characteristics that will help them pick a topology. But it’s not always apples to apples; there are a number of lemons that make it hard to say this approach will work best in this environment.”
*****************************************************************
be sure to visit our Industrial Ethernet Supplement archive.
The level of communications jitter that the system maintains is one of these factors that NI’s integrators use. Others include understanding the level of determinism that’s needed and using the right technologies for distant nodes. Fiber optics may be the way to go for long distances, because shielding can cause signal degradation on long runs, Holmes explains.
Network specialists also note that users must clarify the level of determinism that their systems need. Determinism and real time are relative terms without solid definitions. Understanding the desired level of performance can help engineers pick networks and topologies that meet their needs without driving costs upward, most specialists agree. That’s especially true in switched networks, as eliminating a couple of switches can often save $100 or so. Eliminating hardware generally improves reliability by reducing the number of components that can fail.
Various topologies
Once engineers consider the relevant parameters, they can pick the best scheme. The ring architecture is one of the more common topologies for networks that require determinism. When nodes are connected to a ring, systems simply send signals in the opposite direction when there’s a break in a cable or when a node won’t pass data along. When the groups that manage networks narrow their options to just a couple of topologies, rings are usually one of the choices.
“SERCOS III (for serial real-time communications system) supports two architectures, lines or rings,” says Scott Hibbard, vice president of technology at automation vendor Bosch Rexroth Corp’s Hoffman Estates, Ill., office. “If you bring a cable back to the master to create a ring, you can send two counter-rotating telegrams that move in both directions. This improves jitter even more and gives you redundancy so you can have what a lot of people call burpless recovery.”
But rings are not an ultimate solution that will ultimately lead to a faultless changeover. Alternatives can bring benefits that give them an edge. For example, adding a second cable can effectively double bandwidth. “A lot of people use rings, but they can also use link aggregation,” Jansons says.
With link aggregation, at least two cables are run between switches. That can double the available bandwidth during normal operations and provide an alternate route if one cable loses connection. In many networks, less than half the available bandwidth is used, so operators barely notice any speed degradation when these cable faults occur, Jansons adds.
One of the simplest approaches is to simply run cables from one piece of equipment to the next. With daisy chains, new machines or nodes can be added by simply connecting the new gear to one of the other nodes.
“With a daisy-chain architecture for deterministic networks, designers don’t have to figure out what Ethernet switches will or will not work for their devices and network speeds. This would require each device to have at least two Ethernet ports,” says Derek Lee, motion product engineer at automation supplier Yaskawa America Inc., of Waukegan, Ill. If more fault tolerance is needed, it’s easy to convert a daisy chain topology to a ring by adding a cable that completes the loop.
Simplicity is always a key consideration. When determinism was added to Ethernet, ease of installation was a critical aspect of design. Engineers wanted to ensure that tricky tasks such as line termination did not become a problem.
“With EtherCat, you don’t have to do anything special to terminate the lines; the protocol was written to be flexible,” says Matt Spexarth, Industrial Embedded Platform marketing manager at National Instruments.
Another aspect of development is to ensure that slower devices don’t get in the way and slow down time-critical tasks. In some older architectures, speeds fall back to the rate of the slowest device that’s nearby. Many industrial networks link up-to-date equipment to time-tested products, which would create major problems for the newer systems.
“Within a single EtherCat network, the user can select standard devices with no deterministic needs running side by side with devices that have the tightest synchronization requirements,” says Joey Stubbs, a spokesman for the EtherCat Technology Group, in Austin, Texas.
Determining factors
Often, the tradeoffs among topologies are similar to those used to pick the networking architecture—precision and response time. Jitter impacts both the overall system response times and the networking requirements. That’s especially true in demanding motion-control set-ups.
“In motion, determinism and synchronization are both required,” Hibbard says. He notes that the requirements for synchronization will vary widely. For example, the SERCOS III motion control bus has a jitter specification that’s better than one microsecond.
“If you say you’ve got one microsecond, many people will say they need something better. Maybe it’s a maximum of 5 percent of the cycle. Often, they will want 1 percent or even 0.1 percent,” Hibbard says. At some point, the capabilities of microprocessors can trim timing down to levels that have little impact on the operating equipment.
For example, some of the application specific integrated circuits (ASICs) used in real-time networking can process commands in 30 nanoseconds. Few mechanical devices can respond in that timeframe, but there are still reasons that justify constant improvements in system and network speeds.
“It’s not just control information going down the wire, you’ve got diagnostics and other information that is being generated by hundreds of devices,” says Chuck Lukasik, director of the CC-Link Partner Association, based in Vernon Hills, Ill. “That consumes a lot of bandwidth.”
This need for high precision and high bandwidth underscores the tight demands of today’s full-throttle production speeds. Companies competing in global environments need to keep equipment running at peak speeds without unexpected stoppages.
“If you get into high-speed machine operations that require updates every x milliseconds, meeting that requirement 99 percent of the time doesn’t cut it. Having a lack of determinism will make that network architecture unusable,” Lukasik says.
Networking specialists note that the network and hardware won’t do much without the right software. In many instances, real-time operating systems play a critical role in deterministic networks. “We support real-time operating systems like Phar Lap and VxWorks,” NI’s Spexarth says.
