05 February 2001
Changing Control Landscapes
by John Guite
What does the future of industrial control hold in store?
In the ever-changing field of industrial control, one thing is certain: Industrial control users are continuously improving their equipment, and in doing so they require a communication protocol that can meet their needs now and in the future.
With the vast array of options available in today's market, choosing a communication standard has become more difficult than ever. Some of the more popular choices are fieldbuses (e.g., DeviceNet, Profibus-DP, and Profibus-FMS); PC buses such as peripheral component interconnect (PCI); and more general-purpose networks such as Ethernet. All three have their place within industrial control, but only Ethernet offers scalability and worldwide acceptance. Ethernet offers an attractive alternative to both existing fieldbuses and PCI-based motion control systems, due in part to its ability to transfer both device and high-level information at relatively high speeds.
Ethernet is also more accommodating than its industrial counterparts in that TCP/IP support is commonly provided with most off-the-shelf, real-time operating systems (RTOSs), and the hardware is readily available on most PCs today. For those worried about the lack of an Ethernet standard or its lack of determinism, a new application layer is currently being specified that will provide both a standard and deterministic performance. That being said, Ethernet isn't a panacea for all the problems associated with industrial control. Without the new application layer, determinism is still a question mark, and for pure high-speed data transfer it cannot compare with the performance of a PCI bus.
Until recently, there's been no better way to eliminate complex wiring and transmit device information than to use an industrial network such as DeviceNet or Profibus-DP. Each of these is designed to efficiently relay device information between a host (or master) and a slave device. Bandwidths ranging from 500 Kbps to 12 Mbps are, in most cases, adequate for transferring the limited information required by the slave devices on the network. Additionally, the major programmable logic controller (PLC) manufacturers offer available Profibus and/or DeviceNet master capabilities. Many third-party slave devices are available for both protocols, and the fact that each protocol has a standard administered by a governing body ensures device compatibility.
Unfortunately, each protocol is largely proprietary and restricted to industrial applications only. Although the protocols themselves are considered open standards, they lack worldwide acceptance, which is required to gain economies of scale and significantly reduce costs. As a result, it can be quite costly to implement even a very basic network using Profibus or DeviceNet. Couple this with the learning curve required in supporting such a network, and the costs may be prohibitive. Lastly, while the architecture associated with each network is specifically adapted for transferring device data to the host and vice versa, it isn't suitable for transferring information at the supervisory control and data acquisition (SCADA) level. Most SCADA networks use TCP/IP over Ethernet as the network protocol and physical layer. This means that a PLC must act as a gateway between the device level using a fieldbus and the SCADA level using Ethernet TCP/IP. This, in turn, requires the ability to support both types of network wiring and communication protocols, as shown in Figure 1.
A simple solution to this problem exists in Ethernet TCP/IP. Both Ethernet (as a physical layer) and TCP/IP are worldwide standards. They've achieved the economies of scale required to reduce their costs to a fraction of those associated with industrial networks, and engineers, technicians, and information science personnel alike already understand the protocol. Lastly, all of the types of devices that supported fieldbus protocols, such as PLCs, vision, industrial I/O, and motion controllers, now support Ethernet TCP/IP. It's no longer necessary to use a PLC as a gateway between dissimilar networks when both the device level and the SCADA level can be implemented using Ethernet TCP/IP (Figure 2).
Currently, there's no higher-speed method of transferring data between peripheral devices and the PC than using PCI bus architecture. With bandwidths of 266 Mbps, PCI is the obvious choice for transferring large amounts of information between a peripheral device and a PC host. All new desktop PCs and industrial PCs are equipped with a PCI expansion bus, so there's no problem finding a suitable host machine. Additionally, many of the industrial devices on the market today support the PCI standard, including vision systems and motion controllers.
Once again, however, there are limitations to even PCI architecture's capabilities. First, because PCI devices must be embedded within a PC, they can be very cumbersome and difficult to maintain and support. In the event of a hardware failure, the entire host PC must be shut down to replace the device. Second, while PCI devices are capable of handling a tremendous amount of data between the PC and the peripheral card itself, this causes a problem: Exactly how do we get the data from the outside world to the peripheral card? This often must be done using a difficult and complex wiring scheme. And last, the question of scalability comes into play when using PCI. How many available PCI expansion slots does the PC have? Is there physical room in the PC to handle another PCI device? Of course, this can be alleviated by carefully choosing components and a host PC up front, but isn't this time better spent working on the machine design than on the control system design?
Using Ethernet TCP/IP can help to eliminate the problems inherent with PCI architecture. Ethernet devices are stand-alone and reside outside the PC. This means that the device is accessible in the event of a hardware failure. Additionally, the entire host system need not be shut down to maintain or support an Ethernet device. Simply unplugging it from the network will suffice. Again, because the Ethernet device is external, there's only the issue of connecting the Ethernet connection to the PC or switched hub to transfer information between it and the PC. And the last, but perhaps most compelling, argument for Ethernet is its inherent scalability. Because Ethernet TCP/IP was designed to network factories with many subnetworks, it's inherently scalable. Adding additional Ethernet devices requires only that there's an available port on the switching hub and a unique IP address available for the device.
An RTOS is an off-the-shelf operating system that's traditionally used in embedded and custom-designed control systems. Many RTOS packages are currently available and the choice of one depends mainly on the complexity of the application and control system being developed. Users of fieldbus networks or PCI devices may find them difficult to implement with an RTOS, however, because most device drivers, libraries, and front-end software are typically written for Microsoft Windows. On the other hand, a TCP/IP stack is commonly supported in all major RTOS packages. With sufficient knowledge of the application protocol used by an Ethernet TCP/IP device, a developer can quickly create an interface from the RTOS to that device.
Ethernet offers several advantages over fieldbus protocols and PCI, but it cannot truly become the de facto motion control communication protocol until a single application layer can be agreed upon. Currently there are many Ethernet devices on the market, each using a different application layer protocol for communication. These range from proprietary packet structures in some motion controllers to Modbus protocols in Ethernet I/O devices. Although Modbus is the closest that Ethernet devices have come to an application layer standard, all manufacturers still haven't adopted it.
In the meantime, while waiting for an application layer to be standardized, several manufacturers have created proprietary Ethernet drivers to provide peer-to-peer communication among different Ethernet devices. The latest attempt to standardize an Ethernet application layer protocol is being created by the Open DeviceNet Vendors Association and is termed EtherNet/IP (for industrial protocol). The EtherNet/IP specification is being written to adopt the ControlNet protocol over TCP/IP in order to provide deterministic control over Ethernet. If this comes to pass and all manufacturers eventually support this protocol, then Ethernet is well on its way to being established as a standard communication protocol for industrial control applications.
Although Ethernet is gaining in popularity and looks to become a more dominant network architecture for motion control, it will never replace the need for application-specific uses of fieldbus, PCI, or other proprietary communication protocols. There are three reasons for this. The first is that Ethernet clearly lacks deterministic abilities. Although work is being done to create a deterministic standard over Ethernet, it isn't yet complete or adopted. Until this happens, fieldbuses such as DeviceNet or Profibus will continue to occupy a prominent place among industrial control networks. Additionally, until this unifying standard is approved and adopted, it will be up to each manufacturer to create device drivers to allow its devices to communicate directly with other Ethernet devices such as PLCs. Lastly, Ethernet's speed is currently limited to 10 Mbps in many devices, with 100 Mbps support on the way. PCI is already capable of 266 Mbps and is positioned to increase its speed even more in future chip sets. In applications where time-critical exchange of large amounts of data is required, PCI is still a viable option.
Users of industrial control equipment, including vision systems, PLCs, motion controllers, and industrial I/O, are faced with many options for communicating with these devices. Although Ethernet faces some limitations to its success in all applications, it's remarkably well suited as a general-purpose communication protocol for both the device and SCADA levels. Considerable effort is being made to create a single application layer protocol for acceptance by all manufacturers of industrial Ethernet devices that will provide interoperability among Ethernet devices and also provide deterministic performance. At the same time, Ethernet device manufacturers are creating individual drivers to enable communication among specific Ethernet devices. Embedded developers are finding it easier to incorporate Ethernet into their industrial control systems with TCP/IP stacks provided in most RTOS packages.
While PCI and fieldbus serve very specific purposes within industrial control, they both have drawbacks. High cost and complexity plague fieldbus networks, and PCI suffers from difficult support and maintenance. In the short run, we're witnessing a change in the landscape of industrial control. More and more devices with Ethernet compatibility are being introduced every day. There's no telling what the future of industrial control holds, but it's hard to imagine that Ethernet won't play a large part. MC
Figures and Graphics
John Guite is the controls engineering manager for Parker Hannifin's Compumotor division, a developer and manufacturer of full-spectrum, computer-based motion controllers and related drives and servo/step motor systems. He received his B.S. in engineering mechanics and a B.S. in mathematics from the University of Wisconsin. He is a member of the Industrial Automation Open Network Association and SERCOS North America. Contact him at 5500 Business Park Drive, Rohnert Park, CA 94928; tel: (707) 584-2596 or (800) 358-9068; www.compumotor.com.