Fast forward with Ethernet

Often heralded as the future of industrial networking, high-speed industrial ethernet appears to be gaining in popularity. Geoff Lock explains the benefits it brings operators in terms of maintenance and accessibility.

With networking at the heart of most of today’s projects there still seems to be uncertainty about the choice of Ethernet as a suitable standard for use in high-speed applications. Engineers want to know if it will be fast enough and if it can work in real time. So you must decide if the network is suitable for your needs.

Ethernet itself is just a physical network: wire, glass fibre or wireless connections between devices on a local area network. With LAN-based communications there can be more than one consumer of a piece of information, but protocols are needed to get information on and off the network.

The best known of these are transmission control protocol (TCP) and internet protocol (IP), which underpin the internet.

When you go to a web page, your browser uses the hypertext transfer protocol (HTTP) to talk to TCP which organises your message into packets of information that are passed to the IP for sending to their destination. There, IP receives the message off the network and TCP reconstructs it for passing via HTTP at application level. In the case of the internet this is via a web server.

TCP is very reliable and highly monitored and messages can be guaranteed to arrive intact. but there is a software and time overhead involved in doing this which puts some additional load on the network and its components. An alternative way to send messages over IP is via user datagram protocol (UDP) which offers low overhead ‘fire and forget’ messages but there is no in-built guarantee that they will reach theirdestination.

TCP and UDP are standard components of the TCP/IP protocol suite and are complementary methods for sending packet-based messages over IP. If packets get lost or corrupted, error correction within TCP will remedy the fault, but there is no such feature in UDP. For connecting devices in industrial applications, an application protocol layer is still required over and above TCP/IP. Some examples of different approaches to this are Modbus/TCP, Ethernet/IP and Fieldbus Foundation.

Ethernet systems can also be almost any size, depending on how segments are connected together. The simplest systems use a common network cable shared among many devices, but if packets collide – in other words, two devices try to send at the same time – low-level software will reschedule their messages so there is only one sender at any time. Segments of shared Ethernet can be joined using bridges or hubs, but this increases the scope for collisions, which can lead to unpredictable performance. Collisions can be avoided by using switches, which connect a number of segments, each having only one device, so there can never be any contention.

Modern cabling allows Ethernet to operate in full duplex mode, both sending and receiving at the same time.

An open standard, Ethernet is claimed to be able to deliver – using the latest in cabling and electronics – high-speed networking to almost any application. So is it suitable for use in modern industrial applications such as controlling multiple variable speed drives of high-speed equipment?

Recent research by John Morse of IMS shows an increasing use of Ethernet in industry, and a general desire to continue moving in that direction. But the move is only just beginning, with announcements at the Hannover Fair from Profibus and the Open DeviceNet Vendor’s Association (ODVA) of real time control initiatives.

By combining the properties of cabling, switches, application level protocols and message types it is now possible to design a network based on Ethernet which will offer the right degree of determinism (predictability) and high speed for it to be used in most industrial situations.

‘The point about Ethernet is not that it has to be deterministic, but that it can guarantee predictable performance,’ said Mark Daniels, product manager for Drives at Rockwell International.

‘You have to do the maths to choose the right set of products, but we see no reason not to use standard TCP/IP with currently available Ethernet. You use UDP messages for continuous I/O data, status messages and speed references where a lost packet doesn’t matter, and failure to receive packets is noticed by software. For fault conditions, one-off requests and asking for specific values, you use TCP.’

Doing the maths means getting the performance right for the particular application. Real time in brewing might be a reliable two-second response. In a high-speed packaging application it might be measured in microseconds.

But Stuart Williams, a network specialist working on variable speed drives systems for Schneider Electric feels the real time and determinism arguments are just a red herring.

‘You can engineer the loadings and response times to suit a situation and you may end up getting better performance from 100Mbit Ethernet than from a 5Mbit proprietary fieldbus connection ,’ he said. ‘Switches might add 20 per cent to the cost for each device, so if collisions can be tolerated you can use hubs.

‘Most Ethernet users are just concerned that they get good performance alongside its other benefits. Being able to connect to a drive across TCP/IP makes problem diagnosis easier, and therefore restoration after a fault is much easier.’

For those applications where real time and absolute accuracy are demanded – in large-scale motion control for example – recent announcements at Hannover show what is possible. Rather than produce a version of Profibus for running directly over TCP/IP, Profinet uses Microsoft’s DCOM standard to make Profibus available over Ethernet, effectively combining the functions of a fieldbus with the benefits ofIT-standard connectivity.

Speaking at Hannover Helmut Gierse, president of Siemens Automation & Drives, said: ‘The new Profinet V3 standard will incorporate Isochrone Realtime-Ethernet based on technology being developed by Siemens A&D.

‘In motion control tasks the cycle times of driving axles need to be synchronised very precisely. Today, using conventional Ethernet or real-time Ethernet controlled by software it takes several milliseconds to do the job, or you can only synchronise a maximum of 30-40 drives.

‘But in the future it will be possible to synchronise over 100 driving axles – or several hundred axles in modular machine operations – within less than a millisecond and a jitter of less than a microsecond.

The IRT technology of Profinet V3 is based on 100Mbit/s switched Ethernet and adds a deterministic channel to IEEE standard 802.3 Ethernet for extremely accurate time synchronisation. The line capacity is split into two chronologically separate channels; the deterministic for cyclical high-speed telegrams, and an open channel for standard TCP/IP.

The real time extensions have no effect on the standard Ethernet functions.Another approach to synchronisation announced at Hannover comes from the ODVA, which has enhanced its common industrial protocol (CIP) to add time synchronisation services based on open standard IEEE 1588 of 2002. A prototype of CIPsync has been produced based on 100Mbit switched Ethernet that can synchronise devices with an accuracy of less than 500 nanoseconds.

The automation industry can now demonstrate reliable, high-speed Ethernet networks in many situations, but it still has to reach some of the more testing applications in any quantity. Hardware and connectivity are still being worked on, while design expertise is building up alongside some very advanced developments that will be suited to motion control. Some of the major manufacturers have yet to bring their Ethernet-connected products to market.

Ethernet-based systems are only going to get better, but the main reasons behind its use today is the benefit it brings operators for accessibility and maintenance, which has nothing to do with determinism or real-time performance. This is still a growing marketplace.