Industrial Communication Systems

Where it pays off

Communication work reaches us in one of two ways: either a new line needs its network built from scratch, or a network that has run for years has started producing downtime nobody can explain. The second case is more common, and the symptoms are always similar — a drive that drops out once a week, a remote I/O station that occasionally raises a “communication lost” alarm, an operator panel that freezes at shift change. None of these looks like a fault on its own; together they destroy the line’s credibility.

The work we see most often:

• New line networks. Carrying control traffic, camera streams and reporting data over the same physical infrastructure without them colliding.
• Network troubleshooting. Narrowing down the source of dropouts in a running plant — cabling, switch configuration, addressing or load.
• Segmentation and security structure. Breaking a flat single network into zones and separating production traffic from office traffic.
• Remote sites. Bringing pump stations, wells, tank farms and energy sites — anywhere a cable cannot reach — back to a central system over cellular links.
• Legacy device integration. Moving meters, scales and drives that speak over serial links up to the higher layers through gateways.

How we work

We start with a cycle time budget. Which device has to be updated every few milliseconds, which data is fine once a second — without that distinction, a topology choice cannot be fixed later by adding switches. A motion control axis typically needs a sub-millisecond cycle with low jitter, while a temperature reading destined for a report is comfortable with several seconds. Carrying both on the same network without separating them means putting the critical traffic at risk for no gain.

Next comes the addressing plan and device naming. Profinet device names, EtherNet/IP addresses, Modbus slave IDs — all in one table, printed inside the panel and version-controlled in the file set. One of the biggest time sinks on site is not knowing what address a replacement part should be given.

Segmentation builds on that. Cell zones on their own VLANs; camera and vision streams on a separate segment with multicast traffic contained by IGMP snooping; a single defined conduit between production and office. Access between zones is opened as far as it needs to be and no further — the reverse approach, opening everything and closing it down later, never actually gets finished.

Redundancy is built with a ring topology where the process needs it, and delivered measured: we physically pull the cable and record the recovery time. Redundancy that has not been measured is an assumption, not redundancy.

The last step is commissioning tests: packet loss under load, multicast behaviour, ring failover and a check of the remote access path. Every topology works on an idle network; the differences appear when the line runs at full capacity.

What we watch for

Office switches do not belong on the floor. An unmanaged switch cannot contain multicast traffic, cannot prioritise and offers no diagnostics — and its lifetime inside a panel, at panel temperatures and with vibration, is short. A device that cannot answer “what is passing through which port” during a fault makes the network invisible.

Shielding and grounding are part of the network. A cable whose shield is connected at one end only, or not at all, raises the error counters the moment a drive starts. Faults of this kind are found in the cable route and the ground bar connection, not in the switch configuration.

Address conflicts are silent. Giving two devices the same IP does not bring the network down; it produces intermittent, non-reproducible behaviour. In plants without an addressing plan this is usually the root cause, and it is usually looked for somewhere else for weeks.

One-way data flow is preferred wherever it is possible. If only data needs to leave for the upper layers, opening write access in the return direction requires a justification. Reporting and monitoring need read access, nothing more.

How this connects to the other areas

The network is the infrastructure the control layer stands on: a correctly written PLC program still produces downtime on a poorly designed network, which is why we produce the network diagram as part of the industrial automation project rather than after it. Above the control layer the network is the carrier of data — every record flowing over OPC UA or MQTT becomes an input to the data acquisition and analytics layer. The physical side, meaning cable routing, shielding and panel layout, is planned together with electrical and electronics work.