What a TSN Certification Mark Won't Tell You

Avnu and the OPC Foundation have aligned testing for OPC UA FX over TSN; five claims about what that settles on a plant network, and what it leaves open.

Listen - AI recap

On 30 September 2026, with a Brussels dateline, the Avnu Alliance and the OPC Foundation announced that they had signed a collaboration agreement to align testing, certification and practical integration of OPC UA FX running over Time-Sensitive Networking. Both sat among the main contributors to TIACC, the collaboration formed to produce one shared conformance test plan for the IEEE/IEC 60802 profile, and the announcement names two jobs for that plan: proving out the test tools themselves, then opening a joint certification path that covers bridges as well as end stations. The same announcement points back at the days just before it: on 28 and 29 September, participants brought hardware to a two-day industrial plugfest in Stuttgart, hosted by the University of Stuttgart.

That closes a sequence whose first half landed in June, when IEC/IEEE 60802 reached published status on 29 June 2026. Three pieces have to exist before a datasheet claim about deterministic Ethernet means anything to a buyer: a profile stating which features a device must implement, a test specification stating how you prove it, and a body willing to sign the result. All three now sit in the same place, which drags "TSN-ready" out of the marketing column and toward something a tender can ask for. It also arrives wrapped in claims that run a year or two ahead of the silicon. Five of them are worth arguing with.

"A certification mark means two vendors' boxes will work together"

Conformance and interoperability get tested differently because they answer different questions. A conformance test puts one device on a bench against a specification and asks whether its behaviour matches. Interoperability is a property of a set of devices under one configuration, on one topology, carrying one set of streams. Passing the first doesn't hand you the second.

In a TSN network, the thing that fails usually isn't the frame format. It's the schedule. A time-aware shaper opens and closes gates per traffic class on a repeating cycle, and the gate list in every bridge along a path has to agree with the gate lists in all the others. That agreement is computed, not negotiated: out of the topology, the per-hop bridge delays, the stream requirements and the cycle time. Feed the calculation optimistic per-hop delay numbers and the schedule validates cleanly on the screen, then misses on the wire, which turns into a commissioning week that never made it into the project plan. Getting those delay numbers into a comparable form across vendors is a large part of why 60802 exists at all.

Then there's the option problem. A profile selects features, but it still permits choices, and two devices can each pass against the same specification while supporting option sets that barely overlap. So read the scope of a certificate rather than the fact of one: which profile, which options, which link speeds, which redundancy mechanisms, which configuration model.

There's a layer underneath all of this that rarely surfaces in a sales meeting: the test tools have to be right themselves. A conformance verdict means something only if the instrument producing it reads the profile the way the profile's authors did, which is why the Brussels agreement puts validating test tools alongside certifying products. Two labs running different tools against the same bridge and reaching different verdicts would set multi-vendor TSN back further than no programme at all.

Configuration is the thinnest part. IEEE 802.1Qcc describes a fully centralized model, with a network configuration entity computing schedules and pushing them into the bridges; the data models are standardized, the vendor extensions layered on top are not, and the engineering tool that can talk to every bridge in your mixed estate may simply not exist yet. Ask for the test report, the option list, and the name of the software that configured the network it was tested on.

"Our plant needs deterministic Ethernet"

Some duties genuinely do, and they're narrower than the brochures suggest.

Coordinated motion is the real customer. A filler and capper holding phase at several hundred containers a minute, a flying shear cutting strip to length, two mill stands trading tension across the gap, a palletizer interpolating four axes at once: those run control cycles in the hundreds of microseconds, with jitter budgets tighter than the cycle. A missed update shows up as a scrap cut, a jam, or a mark on the product.

Now the duties that fill a process plant. Drum level on a boiler. Secondary air damper position on a waste-to-energy furnace, conductivity on a CIP return line, pump speed trimming to a pressure setpoint, hold-tube temperature on a pasteurizer. These close in tens to hundreds of milliseconds, and 100 Mbit/s switched Ethernet with priority queueing has carried them for two decades. Gigabit links with a sensible QoS plan have margin to spare. Buying a scheduled network to serve a 200 ms loop means paying for a property you'll never measure.

Determinism isn't free in bandwidth, either. A gate window reserved for a motion class stays unavailable to everything else for its slice of every cycle, whether the talker sends a frame or not, so a converged link returns less best-effort throughput than its line rate advertises. Plants that reach for TSN are often chasing a different shortfall: not enough headroom for the camera and vibration channels somebody wants to add, or timestamps they can't defend. Both of those have cheaper answers.

One piece of TSN pays off on a plant with no coordinated motion anywhere: the clock. gPTP, defined in IEEE 802.1AS, disciplines each hop with hardware timestamping and holds a common time base inside a microsecond across a handful of hops. NTP over the plant LAN lands in the millisecond range on a good day (and drifts further whenever a switch gets busy). For a historian trending tank level, milliseconds are irrelevant. For comparing vibration phase between a gearbox's input and output shaft, or lining a motor-current signature up against an accelerometer on the same compressor, a shared microsecond-class time base decides whether the cross-channel comparison means anything or whether you're eyeballing two traces that drifted apart.

Condition-monitoring traffic is undemanding in every other respect. It's a burst of sampled data per asset per interval, it tolerates seconds of delay, and it survives a dropped uplink provided the gateway buffers locally and replays when the link returns. So reserving a gate window for that class of traffic wastes a schedule on data that never needed one.

MechanismWhere it's definedWhat it buys a plant duty
Time synchronization (gPTP)IEEE 802.1ASA common clock inside a microsecond across hops; makes cross-asset phase comparison possible
Scheduled traffic (time-aware shaper)IEEE 802.1QbvA reserved gate window per class each cycle; bounds latency for motion
Frame preemptionIEEE 802.1Qbu with IEEE 802.3brLets an express frame interrupt a long best-effort frame mid-transmission
Frame replication and eliminationIEEE 802.1CBSends one stream over two disjoint paths and drops the duplicates at the far end
Per-stream filtering and policingIEEE 802.1QciCaps what a single talker can inject, so one faulty node can't eat the schedule
Centralized configurationIEEE 802.1QccThe model under which a configuration entity computes schedules and pushes them to bridges
Feature names as defined in the IEEE 802.1 and 802.3 standards. The 802.1Q amendments listed now live in the base IEEE 802.1Q standard, and 802.3br in IEEE 802.3; IEEE 802.1AS and IEEE 802.1CB remain standalone.

"TSN replaces the fieldbus"

No. TSN is a set of layer-2 mechanisms in bridges and end stations, and it carries industrial protocols rather than displacing them. PROFINET, CIP, CC-Link IE and OPC UA FX all map onto the same 60802 profile, which is why the organisations behind those protocols went to the trouble of one shared test plan instead of several private ones. A bridge certified against the profile doesn't care which of them rides on top.

What OPC UA FX brings down to the field level is a typed information model and controller-to-controller exchange as a first-class case, instead of a cyclic image of bytes whose meaning lives in somebody's engineering tool. But that matters more for the next plant than for this one. Your installed base keeps the protocol it has.

So the practical consequence is dull and worth stating anyway: nothing gets ripped out. TSN supplies a shared lower layer and migration happens as islands. The boundaries are where attention belongs, because a non-TSN bridge sitting in the middle of a TSN path doesn't degrade politely. It breaks the delay budget and the time chain at the same moment. An unmanaged five-port switch zip-tied inside a conveyor junction box will do precisely that, and it won't log anything while doing it. Draw the TSN domain deliberately, give it its own schedule, and bridge it to everything else through a device that knows it's standing on a boundary.

The schedule also becomes a document with an owner, and the maintenance organisation feels that before anyone else does. Add a vibration channel to a gearbox two years after commissioning, hang a camera off the capper, move a drive to a spare port, and the gate schedule for that domain has to be recomputed and pushed again. On a priority-queued network the same job is a patch lead and a switch port. Under a scheduled network it's a change-control event with a calculation behind it, so the plant needs either a named person who can run that calculation on a Sunday night (with a licence for the tool that does it) or a written rule that nobody touches the domain until the integrator is back on site. Both are defensible positions. Neither one appears by itself.

Diagnostics shift as well, and the tooling lags the standards. On a priority-queued network trouble announces itself as a discard counter climbing or a port flapping. On a scheduled network the characteristic failure is a frame landing in the wrong window, which a bridge may forward quite happily while the end station rejects it, and a routine counter sweep shows you nothing at all. Budget for a measurement capability: a TSN-aware capture point, or at minimum end stations that count late and early arrivals and publish the counts somewhere the historian can reach them.

Modbus RTU on a 20-year-old RS-485 run is untouched by any of this and will outlive several rounds of it.

"Converging onto one network tightens up security"

It loosens things first. Separate cabling used to supply separation for free: undocumented, unmanaged, but real. Collapse motion traffic, process I/O and plant analytics onto shared infrastructure and the separation you have afterwards is exactly the separation you specified, configured and verified. IEC 62443 frames that exercise the way it always has, and a shared TSN backbone carrying two zones' worth of traffic is one conduit until somebody writes down otherwise.

Per-stream filtering and policing gets misread in this argument. IEEE 802.1Qci caps what a single talker can inject, which protects the schedule from a babbling node and keeps one failure from becoming a plant-wide timing event. That's availability engineering. It establishes nothing about identity: a device holding a gate window holds a gate window, whoever is behind it.

OPC UA FX carries the OPC UA security model down to the field, with certificates, signing and encryption, and real key management behind all three. That's an application-layer protection sitting on a layer-2 network that authenticates nothing, which is a perfectly sound arrangement as long as everyone understands which layer does what. Key management at field level carries its own operating cost, though. Certificates on several hundred devices expire on a schedule of their own, and a drive that falls off the network over a lapsed certificate is a failure mode the maintenance team hasn't met before.

Which raises the asset that tends to be missing from the register: the network configuration entity. It computes gate schedules and writes them into every bridge in the domain, so compromising it lets an attacker retime the plant's motion traffic without going near a PLC. Does your asset inventory have a line for that box? It deserves the treatment IEC 62443 reserves for control-capable systems rather than the treatment a switch-management VLAN gets by default: a hardened host, authenticated and logged configuration changes, credentials separate from the IT estate, and a tested path back to a known-good schedule (restoring a schedule is not the same job as restoring a switch config).

And a conformance logo speaks to none of this. Testing a bridge against the 60802 profile asks whether the bridge implements the profile. It doesn't ask whether the device ships with a default credential, whether its management plane answers from the business network, or whether its firmware is still supported. Those are separate programmes with separate evidence, and the arrival of the first doesn't supply the second. Anyone presenting a TSN certificate as a security argument has made a category error.

"There's nothing to do until certified products ship"

Most of the preparation that decides whether a TSN project goes well contains no TSN at all.

  • A topology and cable-plant record that matches what's installed. The schedule is computed against the topology, so an as-built drawing three retrofits out of date turns into a commissioning problem rather than a documentation problem.
  • An inventory of the managed switches you own: what the silicon supports, what firmware they run, what QoS they're configured for today. Shaping needs hardware support, and a software queue won't hold a gate window.
  • A time story. What disciplines your clocks now, how far apart two timestamps from two gateways can sit, and whether anybody checks.
  • A written answer to who owns the schedule after handover. If the integrator computes it and goes home, the first sensor anybody adds becomes a service call.
  • Traffic measured per segment rather than estimated. You can't reserve a slice of a link you've never characterised.

None of that waits on a certificate, and all of it costs less before a project than during one.

When products do start carrying a mark, ask for interoperability evidence alongside the certificate. A plugfest result says a device shared a wire with other vendors' devices and behaved; a certificate says it matched a specification on a bench. Request both, name the configuration tool chain that was used, and read an evasive answer as information.

Where the new certification path earns attention is the site where cycle-critical motion and plant-wide data must share one infrastructure: a high-speed packaging hall, a rolling mill, a greenfield line being specified this year. Those projects can start asking which profile a switch was tested against, by whom, and with which options enabled. And compared with "is it TSN-capable", that's a question with an auditable answer.

For a single-line creamery closing its loops in 200 ms over three Modbus trunks, with two people in maintenance and a managed switch stack bought a decade ago, none of this is the next capital decision or the one after it. Its constraints sit somewhere else entirely. The motor control centre, the spares shelf, who covers the second shift: a test specification for a TSN profile reaches none of them, and won't for years. Nor does it help the plants where the network is genuinely fine and the data is the problem: a converged deterministic backbone won't tell you why the dryer's throughput falls off every Thursday.

Notes

The limits here are mostly limits of scope. A conformance certificate covers one device against the profile options it was tested with, not the network you assemble out of several of them, and not the tooling that configures it; interoperability still has to be shown on your topology, at your cycle time, with your stream set. The 60802 profile also makes selections that assume particular link speeds and redundancy choices, so a device certified as a good fit for one plant's design can be the wrong part for the next. Certification programmes move as well: a mark issued against this year's test specification says nothing about options added in a later revision, so an early certificate can mean a retest rather than a firmware note.

References

  1. OPC Foundation and Avnu Collaborate to Advance TSN in Industrial Automation through Testing, Certification, and Interoperability (30 September 2026)
  2. IEC/IEEE 60802 TSN Profile for Industrial Automation — project status, IEEE 802.1

Reuse & license

This article is published by Zoniax OÜ under a Creative Commons Attribution 4.0 International (CC BY 4.0) license. You are free to share and adapt it for any purpose, including commercially, as long as you give appropriate credit to Zoniax and link back to the original article.

Disclaimer

These Field Notes are general technical information, published as-is for industry peers. They are not professional, engineering, safety, legal, or financial advice, and nothing here is a recommendation to buy, sell, or act. Figures are cited from public sources believed reliable but are not independently guaranteed - verify them against the primary sources and your own plant conditions before acting. Zoniax OÜ and the author accept no liability for decisions made from this content. Naming a standard, product, or vendor is not an endorsement.

Cite this article

Nõmm, A. (2026). What a TSN Certification Mark Won't Tell You. Zoniax. https://zoniax.com/blog/posts/tsn-certification-opc-ua-fx