On every industrial line, 3–7 different protocols run simultaneously. A client who asks "can we have it all in OPC UA?" is asking for something not technologically solvable — fieldbus, motion bus, and IT/OT integration are three different worlds with different requirements for latency, determinism, bandwidth, and lifecycle. This article describes when to use which protocol and when the choice is constrained by an earlier technology pick (PLC, motion controller, sensor) rather than by engineering decision.
Three layers, three worlds
Field/sensor layer (lowest 5–50 μs latency)
- **EtherCAT** — Beckhoff, KEB, some ABB motors. Sub-100 μs jitter, fully deterministic. Best for motion control with more than 4 coordinated axes.
- **Profinet IRT / Isochronous** — Siemens core, Phoenix Contact, Wago, Bosch Rexroth. 100–250 μs, high determinism. For most machines in Europe = the de-facto standard.
- **Sercos III** — Bosch Rexroth, KEB, niche. Excellent motion bus, but rapidly retreating in favor of EtherCAT.
Cell/machine layer (1–100 ms)
- **Profinet RT** — common Siemens machine. 1–10 ms RTT, suitable for 95% of manufacturing machines.
- **EtherNet/IP** — Allen-Bradley / Rockwell ecosystem. Less widespread in Slovakia, but dominant in USA / automotive Tier 1.
- **Modbus TCP** — the simplest. No determinism, but 200+ existing devices from dozens of manufacturers speak Modbus.
Plant/IT layer (latency doesn't matter, content does)
- **OPC UA** (server/client + PubSub) — interoperability with SCADA, MES, ERP, cloud. This is where OPC UA lives.
- **MQTT (most commonly Sparkplug B)** — IIoT, telemetry to cloud, edge gateway → cloud.
- **HTTP/REST + WebSocket** — modern dashboards, Grafana, custom UI.
A client wanting "everything in OPC UA" wants to achieve plant-level interoperability between 5 PLC vendors — which OPC UA solves perfectly. Using OPC UA between a servo motor and a controller is technological nonsense: 5–20 ms latency is 100× slower than motion control requires.
Modbus — when it still makes sense in 2026
Modbus RTU (RS-485) and Modbus TCP are the simplest protocols industry knows. Register table, function code 03/04 (read), 06/16 (write), CRC. Implementation in PLC or Python in 2 hours.
**Pros:** - Universal — every manufacturer of sensors, frequency drives, energy meters supports it - Trivial to debug (Wireshark with Modbus dissector, Modpoll command-line) - No licenses, no integration kits, no EDS/GSDML files - Cheap — Modbus gateway for €80, OPC UA gateway for €800–2,000
**Cons:** - Master-slave (request-response), no event-driven data - No determinism — Modbus RTU at 19,200 baud takes ~150 ms to read 100 registers - No security (RTU not at all, TCP only via TLS tunnel) - No semantics — register 40001 is "whatever you put there". Without documentation, reading the value is meaningless.
**When to use:** cheap sensors, energy meters, simple controllers (Carel, Eliwell, Belimo, IFM), legacy devices (2015– frequency drives), DIY and maker setups. Any case where 1 Hz polling is enough and you're saving budget.
**When NOT to use:** any setup with more than 50 devices on one network (Modbus master scaling is a problem), any real-time application (motion, fast safety), any IT/OT integration (no semantics).
Profinet — the workhorse of European lines
Profinet (Process Field Network) is in the hands of Siemens and the PROFIBUS & PROFINET International (PI) consortium. Three versions:
- **Profinet TCP** (acyclic) — 100+ ms, for configuration and diagnostics
- **Profinet RT** (Real Time) — 1–10 ms, for normal PLC ↔ I/O communication
- **Profinet IRT** (Isochronous Real Time) — sub-1 ms, for motion control
**Pros:** - De-facto standard in the EU (Siemens dominance) - Rich ecosystem — Beckhoff, Phoenix Contact, Wago, Festo, SICK, Pepperl+Fuchs, B&R speak Profinet - Web-based diagnostics in every IO module - TSN-ready (from Profinet 2.4) - Security via PROFIsafe profile (for safety relays)
**Cons:** - More complex than Modbus — requires GSDML files, IP addressing, device naming - License context (Siemens TIA Portal, OpenPN stack for 3rd party) - Switches must be "Profinet conformance class A/B/C" — a regular Cisco switch often doesn't suffice at IRT - Topology constraints — IRT requires line or star (not ring)
**When to use:** any modern machine in the EU with more than 5–10 IO modules, motion control above 4 axes, integration with Siemens SCADA (WinCC, PCS 7), cases where the client already has 5+ Siemens machines and wants a homogeneous ecosystem.
**When NOT to use:** highly heterogeneous environments (where half the devices are Allen-Bradley → EtherNet/IP is the better choice), small machines with 3–5 IO points (Modbus RTU is enough, cheaper), USA clients (where Rockwell dominates).
OPC UA — interoperability for IT/OT integration
OPC UA (Unified Architecture) is not a fieldbus. It's an **application layer** on top of TCP/IP that solves: how different machines, SCADA, MES, and cloud services exchange **semantically rich** data.
**Pros:** - Platform-agnostic (PLC, embedded Linux, Windows server, Docker container) - Vendor-agnostic (Siemens, Rockwell, B&R, Beckhoff, Honeywell — all have an OPC UA server) - Rich semantics — types, hierarchy, metadata, timestamps - Security as a first-class citizen — X.509 certificates, role-based access, audit trail - Companion specs for specific domains (OPC UA for Robotics, for Machinery, for Pumps) - PubSub model (MQTT-based or UDP multicast) for cloud + brokerless scenarios
**Cons:** - Latency 5–50 ms per cycle — unusable for motion control - CPU-heavy (on a typical embedded device, 20–40% load just for the OPC UA server) - Complex — security stack, certificate management, address space modeling — it's not "read register 40001" - For the same task, 5× more code than Modbus
**When to use:** SCADA, MES, ERP integration. Bridging between different vendors. Cloud telemetry. Any case where analytics, ML models, or BI dashboards will be plugged in later. Machines integrating into a MOM/MES system (Apriso, Critical Manufacturing, GE Plant Apps).
**When NOT to use:** between PLC and sensor (Profinet/EtherCAT). For simple telemetry to the cloud (MQTT Sparkplug B is 10× cheaper to implement). Between two PLCs that can speak via native fieldbus (Profinet PLC-to-PLC is faster and more deterministic).
Most common architectural pattern 2026
A machine in Europe in 2026 typically runs:
``` Sensor/I/O ←Profinet RT→ PLC ←Profinet RT→ Drive (motion) ↓ OPC UA Server ↓ SCADA (Ignition, WinCC) ↓ OPC UA / MQTT Sparkplug B ↓ MES (Apriso, Critical Manufacturing) ↓ ERP / Cloud (SAP, AWS IoT, Azure IoT Hub) ```
Modbus appears on the edge — at cheap sensors, energy meters, frequency drives. We talk to it via a Modbus → Profinet gateway (Hilscher netTAP, Anybus X-Gateway, Phoenix Contact AXL F BK) or via an OPC UA server with a Modbus driver (Kepware, Matrikon, ignition.com).
Three decisions that don't have a technologically correct answer
1. Should I use the OPC UA Companion Spec for Robotics?
For KUKA, ABB, FANUC integration with MES it's theoretically the right choice. In practice: companion spec implementation is 12–18 months behind the robot firmware, so the client sometimes has to use a proprietary REST API or OPC UA with custom semantics. The choice depends on **the specific robot firmware version** — verify it with the vendor before committing in the project.
2. Should I migrate from Profibus DP to Profinet?
Profibus DP is the massive foundation of older lines (2005–2018). Migration to Profinet costs 30–50% of installation CAPEX and is often only worth it when you also migrate the PLC. Realistically: if your line is stable and the Profibus master is still available (Siemens sells spare parts through 2030+), don't migrate unnecessarily. Migration makes sense when combined with a new PLC or expansion.
3. Should I migrate to TSN (Time-Sensitive Networking)?
TSN is an IEEE standard that brings deterministic Ethernet to any industrial protocol. Profinet, EtherCAT, EtherNet/IP — all will gradually become TSN-aware. Realistically in 2026: unless you're building a new flagship line, wait 2–3 years. TSN switches are still 3–5× more expensive than regular managed switches, the ecosystem is immature.
Practical checklist for a project
Before choosing protocols: 1. What PLCs are already in the shop? (Siemens → Profinet default, AB → EtherNet/IP, Beckhoff → EtherCAT) 2. What is the SCADA? (WinCC → Profinet + OPC UA, Ignition → OPC UA + MQTT, FactoryTalk → EtherNet/IP + OPC UA) 3. What is the MES? (Apriso → OPC UA, Critical Manufacturing → OPC UA, custom app → REST + MQTT) 4. Is there a cloud strategy? (Yes → MQTT Sparkplug B or OPC UA PubSub, No → OPC UA server is enough) 5. What is the estimated device count? (< 20 → Modbus is enough, 20–100 → Profinet, 100+ → Profinet + OPC UA hierarchy) 6. What is the required cycle time? (> 100 ms → OPC UA, 10–100 ms → Profinet RT, < 10 ms → Profinet IRT / EtherCAT, < 1 ms → EtherCAT or Sercos III)
These six questions sum up 90% of the decision. The remaining 10% is vendor relationships, existing contracts, and personal preferences of the team that will maintain the line for 10 years.
---
*We do PLC programming and IT/OT integration across protocols (Siemens, Beckhoff, Allen-Bradley, B&R, OPC UA, MQTT). If you're considering a new line or modernizing an existing one, the first consultation (60 minutes) walks through protocol architecture before the electrical project is even opened.*