Skip to content

Protocols

Protocol choices often reshape the hardware shortlist more than teams expect. This section explains how protocol fit affects interoperability, data handling, engineering effort, and long-term architecture boundaries.

The expensive protocol mistake is not choosing an unpopular standard. It is putting the wrong protocol at the wrong boundary. A serial device may still be perfectly useful if the boundary is a converter or gateway. OPC UA may be the right plant-side integration layer but the wrong cloud delivery pattern. MQTT may be excellent for publish-style telemetry but weak if the team still has not modeled the machine states it wants to publish.

Protocol or patternBest-fit boundaryWeak-fit boundary
Modbus RTU / TCPSimple register access, meters, drives, legacy devices, straightforward pollingRich semantic data models or broad enterprise context
OPC UAPLC, SCADA, historian, and structured plant data integrationUltra-light remote telemetry where overhead and configuration burden dominate
MQTTPublish/subscribe telemetry, edge-to-cloud movement, event deliveryRaw device interrogation without an event or topic model
Serial bridgingPreserving useful RS-232/RS-485 assets without replacing equipmentComplex multi-consumer data architecture
PollingStable low-frequency reads where freshness expectations are modestHigh-speed event capture or loss analysis
Publish/report-by-exceptionState changes, alarms, and bandwidth-aware telemetryWorkloads that still need broad historical sweeps

The shortlist should be driven by where translation happens and who owns it after commissioning.

Before a gateway, converter, or edge device belongs on the shortlist, define these protocol facts:

  1. Machine-side protocol burden: serial, Modbus TCP, EtherNet/IP, Profinet, OPC UA, vendor driver, file export, or mixed.
  2. Upstream consumer: historian, SCADA, MES, OEE tool, cloud broker, maintenance system, or custom application.
  3. Freshness requirement: seconds, minutes, shift-level summaries, or event-first capture.
  4. Data shape: raw tags, calculated states, event records, alarm sequence, batch context, or traceability evidence.
  5. Ownership: controls team, plant IT/OT, integrator, enterprise platform team, or vendor support.

Those facts decide whether the device class should be a protocol converter, industrial gateway, edge computer, remote I/O layer, or something smaller.

Failure modeWhat it looks likeBetter approach
Protocol-first buyingA gateway is bought because it supports many protocolsStart with the two or three protocols actually crossing the boundary
Cloud-first abstractionMQTT is chosen before machine states are modeledDefine event and state semantics before topic design
Historian sprawlEvery readable register becomes a tagSeparate durable signals from diagnostic noise
Serial replacement biasLegacy serial devices are replaced before their value is understoodBridge or convert if the device is stable and useful
OPC UA overreachOPC UA is expected to solve ownership, naming, and context by itselfTreat OPC UA as transport and information modeling, not governance
  1. Confirm which systems must talk to each other and who owns each endpoint.
  2. Separate field protocol needs from upstream data delivery needs.
  3. Decide where translation should happen and which device class should own it.
  4. Compare the operational consequences before locking into vendor-specific hardware.