Machine connectivity protocols are the languages equipment uses to share data. On the floor you will meet Modbus, EtherNet/IP, and Profinet; between machines and software, OPC UA and MQTT dominate. Almost no plant runs just one, so the real job is translating a mix of protocols into a single usable stream. The good news: every protocol that matters is open, documented, and supported by mature tooling. You do not need to standardize your machines. You need a layer that speaks all of them.

This guide maps the protocol landscape in plain English: what each protocol is for, how data actually moves, how to choose a path for your own plant, and what to do about machines that speak nothing at all. It pairs with our guides to machine data collection methods and OPC UA for machine connectivity.

What is a machine connectivity protocol?

A protocol is an agreement about how two devices exchange data: how a value is addressed, encoded, transported, and secured. When software asks a PLC for the machine state, the protocol defines how that question is framed and how the answer comes back. Protocols matter to non-engineers for one reason: they decide how hard, and therefore how expensive, it is to get data out of a given machine. A machine with an OPC UA server is an afternoon of configuration. A machine with a proprietary serial port from 1993 is a puzzle. A machine with no port at all is a retrofit sensor. All three are solvable; the protocol just sets the path.

Which protocols will you actually meet on the floor?

Four families cover most plants:

ProtocolStewardWhere you meet itCharacter
Modbus (RTU/TCP)Modbus OrganizationDrives, meters, older PLCs, utilitiesBorn 1979. Simple registers, no data typing, no built-in security. Everywhere.
EtherNet/IPODVARockwell/Allen-Bradley environmentsIndustrial Ethernet carrying CIP. Common across North American plants.
ProfinetPROFIBUS & PROFINET InternationalSiemens environmentsIndustrial Ethernet, strong in European-built machinery.
OPC UAOPC Foundation (IEC 62541)Modern PLCs, machine tools, packaged linesRich typed data model, discovery, built-in security. The lingua franca upward.
The floor-level protocol families most plants encounter, plus OPC UA as the bridge to software. MQTT (below) is the transport that carries data beyond the plant network.

Beneath these sit serial links, proprietary interfaces, and plain discrete I/O. Above them sits MQTT, which is less a machine protocol than a delivery service: a lightweight publish-subscribe transport that moves whatever payload you give it. The Sparkplug specification adds a standard topic structure and payload format so MQTT data is self-describing rather than ad hoc.

Where each protocol lives, floor to software The protocol stack, floor to software OPERATIONAL LAYER / SOFTWARE one stream: states, counts, process values, context OPC UA (IEC 62541) MQTT / Sparkplug EDGE / GATEWAY: translate + normalize polls the floor protocols, publishes upward MODERN PLCs EtherNet/IP · Profinet OLDER DEVICES Modbus RTU/TCP · serial NO NETWORK retrofit sensors · I/O mixed floor below, two open standards above: that is the normal pattern
A typical plant: mixed protocols at machine level, an edge layer that translates, and open standards (OPC UA, MQTT) carrying one normalized stream to software.

What do OPC UA and MQTT do differently?

OPC UA and MQTT are complements, not rivals, and most modern architectures use both. OPC UA is a full information model: a client can browse a server, discover that a tag is a temperature in Celsius on machine 4, and read it with security built in. It answers the question "what data exists and what does it mean?" MQTT answers a different question: "how do I move data efficiently to many consumers?" It is a publish-subscribe transport where devices push updates to a broker only when values change, which is light on bandwidth and ideal for sending plant data to servers and cloud systems; see connecting PLC data to the cloud. Sparkplug closes MQTT's gap by standardizing topics, payloads, and birth/death certificates so subscribers know what they are receiving and whether a device is online.

How does data actually move: poll or publish?

Two patterns move nearly all machine data. Polling asks on a schedule: a gateway reads a Modbus register or PLC tag every second and passes the value along; it is simple and predictable but generates traffic whether anything changed or not. Publish-subscribe reports by exception: the device announces changes as they happen, so a machine state that flips twice an hour sends two messages, not 3,600. Report by exception scales far better across a plant, which is why the MQTT/Sparkplug pattern has become the default for moving data beyond the machine network, while polling remains the workhorse for extracting data from the machines themselves. In practice the edge does both: poll the floor, publish upward.

Polling versus report by exception Two ways data moves POLLING (ask on a schedule) every tick, changed or not · simple, predictable, chatty REPORT BY EXCEPTION (publish on change) stopped running stopped only transitions are sent · efficient at plant scale · the MQTT/Sparkplug pattern
Polling reads values on a fixed clock; report by exception sends only changes. Edges typically poll the machines and publish changes upward.

How do you choose a protocol path for your plant?

You mostly do not choose protocols; your machines already did. What you choose is the extraction path per machine and the standard you normalize to. A practical sequence:

  1. Inventory what each machine already speaks. Model and year of the PLC or controller answers most of it. Note ports, protocols, and whether tags are documented.
  2. Take the easiest honest path per machine. OPC UA server if present; native driver (EtherNet/IP, Profinet, Modbus TCP) if not; serial gateway for legacy links; retrofit sensor when there is nothing to talk to. Do not force elegance; see how to connect legacy machines.
  3. Normalize at the edge. Translate everything into one namespace with consistent names, units, and timestamps. This is where tag mapping discipline pays off for years.
  4. Publish upward over an open standard. OPC UA or MQTT/Sparkplug out of the edge, so no software above the plant ever needs a proprietary driver again.
  5. Design security in from day one. Segment machine networks, keep data flow outbound, and treat the edge as read-only; our guide to machine connectivity and security covers the pattern.

What about machines with no network at all?

A machine with no controller and no port still shows its state physically: a motor draws current, parts break a photo eye, a stack light glows. Retrofit sensors turn those physical facts into signals: a current sensor on the motor gives run/stop, a proximity sensor gives cycle counts, a tap on the stack light gives machine state. These signals are modest, but as our post on connecting machines for OEE shows, run state plus counts is most of what daily improvement needs. Retrofit machine monitoring covers the options in depth.

What do the standards bodies say?

Every protocol above is an open, published standard with a steward you can read directly:

The takeaway for a plant manager: none of this is vendor lock-in territory anymore. The protocols are public, the tooling is commodity, and connectivity cost has fallen accordingly; our post on the ROI of connecting machines works through what that means for payback.

Where does Harmony AI fit?

Harmony AI treats protocol translation as table stakes, not the product. The platform connects to what your floor already speaks, modern PLCs over their native protocols, older devices over Modbus or serial gateways, unnetworked machines through retrofit sensors, and normalizes everything into one operational layer where machine data meets schedules, paperwork, and people. No rip-and-replace, and no requirement that your machines share a brand or a decade. Deployment happens in person: the field team typically visits once or twice, walks the floor with your electricians, and maps the extraction path machine by machine. The CLS case study shows the end state on a real line: mixed equipment, one live picture. For the wider context on plant networks, see what is SCADA and IIoT.