OPC UA and Edge Computing: Connecting Machines to Systems

What OPC UA is and what edge computing does

OPC UA (IEC 62541) is the open standard that lets machines from different vendors expose their data in one common format, with data types, names and built-in security. Edge computing is the computer installed close to the machine that collects this data, filters it, converts protocols and only then sends what matters to the MES, the ERP or the cloud.

Together they solve the usual problem: every machine speaks its own protocol, every system wants its own format, and the integrator rewrites the same interface on each project. With OPC UA at the machine and an edge node in the middle, integration becomes configuration instead of custom programming.

Visual map: machine, gateway, edge and cloud

The map shows the four layers, the path of the data and the volume calculation from the worked example below.

The 4 integration layers

Many new machines ship with an embedded OPC UA server and the gateway disappears. On older machines the gateway is what brings them into an IIoT architecture without replacing the controller.

How OPC UA organises data

• Address space: the machine publishes a tree of nodes (objects, variables and methods). Each node has an identifier (NodeId) and a namespace.
• Information model: besides the value, the server returns type, unit and description. This is what separates OPC UA from a plain Modbus register.
• Client/server and subscriptions: the client subscribes to variables and receives only the changes. Two parameters matter: the sampling interval and the deadband, which ignores changes smaller than a threshold.
• PubSub: a publish mode without a direct connection, for example over MQTT, for many consumers.
• Security: application certificates, user authentication and the Sign and Sign&Encrypt modes. Avoid the None mode outside a test bench.
• Companion specifications: ready-made models per machine type, such as those for machine tools (umati) and injection moulding machines (Euromap). Where they exist they reduce manual mapping.

Worked example: how much data reaches the cloud

Illustrative example: 40 machines, 25 variables each, sampled at 1 Hz, with 50 bytes per value (headers included).

Rate = machines × variables × samples/s × bytes = 40 × 25 × 1 × 50 = 50,000 B/s

Sending raw data would take 20% of a 2 Mbit/s link (0.4 ÷ 2) for telemetry alone. With filtering at the edge, volume drops 92% with the deadband and 99% in the events scenario. The 8% and 1% figures are assumptions of the example: measure your own process before sizing anything.

What the edge node should do

• Filter: deadband, report by exception and a different sampling rate per variable.
• Aggregate: minimum, maximum and average per minute, part count per cycle, machine state per interval.
• Buffer: store data when the link drops and resend it later, with no gaps in the history.
• Add context: attach order, lot, shift and machine to the data. This feeds OEE and traceability without spreadsheets.
• Decide locally: an alarm, stop or adjustment that cannot wait for a round trip to the cloud.

Step by step: connecting the first line

1. Pick one use case with an owner, for example automatic production and downtime logging.
2. List the minimum variables and the sampling interval of each. Fewer variables, more value.
3. Check machine by machine whether there is a native OPC UA server or a gateway is needed.
4. Separate the machine network (OT) from the corporate network (IT) with a firewall and a demilitarised zone.
5. Install the certificate of each application and disable the unsecured mode.
6. Configure subscriptions with deadband and validate with one week of data against manual logging.
7. Publish to the production dashboard and only then roll out to the remaining machines.

With clean history in place, the natural next steps are predictive maintenance and the digital twin. For the bigger picture, see Industry 4.0 on the shop floor.

Common mistakes and limitations

• Sending everything to the cloud: traffic and storage cost for data that mostly goes unused.
• Leaving security on None to "get it working" and never reviewing it.
• No naming standard for variables: without an information model every machine becomes a special case.
• Ignoring time: unsynchronised clocks (NTP) ruin the order of events.
• Using OPC UA for hard real-time control: client/server serves supervision and integration; it does not replace the fieldbus of the control loop.
• Unmanaged edge: updates, backups and monitoring of the node also need an owner.

Frequently asked questions

Does OPC UA replace Modbus or PROFINET?

No. They stay on the fieldbus. OPC UA sits above, standardising data access for supervision and business systems.

Do I need edge if the machine already has OPC UA?

Usually yes, to filter, buffer, add context and protect the network. With few machines and low volume, a direct client may be enough.

What is the difference between edge and cloud?

Edge processes close to the machine, with low latency and offline operation. The cloud holds history and analytics across several plants.

Is OPC UA secure?

It has authentication, certificates and encryption, but it only protects when configured. Avoid the None mode.

Which port does OPC UA use?

The conventional opc.tcp port is 4840, but each server can use another. Open only what is needed on the firewall.

Do OPC UA and MQTT compete?

Not necessarily. OPC UA describes the data and MQTT can carry it in PubSub mode.

About the author

Vagner Soares

Lean Manufacturing & Behavioral Management Specialist

Over 20 years in the automotive and metalworking industries (GM and Dana), Lean Manufacturing practitioner since 2006. SENAI instructor and mentor in Brazil’s Brasil Mais Produtivo program, delivering consulting, training and audits for 50+ companies, combining quality, productivity and people development.

LinkedInInstagramFull profile →