Explainer ยท Practical AI education

Connecting OT and IT for AI: OPC UA, MQTT, Unified Namespace and security zones

An explainer on the protocols, architectures and security zoning that let plant data reach analytics systems without putting production at risk.

Key takeaways

  • Design to get data out once, contextualize it once and serve many consumers, rather than building point-to-point links per project.
  • OPC UA brings rich information models and security; MQTT with Sparkplug brings lightweight publish-subscribe with defined payloads.
  • A Unified Namespace is a pattern that needs naming conventions, data models and governance to work.
  • Use IEC 62443 zones and conduits with an industrial demilitarized zone, and treat any write-back to the process as a separate, risk-assessed decision.

Why the connection is the hard part

Operational technology, the PLCs, drives, robots, SCADA and historians that run production, was built for reliability and determinism, often on networks never designed to talk to the business side. Information technology systems, including ERP, cloud platforms and analytics tools, expect open, routable networks and frequent updates. AI projects need data to move from the first world to the second, and sometimes recommendations to move back. Doing that without exposing the control network is the core design problem.

The goal is an architecture that gets data out once, adds context once, and makes it available to many consumers, rather than a new point-to-point connection for every project.

A sensible first step is an inventory. List the controllers, protocols, firmware versions, network segments and existing data paths on the lines in scope. Many plants discover undocumented connections, such as a vendor modem on a packaging machine or a historian that pulls directly across a flat network. Fixing those is part of the work, and it is far easier to do before an AI project adds new traffic. The inventory also tells you which equipment can speak modern protocols and which will need a gateway.

OPC UA

OPC UA is a platform-independent standard for industrial communication. It goes beyond moving values: it carries an information model, so a server can describe a machine as an object with named variables, units, ranges and relationships. Many modern PLCs, CNCs and robots include OPC UA servers, and gateways can translate older protocols into it. Industry groups publish companion specifications that standardize how specific equipment types describe themselves.

Strengths include built-in security with certificates, encryption and user authentication, and rich semantics. Tradeoffs include configuration complexity and the classic client-server pattern, where each consumer connects to each server. OPC UA also defines a publish-subscribe mode, which helps at scale.

MQTT and Sparkplug

MQTT is a lightweight publish-subscribe messaging protocol. Devices and gateways publish messages to topics on a central broker, and any authorized consumer subscribes to the topics it needs. Because publishers and consumers do not need to know about each other, adding a new analytics application does not require touching the PLC or the gateway.

Plain MQTT does not define what the payload means. The Sparkplug specification adds a standard topic structure, payload encoding, and state management so consumers know when a device goes offline and what its data represents. When evaluating MQTT-based architectures, ask how payloads are structured and how device state and metadata are handled.

Unified Namespace

A Unified Namespace is an architectural pattern rather than a product. The idea is a single, organized place, usually an MQTT broker or a set of them, where the current state of the business is published in a consistent hierarchy, often following ISA-95 levels such as enterprise, site, area, line and cell. Machines publish their data into the hierarchy; MES, ERP and analytics systems publish their events too; and any system can subscribe to what it needs.

A Unified Namespace can reduce the tangle of point-to-point integrations and make it easier to add new use cases. It also requires discipline: agreed naming conventions, data models for common asset types, governance over who can publish where, and a plan for historical storage, since a broker holds current state rather than long history.

Edge gateways and edge computing

Edge gateways sit near the equipment. They read data from controllers using native protocols, normalize and contextualize it, buffer it during network outages, and forward it upstream using OPC UA, MQTT or other methods. Many can also run analytics or models locally, which matters when a decision must happen within a cycle time, when bandwidth is limited, or when raw data such as vibration waveforms or images is too large to send continuously.

FunctionWhy it matters
Protocol translationLets older equipment participate without replacing controllers
Store and forwardPrevents data loss when upstream links drop
ContextualizationAttaches asset IDs, units and hierarchy before data leaves the plant
Local inferenceRuns vision or anomaly models where latency or bandwidth demands it
Outbound-only connectionsAvoids opening inbound firewall paths into the plant

Manage gateways like the critical assets they are: inventory them, patch them, control their configurations, and monitor their health.

Security zones and conduits under IEC 62443

The IEC 62443 series of standards is the most widely used framework for industrial automation and control system security. Its central idea is to divide the system into zones, groups of assets with similar security requirements, and to control all communication between zones through defined conduits. Each zone is assigned a target security level based on risk, and controls are chosen to meet it.

In practice, many plants use a layered layout informed by the Purdue reference model:

  • Field devices and controllers in cell and area zones, with no direct path to the internet.
  • Supervisory systems such as SCADA and local historians in a site operations zone.
  • An industrial demilitarized zone between OT and IT, where historian replicas, brokers, patch servers and remote access jump hosts live, so that no traffic passes directly from the enterprise network to control systems.
  • Enterprise and cloud systems beyond that.

For AI projects, the main rules are simple to state and harder to follow: data should flow outward through the demilitarized zone, connections should be initiated from the more trusted side where possible, every conduit should be documented with its allowed protocols and endpoints, and remote access for vendors should go through a controlled, logged path rather than a modem or a remote desktop tool someone installed on an engineering workstation.

Writing back to the process

Some AI use cases, such as setpoint optimization, need to send values back toward the controls. Treat this as a separate design decision with its own risk assessment. Common safeguards include writing to a supervisory layer rather than directly to the PLC, limiting changes to defined ranges and rates, requiring operator acceptance, logging every write, and keeping a simple way to revert to manual or standard control. Follow your management-of-change process, and involve controls engineering and process safety from the start.

Related use cases

Operations & production

Machine monitoring & OEE analytics

Automatically capture machine states, downtime reasons and output to measure OEE accurately and target the biggest losses.

Industrial IoT & edgeMachine learning
Established8 software companies

IT / OT

OT network anomaly detection & cybersecurity

Passively map industrial networks and detect unusual traffic or commands that may signal attacks or misconfigurations on control systems.

Machine learningIndustrial IoT & edge
Established3 software companies

Operations & production

Process anomaly detection

Models learn normal process behavior across many sensors and flag unusual patterns before they become quality, safety or equipment problems.

Machine learningIndustrial IoT & edge
Proven, scaling7 software companies

Operations & production

Process & yield optimization

Models recommend or automatically adjust process setpoints to keep quality on target while improving yield, throughput and energy use.

Machine learningOptimization & planningIndustrial IoT & edge
Proven, scaling7 software companies

Put AI to work

Find the AI opportunities hiding in your plant.

Genvo, our implementation partner, works with manufacturers to pick the right first project, connect the systems you already run and put AI agents and tools into daily use, with people approving what goes out.

Assessments and consultations are provided by Genvo. How our partnership works.