ADS, OPC UA, MES and OEE: Designing a Reliable Machine-Data Architecture

Publish a deliberate machine contract

Create a stable interface structure containing machine identity, mode, state, production counters, current order or recipe identity, quality events, downtime reason, alarms and communication health. Give every value a clear unit, meaning and update rule. Do not expose the entire PLC namespace and ask downstream developers to infer the process.

Choose the interface for the responsibility

• ADS: high-performance TwinCAT-native communication for engineering applications, diagnostics and tightly controlled services.
• OPC UA: standards-based, modelled and secured interoperability across vendors and supervisory systems.
• Transactional API or message layer: order, recipe, genealogy and acknowledgement workflows that require explicit delivery semantics.
• Historian or time-series path: process values and events intended for analysis rather than machine command.

A site may use several routes. Avoid duplicating ownership or allowing multiple systems to issue competing commands.

Keep control independent

The machine should continue in a defined way when MES, network, DNS, certificate services or a historian are unavailable. Buffer only the data the machine can retain safely, bound queue size, expose backlog health and define reconciliation after reconnection.

External commands should include identity, validity and acknowledgement. The PLC validates them against mode, state and local authority. Never allow a generic data write to bypass the machine command model.

Give OEE the right context

Availability, performance and quality calculations depend on an agreed production model. Define planned production time, ideal cycle, good and reject counts, changeover, starved/blocked conditions and downtime categories. Derive events from clear state transitions and retain the reason source and confidence.

An OEE dashboard cannot repair ambiguous machine states. Validate the event stream against observed production before using it for operational decisions.

Secure and accept the interface

• Inventory endpoints, ports, certificates, users and trust relationships.
• Apply least privilege and separate read-only monitoring from command capability.
• Test normal exchange, duplicate messages, stale commands, disconnects, queue limits and restart.
• Verify timestamp source, time synchronisation and quality propagation.
• Document data ownership, retention, schema/version changes and support responsibilities.
• Retain packet, server and application evidence sufficient to diagnose a failed transaction.