How control, supervision and production execution exchange information without turning the network into a bowl of spaghetti.
A PLC, SCADA platform and Manufacturing Execution System solve different problems. The PLC executes deterministic control logic close to the process. SCADA supervises equipment, aggregates alarms and measurements, and provides operator-facing views. MES connects production execution to orders, materials, equipment states, genealogy and quality workflows. The exact product boundaries vary, but the architectural distinction is important: a higher-level system should not become an accidental substitute for the control system’s responsibility.
Separate control from coordination
Fast interlocks, equipment protection and deterministic machine sequences generally belong in the control layer designed for them. Supervisory applications can request modes, exchange setpoints through approved interfaces and present context, but they should not be the only place an essential safety or machine-protection condition exists. For each command, define who is authorised to issue it, what permissives are checked, how acknowledgement works, and what happens when communication is lost.
Define an interface contract
Every exchanged value should have a name, data type, unit, quality/status, timestamp semantics, ownership, update policy and a behaviour for stale or missing data. “Temperature = 18” is not a complete data contract until the unit, source, status and meaning of that number are known. An MES consuming a tag should also know whether it represents a live measurement, a batch snapshot or an operator-entered value.
Prefer explicit boundaries
- OPC UA or an approved industrial protocol may expose structured, typed data from the control domain.
- SCADA or a historian can provide supervisory context and time-series observations.
- A staging or integration service can validate events, normalise units and isolate transient database failures.
- MES/business interfaces should consume explicit events and records rather than infer business meaning from an arbitrary set of PLC bits.
Not every architecture needs every layer. A small machine may exchange a handful of approved values directly, while a multi-line facility may require an integration service, broker, historian and governed data model. Choose the smallest architecture that meets the reliability, security and maintainability requirements.
Event semantics and recovery
Polling a state variable tells a consumer what the state appears to be now; it may not tell it which transitions occurred while the consumer was offline. If each transition matters—for example, batch start, batch end, rejection or alarm acknowledgement—design an event record with a stable identifier, source timestamp, sequence or version, event type, relevant equipment identity and validation status. Reconciliation logic should deal with duplicate delivery and late arrival explicitly.
Design for disconnection
Network outages happen. Define whether the control system continues its local process, how much data is buffered, when a queue is considered full, and how missed events are reconciled. Recovery should be tested by stopping a service or disconnecting a link, not inferred from a successful ping. If data are replayed, the receiver must have a rule for duplicates and out-of-order timestamps.
Security and change control
Restrict network flows, use named service identities, separate read and write permissions, log remote commands, and apply changes through a controlled process. Read-only collection and write-capable control interfaces have different risks and should not share broad credentials merely for convenience.
The best integration is not the one with the most arrows on the architecture diagram. It is the one whose data meaning, ownership, failure behaviour and recovery procedure are still clear to the engineer who inherits it.