A PLC, a SCADA system and an MES do not represent three competing ways to do the same job. They operate at different time scales, own different kinds of state and answer different operational questions. Most integration failures begin when a system is asked to own data or decisions that belong somewhere else.
Begin with responsibilities, not product names
The familiar automation pyramid is useful as a map, but it should not be treated as a rigid law of physics. A modern site may distribute functions across edge computers, historians, cloud services and modular equipment. The practical question is still the same: who is authoritative for each value, event, recipe, order or production result?
| Layer or system | Primary responsibility | Typical data |
|---|---|---|
| PLC / controller | Deterministic control and equipment state | Inputs, outputs, interlocks, sequence state, measured values |
| HMI / SCADA | Operator supervision, alarms and process visibility | Live values, acknowledgements, trends, operator actions |
| Historian / event service | Time-oriented evidence and retrieval | Time series, alarm transitions, timestamps, quality flags |
| MES / MOM | Execution context and production records | Orders, batches, material genealogy, equipment use, results |
| ERP / planning | Business planning and logistics | Demand, supply, inventory, financial and planning objects |
The boundaries are conceptual rather than universal product definitions. SCADA and MES capabilities overlap in some implementations, and a historian may sit inside a vendor platform or operate as a separate service. What must not overlap accidentally is authority: two applications should not independently decide the final value of the same business record.
A useful event path
Consider a filling line that completes a unit. The controller confirms that the sequence reached its defined completion state and exposes the relevant signals. SCADA presents the state to the operator and records alarms or manual interventions. An integration service collects the production event, validates its time and quality, and attaches equipment identity. MES associates that event with the active order, batch or unit genealogy and applies the site’s execution rules. The resulting production record can then support reconciliation, quality review and reporting.
That path should not depend on a dashboard being open at the right moment. User interfaces are for people; integration pipelines must run according to explicit delivery, retry and recovery rules.
Define the contract at each interface
- Identity: use stable equipment, tag and event identifiers. A display label is not a durable key; someone will eventually rename “Mixer 1”.
- Time: specify the timestamp’s source, timezone, precision and meaning. “Time the database received it” is not always “time the process occurred”.
- Quality: distinguish a valid measured value from stale, substituted, unavailable or bad-quality data.
- Units and meaning: include engineering units, enumeration definitions and state semantics. A value of
2is not self-documenting. - Delivery semantics: decide what happens when an acknowledgement is lost, a receiver is offline or an event is sent twice.
- Ownership: identify which application may create, revise, approve or close each record.
Do not make a database the real-time control path
SQL is excellent for durable records, joins, reconciliation and queries. It is not a replacement for deterministic controller logic. A controller should not wait for a business database to decide whether a safety-related interlock may be bypassed. Similarly, a database trigger should not become an undocumented control program merely because the first prototype was quick to write.
Prefer asynchronous integration for non-real-time exchange: an edge or middleware component reads data, validates and normalizes it, writes a durable event or queue record, and retries delivery independently. If the business receiver is unavailable, the production process should follow a risk-assessed fallback policy rather than improvising one at 02:00.
Observability is part of the architecture
Monitor the age of the last successful event, queue depth, rejected records, retry count, acknowledgement delay and reconciliation difference. A green “connected” icon proves that a socket is open; it does not prove that the last 800 production events arrived intact. A useful dashboard separates transport health from data quality and business completeness.
Commission the failure path
Test normal operation, duplicate delivery, missing sequence numbers, wrong equipment identity, stale timestamps, database unavailability, service restart and recovery after a long outage. For each case, document expected behaviour and evidence. If the system cannot explain which records were accepted, rejected or replayed, it is not yet ready for a reliable handover.
Closing principle
Use the PLC to control the process, SCADA to supervise it, the historian or event service to preserve its timeline, MES to put it into production context, and enterprise systems to plan the business. The exact product boundaries can vary. Clear responsibilities, explicit contracts and tested recovery paths cannot.