5 Patterns to Integrate with MES: OPC UA, MQTT, REST, File Drop, Shared Memory

All 13 of MVCreate's products integrate with customer MES (Manufacturing Execution Systems) — pushing inspection results (defect type, location, confidence, pass/fail) in real time so MES can trigger sorting, marking, statistics, and traceability. But customer MES varies enormously: some run SAP ME, others Siemens Opcenter, others Yonyou or Kingdee, others a homegrown stack. No single integration pattern fits all. Here are the five we use, and where each one belongs.

1. OPC UA: industrial gold standard

Use when: factory-grade MES + mature OT network + data security matters.

OPC UA (Unified Architecture) is the de-facto standard of the Industry 4.0 era; supports both pub-sub and client-server.

Advantages:
- Standardization: all major MES (SAP ME, Opcenter, Wonderware) support it natively;
- Type system: strong typing carries quality codes, timestamps, units along with values;
- Security: TLS + client certificates;
- Mixed discrete/continuous: events (single-wafer fail) and process (line cycle metrics) on the same channel.

Disadvantages:
- Complex: deep stack, expensive to implement from scratch;
- Hard to debug: arcane error messages; specialized diagnostic tooling needed;
- Customer infra required: an OPC UA Server or gateway.

Our libraries: open62541 (C) + opcua-asyncio (Python). Our products embed a built-in OPC UA Server; customer MES connects and subscribes.

2. MQTT: lightweight IoT default

Use when: distributed deployment + high-frequency data + small-to-mid-scale MES.

MQTT is the IoT industry's de-facto standard — lightweight + pub-sub:

• Tiny message overhead (tens of bytes);

• One-to-many distribution out of the box;

• Auto-reconnect and QoS guarantees.

Advantages:
- Lightweight: protocol header is tiny — ideal for high-frequency data;
- Decoupled: pub/sub separates upstream and downstream — MES upgrades don't disturb inspection equipment;
- Mature ecosystem: Mosquitto, EMQX, HiveMQ.

Disadvantages:
- No native type system: payload is bytes; types via JSON/Protobuf by convention;
- No transport reliability (QoS 0) — or expensive (QoS 2).

Typical deployment: SC-DEL-Drone runs an MQTT broker on its ground station; the drone publishes results, MES subscribes.

3. REST API: simplest, but high latency

Use when: asynchronous reporting + low-frequency data + cross-network integration.

REST (HTTP + JSON) is the easiest pattern: the inspection device acts as HTTP client and POSTs each result to a MES endpoint; MES returns 200.

Advantages:
- Simple: five lines of Python;
- Cross-network: HTTPS traverses firewalls easily;
- Universal: every modern MES supports it.

Disadvantages:
- High latency: handshake + request takes 50–100 ms even on a good link;
- No pub-sub: MES can't push back; client polls;
- Batching is awkward: high-frequency data needs client-side buffering.

Typical deployment: SC-IV-Portable on a plant uses 4G uplink and REST-POSTs to cloud MES every 30 s (cycle permits).

4. CSV file drop: the realist choice for legacy plants

Use when: legacy MES + offline lines + no network.

Many customer MES installations froze in the 2010s — direct DB ingestion, no API — but include "directory monitoring". The simplest integration is CSV files.

Advantages:
- Zero dependency: device just writes files; MES just watches a directory;
- Disconnect-safe: MES restarts don't lose data; files accumulate;
- Audit-friendly: CSV is its own audit log.

Disadvantages:
- High latency: MES poll cycles 30 s – 5 min;
- File-write conflicts: multi-device cases need careful directory layout;
- Coarse typing: everything is string; type conversion lives in MES.

Our implementation: each device writes per-minute CSVs to a shared SMB/NFS directory; filenames carry timestamp + device ID; MES file-watcher triggers ingest.

5. Shared memory: ultra-low-latency line integration

Use when: inspection device and a local PLC/MES service share a host + cycle<100 ms.

Some ultra-high-speed lines (>10,000 wafers/h) consider even OPC UA's 5–10 ms slow — they need the most aggressive option: shared memory.

Implementation:
- Inspection software and MES service run on the same industrial PC;
- Share a POSIX shared memory segment (Linux) or memory-mapped file (Windows);
- The device writes a result struct directly into shared memory;
- The MES service reads via polling or semaphore notification.

Advantages:
- Microsecond latency: pure memory I/O,<10 μs;
- Massive throughput: limited only by CPU cache bandwidth;
- Zero serialization overhead.

Disadvantages:
- Tightly coupled: same host required; MES upgrade affects inspection;
- Cross-language pain: C/C++ struct compatibility with Python takes care;
- Operationally heavy: shared-memory leaks, semaphore deadlocks need expert troubleshooting.

Typical deployment: SC-MC-W (6,000 wafers/h) has a 200 ms cycle SLA on a particular line; we use shared memory to push results into the customer's PLC sorting signals — end-to-end latency<2 ms.

6. Comparison

Decision tree:

1. Cycle<100 ms and same host? → shared memory

2. Customer MES OPC UA-capable? → OPC UA

3. Distributed + high-frequency? → MQTT

4. Cross-network + async? → REST

5. Legacy MES with no API? → CSV drop

7. MVCreate's "integration default"

Our 13 products ship OPC UA + MQTT + REST simultaneously by default; the customer enables what fits their MES. CSV and shared memory are project-customized.

The design philosophy: inspection devices should not be locked to customer MES choices — MES turns over every ~5 years; inspection equipment runs ≥8 years. MES upgrades must not require device rework.

For an MES-integration assessment or on-site integration service, contact MVCreate at +86 159-5048-9233.

Originally published by Vision Potential (Nanjing MVCreate Intelligent Technology Co., Ltd.). Reproductions must credit the source.