Vision inspection is a data producer. An inspection station running at 400 parts per hour generates 400 pass/fail verdicts per hour, plus defect images for every flagged part. The question is not whether that data is useful — it is — but where it goes and how fast it needs to get there. Three integration architectures answer that question differently.
Why Integration Architecture Matters Before You Buy
A vision inspection system that writes verdicts to a local database and requires manual CSV export at end-of-shift is not integrated — it is isolated. The value of automated inspection multiplies when the verdict reaches your line control system in time to act on it, and reaches your MES in time to record it against the part serial number or batch lot before the part moves downstream.
The three integration paths have different latency profiles, different complexity costs, and different dependencies on your existing plant infrastructure. The right choice depends on answers to three questions: Does your PLC need the verdict before the ejector fires? Does your MES need part-level traceability or just shift-level aggregates? Is your plant network OT-IT converged or air-gapped?
Path 1: Discrete I/O — Simple, Fast, No Network Required
Discrete I/O integration means the inspection system signals verdict to the PLC via a 24V digital output line. Pass = signal high. Fail = signal low. The PLC reads the signal and fires the reject ejector, stops the conveyor, or flags the part for operator divert — whatever your line logic requires.
Latency: Signal transition in under 5ms after verdict is generated. If the vision system decision latency is 120ms, the PLC receives the verdict at 125ms. That is fast enough for a conveyor running up to 60 parts per minute with standard part spacing.
What it does not give you: Defect type, defect location, defect image, part ID, or any structured data. The PLC knows pass or fail. It does not know whether the fail was a surface scratch, a dimensional deviation, or a missing feature. For MES traceability, you need a second data path.
Discrete I/O is the correct choice when your PLC is the only consumer of the verdict, your MES traceability is at the batch level (not part level), and you are adding inspection to an existing line that was never designed for structured data output from an inspection station. It is also the correct fallback path if your MES integration fails — the line keeps rejecting bad parts even if the network is down.
A Tier 2 automotive interior trim supplier in central Ohio running a 2-shift stamping press used discrete I/O as the primary integration path in their initial Eolvision deployment. Their Rockwell ControlLogix PLC handled the reject logic; the inspection system's digital output wired into an available spare I/O card. Total electrical integration time: four hours. No network configuration, no MES connector work, production the following morning shift.
Path 2: OPC-UA — Structured Data to MES and SCADA
OPC-UA (IEC 62541) is the industrial standard for secure, structured machine-to-machine communication on plant networks. An inspection system publishing verdict data over OPC-UA exposes a node structure that a MES client or SCADA historian can subscribe to. Each inspection event publishes a structured object: part ID, timestamp, verdict (Pass / Fail / Review), defect class (if fail), defect confidence score, and optionally an image path or thumbnail.
Latency: OPC-UA subscription update rates are configurable. A 100ms sampling interval is standard; 50ms is achievable on most plant networks. This is not fast enough to drive a high-speed ejector directly — use discrete I/O for the ejector, OPC-UA for the data record.
OPC-UA works well when your MES or SCADA historian is already running an OPC-UA server. Rockwell FactoryTalk, Siemens Opcenter, and Inductive Automation Ignition all have native OPC-UA client capability. The inspection system's OPC-UA server connects to your MES's OPC-UA client, and every verdict lands in the MES as a part-level record with full defect data.
We're not saying OPC-UA is always the right answer — it requires a plant engineer to configure the node namespace mapping, test the subscription connection, and validate that the MES is correctly ingesting the inspection object schema. On a plant with a mature automation engineering team, that is a 2-day integration project. On a plant where the MES was installed by a system integrator five years ago and nobody internally has touched the OPC-UA configuration since, it may be a week. Factor that effort into your integration timeline.
Security consideration: OPC-UA over TCP requires the inspection server to be accessible from the MES network. If your plant has strict OT/IT network segmentation, the inspection system's OPC-UA traffic needs to cross the inspection VLAN boundary. Work with your plant IT/OT team to open only the required port (default: 4840) and enforce OPC-UA security mode. See Eolvision's integration page for network architecture details.
Path 3: REST API — Maximum Data Richness, IT-Side Integration
A REST API integration means the inspection system exposes an HTTP endpoint. After each inspection event, it POSTs a JSON record to your MES's REST API — or your MES polls the inspection system's endpoint for new events. The JSON payload carries the full inspection record: all the fields available in OPC-UA, plus the full-resolution defect image encoded as a URL or base64 blob.
When to use REST: Your MES has a documented REST API (SAP ME, some configurations of Siemens Opcenter). Your IT team is more comfortable with HTTP than with OPC-UA namespace configuration. You want to route inspection data through a plant-level data bus (MQTT broker, Kafka, or similar) that already speaks HTTP/REST. You are doing cloud sync of aggregate inspection metrics to a remote dashboard.
Latency: REST over TCP on a plant LAN adds 10–30ms of overhead per request compared to OPC-UA subscription. For most MES integration purposes, this does not matter. For PLC-level line control, REST is not the right path — still use discrete I/O for the ejector signal.
Payload example: A typical Eolvision inspection event POST to a MES REST endpoint looks like:
{
"event_type": "inspection_result",
"timestamp": "2026-02-27T08:14:33.421Z",
"station_id": "EOL-STATION-03",
"part_id": "PN-44821-B",
"serial": "SN20260227-08141",
"verdict": "FAIL",
"defect_class": "SURFACE_SCRATCH",
"defect_confidence": 0.94,
"defect_bbox": [124, 88, 212, 147],
"image_url": "http://10.1.12.41/images/SN20260227-08141.jpg"
}That record carries everything a quality engineer needs for a SCAR or audit response: part identification, exact timestamp, defect classification, and a link to the defect image stored on the plant server.
Combining Paths: The Typical Production Architecture
Most production deployments do not use a single integration path exclusively. The architecture that works across the widest range of plant configurations uses all three in parallel:
- Discrete I/O → PLC: Drives ejector/divert hardware. Latency-critical. Always active.
- OPC-UA → SCADA historian: Logs part-level verdicts to time-series database for shift reporting and trend analysis.
- REST API → MES: Posts full inspection records with defect images for traceability and audit records.
This architecture means the line does not stop if the MES REST endpoint is temporarily unavailable — the discrete I/O ejector keeps working, and the OPC-UA historian keeps recording. When the MES comes back online, the inspection system can replay buffered events from local storage.
For plant automation engineers planning a new inspection station install, the integration question should be settled before the hardware arrives on-site. If your MES is Rockwell FactoryTalk with an existing OPC-UA data server, the integration path is clear. If your MES is SAP ME with a REST API, the path is also clear. If you are running an older Wonderware (AVEVA) historian with Modbus TCP as the primary plant bus, you need to verify whether the historian's Modbus TCP driver can ingest the inspection system's data model — that configuration is non-standard and will add time.
Questions about your specific plant stack? The Eolvision integration team reviews plant network diagrams before any demo commitment. Bring your MES version number and your current PLC model to that conversation — most integration questions resolve in a 30-minute call. More detail on supported protocols and MES connectors is on the integration page.