Updated July 2026 · 8 min read · Adente Vision Engineering Team
What does quality traceability actually require on a line?
Traceability is the ability to reconstruct, after the fact, what was inspected, what the result was, and what happened to a part that failed. In practice a quality team needs four things per part: a pass or fail result, a measurement or classification behind that result, a timestamp, and a record of the reject action taken. If any of those is sampled rather than recorded per part, the audit trail has gaps exactly where an escape would hide.
This is a general expectation across regulated and quality-managed manufacturing, and the traceability principle sits at the centre of quality-management standards such as ISO 9001. The point of an inspection unit, from a traceability standpoint, is not only to decide pass or fail but to leave a checkable record of every decision without adding manual paperwork.
A second, quieter requirement is where that record lives. If the evidence is retained on-site and retrievable, an audit is a query. If it has been shipped to a vendor cloud, the audit now depends on a third party's retention policy and export path.
A third requirement is easy to overlook: the record has to be complete without extra manual entry. Any step that relies on an operator logging a result by hand is a step that gets skipped under line pressure, and the gap in the trail appears exactly on the busy shift. Traceability that survives real production has to be produced by the inspection itself, per part, as a by-product of the decision.
How does an edge AI unit create a per-part quality record?
An edge AI unit builds the record at the moment of inspection, on the line, for every part rather than a sample. Adente Vision is an edge-AI visual inspection unit built by ADENTE Advanced Engineering Technologies, part of the Aden Group, sold through automation system integrators, and it produces a per-part pass/fail result, a measurement where the task calls for one, and a reject signal, all on-device.
The measurement half of that record is concrete. In a delivered geometry and position case, the unit reports feature coordinates and dimensions per part in millimetres, so the traceable evidence is a number a quality engineer can check, not just a pass flag. The unit runs four inspection modes, Anomaly, Defect, Counting and Quality, and combines classical computer vision for measurement with AI inference for judgement, so a result is auditable rather than a black-box verdict. Each decision can be tied to a timestamp and surfaced on the unit's web-based dashboard.
Why does keeping the data on the line matter for an audit?
Keeping the record on the line means the evidence an auditor asks for is already inside the plant, under your retention rules, not on a supplier's servers. Inference runs on a fanless Jetson-class board in the enclosure, so raw part imagery is processed where it is captured and does not leave the line. Model updates arrive by USB stick, which supports fully air-gapped operation with no inbound internet path.
For traceability that changes the risk profile. There is no cloud egress of process imagery to reason about in a data-handling review, and no dependency on a vendor's uptime to retrieve last quarter's records. For the fuller argument on why on-device inference beats a cloud round-trip on both latency and data control, see the sibling post on edge versus cloud visual inspection.
What does each traceability need map to in the unit?
Each traceability requirement has a concrete counterpart in what the unit records and where that record goes.
| Traceability need | What the unit records | Where it goes |
|---|---|---|
| Per-part pass/fail | A result for every part, not a sample | Pass/fail to the PLC, reject over a 24V output |
| Dimensional evidence | Feature coordinates and dimensions in millimetres | Logged per part for the quality record |
| Timestamped event | The result tied to the moment of inspection | Web dashboard, retrievable on-site |
| Reject action | Which parts were routed out, with a reject reason | Discrete 24V output plus the reason |
| Data location | Raw imagery for later review | Stays on the line; nothing shipped off site |
How do the results reach your quality system and PLC?
The results leave the unit in two forms your systems already understand. A per-part pass/fail and any classification travel over the protocol your controller speaks, PROFINET, EtherNet/IP, Modbus TCP, EtherCAT or OPC UA, and discrete events use the 4 inputs and 4 outputs at 24V, so a reject is both a logged record and a physical actuation. Capture is locked to the part by an encoder pulse, a photoelectric sensor or a fixed interval, so every record corresponds to a real part, not a timer tick.
Alongside the fieldbus path, the web-based dashboard gives the quality team a human-readable view of results without a separate PC on the line. That combination, a machine-readable stream to the PLC and MES plus a human-readable dashboard on-site, is what turns per-part inspection into a traceable record rather than a transient decision. For the metric side of that record, how to read the false-negative rate and why escapes are the number that matters, see the sibling post.
What do the CE, IP54 and IEC marks tell you, and what should you verify?
Standards literacy is part of traceability, because an auditor will ask what the marks on the datasheet actually declare. The standard unit declares CE, IP54 and IEC. IP54 is an ingress-protection rating defined by IEC 60529, describing protection against dust and splashing water; the washdown and hazardous variants carry higher ratings for their environments. CE is a declaration that the unit meets the applicable directives for the European market. These are declared marks, and the honest position is that a mark is not a certificate you can wave at every requirement.
So verify, do not assume. Ask for the certificate for the exact variant you deploy, because ratings differ across the AV-S100, AV-W100, AV-H100 and AV-X100 enclosures, and the hazardous-area certification in particular must be confirmed against the current document rather than a general claim. Just as important: the unit gives your quality team a per-part record and a data-on-site architecture that supports your traceability requirements, but it does not by itself certify a regulated process. Compliance is a property of your validated process, of which the inspection record is one input. Treat the unit as evidence-generating infrastructure and keep the compliance determination with your own quality system.
This post is a spoke of the pillar guide on AI visual inspection; to see where per-part traceability sits alongside the other tasks the unit runs, browse the real applications.