Integrating what you already have: laser marker and weld cell

The problem

One wrong revision digit turns a full shift into misidentified parts.

This is not a technical problem: integrating it properly has simply never climbed high enough up the priority list. The price of not having done it shows up in two different places. The first is marking. The marker is the last physical step before the customer: one wrong revision digit turns a full shift of production into misidentified parts, which at best are caught at the vehicle maker's goods-in and at worst are fitted into vehicles. The second is evidence. On parts with special characteristics — an instrument panel with an airbag tear seam, a structural weld — the process curve is the evidence of conformity. If it only lives inside the cell, as far as the auditor is concerned it does not exist, and conformity has to be demonstrated by other, usually more expensive, means.

• The marker that engraves the Data Matrix on every part is ten years old and has a perfectly usable interface. And yet somebody types part number and revision on its panel at every changeover, with the same data already in the ERP.
• The marker is the last physical step before the OEM: a wrong revision is caught at the vehicle maker's goods-in at best, and fitted into vehicles at worst.
• The vibration welding cell publishes its curves over OPC UA — melt-down depth, pressure, time — and they stay inside the cell.
• On a part with a special characteristic, such as an instrument panel with an airbag tear seam or a welded airbag chute, that curve is the evidence of conformity. If it only lives in the cell, for the auditor it does not exist.

How it fits the IRIS system

Connect stitches two modern systems through their own interfaces, replacing neither.

Connect stitches two modern systems through their own interfaces. When the ERP declares an order start, it automatically publishes part number, revision and traceability data to the marker: changing part number becomes a change in the ERP, not a keystroke on the panel. In the other direction, Connect reads each weld curve over OPC UA and ties it to the serial number engraved on the part, so every unit leaves with its evidence attached and searchable. Neither machine is replaced or modified.

Changing part number becomes a change in the ERP, not a keystroke on the marker. And what the weld cell says it did comes back attached to the serial number it did it to, so every unit leaves with its evidence searchable. Neither machine is replaced, modified or taken out of production to do it.

See the full IRIS architecture →

Before and after

The isolated cells versus the stitched cells

Aspect	Today	With iLEAN Connect
Part number and revision on the marker	Typed at every changeover	Published by the ERP at order start
Keying errors in marking	Two or three a month	Zero of human origin
Vibration weld curve	Inside the cell	Tied to the engraved serial number
Evidence on the airbag tear seam	Demonstrated by other means	Per serial number, on demand
Changeover on the marker	Minutes on the panel	A change in the ERP
Either machine	—	Neither replaced nor modified

from two or three keying errors a month, some reaching the customer, to zero human-origin errors in marking · from weld curves that only exist inside the cell to evidence per serial number available to quality and to the auditor.

Impact estimate

Impact estimate — to be validated with your numbers.

The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.

• Estimated payback 3-6 months, the fastest of the twelve.
• A sharp reduction in the risk of a misidentified lot reaching the OEM, which is the failure that turns a full shift of instrument panels into a containment.
• Weld curves that become evidence per serial number, available to quality and to the auditor without a second, more expensive demonstration.
• And it enables the traceability ring that the zero line-stop flagship depends on.

fast payback, estimated at 3-6 months; a sharp reduction in the risk of a misidentified lot reaching the customer, and it enables the traceability ring that case 12 depends on. *Estimate to be validated.*

And the fair question from the production manager

"Isn't a classic integration cleaner?" — it is, the day the project reaches the top of the list; the problem is that it never does. This stitching uses the interfaces both machines already have, asks for no shutdown window and replaces nothing. Where a value is read rather than published — a label, a panel — it is an anchored task where the best models drop below 1.5% error [1], and the marker confirms what it actually engraved.

^[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.