The full system against your plant's recall risk

The problem

A weld recall starts as a small mismatch nobody crossed.

a recall or field nonconformance from a mislabeled weld or lot is the sector's biggest reputational and financial risk, and it puts the IATF 16949 certification itself at risk.

• The capital pain of any Tier 1/2 precision tube supplier is the same: a recall or field nonconformance from a mislabeled weld or lot.
• It is rarely one big failure, and that is why isolated controls miss it. It is a parameter typed wrong, a lot tag swapped at cut-to-length, a changeover signed without evidence — each small, each passing its own control.
• On steering columns, shock absorbers, drive shafts or stabilizer bars, a bad seam is a safety issue, not a cosmetic one.
• Beyond the direct cost, it puts the IATF 16949 certification itself and the OEM relationship at risk, and with them the award of future programs.

How it fits the IRIS system

Four verification rings — if they disagree, the tube does not ship.

Ring 1 (ERP) sets the parameter and program; Ring 2 (weld cell) applies it with no typing; Ring 3 (Edge) verifies the real in-line result and triggers JIDOKA AI on mismatch; Ring 4 (evidence pack) certifies the whole cycle. SMED AI speeds up changeovers between programs.

No ring is new on its own: the ERP, the cell, the camera and the dossier all exist in the cases above. What changes is that each one checks the others, so a mismatch between program, parameter, weld and lot is caught in-line, not at the OEM. The strength is in the cross-check, because four independent sources stop agreeing exactly where a mislabeled weld or lot hides.

See the full IRIS architecture →

Before and after

The four rings against a weld or lot recall

Ring	What it does	Built on
1 · ERP program	Sets the program and the qualified weld parameter	Travel card (1) + steel receiving (7)
2 · Weld cell	Applies the parameter with no typing and reports it back	Weld cell stitching (8) + tablet (5)
3 · Edge	Verifies the real weld in-line; JIDOKA AI stops on mismatch	Weld vision (9) + gauge changeover (10)
4 · Evidence pack	Certifies the whole cycle per lot and per OEM	PPAP dossier (11) + metrology (6)
SMED AI	Shortens changeovers between programs without skipping a check	Gauge changeover (10)
Line stop	Holds the tube when program, parameter, weld and lot disagree	All four rings

several containment/recall-risk incidents a year → trending to zero. Discrepancies caught in-line, not at the OEM.

Impact estimate

Estimated impact — to validate with your own 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 6-12 months against the first nonconformance avoided, not against a productivity line.
• Protection against a recall or a lost program, where a single event can exceed several hundred thousand dollars once sorting, freight, rework and penalties are added up.
• From several containment or recall-risk incidents a year to a trend toward zero, with each layer paying on its own along the way.
• Discrepancies caught in-line, not at the OEM, when fixing them still costs minutes instead of a campaign.

protection against a recall or lost program (potential single-event impact > several hundred thousand USD), estimated payback 6-12 months against the first nonconformance avoided. *Estimate to validate*.

And the fair question from the production manager

“Do we have to deploy all twelve cases before we see any of this?” — no. The rings are built in layers and each case pays on its own; with the weld cell stitched to the ERP and the camera on the seam, rings 1 to 3 already cross-check. And JIDOKA AI stops on a verifiable mismatch between sources — reading them is an anchored task, where the best models drop below 1.5% error [1] — not on a model's doubt. When the doubt is the model's, a person decides.

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