A pore in airbag fabric is not a defect — it is a recall on the finished vehicle.

The problem

What slips through at the loom ends up stitched into the cushion — and by then it is a recall.

Airbag fabric carries a quality constraint the rest of the textile world does not have: once it has been cut and sewn, visually inspecting for the defect becomes practically impossible and extremely expensive. A pore in the OPW that changes the permeability of the cushion as it inflates cannot be seen from the outside. An irregular seam in the critical zone of the OPW alters the deployment pattern. The tier-1 audits PPM and asks for a dossier — and a defect that reached a vehicle means a recall or, in the worst case, an accident.

The three realities almost nobody cross-references in the plant in time:

1. What defect the fabric has at this meter — the operator sees it if they happen to be looking. If not, the fabric keeps moving.
2. Exactly where the defect is on the roll — so that cutting downstream can avoid it. A piece of data that lives in the shift notebook.
3. Which tier-1 and which OEM are expecting that roll — and therefore which tolerance applies. ERP data that comes into play too late.

The quality standard in automotive is on the order of 25 PPM according to industry references (Symestic). The classic system (operator + final inspection on a light table) works 99% of the time. The remaining 1% is what triggers the tier-1 audit and the penalty.

How it fits the IRIS system

iLEAN Vision does not replace the inspector — it puts their eyes on the meter coming out right now.

The problem with airbag defects is not ignorance of what to look for: the inspector knows perfectly well. The problem is weaving speed + human visual fatigue. iLEAN Vision acts as the putty between the loom output and the light table downstream, without asking you to change the loom, the traceability system or the cutting flow.

Edge watches the fabric at the loom output. The CNN detects pores, irregular seams and weft defects. It marks the meter, holds the roll or alerts the operator. The person validates — the batch does not move on to cutting by itself.

The iLEAN pieces applied to defect control in airbag fabric:

• iLEAN Vision (Edge) — a terminal with a high-resolution camera over the output of the OPW or conventional loom. The CNN is trained on your own defects during the first weeks (transfer learning on top of the quality baseline). It fires an actuator (stack light, mark on the roll, alert to the operator's earpiece). It works with no network.
• Connect — captures the incoming yarn batch, the silicone batch used for the coating, the warp changeover, the tier-1's notices about special tolerances for a specific OEM, everything that arrives by email, WhatsApp or delivery note. At second zero.
• Agent — links the detected defect to meter, position, loom, batch, shift and yarn lot. It assembles the PPM dossier for the tier-1 and proposes to cutting downstream which meters to scrap. A defect repeating at the same position on the same roll triggers a loom maintenance review — before the major breakdown.

See the full IRIS architecture →

Before and after

Human inspection on the light table vs. iLEAN Vision at the loom output

Aspect	Classic human inspection	With iLEAN Vision at the loom
Inspection point	Light table downstream — the fabric is already finished	Loom output — the fabric is still being formed
Defect types	Whatever the inspector can see at table speed	Pores, weft, OPW seams and stains, in milliseconds
Defect location	Estimated in meters from the end	Exact meter + coordinate across the width
PPM traceability to the tier-1	Rebuilt by hand for the audit	Automatic per-roll dossier with image
Operation without a network	n/a	Edge keeps inspecting and marking
Pattern by loom/batch/yarn	The veteran remembers it, if they remember it	The agent cross-references history and proposes an inspection

Impact estimate

Impact estimate for your plant — to be validated with your numbers.

The block below is an estimate to be validated with the specific data of your plant. We put it forward so the committee has an order of magnitude; we refine it during the diagnostic.

• Tier-2 airbag weaving mill with 4-12 Dornier/Picanol/Itema OPW looms or conventional looms for sewn fabric, supplying European or American tier-1s.
• Vision pilot on 1-2 looms (high-resolution camera + CNN trained on your own defects + integration with the traceability system). First value expected within a few weeks.
• Indicative payback between 4 and 9 months depending on monthly volume, roll value, current PPM reported to the tier-1 and the average cost of a nonconformity/return.
• Reduction in PPM caused by fabric defects on the order of ≥30% already during the pilot.
• The hard lever is threefold: PPM to the tier-1 (with dossier and traceability), scrap avoided at the light table, and a lower cost for any nonconformity or recall downstream.

And the quality manager's reasonable doubt

“What if the vision system fails and lets a pore through?” — reliability when the task is anchored (the CNN compares the loom output against the calibrated pattern of that same loom) brings error below 1.5% ^[1]. Hallucinations are a problem of free AI, not of anchored AI. And even so, the system holds; the person signs. iLEAN does not relieve the quality manager of responsibility: it gives them the eyes the human eye cannot sustain for 8 hours. The traceability of the decision (what Edge saw, what the person validated) stays in the dossier for the tier-1.

^[1] OpenAI paper “Why Language Models Hallucinate”, 2025 — on the reliability of AI in tasks anchored to the source. 25 PPM standard in automotive: Symestic reference.