Outdoor membrane lamination control with AI vision — the bubble the eye cannot see is found by the customer's first wash.

The problem

Bubbles and delamination that show up when nothing can be fixed anymore.

Lamination of a 3L outdoor garment bonds face fabric + membrane (ePTFE, PU or nanoporous polyester) + backer fabric with hot-melt adhesive applied by a calender. Conformity depends on three things that drift over the course of a shift:

1. Roller temperature and pressure — they drift with wear and with ambient temperature.
2. Adhesive flow and distribution — they vary with the adhesive batch, with humidity and with every fabric roll change.
3. Web tension — the mechanical setting slips shift after shift and wrinkles or poorly bonded areas appear.

Human inspection at the calender exit is done by sampling, and the defects are subtle: a tiny bubble, a patch with missing adhesive, a cold spot on the roller. The garment passes the initial check, reaches the store — and the customer's first wash uncovers it. The complaint costs material, yes, but the real cost is the technical brand of the premium outdoor customer, which is what pays the price. One bad run in the wrong place and the following year that brand changes supplier.

How it fits the IRIS system

iLEAN Vision does not replace the inspector — it gives him the camera his eye cannot sustain.

The lamination problem is not that the inspector does not know his job: it is that nobody can watch 100% of the roll at calender speed, shift after shift. iLEAN acts as the putty that fills the gap between the veteran's eye (which does know a good lamination when it sees one) and the reality that you cannot see everything — without asking you to change the calender, the adhesive or the membrane brand.

Vision watches the whole roll. Connect reads the calender. The agent cross-references defect pattern with parameter drift and proposes the adjustment. The person holds the roll and signs. Root cause appears almost immediately, not three shifts later.

The iLEAN pieces applied to outdoor membrane lamination:

• Vision (an Edge device with a CNN trained on lamination defects) — installed at the calender exit. It watches 100% of the roll at line speed and detects bubbles, delamination, missing adhesive, roller cold spots and wrinkles. It marks the defect on the roll with service ink and alerts the operator before cutting. It works with no network: the detection cycle does not depend on WiFi.
• Connect — reads the calender parameters wherever they live: roller temperature, pressure, speed, adhesive flow, roll changes. If the machine is old and isolated, Connect hooks into the local computer; if the panel is analog, it reads it from a photo every N minutes. Integration does not force you to replace the calender.
• Agent — cross-references the defect pattern detected by Vision with the drift in the parameters read by Connect and with the batch history (which adhesive went in, which membrane, what ambient temperature). It proposes the adjustment to the operator — raise by 2 °C, replace a roller, check the flow — and learns to anticipate the next failure at the following roll change.

And the safety boundary: Vision marks and holds; the inspector and the quality manager sign. The three rings of the IRIS system guarantee that no roll flagged as faulty slips through to cutting without human validation.

See the full IRIS architecture →

Before and after

Sampling inspection vs. in-line control with iLEAN Vision

Aspect	Sampling inspection	With iLEAN Vision + Connect + agent
Roll coverage	1 out of every N meters, by eye	100% of the roll, at calender speed
Detection of a subtle bubble	Slips through until it shows after washing	Marked in line, roll held before cutting
Adhesive roll change	The operator writes it down, or not	Captured by Connect, linked to the roll
Root cause diagnosis	Three shifts of trial and error	Defect pattern crossed with parameters, almost immediately
Complaint from the outdoor customer	“We don't know when this roll was laminated”	Roll file with photo and calender parameters
Isolated roll with no network	n/a	Vision keeps watching and marking

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.

• 3L lamination plant for a premium outdoor brand, 1-2 calenders, ePTFE or PU membrane, multi-customer, export-oriented.
• Vision pilot on one calender (camera over the outgoing roll + marking actuator + integration with the MES and the calender parameters). First value in a few weeks.
• Indicative payback between 4 and 9 months, depending on volume and on the frequency of delamination complaints documented in recent years. Expected reduction in faulty roll delivered to cutting ≥ 30%.
• The hard lever is not only the material recovered: it is a single complaint from the premium outdoor customer avoided and the technical brand image preserved.

And the production manager's reasonable doubt

“What if Vision flags a good roll as faulty?” — hallucination is a problem of free generation, not of anchored tasks. Vision does nothing more than compare the real roll against the conforming pattern learned on your own good rolls. In anchored tasks, the best models brought error below 1.5% ^[1]. And even so, no flagged roll is scrapped without the inspector validating and the manager signing.

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