Textile adhesive control with AI — one badly formulated drum means an entire collection rejected.

The problem

A piece that does not survive the wash does not come back on its own — the whole collection comes back.

The plant manager of a textile adhesive operation lives with a demanding customer: the garment maker who assembles the finished product and the retail brand that puts its name on the label.

1. The polymer raw material changes from batch to batch — the SBR latex may arrive with a slightly different solids percentage, the acrylic with a lower molecular weight, the hot-melt PUR with a softening point half a degree lower. The supplier's analysis is in the LIMS or in a PDF that arrived by email, but it is rarely cross-referenced with the recipe in time.
2. There is no single garment maker spec — one retailer asks for a minimum peel after 5 industrial washes, another asks for Oeko-Tex with zero heavy metals, another asks for ZDHC MRSL compliance on every batch. The spec sheets live in PDFs, in emails, in the sales manager's head.
3. The failure does not appear in the plant — the finished piece moves down the chain and blows up weeks later: the bond gives way at the first wash, the plasticizer migrates into the color and stains the fabric, the drum's VOC exceeds the limit the retailer has just updated. The return arrives when the plant has already run another twenty batches.

The lab measures by sampling and returns the result hours later. By then the drum is packed and, if you are lucky, it goes to a less demanding customer; if you are not, the collection comes back. The problem is not the recipe, it is that the classic system always arrives late — and every return from a repeat garment maker shows up in the full-year accounts.

How it fits the IRIS system

iLEAN does not change your mixer — it adds the intelligence that was missing on the data you are already generating.

The problem with textile adhesive control is not a lack of information: it is information living on islands — the LIMS knows how the latex came in, the mixer's SCADA knows what is happening right now, the ERP knows which garment maker is waiting for that drum and to which spec, but none of them talks to the next. iLEAN acts as the putty that fills those gaps, without asking you to change the SCADA, the ERP or the packing line.

Edge sees viscosity, solids and VOC in line. Connect reads the LIMS analysis and the garment maker's spec wherever they come from. The agent cross-references them with the expected window and, if something does not add up, holds the drum before the pallet. The person signs — never the other way round.

The three iLEAN pieces applied to textile adhesive control:

• Edge — a terminal with machine vision (CNN) and in-line signal capture: an in-line Brookfield viscometer, a refractometer for solids, a PID/FID sensor for VOC in the booth, a camera over the filling head. If one of the four drifts outside the SKU window, it triggers a light stack and notifies the operator before the drum is closed. It works with no network.
• Connect — captures the raw material analysis whether it comes from a modern LIMS, the lab's spreadsheet or the supplier's PDF. And it also captures what arrives from outside (the retailer's Oeko-Tex PDF, a ZDHC MRSL update, an email from the garment maker with a spec change) at second zero, with nobody having to forward anything.
• Agent — cross-references the batch recipe, the raw material analysis, the mixer's real curve and the garment maker's spec. If there is a deviation, it holds the drum and alerts the quality manager on whichever channel they use. It also generates the conformity dossier — which raw material, which VOC, which projected peel — ready for an Oeko-Tex/ZDHC audit. The person validates and signs; the pallet does not leave on its own.

See the full IRIS architecture →

Before and after

Manual control vs. cross-referenced control with iLEAN

Aspect	Manual control + LIMS by sampling	With iLEAN Edge + Connect + Agent
Detecting a viscosity deviation	When the lab returns the result (hours)	In line, during packing
Solids content / VOC	A spot sample every shift	Continuous in-line sensor
Oeko-Tex / ZDHC MRSL compliance	Retailer spec in a PDF, reviewed by hand	Automatic cross-check of the batch against the current MRSL
Projected peel for the garment maker	Lab prediction after the batch	In-line estimate from viscosity + solids
Conformity dossier	Rebuilt by hand	Generated automatically by the agent
Operation with no network	n/a	Edge keeps running on panel power alone

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.

• Mid-sized textile adhesive plant (SBR/acrylic latex, hot-melt PUR, hot melt for nonwovens), 2-3 drum/IBC packing lines with multiple SKUs and multiple retailers.
• Edge pilot on one packing line (in-line viscometer + refractometer + VOC sensor + integration with the batch recipe in the ERP/MES and the garment maker's spec). First value expected within a few weeks.
• Indicative payback between 4 and 9 months, depending on the frequency of garment maker returns and Oeko-Tex/ZDHC non-conformities documented over recent years.
• The hard lever is a single return avoided on a collection: drums collected, reverse haulage, reprocessing, damage to the relationship with the brand. The expected reduction in in-line rejects is ≥ 30% against the current baseline.

And the quality director's reasonable doubt

“What if the AI gets the peel prediction wrong and holds a drum that was fine?” — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely compares a real measurement against a spec window trained on your own batches, the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN holds and the person signs. The three safety rings exist precisely for this.

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