Seam control on EN ISO 20345 safety footwear with AI — the skipped stitch reaches the customer.

The problem

What decides the test is not the style — it is three millimeters of stitching.

An EN ISO 20345 safety boot is designed to withstand 200 J at the toecap and a sole bond strength that leaves no room for improvisation. But the test does not fail because of the rubber, nor because of the composite toecap — it fails because of the seam.

1. Skipped stitches — the machine completes the cycle, the operator does not notice the gap, and the seam closes with two stitches missing. In the bond-strength test the stitch line is the first weak point.
2. Stitch line drifting off the edge — the panel is not where the pattern expects it and the stitch runs off the reinforcement. Visually it looks fine; mechanically, it is not.
3. Badly managed SKU changeover — the order changes, the thread color does not, the stitch program does not, and the first dozen pairs come out with the previous recipe.

The quality manager knows this, but cannot watch every stitch of every pair at 600 pairs per hour. The classic system (visual check at the end of the line + destructive lab sampling) works 99% of the time. That 1% is the distributor's NCR returning 800 pairs, or the batch held at customs because the notified body's type test came back abnormal — and the cost of a single return pays for the line.

How it fits the IRIS system

iLEAN does not change your Pfaff or Adler machines — it puts eyes where nobody reaches today.

The problem with critical seams is not a lack of procedure: it is information living on islands (the production order in the ERP, the stitch recipe in the supervisor's head, the veteran's eye at the station) that at the critical moment (the first pair after a changeover, the last pair of the night shift) does not arrive in time. iLEAN acts as the putty that fills those gaps, without asking you to replace a single Pfaff.

The camera watches the seam in line. The agent cross-references it with the style recipe. The suspect unit is set aside before assembly. The person signs.

The three iLEAN pieces applied to seam control in safety footwear:

• Edge (Vision) — a terminal with a CNN camera over the output of the sewing machine. It recognizes the correct pattern trained on your own shop floor for each part number, and detects skipped stitches, drifting stitch lines, abnormal density and thread changes. It works with no network. If the plant loses WiFi, Vision keeps inspecting and flagging.
• Connect — captures the production order from the ERP/MES and learns about the model changeover at second zero. It also captures what arrives from outside (a new specification from the retailer's customer, an additional requirement from the notified body) and puts it in front of the quality manager without anyone having to forward anything.
• Agents — cross-reference the visual inspection with the order, the station history and the thread batch. If there is a deviation, they flag the unit and alert both the station and the quality reviewer. They assemble the per-batch inspection dossier that the distributor's audit asks for, on a single sheet. The person validates and signs; the unit does not return to the flow on its own.

See the full IRIS architecture →

Before and after

Final visual inspection vs. in-line AI inspection with iLEAN Vision

Aspect	Final visual inspection + sampling	With iLEAN Vision + Connect + Agents
Seam coverage	Batch sample, reviewer's eye	100% of units, every critical seam
Skipped stitch	Sometimes caught, sometimes not	Flagged within a second, pair pulled out
Model changeover	First pairs run with the previous recipe	Vision model switches with the order, in line
Off-pattern stitch line	Moves on to assembly	Detected at the machine output
Operation without a network	n/a	Vision keeps inspecting on cabinet power
Dossier for the distributor's audit	Rebuilt by hand, days	Per batch, automatic, ready to send

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.

• EN ISO 20345 safety footwear plant with 2-4 sewing lines (upper stitching and Goodyear/Strobel joining), multi-model and multi-thread-color volume.
• iLEAN Vision pilot at one critical station (upper-to-bottom joint or rear reinforcement stitching): camera + lighting + integration with the production order from the ERP/MES. First value expected within a few weeks (detection of skipped stitches and drifting stitch lines above 90%).
• Indicative payback between 4 and 9 months, depending on the history of distributor claims, the number of batches held at customs after abnormal type tests and the average cost of rework.
• Hard levers: a reduction ≥ 30% in seam-related rework, a per-batch inspection dossier that closes the distributor's audit without argument, and a single avoided batch return — which on its own pays for the pilot.

And the quality manager's reasonable doubt

“What if the AI lets skipped stitches through or blocks good pairs?” — hallucination is a problem of free generation, not of anchored tasks. Identifying a stitching pattern already trained in your own plant is a task anchored to the reality of your product, your thread and your machine. In that family of tasks the best models brought error below 1.5% ^[1]. And even so, what is critical is never decided alone: iLEAN flags and sets aside, and the station reviewer signs before the unit goes back into the flow. The three safety rings exist precisely for this.

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