An undetected yarn break is paid for in meters of scrap fabric — not on the quality report.

The problem

The loom keeps weaving without knowing it — and the piece becomes one continuous defect.

On a serious weaving floor, one operator runs four, six, eight looms at a time. When a warp yarn breaks on one of them, the operator sees it if they happen to be looking at that particular loom in that particular second. If they are running another batch on a different loom, or have gone down to the creel to change a spool, the loom with the broken yarn keeps weaving. And every meter it advances is continuously defective fabric, not a one-off flaw.

The three realities almost nobody manages to cross-reference on the floor in time:

1. Which yarn broke — and in which warp position. Data that lives in the operator's eye.
2. How long it has been weaving like that — and therefore how many meters of the piece are already scrap. Data that surfaces when someone inspects the finished roll.
3. Which customer was expecting that piece — and whether they accept rework or demand a replacement. ERP data, which enters the scene late.

The classic system (look, stop, tie) works 99% of the time. The remaining 1% is the entire roll that reaches final inspection as scrap and drags lead time, raw material and margin down with it.

How it fits the IRIS system

iLEAN Vision does not replace the operator — it gives them eyes on the looms they are not looking at.

The yarn break problem is not a lack of knowledge: the operator knows exactly what to do. The problem is presence: they cannot watch four looms at once. iLEAN Vision acts as the putty between human attention and the loom that needs attending to right now, without asking you to change the loom, the ERP or the shift reporting system.

Edge watches the shed. If the gap of a missing yarn appears, it triggers the loom stop before the meter advances. It alerts the operator through the channel they already use. The person validates and restarts — never the other way round.

The iLEAN pieces applied to yarn break detection:

• iLEAN Vision (Edge) — a physical terminal mounted over the reed outlet or over the shed. It runs a CNN trained to tell an in-order weave from a weave with a missing yarn. It fires an actuator (dry contact to the loom brake, andon light, alert to the operator's earpiece). It works with no network: if the plant loses WiFi, Edge keeps watching the shed and stopping the loom — what is critical cannot depend on connectivity.
• Connect — links the Edge alert to whichever operator is on shift (earpiece, tablet, phone) and to the production reporting system, so the meters woven at the moment of the stop are tied to the specific batch and the incident does not get lost in a notebook.
• Agent — cross-references break frequency by loom, by shift and by yarn lot. When the same loom accumulates repeated breaks in the same position, it proposes it to maintenance for inspection before the major stoppage. It is the hot bed: the shift comes in and the alerts are already lined up.

See the full IRIS architecture →

Before and after

Operator watching by eye vs. operator assisted by iLEAN Vision

Aspect	Classic watching by eye	With iLEAN Vision on the loom
Break detection	The operator sees it as they walk past — or does not	CNN over the shed, detection in milliseconds
Loom reaction	Operator walks to the loom and stops it by hand	Actuator on the loom brake, automatic stop
Meters of scrap fabric	Variable — depends where the operator was	Bounded by the Edge cycle (milliseconds)
Operation with no network	n/a	Edge keeps running on its own frame light
Incident traceability	Written by hand in the shift notebook	Tied to the batch in the reporting system
Loom/yarn lot pattern	Hard to see — lives in the veteran's memory	Agent cross-references history, proposes 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.

• Mid-sized weaving mill with 16-30 active looms, technical or automotive fabric, one operator running 4-6 looms at a time.
• Vision pilot on 1-2 looms (camera over the shed + actuator on the brake + integration with the production report). First value expected within a few weeks: the first automatic stop that saves an entire roll of scrap.
• Indicative payback between 4 and 9 months, depending on the number of looms, shed speed, the value of a woven meter and the documented frequency of break-related incidents.
• Reduction in scrap from undetected breaks in the order of ≥30% already in the pilot.
• The hard lever is twofold: scrap avoided (meters + yarn raw material) and higher loom OEE (an automatic stop is shorter than an unplanned round).

And the weaving manager's reasonable doubt

“What if the vision system gets it wrong and stops the loom for nothing?” — hallucination is a problem of free generation, not of anchored tasks. Break detection is a task anchored to the image of the loom itself (the AI is not asked to invent anything, only to recognize a visual pattern against the calibrated pattern). On anchored tasks, the best models brought the error below 1.5% ^[1]. And even so, restarting the loom is decided by the person — not by the AI. The three safety rings exist precisely for this.

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