The drawing line inspected continuously, with local CNN vision
At the speed of a wire drawing line or a rod mill the human eye cannot keep up: meters of wire pass per second and the surface defect — mark, scratch, ovality, doubled strand, incipient oxidation — slips through where nobody is looking. iLEAN Edge places a high-speed industrial camera over the line, infers in milliseconds per meter with a local CNN and marks or ejects the defective stretch before the coil is collected. Pure jidoka AI: the process detects its own defect and reacts.
100% human visual control is impossible at real line speed.
On a drawing line or a rod mill, meters of wire pass per second in front of any inspection station. At that speed, asking a person to guarantee the surface of every meter is not demanding: it is physically impossible. What exists in practice is sampling inspection every N meters — or coil by coil, on the ends — and sampling has two structural problems:
- It leaves gaps — quality checks a minimal fraction of the footage. The roll mark, the longitudinal scratch, the out-of-tolerance ovality, the doubled strand or the incipient oxidation in the cooling area slip through exactly in the meters nobody looks at. In Lean vocabulary: there is no poka-yoke, there is statistical hope.
- It travels to the customer — the defect sampling did not catch is found by the customer when uncoiling. And it is not one lost stretch: it is an expensive claim — coil return, inspection of stock in transit, a technical argument with no per-meter evidence — that erodes approval as a supplier.
The consequence is the worst muda of all: the defect that consumed drawing, annealing and transport before being discovered. In long steel for demanding customers — automotive, fasteners, springs — the cost of the defect leaving the plant runs far above the cost of internal scrap.
iLEAN Edge — high-speed camera, local CNN and marking or ejection before the coil is collected.
The problem is not judgment — the quality manager tells a mark from a rolling shadow perfectly well — it is speed and sustained attention, exactly where machine vision wins. iLEAN Edge replicates the veteran inspector's judgment at line speed, meter by meter, without fatigue. It is the poka-yoke sampling could never be.
Edge sees every meter of wire before the coil is collected. The local CNN tells apart mark, scratch, ovality, doubled strand and incipient oxidation. The actuator marks or ejects the defective stretch without stopping the line. It works with no cloud and without sending a single image outside the plant — even when the mill sits in a hall with limited connectivity. Every stretch is recorded with a timestamp, traceable to the heat and the coil.
The specific iLEAN pieces for a drawing line or a rod mill:
- Edge — a physical terminal with a high-speed industrial camera installed over the line, before the coil collection point. It carries a local CNN on an industrial GPU, trained with examples of the specific gauge and the specific customer finish. It inspects roll or guide marks, longitudinal scratches, out-of-tolerance ovality, doubled strand and incipient oxidation in the cooling area. Inference is local, with no image sent to the cloud — no latency, no bandwidth cost, and no dependence on connectivity that in many EAF plants is limited exactly at the mill —: on detecting a defect it triggers the marking or ejection of the stretch before the coil is collected. Jidoka AI: the process reacts to its own defect in the instant, not at the audit.
- Connect — captures the production order, the heat, the gauge and the coil in progress, whether from the ERP or the MES. Every stretch marked or ejected by Edge is signed with a timestamp and tied to the heat and the coil, so the traceability the customer demands is automatic, not a manual reconstruction.
- Agent — lives in Central, cross-referencing the Edge history (meters marked per hour, by defect type, by shift) with the order and gauge from Connect. If marks always spike after a specific die change, it does not send an email at midnight: it presents the already cross-referenced hypothesis to the quality manager, who validates and decides. The person supplies the judgment; the system does the gemba walk through the data.
Sampling inspection vs. inspection with iLEAN Edge
| Aspect | Sampling inspection | With iLEAN Edge on the line |
|---|---|---|
| Inspection coverage | A 1-2% sample of the footage, every N meters | 100% of the footage, at real line speed |
| Roll mark / longitudinal scratch | Depends on it landing in the sample | A local CNN on every meter, without exception |
| Ovality / doubled strand | Discovered when uncoiling, at the customer | Stretch marked or ejected before the coil is collected |
| Incipient oxidation in the cooling area | Appears weeks later, at the customer receiving | Detected on the line as a shade change, before collection |
| Claims for surface defects | Coil return + stock inspection + approval at risk | Drastic reduction (estimate to be validated) |
| Coil certification | Certified on a sample, with no per-meter evidence | Every coil ships with vision certification of 100% of its footage |
Impact estimate for your plant — to be validated with your own numbers.
The block below is an estimate to be validated against your plant's actual data. We put it forward so the committee has an order of magnitude; we refine it during the assessment.
- Wire drawing plant or rod mill with lines moving meters of wire per second and current control by sampling every N meters or on the coil ends.
- Edge pilot on one line — a high-speed industrial camera plus a marking or ejection actuator before the coil collection point, with no machining work. First value expected within a few weeks.
- Indicative payback of 5 to 12 months, depending on the current scrap ratio and the weight of customer claims. Estimate to be validated against your history.
- Expected reduction in claims for surface defects: drastic on moving from a 1-2% sample to 100% of the footage. (Estimate to be validated against your history.)
- The strategic lever is vision-certified shipping: every coil leaves with 100% of its footage inspected and documented — a growing requirement from automotive Tier 1s towards their long steel suppliers. It enables business, it does not only save scrap.
And the fair question from the quality manager
"What if the AI gets it wrong and lets a mark through?" — hallucination is a problem of free generation, not of anchored tasks. In tasks where the AI merely classifies an image against known patterns (this meter of wire matches the "good" pattern or the "defect" pattern), the best models brought the error below 1.5% [1]. And even then, nothing is decided in a vacuum: the quality manager sees each shift's history, validates false positives in the Edge interface itself, and retraining enters with every model version documented and approved. The line does not stop while training happens. The person supplies the judgment; the machine keeps the cycle turning.
[1] OpenAI paper "Why Language Models Hallucinate", 2025 — on the reliability of AI in anchored tasks.
What people ask about the vision poka-yoke in wire drawing
Which specific defects does iLEAN Edge detect in drawing and rod?
The ones that end in a customer claim. On the surface: roll or guide marks, longitudinal scratches and incipient oxidation in the cooling area, visible as a shade change before the rust is evident to the naked eye. In geometry: out-of-tolerance ovality and doubled strand. The CNN is trained with real examples of the specific gauge and the specific customer finish, so the acceptance criterion is yours, not a generic one: the defective stretch is marked or ejected before the coil is collected.
At what real speed does Edge inspect without becoming the line bottleneck?
At the speed of the line itself. A drawing machine or a rod mill moves meters of wire per second, and the local CNN infers in milliseconds per meter on an industrial GPU: vision is never the bottleneck — the limit is set by the marking or ejection actuator, not by the model. The defect signal arrives before the coil is collected, the line does not slow down and no meter passes uninspected. It is jidoka in its literal definition: the process detects its own defect and reacts without depending on a person's attention.
How is the CNN trained on my line's own gauge and finish?
With real examples of your product, not with a generic steel library. The quality manager marks good stretches and stretches with each defect type — mark, scratch, ovality, doubled strand, incipient oxidation — of the specific gauge and the specific finish of the line (bright, annealed, galvanized…). The CNN learns the visual pattern of that wire under that lighting and that speed. When the gauge changes or a new finish comes in, it is retrained with the new samples and every model version is documented, validated against a known set of stretches and approved by quality before entering production, with a record of which version was active on each coil.
Does Edge work with no cloud? My mill has limited connectivity.
Yes — and that is the design case, not the exception. Edge is a physical on-premise terminal with the CNN loaded onto an industrial GPU in the device itself: inference is local and no image leaves for the cloud, so there is no latency, no bandwidth cost and no network dependency. Many EAF plants have limited connectivity precisely in the mill hall; Edge keeps inspecting every meter and triggering the marking or ejection even with no network. When the connection returns, the signed record of every stretch uploads to be cross-referenced with the heat and coil traceability.
What is vision-certified shipping and why do Tier 1s ask for it?
That every coil leaves accompanied by a verifiable record: 100% of its footage has passed vision inspection, with timestamp, CNN model version and image evidence of every marked or ejected stretch. It is a growing requirement from automotive Tier 1s towards their long steel suppliers: against a certificate based on sampling, vision certification documents the complete footage, not a fraction. And when a claim arrives, per-meter evidence narrows the discussion to the specific stretch — with its image and its timestamp — instead of assuming the return of the whole coil.
Keep the defect out of the coil — we will send the estimated ROI of this AI project for your drawing line within 48h.
We work on your plant's real data, not ours. Assessment with no commitment.
Request estimated ROI within 48h ‹ See all 12 long steel cases See steel