Steel Coil Surface Inspection — Scratch & Rust AI

By James Smith on July 11, 2026

ai-vision-steel-coil-surface-scratch-rust-gauge

A quality manager on a hot strip mill knows exactly how unforgiving coil surface inspection is at production speed: strip can move past a viewing point at rates exceeding 1500 meters per minute, which makes catching a scratch, a rust spot, or a gauge deviation with the human eye essentially impossible beyond a sampled spot check. A single missed surface defect that reaches a customer, particularly in exposed automotive or appliance applications, can trigger a claim that costs far more than the coil itself. Deep learning AI vision inspection was built specifically for this kind of high-speed, high-consequence defect detection, classifying scratches, rust, dents, and gauge deviation continuously across the full coil rather than a sampled section. iFactory's steel surface inspection system runs at full line speed, and you can book a demo to see it classify defects against your own coil surface footage.

AI VISION CAMERA · STEEL COIL SURFACE INSPECTION

At 1500 Meters a Minute, the Human Eye Isn't Actually Inspecting the Coil. It's Sampling It

iFactory's deep learning vision system inspects the full coil surface continuously at full line speed, classifying scratches, rust, dents, and gauge deviations and auto-grading coils in real time.

WHY MANUAL COIL INSPECTION CANNOT KEEP UP WITH LINE SPEED

A Full Coil Surface Passes an Inspection Point Faster Than a Human Eye Can Track It

Steel production lines, whether hot rolled, cold rolled, galvanized, or tinplate, run at speeds where the full width and length of strip surface simply moves too fast for continuous human visual inspection to be physically possible. What actually happens on most lines is a sampled inspection, a technician watching a section of strip through a viewing window, or reviewing periodic still images, which means most of the coil's actual surface is never directly observed by a human at all. Surface defects like scratches, rust, and localized gauge deviation can easily occur in a section that was never sampled, and the first time anyone finds out is when a customer complaint traces a coil-related quality issue back to the mill.

WHAT DEEP LEARNING VISION ACTUALLY CLASSIFIES

The Full Range of Surface Defects, Classified Continuously Across the Whole Coil

Scratches

Linear surface marks classified by depth and length across both hot rolled and cold rolled surfaces.

Rust & Corrosion

Early-stage rust identified on cold rolled and stored coil surfaces before it spreads further.

Dents & Edge Cracks

Localized deformation and edge crack propagation flagged as it develops along the strip.

Gauge Deviation

Thickness variation tracked continuously and correlated against rolling parameters.

Coating Uniformity

Galvanized and tinplate coating consistency checked across the full coil width.

Inspect the Whole Coil, Not Just a Sampled Section

iFactory's deep learning vision system runs continuously at full line speed, so no section of strip goes uninspected.

SAMPLED VISUAL CHECK VS FULL-COIL AI CLASSIFICATION

What Changes When Every Meter of Strip Is Actually Inspected

Inspection Element Manual Sampled Check iFactory Continuous AI Inspection
Coverage Sampled sections of the coil Full coil surface, every meter
Defect classification Judged by technician experience Consistent deep learning classification
Coil grading Manually assigned after review Auto-graded in real time as strip passes
Speed dependency Effective coverage drops as speed increases Maintains full coverage at line speed
FROM DEFECT DETECTION TO AUTOMATIC COIL GRADING

Grading Happens as the Coil Is Produced, Not After It's Already Shipped

Beyond flagging individual defects, the system aggregates defect type, location, and severity across the full coil to assign an automatic quality grade as the coil completes, matching the grading criteria your quality team already uses for customer commitments. This means a coil destined for an exposed automotive application can be automatically flagged and routed for closer review the moment a borderline defect pattern is detected, rather than relying on a technician remembering to apply extra scrutiny to a specific coil hours or days later. Defect location data is also logged precisely enough to support targeted downstream processing decisions, such as identifying exactly where on a coil a defect occurred for slitting or cut-to-length operations.

WHAT QUALITY TEAMS REPORT

Measured Outcomes From Full-Coil AI Surface Inspection

Full Coverage
Every meter of strip inspected at full line speed instead of a sampled section
Consistent
Defect classification that doesn't vary by technician fatigue or shift
Real-Time
Automatic coil grading as strip completes, instead of a manual review after the fact
Fewer
Customer claims tied to surface defects that would have gone undetected in a sampled inspection
FREQUENTLY ASKED QUESTIONS

Questions Quality Teams Ask About AI Coil Surface Inspection

Can this really keep up with line speeds above 1500 meters per minute?
Yes, the vision system is built specifically for high-speed continuous line inspection, using high-frame-rate cameras and processing designed to classify defects in real time rather than batching images for later review. The system is configured against your specific line speed during setup to ensure full coverage is maintained even during peak production rates. Book a demo to see the system running at your actual line speed.
How is this different from the automated optical inspection system we already have installed?
Many existing automated inspection systems use rule-based detection tuned to specific defect signatures, which tends to plateau in accuracy and struggles with new or subtle defect variations, while deep learning vision continues improving as it sees more coil surface data and adapts more readily to new defect types. Facilities upgrading from an older rule-based system typically see meaningfully higher defect capture rates without needing to manually reprogram detection rules for every new defect variation. Contact our support team to discuss upgrading from your current AOI system.
Does this work across hot rolled, cold rolled, galvanized, and tinplate lines?
Yes, the detection model is calibrated separately for each surface type, since hot rolled, cold rolled, galvanized, and tinplate surfaces each have distinct visual characteristics and defect signatures. A facility running multiple line types typically has the model trained across each surface category during initial implementation. Book a demo to review calibration for your specific line types.
How does automatic coil grading actually get applied to shipping decisions?
The grading logic mirrors the quality criteria your team already uses for customer specifications, so a coil automatically graded against a specific standard can be routed to the appropriate customer or application without requiring a manual review of every coil. Borderline cases are flagged for a quality engineer to confirm rather than being automatically shipped or rejected without human oversight. Contact our support team to align grading logic with your existing quality specifications.
Can defect location data support downstream slitting or cut-to-length decisions?
Yes, defect location is logged precisely enough along the coil length to identify exactly where a slitting or cut-to-length operation should avoid a defective section, which reduces the amount of good material scrapped unnecessarily around a defect that only affects a small portion of the coil. This precise location data is one of the most immediately actionable outputs for downstream processing teams. Book a demo to see how defect location data feeds into downstream processing decisions.

Inspect Every Meter of Coil at Full Line Speed, Not a Sampled Section

iFactory's deep learning vision system classifies defects and auto-grades coils continuously, in real time.


Share This Story, Choose Your Platform!