Stamped metal components make up a substantial share of the parts we see Procunit deployed on — brackets, shields, clips, housings, formed channels. The defect taxonomy on a stamping line is more specific than people outside the industry often assume. "Surface defect" covers a lot of ground, and the imaging and detection strategy needs to match the actual failure mechanism. This post describes the six defect types we encounter most frequently, what causes each one, and how it presents under ring illumination.
We're not trying to cover every stamping defect that exists — that would be a metallurgy textbook, not a blog post. This is a practical field guide for quality engineers who are setting up inline inspection on a stamping line and need to understand how each defect type differs as a detection problem.
Edge cracks
Edge cracks form at the shear zone during blanking or trimming operations. As the punch breaks through the material, the shear lip and fracture zone have slightly different microstructures. When the tooling is worn, the clearance between punch and die exceeds the optimal range (typically 5–8% of material thickness for mild steel), and the fracture propagates at a shallower angle, creating a burr on one side and micro-cracks on the other. On high-strength steels and advanced high-strength steels (AHSS), this sensitivity to clearance is significantly greater.
Under ring illumination, edge cracks appear as dark linear interruptions in the bright specular band that runs along the sheared edge. The key imaging consideration is that the camera needs to capture the edge at sufficient resolution to distinguish a true crack from a normal burr or a rollover radius. We typically configure a pixel pitch of 10–15 microns per pixel on the edge zone for reliable crack detection on parts where the critical crack width is above 0.1mm.
Edge cracks correlate strongly with tool wear state. One of the more useful quality-engineering applications of inline crack detection is using the detection rate trend as a proxy for punch wear — a rising crack detection rate on a previously clean run often signals that the tooling is approaching replacement threshold.
Scoring
Scoring (also called galling in some plant contexts) is a linear scratch pattern caused by metal-to-metal contact between the blank surface and the die face. It can originate from debris on the blank, insufficient lubrication, or pickup material accumulating on the die. On automotive-grade cold-rolled steel, scoring typically appears as bright directional marks that run parallel to the metal flow direction during the draw operation.
The directionality of scoring is both a visual characteristic and a training consideration. Scoring marks that run at an oblique angle to the part travel direction on the conveyor look different from scoring along the travel axis. We separate these into distinct defect classes during training rather than treating all scoring as one class — the visual signature difference between them is large enough that a single-class model tends to underfit at least one orientation.
Under dome or diffuse illumination, scoring can be difficult to detect because the directional reflection that makes it visible under ring or directional lighting is washed out. Ring illumination is generally preferred for scoring detection on flat and lightly curved stampings.
Die pickup
Die pickup refers to small fragments of work material that adhere to the die face and subsequently transfer back to the part surface on subsequent strokes. It produces small raised nodules or smeared surface patches that have a different reflective character than the surrounding material. The nodule shape varies — sometimes roughly spherical, sometimes elongated — depending on how the transferred material deforms under subsequent punch pressure.
Die pickup is intermittent in its appearance pattern. A line might run cleanly for several hundred parts, then produce a cluster of pickup-marked parts before the accumulated transfer material is worn back into the flow. This intermittency makes it harder to catch by sampling-based inspection; it needs 100% coverage to reliably intercept the clusters.
Detection-wise, die pickup presents as local bright spots (raised nodules reflect ring light intensely) or as irregular dark smear marks depending on the material composition and deformation state. We often use a two-channel detection approach — a high-threshold bright anomaly channel for nodular pickup and a lower-threshold dark anomaly channel for smear-type pickup.
Orange peel
Orange peel texture is a surface roughness condition that develops during deep draw operations when the grain structure of the sheet material is too coarse relative to the strain being applied. Individual grains rotate and reveal their individual crystal facets, creating a dimpled texture that resembles the skin of an orange. It's more common on softer aluminum alloys and on steel with excessive grain size from improper annealing.
Orange peel is a diffuse surface condition rather than a localized defect, which makes it a different detection problem from the point defects above. It requires capturing the mid-spatial-frequency texture of the surface — ring illumination works but the lighting angle matters more than for detecting cracks or scoring. A ring light mounted at 25–30 degrees off-axis tends to provide better texture contrast for orange peel than a ring perpendicular to the part surface, which can produce too much uniform specular return.
The quality threshold for orange peel is often cosmetic rather than functional — it affects paint adhesion and final appearance in exposed assemblies. This means the accept/reject threshold is more grade-dependent and sometimes application-specific, requiring the detection model to be calibrated against part specifications rather than a universal severity cutoff.
Springback shadows
Springback shadows are an artifact of the inspection process rather than a primary surface defect, but they appear frequently enough in image data that they need to be handled explicitly. When a formed stamping springs back from the die geometry, the part's final shape differs slightly from the die shape — this is the well-known springback phenomenon. In parts with complex geometries, this can result in subtle surface facets or curvature discontinuities that produce anomalous shadow patterns under certain illumination angles.
The challenge with springback shadows is that they are part-to-part variable (springback magnitude depends on sheet property variation within the coil) and can appear visually similar to real surface defects under ring illumination. Training a model to distinguish springback-induced illumination artifacts from true surface defects requires including both positive examples and hard negative examples — parts with visible springback shadows that passed manual inspection — in the training set.
We're not saying springback shadows are never quality concerns. For tight-tolerance assemblies where springback creates out-of-spec geometry, that's a dimensional issue handled separately from surface inspection. The surface inspection concern is specifically the false positive problem: flagging a cosmetically acceptable part because the illumination created a shadow on a radius transition.
Coating skip
Coating skip applies to parts that receive a coating — phosphate, zinc, paint primer, or e-coat — before or after stamping. A coating skip is a region where the coating failed to apply uniformly, leaving bare substrate or a thin-coat area. Under ring illumination, a coating skip often shows as a lighter patch on a coated surface (assuming the coating is darker than the substrate) or as a region with different specular reflectivity.
Coating skip detection benefits from multi-image capture when the illumination geometry can be adjusted — the reflectance difference between coated and bare metal typically increases as the illumination angle becomes more oblique. In single-camera single-shot setups, a coaxial illumination configuration (light path co-incident with the lens axis) often provides better contrast for coating uniformity than ring illumination, because it emphasizes diffuse vs. specular differences rather than surface topography.
One practical complication: coating skip defects on curved surfaces interact with the surface geometry in ways that change the apparent severity depending on viewing angle. Setting consistent detection thresholds for curved stampings requires normalizing for the expected reflectance variation due to geometry, which is why accurate part position registration matters more for coating skip inspection than for point defect detection.
The detection problem differs by defect type
The unifying theme across these six categories is that each defect type has a distinct physical origin, a characteristic visual signature under illumination, and — critically — a different set of hard negatives that need to be represented in training. A model trained only on severe edge cracks will miss the early-stage micro-cracks that matter most for fatigue-sensitive applications. A model trained on scoring from one material will generalize poorly to scoring on a different alloy with different surface finish.
When we set up a new Procunit deployment on a stamping line, the first question we ask quality engineering is: which defect types does your quality plan currently address, and which ones are being caught downstream or not caught at all? The answer shapes which defect classes get their own detection head, where the training sample collection effort needs to be focused, and which illumination configuration gives the best signal-to-noise ratio across the mix of defect types on that line.