Steel coil end face inspection with multi-camera imaging

A steel-coil end face can fill an image while its smallest important defect remains barely visible. Adding cameras helps only when their views preserve detail, cover the entire inspection region, and stay associated with the correct coil. This guide develops a validation plan for integrators specifying multiple area-scan views, controlled lighting, and identification capture. You will calculate sampling and overlap, separate identification from defect decisions, and define commissioning tests before committing to a production layout.

Define the inspection decision before choosing cameras

Start with a defect specification, not a megapixel target. Record the smallest visible width, required location accuracy, and acceptable false-reject rate for each defect class. A narrow crack and a broad displaced layer require different image evidence, even when their overall lengths match.

Separate the station into three decisions:

• Surface inspection: Is a visible edge, layer, or surface feature outside an agreed acceptance limit?
• Identification: Which coil does the captured evidence belong to?
• Disposition: Should the coil pass, stop for review, or follow a defined exception route?

List features that an end-face view cannot establish reliably. Hidden internal damage, overall coil width, and axial displacement may require other viewpoints or depth measurements. Do not turn a visible intensity change into a dimensional claim without validating the geometry.

Use the system requirements selection guide to organize the acceptance requirements. For acquisition architecture, the closest companion is the multi-camera capture design guide.

Calculate sampling from the smallest required detail

Object-space sampling describes how much surface distance corresponds to one image pixel. It is not the same as optical resolution or guaranteed defect detection.

For an initial design, assign several pixels across the smallest relevant defect width. Then test real samples because contrast, focus, surface texture, and noise determine whether those pixels contain useful evidence.

Consider an illustrative requirement for a visible defect width of 1 mm. Choosing an initial target of 4 pixels across that width gives:

This is a planning assumption, not a detection guarantee. A low-contrast crack may need finer sampling or a different lighting direction. A long feature can still be missed when its width is poorly resolved.

Use the camera resolution calculator workflow alongside the calculation below to shortlist sufficient image dimensions, then verify both sensor axes rather than megapixels alone.

Record the required code detail separately. Character stroke width or barcode module width may demand finer sampling than the defect inspection.

Divide the face into overlapping, calibrated views

For an illustrative maximum face diameter of 2200 mm, plan coverage over its bounding square. Include additional allowance for positioning error and coil movement before finalizing the fields.

For equally sized adjacent views, overlap reduces the combined usable width:

Suppose the starting layout uses 3 views across each axis with 15% overlap:

Rounding each field upward to 850 mm gives 3400 pixels per axis at the target sampling. Combined coverage becomes 850 × 2.7 = 2295 mm. The remaining margin is only 95 mm across the total field, so confirm whether that accommodates the actual positioning envelope.

The lens must deliver the required field and usable detail at the available working distance. This relationship is illustrated below.

A square layout would contain 9 views. Some corner regions lie outside the circular face, but removing cameras requires a coverage analysis across every allowed coil diameter and position.

Overlap serves two purposes: preventing coverage gaps and providing common features for registration. Repeated coil layers can produce ambiguous matches, so calibrate camera geometry with a suitable reference rather than relying entirely on surface matching.

Keep native images for defect decisions. A stitched overview is useful for navigation, but resampling can soften narrow features or create misleading seams. Use the lens and lighting test plan to check corner detail and focus before fixing the layout.

Select camera architecture around motion and throughput

Browse area scan cameras using the camera finder, but reject candidates that cannot supply the required dimensions on both axes. A high total pixel count does not establish sufficient vertical coverage.

The MV-CS200-10UM is a 20 MP USB3 camera with a rolling shutter. The MV-CH120-11UC is a 12 MP USB3 camera with a global shutter. These illustrate different shutter and resolution choices; neither is established as suitable for the worked layout without checking its full image dimensions and acquisition behavior.

Compare their catalog specifications before shortlisting either camera:

All 2 share: Gain 0 dB to 24 dB; Binning Supports 1 × 1, 2 × 2, 4 × 4; Decimation Supports 1 × 1, 2 × 2, 4 × 4; Lens mount C-mount; Compatibility USB3 Vision, GenICam; Certification CE, RoHS, KC. Values as printed in each product's data sheet; open a part for its drawing and full specifications.

A rolling shutter can be appropriate when the coil is stationary throughout the frame acquisition. For moving material, a global shutter is generally the simpler starting point because it avoids row-by-row geometric skew. Global shutter operation still needs sufficiently short exposure to limit motion blur.

For an illustrative surface speed of 500 mm/s, sampling of 0.25 mm/pixel, and permitted blur of 0.5 pixel:

Use the fastest expected motion, not the nominal transport speed. Confirm exposure timing, trigger support, and inter-camera synchronization from the selected equipment documentation.

Also budget transfer and storage. An illustrative set of 9 images, each containing 20 million pixels at one byte per pixel, produces approximately 180 MB per capture. Storing each pixel in two bytes increases that to approximately 360 MB, before metadata and processing overhead.

Test lighting direction before increasing intensity

Steel surfaces can return bright reflections next to dark scale and layer shadows. Increasing light output may saturate reflections without revealing the defect. Start by testing direction, apparent source size, and exposure using representative surface conditions.

Bar lights allow directional illumination from different sides. The lighting finder helps separate diffused sources from more directional options. A diffused source may reduce sharp highlights, while grazing illumination may reveal raised edges but hide features behind their shadows.

The LS-BAR-1276 is a diffused white bar light. Treat it as a bench-test candidate, not evidence that one light covers the complete coil face.

Capture opposing lighting directions separately when one direction hides relevant features. Keep the images associated with their lighting state, and compare performance against the same defect samples.

The task-based lighting selection guide provides a useful framework for these trials. If glare remains dominant, evaluate the tradeoffs described in the cross-polarization guide; do not assume polarization will remove every metallic reflection.

Strobing can support short exposures, but it does not inherently remove glare or guarantee lower equipment temperature. Verify permitted pulse operation, repetition rate, electrical compatibility, and thermal limits for the actual light and controller. The optical rise-time measurement guide explains why the commanded pulse duration may differ from the useful illumination interval.

Keep code reading and text recognition distinct

A coil identifier may be a barcode, a printed character string, or an embossed marking. These require different tools. A barcode decoder does not automatically recognize ordinary text, and successful text recognition does not verify that the identifier belongs to the inspected coil.

Use fixed code readers when the task involves a supported machine-readable code and the required field can be covered. The LS-CRD-0028 is a fixed code reader with a global shutter. Its suitability still depends on the actual symbol, module size, reading geometry, and interface requirements.

For sprayed or embossed text, plan a separate text-recognition validation using representative fonts, damaged strokes, contamination, and changing marking positions. Embossed markings may benefit from directional lighting that differs from the defect-inspection lighting.

Assign one inspection transaction identifier when the coil enters the capture sequence. Attach every camera frame, lighting state, decoded value, and decision to that transaction. Do not associate images solely by the order in which processing finishes.

Define disagreement handling explicitly: an unreadable identifier, conflicting identifiers, or an unexpected production record should produce a controlled exception. Use the barcode-to-part verification guide for identity checks and the reader verification guide to distinguish decoding from print-quality assessment.

Validate protective measures as part of the optical system

Where the installation exposes equipment to radiant heat, dust, moisture, or airborne debris, protection becomes part of image formation. A protective window can introduce reflections, reduce contrast, or collect contamination directly in the optical path.

Specify environmental conditions at each camera and light location. Include the expected operating cycle, radiant exposure, ambient temperature range, and maintenance access. Check equipment limits against measured conditions rather than assuming the surrounding air temperature describes every component.

Develop housings, cooling, purging, and window arrangements with qualified equipment specialists. Do not infer a complete assembly's environmental rating from a single component rating. Cooling arrangements must also account for condensation risk and the consequences of a supply failure.

Validate the installed optical path with the final window in place. Capture clean-window reference images, then establish how contamination changes contrast and defect visibility. Define a maintenance alarm before image quality becomes unacceptable.

Commission with a written acceptance matrix

Use samples that represent the permitted surface variation, not only clean examples with obvious defects. Include acceptable coils with confusing texture so that false rejects receive the same attention as missed defects.

Run commissioning in a repeatable order:

1. Measure scale and coverage. Check every camera and the complete positioning envelope, including the inner opening and outer edge.
2. Verify exposure and focus. Test the most reflective and darkest acceptable surfaces without changing unrelated settings.
3. Challenge overlap boundaries. Place relevant reference features where decisions might otherwise depend on a stitched seam.
4. Validate defect decisions. Compare automated classifications with agreed ground truth using a separate validation set.
5. Challenge transaction handling. Introduce delayed frames, missing frames, failed reads, and repeated identifiers.
6. Repeat under operating conditions. Include warm-up, normal vibration, lighting variation, and the defined maintenance interval.

Report misses and false rejects by defect class and operating condition. An average score can hide a complete failure on a rare but important surface condition.

Check ordering assumptions and know when to change the design

Before ordering, complete this short checklist:

• Coverage: Confirm maximum diameter, inner opening, position variation, and required overlap.
• Detail: Specify the smallest defect width and code feature independently.
• Optics: Confirm lens coverage, working distance, aperture, mounting, and installed-window effects.
• Acquisition: Verify image dimensions, shutter behavior, triggering, and simultaneous transfer capacity.
• Lighting: Confirm coverage, mounting space, permitted operating mode, and controller compatibility.
• Protection: Document component limits, maintenance access, and image-quality degradation thresholds.
• Traceability: Define transaction identifiers, exception handling, retained images, and required production interfaces.

More cameras are not always the right answer. A stationary coil may permit sequential views if cycle time allows. A localized marking may justify a dedicated identification view rather than finer sampling across the entire face. Depth-sensitive defects may require another viewing angle or a different measurement method.

Likewise, do not require a visually perfect mosaic when calibrated native views already support reliable decisions. A stitched overview and a measurement image can serve different purposes.

Start the camera configuration process with the required field and sampling, or use lighting configuration after completing the bench tests. For a layout review, send the defect specification, coil dimensions, motion profile, sample images, and acceptance matrix through application support.