Lighting for crack detection on white plastic cylinders

Cracks on small white plastic cylinders are one of the classic trap cases in surface inspection. The part is light, so almost any lighting gives a bright, pleasant image. The crack is narrow and shallow, so almost any diffuse lighting fills it in. The result is a system that passes bad parts for weeks until someone notices. This guide gives you the geometry, the numbers and a test sequence for finding hairline cracks on curved white polymer surfaces, and tells you when the usual dome-first answer is exactly wrong.

The same reasoning applies to any matte light-coloured cylindrical part: connectors, caps, medical components, moulded sleeves. If your part is dark instead, read stereo depth lighting optimization for dark surfaces for the opposite problem.

The decision first

For crack detection on white plastic, start from one rule: diffuse light hides cracks, directional light reveals them.

The physical picture is simple. A crack is a narrow groove. Under grazing, directional light, the flat surface around it reflects the beam away from the camera while the groove walls scatter light into the lens, so the crack appears as a bright line on a dark field, or a dark line on a bright field depending on the geometry. Under a dome, both the surface and the groove floor receive roughly the same intensity, and the contrast collapses.

This is the mirror image of the glare problem on shiny parts, where a dome is the fix; our guide dome lights for shiny, curved and wrinkled parts covers that side. For a deeper treatment of why thresholding fails when the lighting is wrong, see glass bottle defect detection: fix lighting before thresholding.

Why the dome can be the enemy

Dome lights surround the part with diffuse illumination from every solid angle. On a shiny metal pin that removes blinding specular highlights, which is why we recommend them in lighting shiny metal pins without glare: dome, diffuse and polarizer. But the same property works against you here.

A hairline crack perhaps 20 to 100 µm wide reflects almost nothing of its own; its apparent brightness comes from light bouncing off its walls. A dome lights those walls as evenly as it lights the surface, so the grey value of the crack approaches the grey value of the plastic. On white plastic both are near saturation, and a crack a few grey levels deep is unrecoverable by any threshold.

There is still a place for a dome in this task: it is the fastest way to confirm the part is present, oriented and free of gross defects such as stains, dents or short shots. Keep one in the recipe as the first station, not the crack station.

The dome lights category carries sizes from under 50 mm to over 400 mm; pick a dome whose emitting area is at least twice the cylinder diameter so the part sits fully inside the diffusion volume.

The geometry that works: grazing, directional light

For the crack station you want light arriving nearly parallel to the local surface. On a cylinder this means:

That formula is an estimate, not physics-exact, but it captures the trend: as the illumination angle drops toward grazing, the shadow cast by a groove grows, and so does its apparent width in the image.

Practical targets:

• Ring light at low angle (0-15°): each point on the cylinder circumference receives a grazing beam from its own azimuth. A low-angle or dark-field ring is the single most effective part for this job. In our ring lights finder, filter Angle to "Low angle (0-15°)".
• Two opposed bar lights: mount one bar on each side of the cylinder, aimed across the axis. Cracks running parallel to the cylinder axis, which a ring light illuminates end-on, respond much better to this cross-lighting.
• Sequential quadrant or arc lights: an arc or multi-angle light fired one segment at a time, with one image per segment, covers the full circumference without a second camera. See multi wavelength light: sequence colours to get more from one camera for the same sequencing idea applied to colour.

The LS-RNG-0256 is a low-angle ring light: it is the geometry class to look for first. A diffused ring such as the LS-RNG-0426 sits between a dome and a bare ring; useful when the plastic is slightly glossy and the bare ring gives hot spots.

For axial cracks, two bar lights from the bar lights category aimed from opposite sides, at roughly 10 to 20 degrees off the surface tangent, usually outperform any ring.

Worked example: 12 mm cylinder, 50 µm crack

Suppose the part is a white nylon cylinder, 12 mm diameter, 30 mm long, with stress cracks around 50 µm wide whose depth you do not control. Sensor and lens choices are covered in choosing camera resolution for dimensional inspection; here only the lighting arithmetic matters.

1. Size the crack in pixels. With a field of view of 40 mm on a 2448-pixel sensor, one pixel covers 40 / 2448 = 0.0163 mm = 16.3 µm. The 50 µm crack spans about 3 pixels. That is enough to see but not enough to threshold blindly, so the lighting must deliver maximum grey-value separation.
2. First pass: dome. Capture an image. Line profile across the crack: intact plastic at grey 235, crack floor at grey 224. An 11-level gap will not survive LED ageing or a batch of plastic with different pigment loading. Reject the dome for this station.
3. Second pass: low-angle ring. Same part, ring at roughly 10 degrees. Profile now reads intact plastic near grey 210 (some glare lost to scatter), crack at grey 45. A gap of about 165 levels: robust, and the crack is now 4 to 5 pixels wide because the shadow extends beyond the physical groove.
4. Third pass: rotate. Rotate the part 90 degrees. If the crack fades on one side, add the opposed bar lights and capture two images per part, OR-ing the binary results.
5. Freeze the recipe: light model, angle, working distance, exposure and gain recorded, so the next rebuild reproduces the same contrast.

You can check the pixel-coverage arithmetic with the camera resolution calculator.

Choosing between the candidate lights

Use this order when you specify parts:

1. Circumferential cracks or unknown orientation: low-angle ring light, sized so the cylinder fills roughly half the ring's inner opening; smaller parts sit too far from the beam.
2. Axial cracks: two opposed bar lights, each about as long as the cylinder, aimed across the axis at grazing incidence.
3. Both orientations: ring light first, bars added in a second station or second exposure.
4. Slightly glossy white plastic: add a diffuser or step up to a diffused ring; keep the low angle.
5. Marginal grey-value gap: try blue instead of white. Short wavelengths scatter more at narrow openings, which often deepens a crack's contrast; the trade-offs are covered in which LED colour gives the best contrast in machine vision.

For cracks that are genuinely three-dimensional (visible depth, chipped edges), a 3D sensor may be the more honest tool; see 3D scanner prototype planning for limb surface capture for how to scope such a build.

Checks before you order

• Measure the crack width on a worst-case rejected part, in µm, with a microscope if you have one. Everything downstream depends on that number.
• Compute pixel coverage: crack width / (field of view / sensor pixels). Aim for at least 2 to 3 pixels of physical crack, which grazing light will widen visually.
• Confirm the part can be rotated or that two images per part fit your cycle time. Multi-exposure recipes need a controller that sequences lights reliably; run the controller step of the configurator if your current set-up cannot.
• Check colour: white is the default, but pigment changes between suppliers shift grey values; a blue or dual-colour recipe costs nothing extra in hardware if the light supports it.
• Confirm power and strobing: all lights in this class run at 24 V, and strobing freezes motion on a moving line while multiplying apparent brightness.

When the usual answer is wrong

Three situations break the recommendations above:

• Very glossy white plastic. Grazing light produces a bright specular band that wraps the cylinder and blinds the camera at one azimuth. Add a diffuser, raise the angle slightly, or cross-polarize; expect to trade some crack contrast for glare control.
• Cracks that are really colour or material change, not geometry. A stress whitening zone scatters differently but has no groove. Directional light adds little; colour imaging or a UV-excited fluorescence recipe is the better path, since many polymers fluoresce at the whitened zone.
• Parts that cannot be rotated and give you one image. A single ring exposure leaves blind zones near its own shadow. Either accept a two-exposure recipe, or move to four-quadrant sequential lighting within one trigger.

If cracks were caused by a machining or moulding process upstream, fix the source; the inspection system then only has to catch the rare escape. For threshold strategy once the lighting is right, scratch defect detection with lighting and dynamic thresholding walks through the image-processing half.

Quick checklist

• Dome station: presence, orientation, gross defects only.
• Crack station: low-angle ring for circumferential cracks; opposed bars for axial.
• Crack spans at least 2 to 3 physical pixels at the working distance.
• Grey-value gap between crack and plastic is at least 30 levels, measured, not eyeballed.
• Part rotated or multi-exposure recipe proves coverage of the full circumference.
• Recipe frozen: light models, angles, distances, exposure, gain.

Next step

Run the lighting step of the configurator with your cylinder diameter and crack width to get a matching light shortlist, or browse the low-angle section of the ring lights finder. If the crack spec is ambiguous ("hairline, visible to the eye"), send sample images through the contact page and the applications team will recommend a geometry and a test plan.