Plastic part inspection lighting: angle and wavelength by defect

Plastic parts hide defects from light that is not chosen for them. Uniform bright lighting makes a housing look clean while molded-in scratches stay invisible. A white light passes straight through a clear adhesive bead and shows nothing. A deflected transparent wall splits the image into bright and dark bands. The fix is not a stronger light but a matched one: grazing angle for relief, 365 nm UV for anything fluorescent, wide diffusion for anything transparent, and multi-directional sequences for curved or engraved surfaces. The sections below give the decision rule for each defect class, the stand-off arithmetic that decides whether a low-angle ring survives your geometry, and the bench sequence that confirms contrast on your own parts before you order.

What the defect does with light decides the light

Every defect on a plastic part interacts with light in one of three ways. Relief — scratches, dents, gate scars, engraving — redirects light by geometry, so it wants a shallow, defined angle. Fluorescent marks — adhesive beads, invisible codes — absorb UV and re-emit visible light, so they want the right wavelength, not the right angle. Everything transparent — walls, films, contents — wants a large, even, diffused field so curvature and deflection average out. Decide the mechanism first; the product class follows.

One station can carry several of these lights. Sequence them and the same camera inspects relief, fluorescence and contents without moving the part. For the general angle decision behind the first row, see machine vision lighting angles: surface or silhouette.

Grazing light makes relief defects bright

Relief is geometry, and geometry obeys the law of reflection. Light arriving at 10-30° above the surface leaves at the same angle and misses the camera, so a flat, semi-gloss face images nearly black. A raised scratch tilts the surface locally by a few degrees — enough to throw light into the lens — and the defect images as a bright streak on a dark field. The same physics finds dents in metal, as in low-angle dent and pit detection, and pulls cracks out of white plastic in the crack-lighting guide.

Even lighting does the opposite, which is why the intuitive set-up fails here. A diffuse ring at 90° drives every point of a semi-gloss case toward the same grey, and the scratch drowns in the average. The full comparison is in bright field versus dark field.

The tool is a low-angle ring mounted around the lens.

The LS-RNG-0426 is a 120 mm dia. diffused ring that emits white light at low angle (0-15°) on 24 V. It hugs the bore around a standard lens and grazes the face below it.

The stand-off arithmetic: will a low-angle light reach the part?

Low-angle light works only while the beam is still travelling almost horizontally when it crosses the part. A beam leaving a ring of radius R at θ degrees above the horizontal drops R × tan θ by the time it reaches the axis. For a 120 mm dia. ring, R = 60 mm, and at 15° the fan crosses the bore after 60 × tan 15° = 16 mm of descent. At a 285 mm camera stand-off the light never gets near the face.

Three relations run the whole decision:

Here m is magnification, f the focal length, R the ring radius, r the field half-diagonal and θ the emission angle above horizontal. The ceiling is a first-pass estimate — it ignores beam spread and fixture walls — but it separates workable stand-offs from hopeless ones by a wide margin, which is what the choice needs.

Worked example — top-face scratch inspection on an 80 × 60 mm case:

• Field of view with 10% margin: 80 × 1.1 = 88 mm wide, 60 × 1.1 = 66 mm high.
• Sensor: 2448 × 2048 px at 3.45 µm, a common 5 MP class. Width = 2448 × 0.00345 mm = 8.45 mm.
• Image scale: 88 / 2448 = 0.036 mm/px. A 0.3 mm scratch spans 0.3 / 0.036 ≈ 8 px — comfortable for detection.
• Magnification: m = 8.45 / 88 = 0.096. A 25 mm lens then needs WD = 25 × (1 + 0.096) / 0.096 = 285 mm.
• Field half-diagonal: r = √(44² + 33²) = 55 mm.
• 15° ring: ceiling = (60 + 55) × 0.268 = 31 mm. 30° ring: 115 × 0.577 = 66 mm.

Both ceilings sit far below the 285 mm focus distance. The conclusion is not that grazing fails, but that the light belongs at the part, not at the lens: mount the ring 25-30 mm above the face, around the barrel, and the camera still focuses at 285 mm. Two checks close the decision. The ring's clear inner opening must exceed the field at the ring's mounting height — an 88 mm field seen from 30 mm below needs about 88 × (285 − 30) / 285 ≈ 79 mm of opening. And the height must respect the ceiling: 25-30 mm for a 15° ring here, under 66 mm for a 30° ring. The lens focal length calculator runs the same m and WD arithmetic for any field and sensor.

Multi-angle rings and photometric stereo for curved and engraved parts

Curved and wrinkled parts scatter grazing light unpredictably: the angle that flares a scratch on the flat top misses it on the shoulder. Two methods restore control.

First, change angle instead of light. Shallow 0-30° emission reads relief; steeper 60-85° emission maps texture and embossing across a curved face.

LS-RNG-0730 is listed from low angle (0-15°) to 85°, in a 103 mm dia. body with a 48 mm inner aperture — one light that covers both regimes on fields up to about 48 mm.

Second, capture from several directions and combine the images in software. Each direction shadows a different part of the relief, and the combined gradient image shows engraving that no single frame resolves — that is photometric stereo, and the photometric stereo ring guide shows the capture set-up. On rubber moldings, where surface texture buries characters, this multi-directional sequence is usually the only method that returns readable characters.

Where one fixed direction is enough, LS-RNG-0189 delivers a 30° fan from a 120 × 80 mm footprint on 24 V. In the worked example's geometry, that mid angle covers stand-offs up to about 66 mm.

Grazing has a hard limit: on matte, textured plastic the surface already scatters in every direction, so relief barely moves the grey level. There, move to multi-directional capture or backlight silhouette instead of pushing the angle lower.

365 nm UV for adhesives and invisible codes

Clear adhesives are engineered to vanish under visible light, which is why visible-light stations miss beads, gaps and overflow. Under 365 nm UVA, additives in the adhesive fluoresce: the bead images as a bright line on a dark background while the container stays dark. The same excitation reveals fluorescent inks in invisible codes on plastic plates. The rule: if the mark is invisible in white light, test 365 nm first, then 395-405 nm violet as the softer option. The catalog lists UVA and violet across all three light families — 15 ring lights, 84 bar lights and 22 spot lights at 365 nm.

Keep three things in step. Shade the cell and block ambient: fluorescence is faint, so the UV light and a slightly longer exposure carry the image. Measure the lens as well as the light: C-mount optics coated for the visible band often transmit less at 365 nm, so judge the captured image, not the assumption. And keep the geometry identical to the visible station, so both images register — sequencing a UV and a white light on one camera does this directly.

Diffused wide panels for transparent parts

Transparent walls beat directional light twice: they reflect it specularly at every curvature, and they deviate it as the wall deflects between parts. A large diffused emitting surface fixes both at once — each point of the wall collects light from a wide solid angle, so curvature and deflection average into an even background while stains and foreign material stand out as local darkening. Put the panel behind the container for contents and in-wall stains; in front, at a slight angle, for surface smears. The screen projection method extends the idea with a structured background, and liquid level lighting applies it to fill checks.

LS-BAR-1276 is a 372 × 28 mm diffused white bar on 24 V — a soft window of a strip for linear fields.

Directional bars stay in the kit for the opposite job: relief on elongated parts.

LS-BAR-1047 is a 290 × 30 mm standard, non-diffused bar on 24 V. Mounted low and angled inward, it does across a long profile what the grazing ring does on a case.

If a glossy face still mirrors the panel, add a polarizer in front of the lens and a polarizing sheet on the light — cross polarization strips the specular sheen and leaves the print.

Colour checks and small fields

Colour identity — right cap, right case, right insert — needs colour-faithful light more than bright light: one diffused white light, one fixed exposure, no ambient. The part's colour then becomes a stable measurement. Where contrast is colour-dependent, sequenced red, green and blue captures beat any white-light compromise; the LED colour contrast guide ranks the colours per surface.

For small fields, a short-stand-off white spot keeps the bench tight: mounted a few tens of millimetres above the face, it floods fields of a few tens of millimetres evenly and fits where a ring would collide with fixturing. LS-SPT-0162 is a 24 mm dia. white spot on 24 V built for exactly that.

Confirm the choice on real parts, then order

Lighting is the cheapest item on the bench to test and the most expensive to get wrong in production. Run this sequence before ordering more than one sample:

• Image the extremes. The best part you must accept and the worst you must reject, in the same set-up. If the light cannot separate them, software will not rescue it.
• Sweep angles by hand. Move a handheld light across 0-85° and note the two angles that flare the defect most cleanly.
• Run the numbers. Field, scale, magnification and stand-off as in the worked example; the ceiling (R + r) × tan θ sets the ring angle and height.
• Check the aperture. Ring inner opening wider than the field, measured at the ring's mounting height.
• Test 365 nm on the fluorescent marks, ambient shaded, camera gain fixed at its production value.
• Score the contrast. In the winning set-up the defect should read two to three times the background grey level at production gain. If it needs maximum gain, change the light, not the gain.
• Then buy the station. Choose lights in the light finder, size the rest of the chain in the system configurator, and ask for a quote on multiples through the quote page.

The neighbouring techniques — domes, coaxial light, line scan — live on the lighting topic page.