Copper surface inspection lighting: beat specular reflection

Counting dirt on copper plates with a pixel counter always runs into the same wall: with no light the image is too dark to judge, and with a bar light the polished metal mirrors the LEDs straight into the lens as one clipped hotspot. Both symptoms share a single cause, and the fix is geometry, not intensity. This guide gives you the decision between the two set-ups that work on flat copper — a diffuse dome or shadowless light directly over the plate, and bar lights grazing the surface for dark field — plus the LED colour, a worked optical example with the arithmetic, and bring-up steps that end in a threshold your counter holds all shift. For the wider background, why machine vision inspection succeeds or fails on lighting walks the same decision across other surfaces.

Why copper throws your light back

Polished copper is a specular surface: a ray leaves at the same angle it arrives, so your camera receives an image of the light source rather than of the metal. A bar mounted above and to the side paints its own LED strip across the plate, and wherever that image lands the pixels clip at white and the pixel counter calls it dirt.

Copper's reflectance curve makes it worse. Polished copper returns roughly 90% of deep red and near-infrared light but only about a third around 450 nm. The orange blaze in your image is mostly long-wavelength return; dimming the bar shrinks it without changing the ratio between hotspot and plate.

That 2θ term is why a fixed threshold fails plate after plate. Re-seat a plate 2° off and its hotspot jumps 4°, landing somewhere new. And with the light off, exposure stretches until overhead factory lighting dominates the image, which drifts every time the room changes. The way out is to stop asking the reflection to miss the lens by luck: fill everything the plate can see with diffuse light, or aim the light so the mirror sends it away on purpose. The choice between surface and silhouette lighting angles is the same fork, and diagnosing stray light and low contrast covers the ambient half of the dark-image problem.

The decision up front

For a dirt counter on flat plates, the diffuse overhead set-up with a blue LED is the default answer, because sooner or later someone will want to see the plate, not just the count. Choose dark field when the dirt rate is the only output and you want the simplest threshold there is: bright specks on a black plate.

Set-up A: diffuse light directly over the plate

A dome or flat shadowless light fills the whole hemisphere above the plate with diffuse light, so every point on the copper reflects a uniformly bright field. The specular image becomes the diffuser itself — a smooth, low-contrast gradient with no clipped core — and the dirt becomes the only thing that stands out. Dome lights do this for small parts; shadowless lights do it in flat, low-profile form for plates.

Two sizing rules govern both. First, the emitting area must overfill the field of view by 10-15% on every side, so no point on the plate sees past the diffuser edge into the dark room. Second, with a classic domed light keep the inspection circle at or under half the dome's emitting diameter, because a dome is only truly uniform over its centre.

For small parts, a classic dome is the compact answer:

The LS-DOM-0085 is a Ø166 mm dome with a Ø116 mm emitting area. That suits inspection circles up to about 60 mm across.

For plates, flat and arch shadowless lights apply the same overfill rule in a lower profile; the worked example below sizes one for a 120 × 80 mm plate. On curved copper parts, dome lighting for shiny, curved and wrinkled surfaces extends the same rules to compound curves, and the same overfill discipline underlies flatness inspection with controlled lighting on any flat-part line.

Set-up B: low-angle dark field bars

Tilt the bars down until they graze the plate. At 10-15° elevation the mirror still obeys angle-in-equals-angle-out, but the reflected beam leaves 10-15° above the surface on the far side, nowhere near a camera looking straight down. The copper images near black, and anything that scatters in all directions — dust, lint, flux residue, scratches, oxide — throws some of that light into the lens. Dirt counts become bright-on-black, the easiest threshold a pixel counter can hold.

Aim each bar across the plate, not along it, and use at least an opposing pair. Height sets the stand-off: at 15° elevation a bar head 100 mm above the plate aims at a spot 100 ÷ tan(15°) ≈ 373 mm along the surface, so budget about four times the bar height in reach.

Bar lights come in directed and diffused construction, and the choice follows the job: a directed standard bar makes the sharper grazing beam for dark field, while a diffused bar wraps edges and suits the four-sided tunnel. The LS-BAR-0400 is a 312 × 30 mm high-luminance bar. The LS-BAR-1276 is a 372 × 28 mm diffused bar in white. Two bars show the split:

All 2 share: Cable length 500 mm. Values as printed in each product's data sheet; open a part for its drawing and full specifications.

If the bar cannot move: cross-polarization

When the machine fixes the light where it stands, put a linear polarizer on the bar and an analyzer on the lens, rotated 90°. Specular reflection from metal keeps its polarization, so the analyzer blocks it; dirt is diffuse and depolarizes, so roughly half of its light passes. The hotspot disappears and the specks remain. The price is light: a crossed pair keeps only about a sixth of what you had, so plan for 4-6× the exposure or a strobed drive. Cross-polarization and dark field solve the same problem — pick one, because stacking polarizers on already-faint scatter starves the image. The full method, including how to rotate the analyzer, is in cross polarization to reduce surface glare.

The LED colour that maximizes dirt contrast

Blue light around 450 nm turns copper into a dark background on its own: the metal reflects only about a third of it, while dust and residue scatter broadly and stay bright. Deep red and near-infrared do the opposite, feeding the metal the roughly 90% it reflects and pressing dirt contrast down. Use white only when a person also judges tone or oxide colour from the image. For a pure dirt counter, commit to blue. The colour-by-material method behind this is in choosing the LED colour for best contrast.

Worked example: 120 x 80 mm copper plate, 0.25 mm dirt

Inputs, each from the bench:

• Plate: 120 × 80 mm copper, held flat within a few millimetres — a typical plate at a manual inspection station.
• Dirt that must count: specks 0.25 mm across and larger. A pixel counter needs at least 3 pixels across a speck to count it reliably.
• Camera: a common 5 MP machine-vision format, 2448 × 2048 pixels on 3.45 µm pixels.
• Margin: 10% around the plate, so the field of view is 132 × 88 mm.

The optical chain, as formulas and then as numbers:

• Sensor width: 2448 × 0.00345 mm = 8.45 mm.
• Scale: 132 ÷ 2448 = 0.054 mm per pixel, so a 0.25 mm speck spans 0.25 ÷ 0.054 = 4.6 pixels — above the 3-pixel floor.
• Magnification: 8.45 ÷ 132 = 0.064.
• Lens and working distance: with a 16 mm lens, WD = 16 × 1.064 ÷ 0.064 = 266 mm.
• Depth of field at f/8, taking the circle of confusion as one pixel: 2 × 8 × 0.00345 × 1.064 ÷ 0.064² = 14 mm, so a plate bowed ±5 mm stays sharp.
• Light: overfilling 132 × 88 mm by 15% needs at least 152 × 101 mm of emitting area. The LS-SDL-0026 is an arch shadowless light at 204 × 110 mm. It clears both axes with the plate's long side laid across the arch.

The station, then: 5 MP camera, 16 mm lens at f/8, 266 mm working distance, blue arch shadowless light over the plate, 2 ms exposure to start. Set drive so clean copper reads about 120 of 255; specks then land 180 and higher; count pixels at 160 and above, which holds 40 levels of margin above the copper and 20 below the smallest speck. If the dirt rate is the only output, swap the arch for two grazing bars on the same camera and keep a bright-on-black threshold.

On the pixel side, the camera resolution calculator checks the 3-pixel rule for your own speck size.

Bring-up, step by step

1. Square the plate to the camera and lock the fixture; the 2θ rule punishes every loose re-seat.
2. Mount the diffuse light on axis, or the bars at grazing elevation aimed across the plate.
3. Set the lens to f/8, focus on the plate, and lock the focus ring.
4. Choose blue output for dirt counting; white only if a person judges tone.
5. Set drive and exposure so clean copper sits near half of the grayscale, with nothing clipped above 250.
6. Gate the light to the exposure if room light varies; strobed lighting and ambient rejection gives the measurement method.
7. Teach the pixel-counter threshold from a clean plate and a known-dirty plate, midway between their levels with margin on both sides.
8. Re-seat a clean plate ten times; it must count zero every time before the station ships.

Checks before you order

• Emitting area of the diffuse light ≥ field of view plus 15% on both axes.
• Classic dome: inspection circle at or under half the emitting diameter.
• Bars: lit length at least 10% longer than the plate edge they cover, in an opposing pair; four bars for plates that arrive in any orientation.
• Confirm the exact model in blue, because colour options differ by series.
• 24 V drive and strobe current match your controller's channel.
• Working distance leaves room for the light between lens and plate.

Check these numbers against parts in the lights finder, or let the lighting step of the configurator size them with you.

When the usual answer is wrong

• Wavy or bent plates break both geometries: the dome gradient slides and dark field shows bright bands where the surface tilts toward a bar. Flatten the plate mechanically or build a four-bar diffuse tunnel around it.
• Plates wider than about 250 mm outgrow practical diffuse overfill. Use the four-bar tunnel, or move the count to a line-scan camera with a diffuse line light; balancing coverage and brightness over a large field of view covers the sizing.
• Oily copper smears a specular film across the plate that no geometry removes. Cross-polarize and fix the cleaning step; improving reliability on oily surfaces works the film problem directly.
• Small curved copper parts — fittings, pins, contacts — need a classic dome rather than a flat source; lighting shiny metal pins without glare is that case in detail.
• Moving plates need the bars strobed: at 100 mm/s, hold the pulse under 0.5 ms to keep blur to one pixel (0.054 ÷ 100 ≈ 0.5 ms).

First-shift checklist

• Clean copper images near mid-scale with zero pixels clipped at 255.
• A re-seated plate changes the copper reading by less than ±10 levels.
• The 0.25 mm speck spans at least 3 pixels on screen.
• Emitting area overfills the field of view on both axes.
• Bars graze at low elevation in an opposing pair, lit length past the plate edge.
• Threshold sits midway between clean copper and the smallest speck.
• Ten re-seats of a clean plate count zero.
• Counts hold with room lights on and off.

When the plate, the dirt or the rate is unusual, send its size and the smallest speck you must count through the contact page and we will size the geometry with you.