Solder pad inspection lighting: separate copper wire from dark edges

A 3 mm spot-weld pad carrying a copper wire packs three contrast problems into one small image: shiny wire against shiny pad, bright pad against a dark masked border, and curved wire edges against both. Point the wrong light at it and every boundary grows a white halo or a black band. This guide compares the three geometries worth trying on this part — low-angle ring, coaxial and shadowless — explains why each produces the image it does, and closes with the pixel-scale arithmetic that turns a clean image into a wire area number you can defend. Every test here runs at the bench in about half an hour, with one camera position and three lights.

The symptom: three boundaries in one 3 mm image

Name the boundaries before choosing a light, because each one fails for a different reason.

• Wire against pad: both are specular metals of similar brightness under white light. The boundary is a hue and curvature edge, not a brightness edge, so it survives only geometry that ignores angles.
• Pad against the dark border: black solder mask or a beveled dark ring reflects a small fraction of what bare metal does, so this boundary survives almost any light you point at it.
• Wire edges: the round wire returns light in a narrow cone that shifts with wire position, so edges flare or vanish between parts.

The classic bad image shows all three failures at once: a white blob where wire and pad merge, a gray halo where the border should be, and edges that change shape part to part. That cluster indicts the illumination geometry first, not the camera or the algorithm.

Why shiny copper vanishes: specular reflection

A polished copper surface behaves like a mirror: the law of reflection sends each incoming ray away at the mirror angle, so a discrete LED shows up as a white hotspot exactly where your segmentation wants an edge. The copper surface inspection guide works through the same failure on larger copper parts.

Two families of geometry beat the mirror. Diffuse sources arrive from many directions at once, so what reaches the lens is an average over angles, and the average is stable. Dark-field sources go the other way and use the mirror deliberately: rays graze the surface so flat areas read dark and raised edges glow. The bright-field versus dark-field comparison maps which surface angles each family rewards. Which one you want depends on what the boundary is made of — and on this part the boundaries disagree, which is why one light cannot do all three jobs.

Cause 1, most likely: the low-angle ring light

Start here if the pad reads dark and the wire edges glow. A low-angle ring light in the 0-15° class is a dark-field tool: rays skim across the surface, so the flat pad reflects them away from the lens while burrs, dents and raised wire edges send light back. For finding dents and pits on metal surfaces that is exactly the right tool. For separating pad, wire and mask by brightness it inverts the tones you need, and the wire's curved top returns light inconsistently as the wire shifts, so the wire's apparent width — and any area measured from it — moves between parts.

Test: swap the ring out and reshoot the same field under any diffuse light. If the pad turns bright and stable, geometry was the cause. Fix: keep the low-angle ring only if a second, edge-defect check needs it, and never measure wire area from its images.

Cause 2: coaxial light brightens every flat face

A coaxial light feeds its LEDs into the camera's optical axis through a beam splitter, so any face perpendicular to that axis returns light straight back and reads bright. On this part that cuts both ways. The flat pad lights cleanly, which is why coaxial is the first choice for flat shiny planes such as recessed connector pins. But the round wire is perpendicular to the axis only along its top line, so the wire collapses to a thin bright stripe, and a glossy mask top brightens along with the pad — the border you need can merge into the pad it surrounds.

Symptom: pad and border blur together while the wire loses width. Test: compare gray levels across the pad-mask border; if the border tracks the pad, the coaxial return is flattening both. For the flat-plane jobs that remain — pad presence, seam width, flat-mask boundaries — a compact coaxial unit is the sharper tool:

The fix: a shadowless light

A shadowless light wins this geometry, and it is not close. Whether the diffuser is domed or flat, the LEDs fire into it and the part sees illumination from a wide cone of directions at once. Specular reflection of a broad diffuse source looks similar from every viewing angle, so the image stops reporting geometry and starts reporting reflectance: solder bright, copper bright with its warm tint, black mask dark. Every boundary you need becomes a reflectance or hue edge, the most stable kind an algorithm can segment. The dome light guide covers sizing for domed versions, and the shiny-metal-pins comparison benchmarks dome, flat diffuse and polarizer on exactly this kind of surface.

Size the light to overhang the field, not the bench: any unit in the category's under-50 mm class covers a 4.5 mm field with margin to spare. If the camera must look down through the light itself, choose an apertured format:

Flat formats fit tight height budgets between lens and part:

Worked example: pixel scale for wire area on a 3 mm pad

The inputs, with where each comes from:

• Field: 3 mm × 3 mm, the part itself.
• Framing margin: 25 % per side, so part-position tolerance cannot clip the dark border. Field of view: 3 mm × 1.5 = 4.5 mm.
• Camera: a typical 5 MP monochrome sensor, 2448 × 2048 pixels at 3.45 µm — an example input; any current 5 MP part lands within a few percent of it.
• Wire spread for the error budget: 1 mm², about a 1.13 mm circle, so the perimeter is π × 1.13 mm ≈ 3.55 mm.

The rule, in words: pixel pitch on the part equals field of view divided by pixels across. A segmentation boundary is only good to about one pixel, so the measured outline can shift a pixel along the whole perimeter, and the area error is perimeter times pixel pitch.

The arithmetic:

• p = 4.5 mm ÷ 2448 px = 0.00184 mm ≈ 1.84 µm per pixel.
• Smallest cleanly segmentable feature ≈ 5 px: 5 × 1.84 µm ≈ 9 µm.
• ΔA = 3550 µm × 1.84 µm ≈ 6530 µm² = 0.0065 mm², which is ±0.65 % of the 1 mm² wire spread.

Result: a 5 MP camera over a 4.5 mm field reads that wire area to about ±0.7 %. The decision follows the accuracy spec. At a ±2 % wire-area spec, 5 MP passes with roughly 3× margin. At a ±0.5 % spec it fails, because 0.65 % exceeds 0.5 %; the same field on a 12 MP sensor (4096 px across) gives p = 4.5 mm ÷ 4096 = 1.1 µm and ΔA = 3550 µm × 1.1 µm ≈ 3900 µm² ≈ ±0.4 %, which passes. Threshold: 5 MP covers specs down to about ±1 %; tighter than ±0.5 % needs 12 MP at this field.

Re-run the division for your own field and feature size:

The same inputs size the lens. Sensor width is 2448 px × 3.45 µm = 8.45 mm, so the magnification is 8.45 mm ÷ 4.5 mm ≈ 1.9×. Ordinary factory lenses focus well below that, so this is macro and telecentric territory; the lens focal length calculator turns magnification and working distance into a focal length.

Color or monochrome: separating copper from solder

Three cues separate copper from solder, in the order worth trying.

1. Hue: under white light copper reads warm and solder neutral. A color camera separates them with no extra hardware, and at this scale — boundaries several pixels wide — it is a dependable primary cue.
2. Gray level under a short wavelength: copper's reflectance drops toward the blue end of the spectrum while solder's stays high, so a monochrome camera under a blue or violet LED renders the wire darker than the pad. That is a clean gray edge for the area tool. The lighting color wheel generalizes the wavelength pick, and the monochrome contrast guide shows the bright-versus-dark arithmetic.
3. Red or infrared: both metals reflect strongly there, so this choice separates wire from dark mask, not wire from pad. Keep IR for color-clutter problems, using the LED color guide.

The 30-minute bench test

1. Lock the camera and working distance once; only the light changes between shots.
2. Shoot the current light first and keep the file — the baseline names the failure mode.
3. Low-angle ring: a dark pad with glowing edges confirms cause 1.
4. Coaxial: a bright pad whose border tracks its gray level confirms cause 2.
5. Shadowless: expect a bright pad, a warm wire, a dark border.
6. Judge with histograms, not eyes: pad and wire medians should sit at least 25-30 gray levels apart in 8-bit terms at final gain, which leaves headroom for LED aging and part-to-part reflectance spread. The robust-lighting guide explains why thin margins pass at the bench and fail in production.
7. Calibrate scale on a stage micrometer and carry the pixels-per-mm value into the worked example above.

Record wavelength, drive current, exposure and gain with every image you keep. If step 6 shows pad and wire too close, the next lever is wavelength — blue for a gray edge — not more current.

Prevention checklist

• Keep the 25 % field margin in the fixture: a clipped dark border corrupts the area number before any optical symptom appears.
• Reshoot a golden part weekly; copper tarnish shifts hue and reflectance over weeks in ordinary air, so a threshold set in one month drifts by the next.
• Recalibrate scale after any lens, tube, spacer or camera change; pixel pitch belongs to the whole optical chain, not the sensor alone.
• When contrast drifts, restore it with exposure time or strobe energy first and gain last — gain lifts noise into the edge band.
• Log drive current with every inspected image so a dim batch traces to the driver, not to a phantom process change.

And when the verdict is physical rather than optical — a burned, lifted or scraped pad — the rework path is electrical, not a lighting tweak; SMD pad repair: solder mask scraping and bodge wire techniques covers that side of the bench.

To go further: take your field size and wavelength choice to the light step of the configurator, compare formats in the lights finder, or send a pad photo through the contact page and we will help you spec the light.