A wrist camera that finds brass parts on a lathe conveyor runs clean for 30-40 parts, then the tray's oil glaze starts bouncing the light straight back into the lens and the false positives follow. The failure is geometric, so the fix is geometric: tilt the camera 20-30° off the specular path, cross-polarize the light and the lens, and retrain the match on features the oil cannot touch. This guide walks the symptom, ranks the likely causes with a one-minute test for each, works the reflection arithmetic for a 25 mm brass part on a flat blue tray, and names the parts that hold the angle once it is set.
Symptom: clean for 30-40 parts, then reflection noise
The field case is common in turning cells. Brass cylinders, 25 mm diameter by 50 mm long, drop off a Swiss-type lathe conveyor onto a tray painted flat blue. A wrist-mounted camera with its own LED locates them reliably at first. Around part 30-40 the reliability decays. The color-difference view still separates part from tray cleanly, yet false positives keep coming no matter how the exposure and focus are trimmed.
The part-count decay is the tell. The camera and the program did not change; the tray did. Every oily part leaves a film on the paint. Matte paint scatters light in all directions, but a smooth oil glaze behaves like a horizontal mirror, and the mirror grows with each cycle. With the camera axis parallel to the tray and the LED firing along it, the glaze returns the light straight back into the lens — the exact geometry that maximizes reflection. The glints are in focus, so focus changes do nothing, and exposure cuts dim the real features along with the glare.
The color view fails for a second reason: a specular highlight is broadband white, so it survives hue filtering and keeps spawning candidates even when the blue-paint-versus-golden-brass separation is otherwise strong.
Likely causes, ranked, with a test for each
Rank the causes and test them in this order:
| What you see | Likely cause | One-minute test | Fix |
|---|---|---|---|
| Glare grows part by part, worst with the LED on | Mirror return off the oil glaze | Switch the LED off, or tilt the camera about 25° by hand | Tilt 20-30° with the light on the bracket |
| Sparkle survives the tilt and the color filters | Polarized specular glare | Rotate a polarizing filter on the lens through 90° | Cross-polarize film and filter |
| Hue separation weak, parts washed out | White light gives too little color split | Compare a white-lit frame with a blue-lit one | 455 nm blue spot |
| Errors track the wall lights or the sun | Ambient leakage into the cell | Shield the tray or strobe the light | Shielding plus strobing |
Causes one and two remove the noise, cause three sharpens the match, and cause four matters in shops with windows or high-bay lighting. Ambient control in a pick-and-place cell is its own subject; see ambient light variation in pick and place vision once the specular work is done.
Fix 1: tilt the camera 20-30° off the specular path
The law of reflection decides where glare goes: light leaves a flat film at the same angle it arrived, measured from the surface normal. A parallel camera with a near-coaxial LED is the 0°-in, 0°-out case, so the mirror return lands inside the lens. Tilt the whole camera-and-bracket assembly 20-30° off the tray normal and keep the spot on the bracket, so the light tilts with the camera. The return then lands at twice the tilt angle from the viewing axis — 40° to 60° — far outside the lens's acceptance cone. The glaze goes dark while the matte tray and the part edges keep their diffuse brightness.
Two practical notes. Re-teach after changing the angle, because a tilted view rotates and foreshortens the part and the old model stops matching. And stay at or below about 30°: at 30° a 25 mm circle reads as 25 mm × cos 30° = 21.7 mm across one axis, and the silhouette begins to distort.
Where the tilt ends up on the bright field / dark field map
Tilt far enough and the outcome is dark field by geometry: the flat glaze mirrors the light away from the lens, so it goes dark, and only edges, chamfers and the cylinder's sides scatter light back. The spot on the bracket now sits more than 45° from the surface plane, the steep direct-lighting regime. The full comparison of the two regimes, and why the law of reflection decides them, is in bright field vs dark field lighting; direct incident light: when it works and fails covers where the directed approach breaks down.
Fix 2: cross-polarize the light and the lens
If sparkle survives the tilt, or the wrist cannot hold an angle, cross-polarize. Specular reflection preserves polarization; diffuse reflection from matte paint and machined brass scrambles it. Put polarizing film over the light and a polarizing filter on the lens with their axes crossed at 90°. The mirror return from the oil is then blocked at the lens while the diffuse image passes, and it works at 0° tilt as well as at 25°. An oil film is a transparent layer, which is the classic case for this technique — crossed polarized lighting for transparent film defect inspection shows it on film, and the cross polarizer on lens and light guide covers film placement and the rotation test.
Budget the light before you commit: crossing a film and an analyzer discards most of the brightness, so restore it with exposure time or strobe current rather than gain. A controller makes that strobe current repeatable part after part; why your vision light needs a lighting controller runs the argument. How many stops the polarizer pair costs is in the FAQ below.
Crossed polarizers need a source with brightness to spare. The LS-SPT-0175 is a 5 W high-luminance spot that keeps headroom after the analyzer takes its share.
- Size
- Under 50 mm
Worked example: tilting a wrist camera 25° off a blue lathe tray
Inputs, and where each comes from:
- Part: brass cylinder, 25 mm diameter × 50 mm long — the field case.
- Tray: flat blue painted steel — the field case.
- Camera tilt: 25° off the tray normal — the middle of the 20-30° fix.
- Light: spot on the camera bracket, axis within 5° of the lens axis — typical bracket mounting.
- Lens: f/5.6 — an example choice deep enough to hold focus across the part height and tray variation; the depth of field check follows below.
Rule, in words: a flat oil film is a mirror, so the return leaves at the arrival angle measured from the normal. With the source on the camera axis, the return lands at twice the tilt angle from the viewing axis. A lens accepts rays only inside a cone of half-angle arctan(1/(2N)) at f-number N.
Arithmetic:
- Lobe offset: 2 × 25° = 50° from the camera axis.
- Lens half-cone at f/5.6: arctan(1 / (2 × 5.6)) = arctan(0.089) = 5.1°.
- Margin: 50° − 5.1° = 44.9°, so 45° of clearance.
- Threshold: at 10° tilt the lobe sits at 2 × 10° = 20°, still 20° − 5.1° = 14.9° clear of the cone.
Result: from 10° of tilt upward, the mirror return misses the lens entirely at f/5.6.
Decision: 10° is the optical minimum at this aperture. Set 20-30° in practice, because the wrist re-seats the camera a few degrees differently between grips and an oil glaze is an imperfect mirror whose lobe is broader than polished glass. Hold below about 30°, where foreshortening flattens the 25 mm silhouette to 21.7 mm and starts costing match quality.
Once the angle and aperture are settled, confirm focus across the part with the depth of field calculator, using the standoff and f-number you chose.
Fix 3: blue light makes the tray bright and the brass dark
With the mirror path broken, sharpen the match with color. Brass reflects most strongly in the red and yellow — that is why it looks golden — while the tray paint is blue and scatters blue light well. A blue spot near 455 nm lights the tray up and leaves the parts comparatively dark, giving a bright-tray, dark-part image with clean polarity for matching and real hue separation for the color-difference view. White light works, but hands you less separation between the two surfaces. The spot lights range covers the sizes and wavelengths; both spots below run on 24 V and are strobed from a controller.
The LS-SPT-0024 is a 455 nm blue spot with a 32° lens and a 49.9 x 66.6 mm lit area.
- Size
- Under 50 mm
- Voltage
- 24 V
For a longer throw, the LS-SPT-0037 pairs a 505 nm cyan LED with a 21° lens and a 99.9 x 141.6 mm lit area.
- Size
- 50-100 mm
- Voltage
- 24 V
Let the part geometry carry the match
Stop asking the oily surface for features. Bores and cross-holes stay dark because light does not enter them, so a through-bore reads as a stable dark ellipse on every part. The machined end face gives a consistent ring, and the silhouette against the brightened tray is the strongest shape available once the parts read dark under blue light. Train the model on those three and put the tolerances on geometry, not on brightness patches of the glaze. A plain cylinder with no bore still offers the end face and the silhouette; no tray sheen ever is a feature.
Hold the angle: mounting that keeps the fix fixed
A wrist camera re-seats with every grip; a fixed station never re-seats at all. If the cell can inspect the tray before picking, put the camera on camera stands with the tilt built in. The LS-MNT-0040 is a professional slide-table stand, 700 mm high, on a 400 x 450 mm base.
- Type
- Camera stands
Under it, a tilt stage sets the angle once. The LS-MNT-0001 slope-angle slide is a manual tilt stage.
- Type
- CCD camera stand and positioning stage (LS-MNT-S01)
Set 20-30°, lock it, wire the spot to the same bracket through a light extension cable, and the geometry stops drifting between shifts. Cables, diffusers and mounts live on the lighting accessories page. To size the whole station — camera, lens, light, stand — start at the stand step of the configurator.
Prevention: make the oil count stop mattering
With the return path broken and polarizers fitted, the image no longer depends on the glaze, so the 30-40 part cliff disappears. Keep two habits anyway. Wipe the tray each shift: oil migrates into gripper pads and fixtures and causes failures that have nothing to do with vision. And if parts arrive dripping, give them drip time or an air wipe upstream, because less oil on the tray also means less oil in the gripper jaws. The broader playbook for oil on inspected surfaces is in machining oil and false inspection rejects: the lighting fix. More lighting geometries for particular surfaces are gathered on the lighting topic page.
Checks before you call it fixed
- At 0° tilt with the LED on, the hot return is visible; at 25° it must vanish from the frame.
- Rotate the analyzer 90° each way from crossed: the sparkle must be darkest at crossed and brightest parallel.
- After any angle change, re-teach and confirm matches at the near, middle and far parts of the tray.
- Fit the polarizers, then restore brightness with exposure time or strobe current — never with gain.
- Run at least 40 parts between wipes; the match rate must stay flat past the old failure point.
- In the color-difference view, confirm tray and part separate with no white sparkle left.
The decision for each case
- Wrist camera that must stay on the arm: tilt the mount 20-30°, keep the spot on the bracket, cross-polarize film and filter, light blue at 455 nm, retrain on bore, end face and silhouette.
- Fixed inspection over the tray: the same recipe on a stand with a locked tilt stage, so nothing re-seats between shifts.
- Sparkle that survives both tilt and polarizers: move the spot off-axis toward dark field so only part edges return light, and let the flat tray read dark.
If you want the parts list checked against your tray and standoff, send the layout through the lighting step of the configurator or reach us on the contact page.
Questions
How much brightness does cross-polarization cost?
An ideal polarizer passes half of unpolarized light, so the film on the light plus the filter on the lens pass at most a quarter between them, and real films pass less. Plan on losing roughly two stops. Buy the brightness back with exposure time, higher strobe current or a brighter spot — never with sensor gain, which lifts noise along with the signal.
Why did reliability decay after 30-40 parts instead of from the first one?
The first parts landed on matte paint, which scatters light in all directions, so each film was too thin and too broken to act as a mirror. Around 30-40 cycles the overlapping films merged into a continuous glaze with a real specular component, and the return path into a parallel lens opened up. Once the camera tilts off that path, the part count stops predicting anything.
Can I keep the camera parallel and just add polarizers?
Often, yes — crossed polarizers remove most of the mirror return at any angle. But parallel incidence sends the maximum energy into the return path, so the film works hardest exactly when the geometry is worst, and residual sparkle concentrates in the image center where the part sits. We recommend doing both: tilt first, because it is free, then polarize for what remains.
Do I need to re-teach the vision model after tilting the camera?
Every time the angle changes, re-teach. A 20-30° tilt rotates and foreshortens the part in the image — a 25 mm circle at 30° reads as 25 mm × cos 30°, about 21.7 mm, along one axis — so features trained at 0° no longer line up. Confirm the new model still matches after the wrist re-seats the camera a few degrees off the taught pose.
Does blue light hurt color recognition of the brass?
No — it helps. Blue illumination brightens the blue tray paint and darkens the brass, because brass reflects blue more weakly than the red and yellow that give it its color. The hue distance between part and tray grows, which is exactly what a color-difference view needs. Keep a white source in the cell only if you also read ink colors or anodizing on other parts.
When is a fixed camera better than a wrist camera here?
When the wrist re-seats the camera with more than a few degrees of variation, or when inspection time is stealing cycle time. A fixed stand holds one angle and one working distance forever, which makes the teach stable and removes the re-seat variable entirely. Keep the wrist camera for the pick itself and inspect on the tray with a mounted camera first.