Machining oil and false inspection rejects: the lighting fix

A stamped part comes out of the press wet with machining oil, and the inline camera that passed clean samples starts rejecting good parts by the dozen. The usual reaction, a blast of compressed air in front of the camera, trades rejects for a cell plated in oil mist and a line stopping anyway. You can break that loop in an afternoon: confirm the oil is the variable, re-aim the light so the oiled surface reads dark while real defects still scatter light back, and only then deal with the film that remains, captured rather than sprayed. This page gives the reflection geometry, one worked check you can run with a calculator, and an order of attack for a full-inspection vision station on a press line.

Why an oil film reads as a defect

Bright-field lighting, a ring light at 90° elevation or a coaxial column, sends its light down nearly the axis the camera looks along. Bare pressed sheet scatters a little of that light back, so the field reads mid grey and your thresholds sit on that grey. A film of machining oil fills the micro-roughness of the sheet and behaves like a liquid mirror: the law of reflection sends the on-axis light straight back into the lens, and oiled patches saturate white.

The mechanism is the one described in bright field vs dark field lighting, and it produces two failure modes. A uniform sheen lifts every grey level, so size and contrast checks drift through the shift. Drips and droplets create bright blobs that wander from part to part and trip stain and blob detectors.

Air makes it worse, not better. Blow-off rewrites the oil pattern on every stroke, so the background changes each cycle, and an inspection tuned on wiped samples sees a new part every time. That is why retuning never sticks: the geometry is wrong, not the threshold. The general playbook for oily parts holds across industries, as machine vision on oily surfaces lays out; the rest of this page is the stamped-metal version with numbers.

Confirm the oil is the variable first

Three checks, none of which needs new hardware:

1. Run ten good parts as they come off the press, wipe the film off five, and run those again. If the rejects follow the film rather than the part, the optics are innocent.
2. Mark where the reject pixels sit. Specular patches that move between parts are oil. Dark marks fixed to the same coordinates are real features or a failing light.
3. Image one clean reference part at shift start and compare its histogram with the day you tuned. LED output drifts with age, so a fading light can masquerade as an oil problem.

Every false reject also costs more than the part: an operator pull, a re-check and a dent in OEE quality-loss accounting.

The fix: take the light off the axis

Dark field puts the light in at a grazing elevation and lets geometry do the filtering. On a flat, oiled facet the rays mirror across at the same shallow angle and leave the field, so the camera reads near black. Anything that breaks the flatness, a dent, a scratch, a burr, an edge, a wrinkle, tips the local normal and scatters grazing light back up as a bright feature on black. That is the same law that powers low-angle lighting for dents and pits.

The lens only accepts rays within a few degrees of its axis, so a specular lobe leaving 65° or more off-axis never enters the aperture, however shiny the film. Specular bare metal fails bright field the same way; the copper surface lighting guide shows the identical fix on a different part.

To run it, mount the ring with its inner aperture around the lens and emit at 0-15° elevation, the low-angle class in the light finder. You do not have to guess the angle, because one family covers grazing to steep emission, so you can A/B on the line before you commit:

The LS-RNG-0730 pairs a 103 mm outer diameter with a 48 mm inner aperture, runs on 24 V and offers emission from low-angle (0-15°) through 85°, so one housing serves the dark-field test now and a steeper geometry later if a second station needs it.

Worked example: a 15° grazing ring on a stamped blank with ±5° local tilt

Inputs. The camera sits on the surface normal at 0°, the standard overhead set-up. The ring emits at 15° elevation, the top of the low-angle class in the finder, so the incidence angle from the normal is α = 90° − 15° = 75°. The worst local tilt is Δ = ±5°, assumed here as typical for stamped sheet; measure your own worst warp with a straightedge or your flatness data, because that number, not the oil, sets the limit.

Rule. The law of reflection returns each ray at the incident angle on the far side of the local normal, and a facet tilted by Δ steers the reflected ray by 2Δ.

Arithmetic. Flat facet: the lobe leaves at 75° on the far side, 75° away from the camera axis. Facet tipped 5° toward the camera: 75° − 2 × 5° = 65° off-axis. Tipped 5° the other way: 75° + 10° = 85°. The lobe centre crosses the lens only when 75° − 2Δ = 0, which is Δ = 75° / 2 = 37.5°.

Result. At 15° elevation the specular return stays at least 65° off-axis across the whole ±5° warp, and your 5° worst tilt sits 7.5× inside the 37.5° failure point (37.5 / 5 = 7.5).

Decision. Apply the rule α / 2 ≥ 2Δ using your measured tilt. With α = 75° the geometry covers tilt up to ±37.5°, far beyond stamping warp, so pick the elevation for coverage first, then confirm the rule. If a facet can present more than half the incidence angle, a deep-drawn pool of oil aimed at the lens, no fixed geometry saves that part: re-orient it in the fixture or degrease upstream.

Print or codes on the oiled face: dome, then polarizer

Grazing light is the wrong tool when the target is a code or print sitting on the film. Matte ink barely scatters grazing light, so characters fade into the black background. You want even diffuse light, the job of a dome light, which wraps the part in light from nearly every direction and removes the single hotspot; the technique is laid out in dome lights for shiny, curved and wrinkled parts.

A mirror-flat oiled face under a dome keeps one artefact: the dome's own camera hole mirrors back as a soft dark patch. Tilt the part a few degrees or place the read zone so that patch falls outside the code, the same positioning logic used in reading QR codes on reflective and curved surfaces.

If sheen still washes the mark out, add a crossed polarizer, which removes most of the once-reflected glare; set-up and tests are in polarized lighting for screen inspection. The same dome-plus-polarizer combination suits small shiny contacts, as lighting shiny metal pins shows.

A mid-size dome covers a bracket-sized field:

The LS-DOM-0085 gives a 116 mm emitting window inside a 166 mm shell at 24 V, so it suits fields up to roughly that width.

Which geometry for which symptom

For small bracket fields, three 24 V rings emit in the low-angle (0-15°) class:

The LS-RNG-0426 and LS-RNG-0517 are 120 mm diffused rings at 24 V; the LS-RNG-0256 is a 96 mm standard-type ring for tighter standoffs. If one station must catch several defect types, stack geometries rather than compromising one; machine vision lighting selection by inspection task sorts that decision.

Capture the mist instead of spraying it

Fix the light first; blow-off is the last resort, not the first move. When a film genuinely must be thinned, contain it with a push-pull set-up: a low-pressure air knife or a plain fan on one side of the part pushes the stripped film toward an extraction hood on the other, and the hood feeds a mist collector, so the aerosol ends in a filter rather than on your lens. Keep the hood close to the part, pull slightly more air than you push so stray mist drifts to the filter, drain the collected oil to a sump, and service the filter on an interval set by its pressure gauge. Trigger the camera after the burst, because air on thin sheet makes it flutter and flutter reads as blur.

Keep the film off the optics

A transparent film you cannot yet see already costs contrast. Wipe the ring or dome window and the lens on a fixed schedule, and keep the shift-start grey-level check on the clean reference part: five minutes catches film build-up, LED ageing and a knocked bracket before they become rejects.

Debug order from rejects to a stable image

1. Wipe test: confirm the oil is the variable.
2. Measure the worst local tilt across a sample of parts.
3. Pick geometry by target: surface defects, grazing dark field; print or codes, dome plus polarizer if sheen remains; oil pattern itself, bright field.
4. Check the geometry holds: α / 2 ≥ 2Δ with your measured tilt.
5. Bench-run twenty parts as they come, and set thresholds on the oiled histogram, never on wiped samples.
6. Add blow-off only for what the image still shows, and contain it push-pull.
7. Schedule the optics wipe and the shift-start reference check.

When the usual fix is the wrong one

Dark field will surface die marks, handling scratches and edge burr you never classified. That is the technique working, not failing: budget a session to re-train the defect classes, and use lighting for metal surface defect inspection for the fault list. Two cases need a different answer. A deep-drawn pool that faces the lens is a liquid mirror aimed at the camera; no ring fixes it, so re-orient or degrease upstream. And when the oil is the product, a rust-prevention film whose pattern you must confirm, stay in bright field, where oiled patches read dark against scattering bare metal.

Fit it on your line

Run the wipe test, measure the tilt, then bench the grazing option before you commit the station. When the image holds, build the rest of the chain, controller, trigger and stand, in the light configurator, or bring the part story to the team through the contact page. More worked lighting cases sit on the lighting guides topic page.