Machine vision oily surfaces: improve inspection reliability

Oil on a conveyor can change reflections, transfer dirt, or move the apparent edge of a stamped part. These effects need different remedies, even when they produce the same reject signal. This troubleshooting sequence helps you separate optical errors from contamination and handling problems. You will build controlled image tests, compare lighting arrangements, calculate a motion limit, and decide when the inspection needs a process change rather than another threshold adjustment.

1. Identify what fails before changing the lighting

Start with the inspection decision: presence, outline, hole position, surface damage, or printed identification. An arrangement that strengthens an outside edge may hide a shallow surface defect. If the inspection task is still unclear, use the lighting selection guide by inspection task before choosing hardware.

Save original images alongside the decision, trigger identifier, exposure, gain, and relevant measurements. Record whether each failure follows a particular part, belt location, or time since cleaning.

The following order is a practical starting point, not a universal ranking of causes.

Do not use the processed mask alone for diagnosis. A thresholded image can make glare, contamination, and blur look nearly identical.

2. Separate belt contamination from oil on the part

Use representative accepted parts and confirmed defective parts. Keep their identities traceable throughout the test. Stop and secure equipment before positioning samples or cleaning, following the machine's documented isolation procedure.

Build a small comparison matrix:

1. Image a reference part on a clean, stable test background.
2. Image the same part on the normal belt at several belt locations.
3. Image a representative oily part on the clean test background.
4. Image that oily part on the normal belt.

Keep camera position, focus, aperture, exposure, gain, and lighting unchanged. Capture the empty background separately. Repeat the comparison using the minimum, typical, and highest oil conditions that production permits.

If the problem follows belt position, investigate background contamination or local belt appearance. If it follows the part onto the clean background, concentrate on the part's reflections, deposits, or actual surface condition. If only moving images fail, examine exposure and transport stability next.

Cleaning a sample is a diagnostic intervention, not proof that cleaning belongs in production. Removing oil can also remove debris or alter the surface appearance. Record exactly what changed, and use only a cleaning method approved for the part and its downstream process.

3. Move reflections away from the feature

Oil changes the surface's optical behavior. A smooth film can produce a strong directional reflection, while droplets introduce local curved surfaces and moving highlights. Increasing light intensity may simply saturate those highlights sooner.

Begin with an adjustable bar light. Hold the camera and part fixed, then move the source through several positions. For a flat sample, testing illumination at approximately 15°, 30°, and 60° above the surface provides a useful starting sweep. These are trial angles, not product operating specifications.

At each position, inspect both accepted and defective parts. Watch the actual measurement boundary, not just the overall image brightness. The lighting angle guide explains why a source can reveal a feature from one direction and conceal it from another.

Low-angle illumination can make scratches, raised contamination, and edges scatter light toward the camera while a smooth background stays darker. It can also make harmless oil droplets appear highly prominent.

A standard-source candidate for directional testing is the LS-BAR-1047. Confirm the required coverage and mounting geometry before selection.

For a more diffuse starting point, consider the LS-BAR-1168. Its diffuser changes the source appearance, but does not guarantee removal of every reflection.

Compare options in the lighting finder after recording the useful angles. Source dimensions alone do not establish illuminated coverage at your working distance.

4. Compare diffuse light and polarization without hiding defects

A larger apparent source can soften sharp highlights, but a broad bright reflection may still cover the inspection region. Evaluate diffusion by measuring feature separation, not by judging whether the image looks more even.

Cross polarization is another controlled trial. A polarizer on the illumination and an analyzer at the lens can suppress some surface reflections when their orientations are crossed. The cross-polarization guide covers the arrangement and its limitations.

An oily metal part is not a simple nonmetallic surface. Reflections from the oil and underlying metal can behave differently. Expect partial improvement rather than assuming complete glare suppression. Polarizers also reduce available light, which can create an exposure problem on a moving line.

If two illumination directions are needed, evaluate them separately before combining them. Simultaneous opposing lights can fill shadows that were revealing a defect.

5. Check sampling, focus, and motion independently

Once reflection control is promising, verify that the camera and optics preserve the required detail. Browse area scan cameras by shutter type and acquisition requirements, rather than resolution alone.

The MV-CU020-90GC is a global-shutter color camera candidate. Confirm its detailed sampling, triggering, and timing capabilities against the inspection requirements.

A global shutter avoids the row-dependent timing distortion associated with rolling readout, but it does not eliminate motion blur during exposure. The camera finder helps narrow the hardware once those requirements are known.

Calculate an illustrative sampling and exposure budget

The following example is independent of the featured products. Assume a horizontal field of view of 120 mm represented by 2400 image pixels. The smallest target feature is 0.20 mm wide, and conveyor speed is 800 mm/s.

The arithmetic is:

• Sampling: 120 mm / 2400 pixels = 0.05 mm/pixel.
• Feature width: 0.20 mm / 0.05 mm/pixel = 4 pixels.
• For a provisional one-pixel blur allowance: 1 pixel × 0.05 mm/pixel / 800 mm/s = 62.5 microseconds.

Four pixels across a feature is an initial sampling target, not a detection guarantee. Low contrast, poor focus, and an unstable boundary can require more sampling or a different optical arrangement. The resolution and accuracy guide for moving inspection explains why pixel count alone is insufficient.

Use the fastest credible part speed, including slip or acceleration where relevant. If a pulsed light defines the effective exposure, measure usable optical pulse duration and check that ambient light does not add a blurred image. Follow the strobed ambient-light rejection procedure rather than assuming every short pulse freezes motion.

Verify lens coverage and focus

Choose machine vision lenses for sensor coverage, working distance, field of view, and required detail. Check the entire part-height range, not only the nominal top surface.

The LS-FFL-0679 is a C-mount lens candidate. Confirm its detailed optical and mechanical suitability for the selected camera and installation.

Use the lens finder to compare candidates. Do not assume these example parts constitute a validated assembly. Confirm focus range, image-circle coverage, and clearance before ordering.

6. Stabilize acquisition before setting acceptance limits

For controlled testing, disable automatic exposure and gain where the camera permits. Fix aperture and lighting output too. An automatic loop may react to a bright oil patch by darkening the rest of the image, creating a second failure elsewhere.

Check for clipped highlights in the actual inspection region. In an 8-bit image, a saturated value of 255 cannot distinguish moderately bright reflections from much stronger ones. Lower exposure or change the lighting geometry until the feature remains measurable without clipping.

Track a simple feature-separation metric alongside the inspection decision. For a bright feature against a darker local background, a useful normalized measure is:

If representative intensities are 80 and 50, then C = 30 / 130, approximately 0.23. If oil raises the background to 70 while the feature remains at 80, C = 10 / 150, approximately 0.067. Overall brightness has increased, but useful separation has fallen.

Treat this metric as a diagnostic aid, not a universal pass limit. Use fixed measurement regions, avoid saturated pixels, and inspect local variation. For edge measurements, also record edge displacement and repeatability; average intensity can remain stable while the apparent boundary moves.

Save original images throughout commissioning. Apply identical processing to each lighting trial so that software changes do not conceal optical differences.

7. Recognize when optics cannot solve the problem

Lighting is not a substitute for an unobstructed view. An opaque deposit can cover the defect completely. Oil can also move debris, change part seating, or alter the visible boundary enough that the original measurement definition no longer holds.

Escalate to the process owner when:

• The target feature is physically hidden under contamination.
• Part position or height changes because of oil accumulation.
• Background contamination becomes indistinguishable from a real target.
• The oil condition exceeds the range represented in validation samples.
• Different acceptable oil states require contradictory acceptance thresholds.

Belt cleaning, oil application control, part separation, or an approved presentation change may then be necessary. Specify the required outcome, such as a visible edge and a validated contamination range, rather than prescribing a chemical or conveyor mechanism.

Any cleaning agent or equipment modification needs approval for material compatibility, lubricant requirements, downstream processing, worker exposure, and waste handling. Follow applicable safety documentation and retain machine safeguards. Optical commissioning should not depend on improvised cleaning during operation.

8. Verify the fix and order against measured requirements

Before release, run known accepted and defective samples across the permitted oil range, belt positions, orientations, and speeds. Include start-up and the longest normal interval between cleaning events. Keep development samples separate from the final validation set.

Record false rejects and missed defects separately. Zero misses in 200 defective trials, for example, does not demonstrate zero risk. Under independent, representative trials, the approximate rule-of-three upper bound is 3 / 200 = 1.5% at roughly 95% confidence. Correlated repeat images of one defect provide less evidence than varied, representative samples.

Use this final checklist:

• Confirm the defect remains visible under every accepted oil condition.
• Record illumination angle, distance, output setting, and any polarizer orientation so the setup is reproducible.
• Verify feature sampling and motion blur at the fastest allowed speed.
• Check focus and contrast across the full inspection region and part-height range.
• Confirm shutter operation, trigger timing, image identity, and reject tracking.
• Verify electrical compatibility and permitted continuous or pulsed operation from the selected hardware documentation.
• Define what happens when glare or contamination makes the image uninspectable; do not silently classify it as acceptable.
• Establish a reference image check after cleaning, maintenance, or lighting replacement.

Start the lighting configurator once the successful geometry and exposure budget are documented. If the trials remain ambiguous, send representative original images, defect dimensions, field of view, speed, and tested oil conditions through applications support before selecting parts.