Inspection troubleshooting

The light source of surface defects: detection requirements and lighting effects

A dark-field inspection cell only shows what the light source of surface defects is aimed at. This page is for engineers and quality leads who already have a camera and lens, but keep missing scratches, chips and digs on cover glass. It maps each visible symptom to a cause and a fix you can apply at the bench.

Get Instant Quote Talk to an Engineer

Dark-field and bright-fieldAngle 15–85°Ra 0.2–0.8 μm reference

part-surface-finishing-services2 with light source of surface defects setup

On this page

1
Symptom, cause, and what to change
2
Why the light source of surface defects decides the result
3
Choosing light angle, distance and aperture for glass inspection
4
Symptoms that point to the light source, not the camera
5
Step by step: setting up the light source for glass defect detection

Symptom to cause map

Symptom, cause, and what to change

Three columns line up one to one. Fix the cause, not the camera gain.

Symptom on screen	Likely cause	What to change
Fine scratches disappear	Glare sits on the glass surface	Move light to 15–30° from surface
Defect visible, edge blurry	Aperture too wide at low angle	Stop down 2 stops, re-focus
Bright band across frame	Reflection of a ceiling fixture	Shield the cell or use a hood
Same defect, two brightnesses	Illuminance drifts over the shift	Log lux every 2 hours, replace LED
Dust looks like a chip	Particles lit at the same angle	Add a laminar air flow above the stage
Cracks vanish at 45°	Light enters the crack, no scatter	Add a second light at 70–85°
Color shifts after 6 months	LED bin and phosphor aging	Use same-bin LEDs, re-calibrate
Shadow under the edge	Light too far off axis	Move to near-coaxial with diffuser

Fix the geometry before you buy a better camera

If defects are escaping, the light source of surface defects is the first thing to audit, not the sensor. Set the angle, even the field, log the output, and re-check a known sample each shift.

Get Instant Quote Talk to an Engineer

What the light must do

Why the light source of surface defects decides the result

Cover glass is transparent. A scratch, a dig or a micro-crack does not block light. It scatters it. That is the whole problem. If your light arrives at the wrong angle, the camera sees a clean pane and a bright reflection of your own fixture. The defect is still there, just not sending photons toward the lens.

The light source of surface defects has to do three jobs at once. It must put enough energy on the surface to raise the scatter signal above sensor noise. It must keep the specular reflection out of the lens. And it must stay stable across a shift, because a 10 percent drop in output moves a marginal defect below the detection threshold.

In practice, dark-field lighting does most of the work on polished glass. The source sits at a low angle, often 15–30° from the surface plane, so the smooth areas do not reflect into the camera. Only edges, scratches and particles scatter light upward. The result is a dark background with bright defects, which is a much easier image to threshold.

Bright-field lighting still has a place. It shows print, coating coverage and edge chips on the outline, where you want the geometry rather than the scatter. Most production cells run both and compare the two images, because a defect that hides in one geometry usually shows in the other.

1
Scatter, not shadowGlass defects are visible through scattered light, not blockage.
2
Angle beats intensityA 20° shift changes contrast more than doubling the current.
3
Stability is a specOutput drift is a defect escape path, not a maintenance detail.
4
Two geometriesDark-field for scatter, bright-field for geometry and print.

Angle and geometry

Choosing light angle, distance and aperture for glass inspection

Set the angle first, then the distance, then the aperture. Start at 20° from the glass plane with the source about 150–250 mm from the part. At that geometry a 20 μm scratch produces a bright line against a dark field on most cover glass. If you see nothing, walk the angle up in 5° steps until the defect appears, then back off 2° to keep the background clean.

Distance controls uniformity, not brightness. A bar light at 100 mm gives a hot band in the middle and falls off at the edges. Move it to 200 mm and add a diffuser, and the illuminance across a 100 mm glass panel typically evens out within 15 percent. Uneven light creates uneven thresholds, and uneven thresholds create escapes at the panel edge.

Aperture matters more at low angles than most people expect. A wide aperture accepts light from many directions, including stray reflections, and softens the defect edge. Stopping down two stops from wide open usually sharpens the scatter line enough to measure its length. The cost is exposure time, so budget for a longer integration or a brighter source.

Do not chase resolution before geometry. A 5 MP camera with correct dark-field angle will out-detect a 20 MP camera pointed straight down at a glossy pane. Once the angle is right, then spend on pixels.

1
Start at 20°Walk in 5° steps, then back off 2° for a clean background.
2
200 mm plus diffuserAim for illuminance variation under 15 percent across the panel.
3
Stop down two stopsSharper scatter edges, longer exposure, less stray light.
4
Geometry before pixelsCorrect angle beats a higher-resolution sensor.

When it goes wrong

Symptoms that point to the light source, not the camera

If defects are missing at the start of a shift and appear later, the light is drifting. LED output falls as the junction heats, and a cell that has no thermal path can lose 10–15 percent in the first hour. Log the illuminance at the stage every two hours for a week. A curve that sags and never recovers means the fixture needs a heat sink or a constant-current driver.

If the same defect reads as two different sizes depending on where it sits in the frame, the problem is uniformity. The center of the beam is brighter, so the scatter is stronger and the threshold catches more of the scratch. Flat-field correction helps, but fixing the geometry is the real answer.

If dust and chips look identical, both are being lit at the same angle. Dust sits on top of the surface and scatters from a different height than a sub-surface crack. Adding a low-angle source with a shallow depth of field separates the two, because the particle is in focus while the crack is not.

If cracks disappear when the part is rotated, you have only one light direction. A single source misses cracks that run parallel to it. Two sources at 90° to each other, or a ring plus a bar, closes that gap.

1
Drift looks like escapesDefects appear mid-shift when output sags.
2
Center-to-edge variationBrighter beam center changes measured defect size.
3
Dust versus subsurfaceDifferent scatter height, so use depth of field to split them.
4
One direction misses cracksCracks parallel to the source stay invisible.

Bench procedure

Step by step: setting up the light source for glass defect detection

Run this in order. Skipping step 2 is the most common reason a cell never stabilizes.

1
Mount a reference sampleUse a glass panel with known defects: a 20 μm scratch, a 50 μm chip and a sub-surface crack. Keep it in a case, clean it with isopropyl alcohol before each session.
2
Set the dark-field angle at 20°Place the bar light 150–250 mm from the part, 20° from the glass plane. Confirm the background reads near black with no specular band.
3
Walk the angle in 5° stepsRaise the angle until the reference scratch reaches maximum contrast. Back off 2° so the background stays clean. Record the number on the fixture.
4
Even the field with a diffuserMove the source to about 200 mm and add a diffuser. Measure illuminance at five points across a 100 mm panel; keep the spread under 15 percent.
5
Stop down two stopsClose the aperture two stops from wide open, then re-focus on the scratch. Extend exposure time to hold the same gray level.
6
Add a second direction at 90°Mount a second source at 70–85° from the surface, rotated 90° in plan view. This catches cracks that run parallel to the first light.
7
Log illuminance every 2 hoursRecord lux at the stage through a full shift. If the reading falls more than 10 percent, add cooling or a constant-current driver.
8
Re-check the reference sampleRun the known panel at the start and end of each shift. If contrast drops, the light has moved or aged, not the camera.

FAQs

Questions engineers ask about inspection lighting

Can a ring light replace a bar light for cover glass?

A ring light is convenient but it sends light from many directions at once, which raises the background and hides low-angle scatter. On polished glass it usually washes out fine scratches.

Use a ring only for edge and outline checks. Keep a bar or a line light at 15–30° for the surface scatter work, and run them as separate images rather than one combined exposure.

What illuminance do I actually need at the glass surface?

There is no single number, because the requirement depends on camera quantum efficiency, exposure time and the defect size you must catch. What matters is the signal-to-noise ratio at the defect, not the lux value on the datasheet.

A practical approach: raise the level until the reference scratch sits at least 30 percent above the background gray level, then log that value and treat it as your control limit.

Why does my light look fine on a white sample but fail on black glass?

Black or heavily coated glass absorbs much of the scattered light, so the same geometry produces a weaker signal. The background stays dark, which helps, but the defect also dims.

Increase exposure or source output for coated panels and re-check the reference sample. Do not change the angle to compensate, or you will lose the dark-field separation you already tuned.

How often should the light be recalibrated?

Check the reference sample at the start and end of every shift. That catches mechanical drift and gross aging without any instruments.

Full re-calibration of angle and uniformity is worth doing monthly, or after any fixture change, cleaning or lamp replacement. Keep the recorded angle and lux values next to the cell so the next engineer does not start from zero.

Does the light need to match the camera sensor?

Color cameras and monochrome cameras respond differently across the spectrum. A monochrome sensor with a narrow-band source usually gives the best contrast on glass, because it removes color noise from the image.

If you use a color camera for other checks, keep the light white and consistent in bin. Mixing LED bins in one fixture creates color shading that looks like a surface defect.

What about ambient light from the room?

Room light is uncontrolled light. It arrives at angles you did not choose, so it lifts the background and can create false bright spots that look like chips.

Enclose the cell or fit a hood around the camera and lights. Measure the background with the inspection light off; if the frame is not near black, the room is contributing and needs to be blocked.

Send us the part and the inspection problem

We machine and finish cover glass fixtures, inspection housings and light mounts to ±0.005 mm, with 100% inspection before shipment and reports on request.

12-hour quote100% inspectionNDA on request

Get Instant Quote Talk to an Engineer

Follow our work