Fiber End-Face Inspection Lighting: White vs Yellow vs Blue Illumination Selection

This article is also available in 简体中文.

A fiber end face is judged by a camera, so the wavelength you choose decides which defects can build enough grayscale contrast to be seen at all. The direct answer: use white, or a narrowband yellow band, for geometry and edge work (core/cladding boundary, chamfer, cleave angle, coating edge), and blue 470 nm for fine scratches, cracks and particle contamination, because a shorter wavelength lowers the diffraction limit and scatters more strongly from small features. Yellow, however, is not a POMEAS standard coaxial light-source wavelength — see section 7.

1. Quick answer: which wavelength for a fiber end face?

• Geometry, edge, angle, profile → white 6500K, or a narrowband amber/yellow band (industry practice), coaxial bright field.
• Fine scratches, cracks, chips, dust → blue 470 nm with oblique or dark-field illumination.
• Core geometry through the glass → red 625 nm or IR 850–950 nm, where glass transmission is high.
• A photosensitive adhesive or coating in the field → amber/yellow, to avoid short-wavelength exposure.
• Always: changing the illumination angle usually changes contrast more than changing the wavelength. Wavelength is the second lever, not the first.

2. What actually has to be inspected on a fiber end face

A "fiber end face" is not one surface — it is a stack of concentric regions plus a machined geometry, each with its own dominant contrast mechanism. POMEAS does not define your acceptance thresholds: the pass/fail limit for each item must come from your own drawing, process specification or customer standard, and it must be fixed before a defect can be classified reliably.

Two consequences follow: a single wavelength rarely serves all five rows, because geometry wants shape rendering while scratches and particles want scatter collection; and a dark spot should not be called a defect until it is checked against the physical sample and the agreed standard.

3. How wavelength changes end-face imaging

Three mechanisms explain nearly all of the difference between white, yellow and blue.

Diffraction limit. The smallest detail a lens can resolve on the object side is set by its numerical aperture (NA) and the wavelength: d ≈ 0.61 · λ / NA. Shorter wavelength means a smaller d at the same NA. For a 10× bi-telecentric lens with object-side NA 0.18 the limit calculated from this formula is ≈ 1.59 µm at blue 470 nm, ≈ 1.78 µm at green 525 nm, ≈ 1.86 µm at 550 nm, ≈ 1.98 µm at yellow 585 nm and ≈ 2.12 µm at red 625 nm (all calculated, not measured). This matches the published 1.86 µm object-side resolution of the POMEAS LTCM10-65C.

Scattering. Scratches, cracks and particles do not absorb light, they scatter it, and for particles small compared with the wavelength the scattered intensity rises steeply as the wavelength falls (approximately as 1/λ⁴). The same scratch that is faint under red light can therefore become clearly visible under blue 470 nm: the wavelength has not changed the scratch, only the amount of scattered light reaching the lens. A longer wavelength is correspondingly more tolerant of surface haze and slight defocus, at the cost of a coarser diffraction limit.

Chromatic aberration. White light is a broad band, and a lens focuses its blue end at a slightly different plane than its red end; this axial color softens the edge at high NA — exactly the condition under which end faces are inspected. A narrowband source focuses to a tighter edge, which is a real and often overlooked reason a yellow band is specified in industry, separate from any material or exposure argument.

One thing wavelength barely changes: the end-face return itself. Fused silica (refractive index near 1.46) reflects roughly 3.5% per surface, and that value varies little across the visible band, so switching wavelength changes the resolution limit, the scatter contrast and the chromatic behavior — not how much light the glass returns.

4. White, yellow and blue contrast by defect type

This table is the core of the decision. White 6500K and blue 470 nm are POMEAS standard coaxial wavelengths; the yellow band (≈ 580–590 nm) is an industry practice (see section 7).

Read it as "which mechanism do I need to switch on", not "which color is best". Geometry wants white or amber; a 2 µm scratch wants blue plus dark field; a photosensitive adhesive removes the short-wavelength options before contrast is considered.

5. Matching illumination geometry to defect type

Wavelength and geometry are chosen together: a defect that scatters light needs the camera to collect scattered rather than reflected light.

6. Illumination and lens coordination at 10× magnification

High magnification re-weights the three mechanisms above, so the illumination must be coordinated with the lens rather than picked independently. The reference configuration is a 10× bi-telecentric lens with a coaxial illumination port; its parameters come from the existing POMEAS fiber end-face knowledge page and should be cross-checked against the product page before quotation.

Field of view follows the magnification directly: FOV = sensor size / β. A 1.1″ sensor (14.19 × 10.3 mm) at 10× gives an object-side field of about 1.42 × 1.03 mm — the scale at which a single end face or a small fiber array fits. Choose the sensor for the field you need before the wavelength, because magnification fixes the pixel scale.

Four coordination points matter at 10×:

• NA sets the diffraction floor. At NA 0.18 the object-side limit moves from about 1.59 µm (blue 470 nm) to about 2.12 µm (red 625 nm), so near that floor wavelength decides whether a feature is seen at all.
• High NA worsens axial color. With a 0.02 mm depth of field, chromatic edge blur is easily mistaken for focus error, so a narrowband source is more valuable at 10× than at 1×.
• A coaxial port ties the source to the lens. Wavelength is chosen at the light source, so changing it means changing the source (or a filter, if a slot exists), not a camera setting.
• Depth of field dominates mechanics. At 0.02 mm, sample tilt or focus drift degrades the image long before the wavelength does.

For lower magnification with an on-axis channel the same logic applies with a larger field and depth of field — for example the 2.0× coaxial telecentric lens (WD 110 mm, 2/3″), the 1.5× coaxial telecentric lens (WD 110 mm), and the 2.0× / 65 mm working-distance coaxial telecentric lens.

7. Where yellow light fits — and what POMEAS can supply

A yellow band of roughly 580–590 nm sits between the standard green 525 nm and red 625 nm. Industry specifies it for four reasons, each a different argument:

• Photosensitive materials. Photoresist and some UV-cured adhesives respond to short-wavelength visible light as well as to UV, so an amber band removes that exposure while still giving enough light to see and align the part.
• Narrowband sharpness. A 10 nm-class band focuses to a tighter edge than white, for the chromatic reason above.
• Contrast on certain residues. Some cured adhesives and polish residues absorb more in amber than in blue, so the residue darkens against the glass.
• Visual comfort. Human photopic response peaks near 555 nm, so at similar power an amber band looks brighter to an operator aligning the sample.

The trade-off is the diffraction limit: at NA 0.18 a 585 nm band resolves to about 1.98 µm against about 1.59 µm for blue, so yellow does not help find the smallest scratches. Choose yellow for material safety and edge sharpness; choose blue when scatter contrast on the finest defects matters most.

POMEAS supply position (to be confirmed per model). The POMEAS coaxial light-source family lists red 625 nm, white 6500K and blue 470 nm as standard wavelengths, with green 525 nm, infrared 850–950 nm and ultraviolet 365–395 nm optional. Yellow is not listed as either a standard or an optional wavelength. The product page states that colour customisation is supported, so a yellow requirement must be confirmed per model and order — treat it as a custom request (to be confirmed) and do not assume a standard yellow source is available.

8. Selection workflow

1. Fix the defect standard and the smallest feature. Without both numbers, neither lens nor wavelength can be justified.
2. Classify the task: geometry/edge, scatter-type defect, or through-glass geometry.
3. Set magnification and NA from the resolution requirement with d ≈ 0.61·λ/NA (shortest candidate wavelength as the optimistic case), then confirm the field with FOV = sensor / β.
4. Pick the first wavelength by task family: geometry → white; fine scatter defect → blue; through-glass → red or IR; photosensitive material → amber/yellow (custom).
5. Pick the matching geometry: coaxial for specular geometry, oblique/dark field for scatter defects, ring for outer-edge and coating checks.
6. Verify on the real sample at a fixed exposure policy; adjust the illumination angle first, the wavelength second.
7. Confirm supply before design freeze: for yellow, confirm the custom option per model (to be confirmed).

9. Send the sample and the defect standard

If you already have a fiber type, an end-face image and a defect specification, send those three things and we will confirm the magnification, the illumination geometry and a candidate wavelength. Background reading: the telecentric lens selection guide, the existing end-face pages — high-magnification fiber end-face inspection with a 10× telecentric lens and wedge lensed fiber inspection and end-face angle measurement — plus telecentric lenses for dimensional measurement and, for microscope comparisons, microscope 400× versus machine-vision magnification. Share your fiber type, array dimensions, field of view and defect standard via the contact page.

Frequently Asked Questions

Can a white-light coaxial light inspect a fiber end face, or do I need a colored light?

White light is the correct first test for geometry — end-face angle, chamfer, core/cladding step and coating edge — because it renders shape with rich grayscale and shows everything in the field. It is often not the best light for the smallest scratches and particles, because a broad spectrum blurs the edge through axial color and does not concentrate energy at the short-wavelength end where small features scatter most. A practical workflow is to set the geometry and focus with white light, then test blue 470 nm for the defect set and keep whichever wavelength gives the more stable contrast on the real sample. Deciding between them by eye on the bench, without the actual defect, is unreliable.

Why would blue 470 nm see a scratch that red light misses?

Because the scratch does not absorb light, it scatters it, and scattering from features that are small compared with the wavelength rises steeply as the wavelength falls. At 470 nm the same scratch throws more light out of the specular path and into the lens, so it appears with more contrast against the end face. The effect is strongest when the illumination is oblique or in a dark-field geometry, because that geometry is designed to collect scattered rather than reflected light. Red light is still useful on a fiber end face for transmitted-light work and for through-glass geometry, where the longer wavelength passes more easily; it is simply not the contrast-optimised choice for a fine surface scratch.

Is yellow light better than white light for fiber end-face inspection?

It depends on which problem you are solving, and there are two separate arguments. If a photosensitive material such as a UV-cured adhesive sits in the field, an amber band is chosen because it does not expose it; that is a material-safety argument, not an imaging argument. If the goal is a sharper edge, a narrowband amber source focuses to a tighter edge than a broad white spectrum because it carries less axial chromatic aberration — that is an imaging argument, and it can be real at high NA. Against both of these, yellow has a coarser diffraction limit than blue or green, so it will not beat blue for finding the finest scratches. Test both on the actual sample before committing.

Is the POMEAS coaxial light source available in yellow?

Not as a standard item. The POMEAS coaxial light-source family lists red 625 nm, white 6500K and blue 470 nm as standard wavelengths, and green 525 nm, infrared 850–950 nm and ultraviolet 365–395 nm as optional wavelengths; yellow is not on either list. The product page states that colour customisation is supported, so a yellow requirement can be raised as a custom request, but it has to be confirmed against the specific model and order before it can be designed into a station. Treat any yellow coaxial requirement as to be confirmed, and do not assume a standard yellow source exists in the range.

My image looks soft at 10× — is that the wavelength or the focus?

Both are plausible, and the object-side depth of field is only 0.02 mm at 10×, so focus error is the more common cause at that magnification. To separate them, capture the same field with a narrowband source and with white light at the same exposure policy: if the narrowband image is visibly tighter at the edge, the softness has a chromatic component and the white-light result is being limited by axial color rather than by focus. If both images are equally soft, the sample position or the focus is the cause. Because the depth of field is so small, mechanical stability and repeatable sample fixturing matter at least as much as the light source choice.

Can I use the same wavelength for both scratch detection and end-face angle measurement?

Rarely in one shot, and it is better to plan two conditions than to compromise one. Scratch detection wants a short wavelength and a scatter-collecting geometry, while angle and geometry measurement wants a shape-rendering illumination and a stable edge. The same 10× lens and the same field can serve both, but the illumination condition is likely to differ — blue with oblique light for the defects, white or amber with coaxial light for the geometry. If the station has a controllable light source and a controllable light geometry, store the two conditions as separate recipes and validate each against the real defect standard rather than trying to satisfy both at once.