A part that looks flawless under a white ring light can be covered in fine marks at 365 nm. The difference is not sensor quality or exposure time; it is wavelength physics. The texture height at which a finish stops mirroring and starts scattering scales with the illuminating wavelength, so shorter light interrogates a finer surface. This explainer gives you the rule — the Rayleigh roughness criterion — the arithmetic that predicts whether ultraviolet will expose defects on your part, and the light, lens and camera decisions that turn UVA into a working inspection channel.
The one-eighth rule that decides smooth versus rough
Imaging rests on reflection. A perfectly flat face obeys the law of reflection: every incoming ray leaves at the mirror angle, and a camera placed off that specular path sees almost nothing. No real face is perfect. Machining marks, rolling texture and polishing residue tilt microscopic facets, and each facet still obeys the law of reflection, but at its own angle. The energy spreads as diffuse reflection, and part of it reaches the camera. Scratches and texture become visible precisely because they redistribute light this way.
Whether a surface acts as a mirror or as a scatterer is set by the height of those irregularities relative to the wavelength. Two rays striking facets that differ in height h collect a path difference of 2h·cos θ, where θ is the incidence angle measured from the surface normal. The Rayleigh roughness criterion holds the surface polished while that difference stays under a quarter wavelength. Solving 2h·cos θ = λ/4 for the height gives the threshold:
At normal incidence (θ = 0°) this is the classic one-eighth rule: texture shallower than λ/8 leaves the surface polished — a mirror; deeper texture turns it rough — a scatterer. Wavelength works like a zoom on roughness:
| Illumination | Wavelength | Polish threshold (normal incidence) |
|---|---|---|
| Red LED | 630 nm | 79 nm |
| Green / white light peak | 550 nm | 69 nm |
| Violet LED | 405 nm | 51 nm |
| UVA LED | 365 nm | 46 nm |
Arithmetic: 630/8 = 78.75 ≈ 79 nm, 550/8 = 68.75 ≈ 69 nm, 405/8 = 50.6 ≈ 51 nm, 365/8 = 45.6 ≈ 46 nm. Texture hidden inside the specular glare of visible light can therefore scatter strongly at UVA — the same mirror-or-scatter boundary moved 23 nm deeper, which is the entire mechanism behind UV surface inspection.
Worked example: a 55 nm brushed finish at 550 nm and in UVA
Inputs:
- Visible channel: a green-centred white or green LED, λ = 550 nm, the middle of the band a visible camera uses.
- UV channel: a 365 nm UVA LED, the standard machine-vision ultraviolet wavelength.
- Geometry: light beside the camera, both near the part normal, so θ = 0° and cos θ = 1.
- Surface: a brushed stainless face with σ = 55 nm rms roughness from a profilometer — an illustrative value typical of a fine brushed finish, chosen because it sits between the two thresholds.
Rule in words: the finish behaves as polished wherever σ falls under λ/(8·cos θ) and scatters wherever σ exceeds it.
Arithmetic:
- 550 nm: h_r = 550/8 = 68.75 ≈ 69 nm. σ = 55 nm is below 69 nm, so the face is polished: the light leaves in one specular lobe, the camera sits outside it, and the image is uniform glare with no texture.
- 365 nm: h_r = 365/8 = 45.6 ≈ 46 nm. σ = 55 nm is above 46 nm, so the same face is rough: every scratch scatters, and the texture appears as contrast the sensor can threshold.
Result: the finish switches from invisible to visible as the wavelength drops from 550 nm into UVA. The gain matches the criterion: (68.75 - 45.625) / 68.75 = 23.125 / 68.75 ≈ 0.34, so texture about 34 % finer becomes detectable.
Decision: put your own σ against the two thresholds. Between 46 and 69 nm, a UVA bench is the fix. Below 46 nm, UVA shows nothing new — the levers then are a shorter band or illumination geometry. Above 69 nm, visible light already scatters and a wavelength change buys little; work on the geometry instead.
The second gain: finer lateral detail at the same aperture
Diffraction sets the smallest resolvable feature at roughly 0.61·λ/NA:
With a modest NA 0.1 optic, 0.61 × 550 nm / 0.1 = 3.36 µm at green against 0.61 × 365 nm / 0.1 = 2.23 µm at UVA — about 1.5× finer lateral detail from the same aperture. On a laser-marked code or a fine pit pattern, that separates readable from smeared. The gain only arrives if the optics image UVA onto the sensor plane, which is exactly where standard lenses stop cooperating; how lenses change outside the visible band is the groundwork covered in the SWIR and UV lens basics guide.
Where reflected UV pays off
Reflected-UV contrast earns its place wherever the defect is texture or a thin film on a surface that visible light calls smooth: laser-marked codes on polished metal, wafer and photomask faces, rolled or drawn strip, polished plastic housings, machined surfaces with fine pits and stringers. On shiny wafers, UV is often the first band where the mark scatters more than the mirror background; the wafer OCR guide shows a working set-up.
Not every UV application is reflection. Inks, contaminants and some polymers fluoresce under UV — they glow at a longer wavelength — which is a different mechanism with different lights and filters, and the reflected-UV-versus-fluorescence comparison sorts the two apart. Whichever mechanism you target, contrast on the real part decides the build, and the test-contrast-before-buying method costs one afternoon.
Building the bench: UVA light, UV lens, mono camera
Light. UVA LED spot lights are standard catalogue items — 365 nm sits mid-band in UVA (320-400 nm) — and the spot light range carries heads at 365, 375, 385 and 395 nm, with the light finder filtering by wavelength. UV emitters deliver fewer photons per milliamp than white LEDs and silicon is less sensitive in this band, so plan strobed overdrive through a lighting controller; the strobe also freezes motion. For a first test that keeps ordinary glass optics in the path, a 405 nm violet head is the gentlest step: its threshold is already 51 nm.
- Size
- 50-100 mm
- Voltage
- 24 V
- Wavelength
- 405 nm
Lens. This is the part engineers underestimate. Common optical glasses absorb sharply between 400 and 350 nm, so a UVA image through a standard lens arrives dim; and even where some light passes, the glass focuses it at a different plane than the alignment image, so it blurs at the same time. The boundary is printed right on a lens's transmission band: a model specified 400-1700 nm is, by its own datasheet, blind in this band. UV designs use materials and corrections that hold focus in the band — the thermal, SWIR and UV lens category groups them, and the lens finder filters by image circle. For a 2/3" sensor, the LS-SWL-0291 covers 240-1000 nm at F2.0 and 25 mm in C-mount. A 4/3" sensor takes the LS-SWL-0280, which spans 190-700 nm over a 16-22 mm image circle. Larger APS-H formats use the LS-SWL-0269, a 60 mm design with a 30-45 mm circle in EF or C-mount.
- Sensor format
- 2/3"
- Mount
- C-mount
- Sensor format
- 4/3"
- Mount
- C-mount
- Type
- UV
- Sensor format
- APS-H
Camera. Silicon still responds at UVA, but the cover glass and the Bayer colour array on a colour sensor block most of the band, so shortlist mono models — the camera category and its camera finder filter spectrum options including UV-sensitive sensors. Finish the chain with a UV band-pass in front of the lens: it strips visible room light so the UV channel stays clean, and what a window or filter does to the optical path is covered in the protective filter window guide. Ambient that still sneaks in shows up as the low-contrast haze described in the stray light guide.
When switching to UV will not help
Three situations predict disappointment. First, texture below 46 nm: the part is polished even in UVA, so the UV image is as featureless as the visible one. The cheaper lever is usually geometry — steepen the light angle so the flat face's mirror lobe leaves the camera and only defect scatter lands on the sensor. That is dark-field lighting, and the bright-versus-dark-field guide covers the layout.
Note the angle term before reaching for it: h_r grows as 1/cos θ, so at 60° from the normal the threshold doubles to λ/4. Grazing light does not expose finer texture by itself; what it does is steer the mirror reflection elsewhere, which is why dark field works.
Second, fluorescence. UV excites oils, residues and many polymers to glow, and that emission can swamp the scatter contrast or masquerade as a defect. Sometimes the glow is the wanted signal (contaminant detection), sometimes it is noise — clean the part or move the excitation band.
Third, photon budget. UV LEDs give fewer photons per milliamp and the sensor's UV response is modest, so exposures stretch or strobes drive harder. On a fast line, score the budget on the real part before committing the design.
Three checks before you order
- Place your texture against the thresholds. Measure or estimate σ and set it against the visible and UVA values from the table. Only a σ between them justifies the switch; above the visible threshold, fix geometry; below the UVA threshold, a shorter band or contact metrology is the honest path.
- Confirm every optic passes the band: the LED at 365 nm or 405 nm, a lens whose printed transmission range includes it (the 240-1000 nm and 190-700 nm designs above), and a mono camera with a UV-transmitting window. One absorbing element darkens the whole chain.
- Budget the photons. Strobe through a controller, add the visible-blocking band-pass, and score contrast on the real part before the build is fixed — the lens and lighting selection test plan is the template.
When the numbers land between the thresholds, the bench is small: one UVA head, one UV-corrected lens matched to the sensor, one mono camera. Let the configurator walk the chain from the camera outward, or contact the team with your σ reading and part photos. If the part sits below 46 nm, pull the lighting geometry guide next: contrast problems that wavelength cannot fix, geometry often can.
Questions
Does 365 nm imaging work on plastic surfaces too?
Yes — the criterion compares texture height with wavelength, not material. High-gloss plastics with a fine mould polish behave like metal: texture under 69 nm hides at visible wavelengths and can appear at 365 nm. Matte or textured plastics already scatter visible light, so UV adds little contrast there; it still helps with lateral resolution and with residues that fluoresce. Trial on the actual moulding before committing.
Can I use a standard visible-light lens with a 365 nm LED?
Usually not well. Common optical glasses absorb sharply between 400 and 350 nm, and the little light that does pass focuses at a different plane, so the image dims and blurs together. A UV-corrected lens fixes both. The exception is 405 nm violet: most standard lenses transmit it usefully, which is why violet makes a low-friction first test before UV optics are ordered.
Should a new inspection use 365 nm or 405 nm?
Compute both thresholds: 365/8 gives 46 nm and 405/8 gives 51 nm. If the defect texture falls between 46 and 51 nm, only the UVA band can expose it, so build the full UV chain. If the texture sits above 51 nm, start at 405 nm — violet passes standard glass and aligns close to the visible image, and the design can upgrade later without rebuilding the mechanics.
Is UVA illumination safe on an open production line?
Treat UVA like bright blue light: shroud the head so nobody faces the LED at close range, and wear UV-blocking eyewear during extended bench alignment. Incidental skin exposure at inspection intensities is not a concern for most people. UVC is different — below 280 nm, require fixed shielding, interlocks and restricted service access.
Why does a reflected-UV camera need to be monochrome?
A colour sensor's Bayer filter array and cover glass absorb most UV before it reaches the photodiodes, and a colour rendering of single-band UV scatter adds no information. Mono sensors with UV-transmitting windows keep the signal, which is why camera finders list UV among spectrum filters. Pair the camera with a visible-blocking band-pass so room lighting cannot raise the black level.