Diffuse vs Collimated Backlight for Dimensional Measurement

Two backlights with the same emitting area and the same brightness can give measurably different diameters on the same part. The difference is not in the light output; it is in the angular spread, and it shows up as an edge that moves when the part gets thicker.

This note explains where that error comes from, how large it is, and the one mechanical condition that decides whether a collimated backlight can be used at all.

The edge is defined by transmission, not by appearance

A backlight turns a part into a silhouette. The edge in the image is where transmission drops from near 100 % to near 0 %, and a threshold placed on that transition defines the measured boundary. This is why backlighting is the default for holes, diameters, gaps and clearances: the edge does not care about surface finish, colour or oxidation, only about geometry.

It also means the entire measurement rests on the shape of that transition. If the transition is sharp and stationary, the threshold lands in the same place every time. If it is blurred or asymmetric, the threshold slides with exposure, focus and part thickness — and the repeatability that looked fine at the factory drifts in production.

Where diffuse backlight loses the edge

A diffuse backlight emits over a wide angular range, typically tens of degrees. For a thin part that hardly matters. For a part of thickness t, rays arriving at angle α illuminate the side wall, and the wall becomes part of the bright region. The measured edge then sits somewhere on the wall rather than at the true silhouette, and the error scales roughly as t·tanα.

Put numbers on it. At t = 2 mm and α = 30°, t·tanα ≈ 1.15 mm. That is not a lighting nuance; it is a millimetre-scale error on a measurement that may be toleranced in tens of micrometres. Even at α = 10° the estimate is about 0.35 mm. The estimate excludes diffraction and lens aberration, so treat it as a lower bound on the problem rather than a correction to apply.

For thin sheet, film and stamped parts a few tenths of a millimetre thick, the same calculation gives a few micrometres and diffuse backlight is perfectly adequate — provided the emitter is close to the part and the background is uniform.

What collimated backlight changes

A collimated backlight emits near-parallel light, typically within a degree or two. Rays travel axially, so the side wall of a thick part is never illuminated: the silhouette is defined by the front face alone, and the measured edge no longer depends on thickness. The transition also sharpens, because the geometric blur from angular spread is gone.

There is a mechanical price. To fill the whole field with parallel light, the lens entrance pupil must be at least as large as the field of view. A standard C-mount lens has a pupil of a few millimetres to a couple of centimetres, so the condition fails immediately. A telecentric lens places the pupil at infinity and, in practice, provides an aperture large enough to accept the full field — which is why collimated backlights and telecentric lenses are specified together, and why a collimated backlight on a standard lens behaves like an expensive diffuse one.

That pairing is not a marketing convention; it is the condition under which the parallel rays can be collected. It also means the backlight aperture should be sized from the largest field the cell will run, not from the current recipe.

Threshold stability and the exposure budget

With a clean silhouette, the remaining variables are brightness stability and exposure. A constant-current driver matters more here than on a front-lit cell: a 5 % brightness ripple moves the threshold slightly on every frame, and on a 10 µm tolerance that is enough to produce a bimodal measurement distribution that nobody can explain.

Exposure has a second constraint in moving applications. Motion blur equals speed multiplied by exposure time; keeping it below one third of a pixel sets the maximum exposure for a given line speed. This is the point at which lighting and mechanics stop being separable: if the required exposure is shorter than the available light permits, the answer is a brighter emitter, a shorter working distance, or a slower axis — not more gain, which adds noise exactly where the edge threshold is decided.

Wavelength, diffraction and the edge transition

Shorter wavelengths produce a narrower diffraction blur at a given aperture, so a blue backlight gives a slightly sharper edge than a red one on the same geometry. The effect is small compared with the angular-spread error above, but it is free, and it matters when the requirement is at the edge of what the optics can do.

Wavelength also interacts with the material. Silicon is transparent in the near infrared, so a wafer or die can be inspected in transmission with NIR in ways visible light cannot achieve. Red penetrates some coatings and dark plastics better than blue. None of this changes the geometry argument; it changes what the silhouette is made of.

Choosing between the two, in order

• Estimate t·tanα for the actual thickness and the actual emitter angle. If the result is well inside the tolerance, diffuse backlight is the simpler, cheaper and more forgiving choice.
• If the estimate approaches the tolerance, move to collimated — and confirm the lens is telecentric before ordering.
• Check the emitter area against the largest field, with margin. An undersized backlight produces edge falloff, which is a brightness problem that looks like an edge problem.
• Confirm the exposure budget at line speed, and choose a controller with constant-current output and defined trigger timing rather than a plain power supply.
• Validate on the thickest and thinnest parts in the family, not just the nominal one. Thickness sensitivity is precisely what the choice is about.

The POMEAS range covers both sides of this decision: telecentric parallel backlights from 30 mm to 265 mm apertures, including right-angle versions for cells with no room behind the part, and high-uniformity parallel surface lights for general silhouette work. The lighting range lists them alongside the controllers, and the lens side is covered by the LTC05-40 telecentric lens for smaller fields.

If the measurement also has to survive a magnification change — a zoom cell that measures several part sizes — the lighting budget has to be re-solved per step, because image irradiance falls roughly with the square of magnification. That interaction is covered in telecentric measurement lighting, and the edge-blur mechanism itself in why a micro hole edge blurs in backlighting.

Frequently asked questions

Is collimated backlight always more accurate?

For thick parts, yes — it removes thickness-dependent edge error. For thin sheet and film the angular-spread error is already a few micrometres, so the gain may not justify the cost or the telecentric lens requirement.

Can I use a collimated backlight with a standard C-mount lens?

Only in a limited sense. The lens pupil is much smaller than the field, so most of the parallel light is wasted and the image behaves like a diffuse backlight. The pairing only makes sense with a telecentric lens, whose pupil can accept the full field.

How much error does a diffuse backlight introduce on a 3 mm part?

Using t·tanα as a geometric estimate, 3 mm at 30° gives roughly 1.7 mm; at 10° roughly 0.5 mm. Both figures ignore diffraction and lens effects, so they indicate the scale of the problem, not a correction to subtract.

Does the backlight colour affect the measured diameter?

It affects the edge transition width through diffraction and through the lens’s chromatic correction. Shorter wavelengths give a slightly sharper edge. The effect is much smaller than the angular-spread error, but it is worth fixing a single wavelength for repeatable metrology.

Why does the measurement drift between shifts?

Usually brightness or thermal drift moving the threshold, or a change in part thickness within the batch. Lock the current, record the exposure, and re-check the t·tanα estimate against the actual thickness range before adjusting thresholds.

Is a wider aperture a cheaper way to sharpen the edge?

It sharpens the image but reduces depth of field, and on a backlit silhouette the part may sit outside the remaining depth. For a stationary part with a flat fixture that trade is sometimes acceptable; for a moving web it usually is not, which is why the edge is better fixed in the illumination geometry than in the aperture.

What to send for a lighting review

Send the part family with its thinnest and thickest members, the tolerance on the dimension being measured, the lens and camera if already chosen, the space behind the part, and the line speed if it moves. Those five items decide whether the answer is a diffuse backlight, a collimated one, or a different optical arrangement altogether.