Backlight for width measurement on shiny turned steel

A ground or turned steel part that passes under a front flash on a white background looks fine on the bench and then fails on the line: the finish mirrors the flash as a white band, and the width tool snaps to the band instead of to the part. The fix is not more threshold tuning or a different filter glass. It is a change of lighting geometry. Put a diffuse panel behind the part and measure the silhouette, where edge contrast comes from the light and never from the surface. This guide takes you through sizing the panel, choosing its color, wiring the strobe, setting exposure with the arithmetic shown, and proving the width number before you trust it.

Why the white band lies about the edge

Ground and turned steel is a curved, partial mirror. When a front flash fires, every point whose surface normal bisects the flash-to-camera angle returns a specular highlight, and on a cylinder those points line up into a band across the part. On a ground finish that band is typically 1-2 mm wide. Your edge tool sees gray levels that rise into the band and fall again, so the threshold lands somewhere inside the surface, not on the geometry. Move the flash, change the finish lot, let the part rotate a few degrees, and the band moves; the measured width moves with it, easily by a few tenths of a millimetre. That is larger than the ±0.05 mm many width drawings carry.

A polarizer does not rescue this. Metal partially depolarizes reflected light, so crossed polarizers leak part of the highlight on bare steel, and on a cylinder the reflection angle changes along the width anyway; you pay roughly 1.5 stops of light for a partial fix. Where edge contrast actually comes from is the subject of the lighting geometry for contrast guide, and the copper surface inspection article shows the same specular trap on another reflective metal. For a shiny-part measurement example worked from the front-lighting side, these shiny-object measurement notes walk through the same failure and its fix.

Backlighting takes the surface out of the measurement entirely. The camera sees a bright panel interrupted by a clean dark silhouette, and the edge is simply where the light stops. The steel can glare all it likes; the measurement reads where the panel is blocked, not where it reflects.

What you need

Three things: a diffuse backlight panel sized by the rule in step 1, a light controller with a strobe output, and a mono global-shutter smart camera with a lens that covers the field.

For width gauging a flat diffuse panel is the default. These three cover the common cases:

The LS-BKL-1468 gives a 150 × 150 mm red field and is the pick in the worked example below. The LS-BKL-0750 is a 180 × 100 mm high-luminance panel that suits oblong parts and short pulse budgets. The LS-BKL-0830 is a 120 × 120 mm collimated panel for the tightest tolerances, and it has its own section further down.

On the camera side, choose mono rather than color: a color sensor spends its Bayer sites on channels a red panel never lights, while a mono sensor reads the panel at full resolution. That argument is set out in the monochrome camera with optical filter article. Global shutter matters too, because a strobed frame must be exposed all at once. The LS-SMC-0017, for example, is an 8.8 MP smart camera with a global shutter on a C-mount, so the lens stays your choice. If your smart camera has a fixed lens, keep it; the arithmetic below only needs the pixel count.

Set the station up in five steps

1. Size the panel. Make the lit area at least 1.5 times the part's largest dimension, and twice that when parts wander on a belt or in a loose fixture. A 40 mm wide shaft therefore needs a minimum lit width of 1.5 × 40 mm = 60 mm, with 80 mm comfortable. The margin keeps the silhouette clear of the dim border most panels have and absorbs placement error without clipping the shadow.

2. Mount it low and close. Panel face up, part above it, gap under about 10 mm. Penumbra width at each edge is roughly gap × lens aperture ÷ camera height: with a 25 mm lens at f/5.6 (4.5 mm aperture), a camera 120 mm up and a 10 mm gap, the gray smear is 10 × 4.5 ÷ 120 ≈ 0.4 mm. It is symmetric, so the 50% point stays at the true edge, but wider gaps soften contrast. Support the part on rails or a V-block that sit outside the edges you measure.

3. Choose red. For a mono camera a red panel is the default: strong sensor response and efficient LEDs. Move to infrared near 850 nm only when oil film or dust sits on the part, and accept softer edges; choose white only when a color camera shares the station.

4. Drive the panel from a strobe output. Constant-on works for stopped parts but heats the panel, and a hot panel drifts. Strobing freezes motion, holds output constant and stretches LED life considerably; the LED lifetime L70 article gives the numbers. Wire the camera's strobe or trigger output to the controller's trigger input, and keep the light pulse wider than the exposure: the smart camera strobe output guide covers the wiring and the camera strobe timing article covers pulse-to-exposure timing. A four-channel digital box like the one below runs the panel from a camera trigger and leaves channels to spare:

5. Set exposure and threshold. Raise pulse width or drive until the lit background sits at 60-80% of full scale, around 160-200 of 255, leaving headroom so the edge does not bloom. Put the threshold at the midpoint of the edge transition. For moving parts, keep exposure short enough that motion blur stays under 0.1 px; the worked example below computes it.

Worked example: 40 mm shaft, ±0.05 mm tolerance, 100 mm/s belt

Inputs, with where each comes from:

• Part width across the silhouette: 40 mm (the drawing).
• Width tolerance: ±0.05 mm, a 0.10 mm band (the drawing).
• Belt speed: 100 mm/s, parts inspected on the move (the line spec).
• Sensor: mono global shutter, 2448 px wide, the standard width of a 5 MP area sensor (sensor catalogue).
• Placement margin: 10 mm each side (assumed for a guided belt; tighten it if parts are fixtured).

The three rules and the arithmetic:

• Field of view: FOV = 40 + 2 × 10 = 60 mm.
• Pixel scale: 60 mm ÷ 2448 px = 0.0245 mm per pixel, call it 25 µm. The camera resolution guide for dimensional inspection works this the other way, from tolerance to pixels.

• Panel size: 1.5 × 40 mm = 60 mm minimum lit width, 80 mm with the wander allowance. The 150 × 150 mm field of the LS-BKL-1468 clears both figures with room for a mis-placed part.
• Blur budget: 0.1 px × 25 µm per pixel = 2.5 µm.
• Exposure: t = 2.5 µm ÷ 100 mm/s = 0.0025 mm ÷ 100 mm/s = 25 µs. Set the camera exposure to 25 µs and the light pulse to 50 µs so the frame sits fully inside the pulse; pulse drive holds the panel brightness through the short frame.

Decision: each edge locates to about 0.1 px = 2.5 µm, so width repeatability from resolution alone is around 3-4 µm, well inside the 10 µm gauge target (10% of the 0.10 mm band). This station meets a ±0.05 mm width check on the move. Below roughly ±0.02 mm, change the panel, not the threshold; see the next section.

Diffuse or collimated backlight?

Diffuse is the default for width: large area, forgiving placement, and its small penumbra (0.4 mm in the example above) does not move a 50%-threshold edge. Go collimated when the tolerance tightens below about ±0.02 mm, when the part is thick enough that its side walls widen the penumbra, or when the edge rolls over. A collimated panel sends parallel rays upward, so the shadow edge stays sharp over height; the trade is a smaller lit field and a fixed working distance at which the collimation holds. The collimated versus diffuse comparison shows the edge profiles side by side.

Prove it before you trust the number

• Run 30 parts and take the 3σ of the width. Target: 3σ at or below 10% of the tolerance band, which is 10 µm on the ±0.05 mm example.
• Do the wiggle test: nudge a part 0.5 mm sideways and re-measure. The width must not move beyond repeatability. If it does, the threshold is off the transition midpoint or the gap has grown.
• Log the lit-background gray level across a whole shift. A falling gray means panel heating or supply droop; strobe the panel or close the loop on brightness, as the closed loop lighting control article shows.
• Check one part against a micrometer. The vision mean should agree within the gauge target, not just within tolerance.

Mistakes that shift the edge

• Polarizer still in the path. It costs about 1.5 stops, leaks specular on bare metal, and buys nothing once the backlight is in.
• Gap creep. Let the part-to-panel gap grow from 10 mm to 30 mm under the same camera and the penumbra triples, from about 0.4 mm to 1.1 mm; edges go mushy.
• Infrared for fine edges. 850 nm light, plus the lens refocus it needs, softens edges; keep IR for oil-film parts where you accept that.
• Constant-on panel. Heat builds, gray level and threshold drift over the shift, and LED life shortens.
• Threshold off the midpoint. With any residual penumbra, a skewed threshold biases the two edges differently; hold it at the transition midpoint.

The decision, and where to order

Stopped parts, ±0.05 mm or looser: a diffuse red panel at 1.5 times part size, constant-on acceptable, strobe better. Moving belt: the same panel, strobed, with exposure set from the blur formula. Tolerance tighter than about ±0.02 mm, or a thick part with tall edges: a collimated panel. Pick the panel from the backlight finder or step through the lighting configurator; the lighting guides hub collects the rest of the lighting decisions you will need around this station.