Low angle bar lights for scratch detection on moving boards

White fiber-cement boards 1220 mm wide and 3050 mm long, 20 mm thick, move through an inspection tunnel at 1 ft/s (305 mm/s). The vision cell has two jobs: flag surface scratches 50 mm long and 1 mm deep anywhere on the face, and grade the concavity some pressed boards carry. One area-scan camera does both, so the lighting carries the load — long bar lights at a grazing angle for the scratch pass, steep bars for the concavity pass, and a strobing controller to alternate them frame by frame. Every number you need to copy this geometry onto your own tunnel is below: angles, bar lengths, working distance, lens focal length, strobe timing and the channel plan.

Why does a low angle bar light make scratches stand out?

At 10-15° from the surface, the flat board sends almost every ray away from the camera, while a 1 mm deep scratch catches the beam and reads as a bright rim beside a dark shadow about 5 mm long — over 20 pixels to the camera this walkthrough sets up.

The law of reflection does the sorting. A ray that leaves the bar 11° above the surface leaves a flat patch 11° on the far side, well away from a lens looking straight down the normal. Clean surface returns nothing to the camera, so a glossy board images near black; a matte white board scatters a little upward and images as an even, dim field. A scratch breaks that geometry twice. Its near wall faces the bar and scatters light up into the lens — the bright rim. The floor behind it sits in the ray shadow — the dark line. Both scale with depth, not with the colour of the surface.

Depth is the whole multiplier. At 11°, every 1 mm of depth buys 5.1 mm of shadow, so the 1 mm target is seen as a 5 mm feature, twenty-odd pixels wide once the camera is sized. Width helps less: a wide but shallow scuff casts the same shadow as a narrow one of equal depth. The same law applied to slopes instead of trenches is what finds dents and pits in flat steel with low angle lighting for dents and pits in metal, and the full geometry map is in bright field vs dark field lighting: the law of reflection decides.

Turning that shadow into a pass/fail call is the software half; scratch detection with lighting and dynamic thresholding covers the thresholding side. The rest of this walkthrough is the hardware that feeds it.

How long should the bar lights be?

Make the lit run at least 1.3 times the board width — 1586 mm against a 1220 mm board — because the outer stretch of any bar's emitting length runs dimmer than its middle, and both camera edges must sit inside the flat part.

Start in the bar lights category or the machine vision light finder with lit length as the first filter. Reach the target by ganging two diffused units: the LS-BAR-0295 at 1272 x 22 mm with the LS-BAR-1168, a 423 x 40 mm red unit. Overlap the pair 110 mm and 1272 + 423 − 110 = 1585 mm of lit length — on target. Aim the short unit a degree inward so its hot centre covers the long unit's end falloff, then flatten what is left with a reference image.

Diffused faces are not optional at this width. A bare LED row prints its own stripe pattern along a metre and a half of board; a diffused face blends neighbouring LEDs and leaves only slow shading that a flat-field correction removes.

If you later split the width across two cameras, swap the ganged runs for the LS-BAR-0069, a standard bar with a 1097 mm lit length. One covers an 800 mm lane at 1.37 x margin (1097 / 800), with the spare length absorbed by the ends.

Three catalog bars span the lengths this tunnel needs.

Values as printed in each product's data sheet; open a part for its drawing and full specifications.

Stationary, close-up parts push the same length problem much harder; large field of view lighting: balance coverage and brightness works those cases. A monochrome camera makes the red short unit free of colour penalty — light colour and monochrome camera contrast explains when colour buys you anything at all.

Where do the bars mount in the tunnel?

Set the emitting face 35 mm above the board and 180 mm out to the side — arctan(35/180) = 11°, squarely in the 10-15° window — with each bar running across the travel so the moving board sweeps through its lit strip.

One ganged run goes on each wall, throwing at each other across the conveyor. The narrow beam keeps the lit strip short along the travel direction, so every point on the board crosses peak brightness instead of idling in a long glow.

The 11° mount sits between the first two rows. Ride height is the angle's enemy: a board running 10 mm high drops the face to 25 mm and the angle to arctan(25/180) = 7.9°, stretching shadows and dimming the field. Hold the transport flat within ±5 mm, or shim the brackets to the thickest board you accept.

The camera on the normal with the bars off-axis is one of the four classic arrangements in lighting geometry for image contrast.

What camera and lens cover a 4 ft board?

A 16 MP class area camera with 3.45 µm pixels and a 16 mm lens at about 1080 mm working distance puts 0.23 mm on each pixel, so the 1 mm scratch spans 4.4 pixels — above the three-pixel minimum for dependable detection.

Take an example sensor of 5328 x 3040 pixels at 3.45 µm pitch, an 18.4 x 10.5 mm chip, with its long axis across the 1220 mm board:

The scratch's 1 mm width covers 1000 / 228 = 4.4 pixels, and its 50 mm length runs about 220 pixels. Detection wants three pixels across a feature, so there is margin to spare — spend it on threshold height, not on more camera.

Depth of field is not a constraint at this magnification. With a two-pixel circle of confusion (6.9 µm), the estimate 2 × f-number × 0.0069 × 1.015 / 0.0151² works out to about 250 mm at f/4. A concave dish of several millimetres never defocuses; pick the f-number between f/4 and f/8 for corner sharpness and strobe energy, not for focus.

How do you freeze a board moving 1 ft/s?

Strobe the bars for 200 µs and let the camera's exposure bracket the flash: at 305 mm/s the board moves 61 µm in 200 µs — 0.27 of a pixel — so the scratch image is frozen without a fast shutter or a shorter lens.

Continuous light cannot match that. One pixel of blur at 228 µm and 305 mm/s means exposure ≤ 228 / 305 = 0.75 ms, and half a pixel means 0.38 ms — workable, but it wastes the bars and invites ambient leakage. Short pulses also let the diodes run at higher instantaneous output than continuous duty, brightness you spend on stopping down.

Trigger the strobe from the conveyor encoder, not from a free-running timer, so every frame lands at a known board position. Each trigger fires one channel group; the grazing pair and the roof pair never share a frame. When frames come back dark, strobe light on but camera image black walks the timing chain, and why use a machine vision lighting controller makes the case for strobed control over raw power supplies.

One box runs both passes.

The LS-CTL-0070 is a digital control box with 4 channels and 200 W, strobe-capable, addressed over RS232. Put the two grazing runs on channels 1 and 2 and the roof pair on 3 and 4. Where the tunnel panel prefers a mains-fed unit, the LS-CTL-0322 runs from AC100-240V 50/60 Hz mains and serves 4 channels at 16-24V with 8 A total output. Browse the light controllers or the controller finder for other channel counts.

How does the same camera grade concavity?

Give the bright-field pass its own bars at about 45° on separate controller channels, fired on alternate triggers: the grazing pair catches scratches, and the steep wash lights the surface so a concave dish reads as a brightness gradient.

Steep light turns slope into signal. A flat board under a 45° wash images evenly after flat-fielding; a dish tilts its walls toward one light and away from the other, so one rim brightens while the other dims. Grade on the peak-to-peak gradient across the width, and set the reject line from a board shimmed to your concavity limit — press a known good board up by, say, 5 mm at centre and record the gradient it produces.

The roof wash uses the LS-BAR-0400, a 312 x 30 mm high-luminance bar. Four of them across the tunnel put 4 × 312 = 1248 mm of light on the 1220 mm board — just past the width, with the ends absorbed by the flat-field correction that precedes gradient measurement.

Two limits to know. Bright-field shading grades concavity, it does not measure it in millimetres; when the spec asks for numbers, parallel illumination with a telecentric lens converts flatness into an edge shift, as covered in when you need a telecentric lens for gauging — start from the collimated lights category for the source side. And if the line speeds up or you want full coverage without stitching frames, line scan photometric stereo with four-angle line lights does scratches and shape in one pass.

When is a low angle bar the wrong tool?

When the defect has no depth, grazing light hides it: stains, colour blotches and film residues cast no shadow, so give those a flat bright-field wash or a backlight instead.

Depth-free marks are a contrast problem, not a geometry problem. The same goes for sealed, mirror-glossy boards, where the grazing pair gives a black field and one blinding specular streak — tame it with cross polarization to reduce surface glare or the setups in glossy panel scratch inspection lighting. Deeply textured surfaces defeat grazing light too: when roughness shadows as strongly as the defects, the image turns to snow, and the fix is a diffuse wash at a higher angle. A board that flutters on the transport moves its own shadow; flatten the ride before blaming the lights.

Decision per case: matte board with depth-type defects — grazing bars, this walkthrough. Glossy board — polarization or a diffuse dome geometry. Depth-free discolouration — flat wash or backlight. Texture noise — diffuse light, higher angle.

What should you check before ordering?

Run this list before you cut metal: an angle jig at 35 mm over 180 mm, a lit run at 1.3 times board width, both walls on separate channels, and a 200 µs encoder-triggered strobe — those four decisions are the ones tunnels get scrapped over.

• Angle jig — a 35 mm x 180 mm block sets every bar to 11°; brackets without a gauge drift a degree at a time.
• Lit run — 1.3 x board width with 110 mm of overlap in the gang; both camera edges must sit in the flat middle of the beam.
• Diffused faces — matte boards stripe under bare LED rows; the diffused variants in the bar lights range keep the field flat.
• Two walls, two channels — the second run turns one-sided dust shadows into non-events; the AND test in the questions below explains the logic.
• Strobe timing — 200 µs from an encoder edge; check that dark frames mean timing, not dead bars, before re-wiring anything.
• Flat-field reference — capture a known good board at commissioning; every threshold above sits on top of that image.
• Reject rule — a shadow three pixels wide at 228 µm/px is a 0.68 mm shadow, about 0.13 mm of depth at 11°; set the floor there so roughness stays out.
• Concavity calibration — shim a good board to the limit and record the gradient; that recording is the spec.

Eight bars and a controller make a bundle worth quoting — start with the configurator below, or take the board spec to the quote desk. The lighting guides page walks the same numbers for other part families.