Lighting trials for oversized parts: run them yourself

Sometimes the part will not fit in a box. A press-platen assembly, a machine bed still bolted to its foundation, a weldment that only moves by crane — when the sample is too large to courier to a lighting vendor, the usual selection loop stalls. You cannot ship the part, and waiting for a travelling demo burns the calendar you have. You need neither. The two facts that pick a machine-vision light — surface finish and part geometry — are already on your floor. This guide runs the trial on your side of the loading dock: it maps finish and shape to a technique, shows how to cut a sample that carries the evidence, and walks a bench procedure that ends with the exposure, the angles and the part numbers written down. Budget a week, most of it courier time.

Ship nothing: run the trial where the part stands

The decision comes first: buy two or three candidate lights and run lighting trials for oversized parts on a bench beside the machine, because surface finish and part geometry decide the technique — and you can read both in ten minutes with a torch and an angle gauge. A vendor demo answers one question on one afternoon. Lights you own answer every follow-up question: the second defect type, the worn-tool batch, the faster line that arrives next quarter. The shortlist falls under the 90-day returns policy, so the losing candidates go back and the winner stays on the shelf as your first spare. In-stock items ship within one business day, which is what makes a one-week loop realistic. For the camera-and-lens half of the same question, machine vision lens and lighting selection: a test plan sets the wider frame.

Surface finish and geometry pick the technique

Two questions sort every lighting job. First, surface or silhouette: if the camera reads an outline — presence, hole positions, outer diameter — a backlight ends the discussion, and machine vision lighting angles: surface or silhouette draws that split. Second, when the camera must read the surface, ask how the finish reflects. A polished or curved specular face bounces the source straight into the lens as glare; a matte face scatters it back evenly. Three techniques cover most large-part trials:

• Curved and specular. A dome lights unit wraps the part in light from every azimuth at once, so no single reflection dominates. Sizing rules are in dome lights for shiny, curved and wrinkled parts, and the same physics tames shiny metal pins.
• Flat, machined or textured. A directed bar at 30–60° from the surface normal throws relief into contrast. Rotate it around the machining direction and the response can double.
• Scratches and pits on smooth metal. Drop the angle under 15° into dark field: the good surface stays black and defects glow, the geometry behind detecting dents and pits in metal surfaces.

Ring lights in the shop span low-angle 0–15° types through 30°, 45° and 60° to on-axis 90°, so one family covers the directed passes on small fields; bar lights do the same job on long ones. The full map of geometries is in machine vision illumination types.

Cut a sample that carries the evidence

Cut about 100 × 100 mm from the worst-case region — the corner where the tool dwells, the gate mark, the deepest draw — never from a show face. Keep a known-good patch beside the known-bad patch on the same piece, because contrast is measured between neighbours on one surface, not between two parts. Mark the machining or extrusion direction with an arrow: ground finishes scatter anisotropically, and a bar rotated 90° reads a different part. Two finishes on one part means two samples; a dome chosen on the polished face fails on the sandblasted rib. If nothing can be cut, a rejected part or an offcut from the same process stands in — matched by process, not by appearance.

The bench trial, step by step

Work on a rigid table with the room lights off, shrouded if the floor is bright. Run the passes in this order and write every number down as you go.

1. Mount the sample in production orientation. If the part sits tilted 10° in the fixture, tilt the sample 10°; lighting chosen flat does not transfer to tilted.
2. Fix the optics before the lights, because the field of view sizes the light. Add 20% margin to the sample width for fixturing; the arithmetic is in the next section.
3. Run the diffuse pass: dome over the curved face, diffused ring or flat diffuse over matte. Log the exposure that holds a mid-grey background at f/8.
4. Run the directed pass: a bar at 30°, 45° and 60°, swept in azimuth either side of the tool marks, then the low-angle dark-field pass for pits.
5. Score against fixed numbers: on an 8-bit camera, at least 40 grey levels between defect and the neighbouring good surface, and no defect smaller than 3 pixels across. A technique that cannot reach 40 levels is the wrong technique, not a dim one.
6. Check stability: capture 20 frames a minute apart; the mean grey level of the inspection region must hold within ±5 levels. Drift between captures usually lives on the trigger side — stabilizing lighting changes between captures covers those fixes.
7. Write the winner's sheet: model, wavelength, drive current or overdrive setting, exposure, f-number, distances and angles. Order a second unit as the spare on the same purchase order.

The optics that size the light

The light must cover the field of view, so settle the field of view first:

Worked example with typical bench numbers. The sample is 100 mm wide, plus 20% margin, so the field of view is 120 mm. The trial camera is a 2/3-inch unit with 2448 × 2048 pixels of 3.45 µm and an 8.4 mm sensor width. Magnification m = 8.4 ÷ 120 = 0.070. From a 300 mm working distance, f = 300 × 0.070 ÷ 1.070 = 19.6 mm, so fit the 25 mm lens from the drawer and solve for distance instead: WD = 25 × 1.070 ÷ 0.070 = 382 mm. Object-side resolution is 120 mm ÷ 2448 px = 0.049 mm per pixel, so a 0.5 mm pit spans 10 px. A curvature of 15 mm across the sample needs depth of field; the geometric estimate 2 × f-number × pixel size ÷ m² gives about 15 mm at f/11 and about 23 mm at f/16, so stop to f/16. Expect readings of this order: roughly 2 ms of exposure at f/11 under the dome, and roughly 0.3 ms with the bar overdriven — the margin that freezes line vibration later.

The shortlist: what to order for the trial

For curved specular samples, the dome's emitting area sets the limit. The LS-DOM-0085 is a Ø166 mm dome light with a Ø116 mm emitting area.

A 100 mm field of view sits inside that emitting area with margin on every side, and a wider sample steps up one size. The LS-DOM-0062 is a 3.0 dome light with a 194 mm outer diameter; the two compare directly:

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

For flat texture, match the lit length to the field of view. The LS-BAR-1047 is a 290 × 30 mm bar light. A 250 mm field of view fits that lit length with room to spare. Small flat areas on the lens axis take a diffused ring: at 150 mm working distance a 120 mm ring lights a 90 mm field evenly. The LS-RNG-0426 is a diffused ring light, 120 mm in diameter, with a 0° output.

Weld seams and extrusions that run the length of the part take the long bars instead. The LS-BAR-0069 has a 1097 mm lit length. That brackets a metre of seam in one pass, so the camera frames once instead of stitching.

Checks before the order becomes production

• Emitting area of the dome at least 1.15 × the field of view; lit length of the bar at least the field of view. An undersized light is the most common re-order.
• Controller headroom: if the trial needed overdrive, the production controller must deliver those pulses. LED overdrive and current control and the hidden strobe technique cover the drive side.
• Wavelength against the sensor: the right colour band on a monochrome camera beats raw power, as light colour and monochrome contrast shows.
• Mechanics: the dome's camera hole must pass the lens barrel plus its focus travel, and the bracket must hold the working distance you logged.
• Order the second unit with the first: the trial winner is the proven spare, and everything carries a 3-year warranty minimum.
• Volume discounts come by quote, so price the line, not the single unit — request a quote once the winner is known.

If the camera side is still open while the light settles, the guide to selecting a machine vision system for parts inspection runs the wider checklist.

When the usual answer is wrong

A dome needs a specular component to work. On matte black rubber it returns a flat black image; close, large soft sources — or a wavelength shift into the infrared — win instead, and the curved-handle case is worked through in black rubber inspection lighting. Translucent parts hide their features in reflection, so put the light behind them and read the silhouette: detecting small transparent parts with machine vision does exactly that. When a ground finish reads differently along and across the marks, azimuth matters more than angle — fix the bar perpendicular to the marks and re-sweep before doubting the light. Past the largest dome, cross-fire two diffused bars from opposite sides at 30–45°; their overlap approximates a flat diffuse source of any length. And when glare survives every geometry, cross-polarisation removes it but passes only about 20% of the light, so budget four to five times the exposure, as cross polarization in machine vision details.

Trial checklist

Before cutting:

• Worst-case region chosen, never a show face
• Known-good patch beside the defect on the same piece
• Machining direction marked with an arrow
• One sample per distinct finish

On the bench:

• Sample in production orientation, tilt included
• Field of view = sample width + 20%
• Diffuse pass first, then directed at 30°, 45° and 60°, then dark field under 15°
• Pass mark: 40 grey levels, defect at least 3 px
• 20 frames, mean grey within ±5 levels

Before ordering:

• Dome emitting area at least 1.15 × field of view; bar lit length at least the field of view
• Controller sized for the overdrive the trial used
• Wavelength matched to the sensor
• Second unit on the same order
• Winner's sheet filed: model, exposure, f-number, distances, angles

Start the shortlist in the lights finder or at the lighting step of the configurator; more technique guides sit on the lighting topic page. Anything the bench leaves open, contact the applications team with the sample photos.