In-line illumination in machine vision: when specular parts need it

Flat, specular parts — semiconductor wafers, glass, polished metal, CCD faceplates — defeat ordinary lighting: a ring light gives uneven contrast across the field, and a front coaxial unit blocks the very angles a telecentric lens needs. In-line illumination solves both problems by folding the light through a beamsplitter inside the lens itself, so the rays travel the optical axis out to the part and straight back. We cover what in-line illumination is, why a telecentric lens cannot get full-field reflected brightfield any other way, the contrast it gains on specular targets, the exposure price of the beamsplitter, and the cases where a ring light still wins. You will finish able to choose between a built-in light and a coaxial port, size the magnification, and set the exposure with the arithmetic shown.

What is in-line illumination in a machine vision lens?

In-line illumination folds the light source into the optical train with a beamsplitter mounted inside the lens, so the rays travel down the optical axis, strike the part and reflect straight back along the same path onto the sensor. The source is an LED head or a fiber guide coupled into the barrel, and the splitter — usually a 50/50 plate or cube — sits at an angle in the imaging path.

It is brightfield by construction: light leaves along the axis, and a mirror-flat surface returns it into the lens. It does the same job as diffuse axial illumination, where a coaxial light box ahead of the lens carries its own beamsplitter, but the fold happens inside the lens, which is more compact and more directional. The front box diffuses its output, so its rays are a mixture of brightfield and darkfield; in-line rays are purely axial. If the part is transparent and you measure through it, that front style is the one coaxial lighting for measurement through transparent substrates works through.

Why is it the only brightfield option on a telecentric lens?

Because a telecentric lens only accepts rays that arrive nearly parallel to the optical axis, any light source placed in front of it falls outside the lens's field of view and cannot deliver reflected brightfield across the whole field. A telecentric lens images a cylinder of space, and its front element is larger than the field precisely so that off-axis object points still send parallel bundles down the barrel. A ring light, a spot, even a coaxial box in front of the barrel — all of them sit outside that cylinder, so their specular return from a flat part misses the aperture or lights only part of the field.

In-line illumination is the exception because the light enters from inside the cylinder. That is why the telecentric lenses built for gauging carry either a built-in coaxial light or a coaxial port, and why this pairing is the standard set-up in telecentric illumination for round shiny parts. Whether you need telecentricity at all is its own question, answered in when you need a telecentric lens for gauging.

How much contrast does in-line illumination add on specular parts?

About 10% more than front brightfield on a highly reflective target: a chrome-on-glass USAF 1951 comparison measured roughly 10% better contrast with in-line illumination, with even contrast across the entire field instead of patchy hot spots. The reason is ray geometry. A specular background returns essentially all the axial light into the lens, so it images bright; diffuse features on that background — bond wires, marks, dust — scatter the axial light away, so they image dark. Because every point in the field gets the same axial illumination, that contrast holds from center to corner.

A ring light in the darkfield position fails on the same part. The specular background reflects the off-axis rays out of the lens, so the background goes dark; the diffuse wires scatter weakly and unevenly, since each wire's angle and the solder at its tip redirect rays differently, so pixel values vary along the wire. A side-by-side CCD comparison sums it up:

The same even background makes digs and chips on a CCD cover glass stand out as dark defects against the bright field. The geometry behind the two columns is the subject of bright field vs dark field lighting: the law of reflection decides.

In-line or backlight: which shows defects better?

Backlight shows defects in the part more readily, while in-line completely reverses the contrast, so choose backlight when you are gauging the part's silhouette or finding through-defects and in-line when you need surface detail on the reflective face. Photographed on the same chrome-on-glass USAF 1951 target, the two methods produce a complete contrast reversal: chrome bright on dark in one, dark on bright in the other. Defects in the chrome itself read more readily in the backlit image — good for finding them, bad when the busy defect field hides the edges you want to gauge. For silhouette work, the choice between backlight styles is covered in collimated backlight or diffuse backlight for edge gauging.

How much light does the beamsplitter cost in exposure?

Two stops: a 50/50 splitter sends about 25% of the source light to the sensor off a perfect mirror and halves the returning image, so budget roughly four times the exposure, source power or aperture you would use with direct lighting.

A shot that read well at 0.5 ms with a ring light wants about 2 ms in-line. Real specular surfaces return 90% and more, so the penalty lands close to that two-stop mark; a part that scatters instead of reflecting never returns enough light at any exposure. Opening the iris one stop, f/8 to f/5.6, doubles the light and halves the exposure, at the price of depth of field — the trade aperture f-number and depth of field in machine vision cameras works through.

The splitter cannot be taken out of the path, so some source light always reaches the sensor without touching the part. Drive the source at the current the exposure needs, not at maximum, and the haze floor stays low.

When is in-line illumination the wrong choice?

On diffuse, textured or dark parts: the splitter always leaks some source light onto the sensor without it ever touching the part, and that haze lifts every black to grey, which is exactly the failure that matters on low-reflectance objects. The leak cannot be designed out — the fixes that would remove it fully would introduce worse problems in the optical train — so it sets the black floor of every in-line image.

The classic demonstration is a wooden part: a ring light lights it evenly, while the in-line image comes out grey overall with a hot spot in the middle where the leak is strongest. If a grey floor you cannot threshold away appears in an existing set-up, read machine vision stray light: diagnose low image contrast before blaming the lens. Curved specular parts are the other miss: a dome surrounding the part beats axial light from one direction, the case dome lights for shiny, curved and wrinkled parts makes.

How do I size a telecentric lens with in-line illumination?

Start from the field of view and the smallest feature: divide the sensor width by the field to get magnification, then divide the pixel size by that magnification and require at least three to four pixels across the feature you must resolve. Work it through for a sensor die with 10 µm bond wires along one edge.

• Camera: a 2/3", 5 MP sensor with 3.45 µm pixels and 2448 pixels across. Sensor width = 2448 × 3.45 µm = 8.45 mm.
• Field needed: 6.2 mm, wires and pad edge in one image. Magnification = 8.45 mm ÷ 6.2 mm = 1.36x.
• Lens: the LS-TCL-0811, an object-side telecentric lens at 1.36x, 73 mm WD, 2/3", with a coaxial light port.
• Object-side pixel = 3.45 µm ÷ 1.36 = 2.54 µm, so a 10 µm wire spans 10 ÷ 2.54 = 3.9 pixels — enough for edge finding.
• Exposure: 0.5 ms of direct lighting becomes 0.5 ÷ 0.25 = 2 ms in-line, or 1 ms with the iris opened one stop.

The magnification decides before the light does. At 0.5x the same 3.45 µm pixel covers 6.9 µm on the part, and the 10 µm wire falls to 1.4 pixels — under-resolved no matter how clean the illumination. For wider fields, step to a lower magnification with a coaxial light built in; for tight stations, a shorter working distance saves height. The two steps of that arithmetic are spelled out in telecentric lens field of view calculation from magnification and object space pixel size from magnification.

All 3 share: Mount C-mount. Values as printed in each product's data sheet; open a part for its drawing and full specifications.

Port or built-in light? A port takes an external source, so you change wavelength and drive when the part family changes; a built-in light keeps alignment fixed and wiring short, which suits a dedicated station. On a standard C-mount lens the in-line path does not exist — there is no splitter in the train, and a cube added into the converging beam brings aberration and haze — so the flat-part alternative is diffuse axial from the front. The LS-CXL-0036 is a high precision coaxial light, 74 x 50 x 50 mm, 24 V.

What should I check before ordering?

Run five checks: sensor format against the lens's image circle, working distance clearance over the fixture, source wavelength against the part's reflectance, a controller channel for the source, and a haze test with the part's darkest surface in the field.

• Format coverage. The lens's image circle must cover the sensor diagonal, or corners vignette and your even illumination stops at the corner pixels.
• Working distance. Set the barrel over the nest, fixture and any cover glass, and confirm the stated distance plus a few millimeters of focus trim.
• Wavelength. Coatings and thin films shift contrast between red, green and blue; test the worst-case part under two source colors before fixing one.
• Controller. The source needs its own strobe channel, because in-line exposures run short — drive it pulsed, not continuous.
• Haze floor. Put the darkest surface that will ever enter the field under the lens and record its grey level at working drive; that number is your threshold floor.

Size the optics in the lens finder, run the station through the configurator from the lens step — the light step picks the source for a port — and take open questions to the contact page. More on lighting geometry sits under the lighting topic.