Crossed polarizer backlight for glass and plastic stress inspection

Clear glass and plastic parts hide their worst flaws from an ordinary backlight image: the part is a faint silhouette, and molded-in stress, heat-treatment patterns and hairline cracks do not appear at all. Cross one polarizer over the backlight with a second on the lens and that changes: the empty field goes dark, and everything birefringent — stressed glass, most clear plastics, cracks under load — lights up as fringes. This guide is for engineers qualifying tempered glass, acrylic lenses, molded housings and medical parts. You will set up the polarizer pair, budget the light the films absorb, size the camera and lens, and learn where the method gives false answers.

What does a crossed polarizer backlight reveal in clear parts?

A dark field in which stressed, birefringent material appears bright: residual molding stress, stress concentrating at cracks, and the cracks themselves all rotate the light's polarization and appear as fringes, while an unstressed clear part stays nearly as dark as the background.

The arrangement is a plane polariscope, and it needs three elements: polarized light, the part, and a second polarizer — the analyzer — crossed 90° to the first. Stress makes a transparent material birefringent: the refractive index along the stress axis differs from the index across it, so the two field components leave the part out of phase by a retardance Δ that grows with stress and thickness:

Here σ is stress, d is thickness, λ is the LED wavelength, θ is the angle between the part's stress axis and the polarizer axis, and T1 and T2 are the transmissions of the two films. Fringes are contours of equal stress times thickness: where they bunch together, the stress gradient is steep.

The magnitudes work in your favor. Tempered glass carries roughly 100 MPa of surface compression; with C at about 3 × 10⁻¹² Pa⁻¹ for soda-lime glass and d = 4 mm, Δ = 3 × 10⁻¹² × 1 × 10⁸ × 0.004 m = 1.2 µm, about 1.8 waves of red light — the familiar zebra pattern of heat-treated glass. Polycarbonate is far more sensitive (C near 70 × 10⁻¹² Pa⁻¹ in typical handbook tables), so PC housings show fringes from ordinary handling stress.

Two of the stress sources you will chase have design-side calculations of their own. Molding leaves residual stress wherever a plastic wall cooled unevenly, the gradient that plastic wall thermal stress calculations predict. Assembly strain concentrates under fasteners, in the contact patch that contact stress under a bolt head analysis sizes. Both appear in the fringe image without sectioning the part.

If the problem is only that the part is hard to see at all, start from how to detect small transparent parts with machine vision; the crossed-polarizer rig is the tool once the inside matters, not just the outline. It builds a dark field the way bright field vs dark field lighting does, but with polarization instead of ray angle.

How do you set the two polarizers up?

Cover the whole lit area of the backlight with one linear polarizer, mount the second in front of the lens, and rotate it until the empty field is darkest — that is the 90° crossed position, and a few degrees of setting error still leaves the background near black.

Work through it in this order:

1. Polarizer on the light. Film or glass that covers every millimetre of the lit surface, corners included — light leaking past the edge is unpolarized and lifts the black floor. Mark the axis with a pen line once it is fixed.
2. Analyzer on the lens. It has to clear the full aperture cone; a small threaded filter vignettes the corners of a wide lens. Threaded filters, film sheets and holders are in lens accessories and the accessories finder.
3. Cross with the part out. Rotate the analyzer to minimum brightness on the empty field, then lock the ring.
4. Bring the part in. If it goes dark instead of showing fringes, its stress axes line up with the polarizer axes; turn the analyzer 45° and the fringes peak.

Room light sets a floor under your dark field: the analyzer passes its usual share of any unpolarized ambient, so a bright factory ceiling adds gray. Shroud the station or handle ambient the way ambient light variation in pick and place vision prescribes. The rotation hardware and drift checks are the same as in polarized lighting for screen inspection, and the front-light cousin of this rig — a crossed pair exposing a transparent film on metal — is covered in crossed polarized lighting for transparent film defect inspection.

How much light do the crossed polarizers cost?

One to two stops per polarizer — each film passes 25–40% of unpolarized light — so the brightest fringe a crossed pair delivers is 6–16% of the bare backlight. Budget four stops: sixteen times the exposure, or a brighter light.

The arithmetic sits in the intensity formula. The first film transmits T1 of the backlight; a part at 45° with enough retardance turns the polarization fully, and the analyzer passes T2 of what arrives. At T = 0.25 per film the best case is 0.25 × 0.25 = 0.0625, or 6.25% of the bare light — four stops down. The empty field is far darker: crossed films hold extinction ratios near 1000:1, so the background sits at about 0.1% of the aligned-pair brightness and a full fringe rides on it with roughly a thousand to one contrast.

Four ways to pay the four stops back:

• Exposure ×16. One millisecond becomes 16 ms. Costs nothing until the part moves.
• Aperture. f/4 to f/2 buys two stops and costs depth of field on domed or tilted parts.
• Gain. Two or three stops of gain bury faint fringes in noise; buy light instead.
• A brighter, strobed light. The clean answer on moving lines, within the pulse and duty-cycle limits covered in LED strobe duty cycle.

The stop-by-stop accounting for polarizer losses is laid out in polarizer light loss in f-stops; the crossed backlight applies it twice and then deliberately extinguishes the background.

Which backlight works best behind the polarizers?

A flat backlight whose lit area exceeds the part's largest dimension by about 20%, in red for threshold work — with a collimated unit for thick glass and a high-power strobed unit for fast lines, to buy back the four stops.

The sizing rule first: lit area at least the part size plus 20%, so the whole part plus its fringe context sits on light. For a 100 mm disc that means at least 120 mm of lit area in every direction.

Color: a red LED is close enough to a single wavelength that each fringe is a clean dark or bright band, exactly what one threshold wants. White light spreads each fringe into a color sequence — readable to a trained eye, awkward for a single threshold, and it forces a color camera. Choose red, or whichever color the material transmits best.

Diffuse versus collimated: in a thick part, rays from a diffuse backlight cross the stress field at different angles and the fringes smear; parallel rays keep them sharp. Past about 10 mm of glass, move to a collimated unit. The LS-BKL-0830 is a 120 × 120 mm collimated backlight in that family. The trade between the two families is the same one made in collimated backlight or diffuse backlight for edge gauging.

When the exposure budget will not close, start from more optical power. The LS-BKL-0865 is a 400 × 300 mm high-power backlight rated 92.2 W — enough lit area to carry a tray of parts behind one pair of films. Strobed, it delivers the sixteen-fold headroom as a short pulse instead of a long exposure.

Three builds cover most needs:

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

What camera, lens and exposure does the fringe image need?

For a 100 mm glass disc, a 2/3-inch camera with 2448 × 2048 pixels and a 25 mm lens at 380 mm working distance cover a 120 mm field at 0.049 mm per pixel, and the crossed pair turns a 1 ms exposure into 16 ms for a static part.

Worked example, with the arithmetic shown:

• Field of view. A 100 mm disc plus 10 mm of margin per side needs a 120 mm horizontal field.
• Camera. A 2/3-inch monochrome camera with 2448 × 2048 pixels of 3.45 µm: sensor width 2448 × 3.45 µm = 8.45 mm, height 2048 × 3.45 µm = 7.07 mm. The pixel-count logic follows choosing camera resolution for dimensional inspection.
• Magnification. m = 8.45 / 120 = 0.0704.
• Lens and distance. With a 25 mm lens, WD = f × (1 + m) / m = 25 × 1.0704 / 0.0704 ≈ 380 mm. Vertical field = 7.07 / 0.0704 = 100 mm, so the disc just fits.
• Pixel scale. 120 mm / 2448 px = 0.049 mm per pixel. A 0.15 mm crack spans 3 px; a 1 mm fringe band spans 20 px — enough to count bands, fine enough to see crack tips.

The light for this example is the LS-BKL-1468, a flat backlight with a 150 × 150 mm lit area in red. That lit surface leaves 15 mm of margin around the 120 mm field on every side.

Exposure: take 1 ms at f/4 as the bare-light reference for mid-gray. The crossed pair needs sixteen times that, so 16 ms at f/4 — comfortable for a static tray part on a 30 fps line with its 33 ms period. A domed part that forces f/8 for depth of field takes this to 64 ms, which is where you order the high-power light instead.

Read the result as two peaks on the histogram — dark field, bright fringes, one threshold between them — the same reading backlight edge inspection with histogram methods applies to silhouettes.

Why does the part disappear when you rotate the analyzer?

Fringe brightness follows the sin²(2θ) term: at 0° and 90° between the part's stress axes and the polarizer axes the part goes fully dark, and it peaks at 45°. A part that vanished is sitting on an extinction position, not proof that it is unstressed.

The working procedure: lock the backlight film, put the part in place, and rotate the analyzer until the part is at its darkest — that is the axes-aligned position. Mark the ring, turn 45°, an eighth of a turn, and lock. Impact cracks and weld lines stress material in many directions at once and show something at every angle; a part with one dominant axis — drawn polymer strip, a bent bracket — swings from invisible to maximum within 45° of rotation.

Two ways to make that robust in production: fixture the part and lock the analyzer at the 45° setting, or capture two frames with the analyzer shifted between them in the same trigger cycle, the channel-sequencing pattern described in sequential machine vision lighting. If the mechanics must stay untouched, a polarization-resolving sensor measures the angle per pixel instead; its reach and limits are covered in polarization camera working principle and inspection limits.

When does a crossed polarizer backlight lie to you?

Four ways: scattering surfaces destroy the polarization (frosted parts glow gray), annealed low-stress parts show almost nothing, curved or variable-thickness geometry makes fringes from shape alone, and a 16 ms exposure smears a moving part.

• Scattering. A frosted, textured or translucent surface scatters light into every polarization state; the whole part glows evenly and stress contrast drowns. That part needs the plain-backlight silhouette treatment, not this rig.
• Truly annealed parts. No birefringence means no signal — a correct result, not a setup failure. Prove the rig's sensitivity with a deliberately bent coupon of the same material before trusting a clean image.
• Geometry fringes. Edges, holes and wall-thickness changes make retardance of their own. Compare each design against its own golden sample; absolute fringe counts do not transfer between wall sections.
• Motion blur. At 150 mm/s, a 16 ms exposure smears 150 × 0.016 = 2.4 mm — 49 px at the 0.049 mm scale above. Allowing 2 px (0.098 mm) caps the exposure at 0.098 / 150 = 0.65 ms. From 16 ms at f/4 that is a 25× shortfall; opening to f/2 recovers 4× and a strobe pulse at about 6× the continuous drive closes the rest — inside the overdrive window most strobed lights accept for sub-millisecond pulses, duty cycle permitting.

For the pass rule, commit to regions: sum fringe brightness inside the known high-stress zones of the part, calibrate the limit on a known-good batch, and alarm on the sum. A whole-field fringe count mixes edge effects with service stress and wanders with thickness tolerance.

What should you check before ordering?

Six items decide the order: film coverage of the lit area and the lens aperture, an LED color the part transmits, 20% lit-area margin, four stops of light headroom, a heat path for the film, and locked rotation axes — proven against one known-stressed sample before the line runs.

• Film one covers the lit area, corners included. Edge leak is unpolarized and raises the black floor.
• The analyzer clears the aperture. Check the image corners at full aperture for vignetting before the mount is fixed.
• One transmission test. Image the part over a bare light in each candidate color and pick the one that passes.
• Lit area at least part plus 20%. Margin keeps edges and fringe context in the lit field.
• Four stops of headroom somewhere. Exposure, aperture or a high-power strobed light, with a controller sized for the pulse current.
• A heat gap or a glass polarizer over any high-power backlight.
• Marked, locked axes at the 45° setting. A few degrees of drift quietly rescale fringe brightness.
• A known-stressed coupon. The only proof that a fringe-free image means good, not blind.

Next step: size the whole chain in the lighting configurator, filter candidates in the backlight finder, or send the sample question through the contact page.