A 30 µm sink mark on a glossy painted panel barely moves the histogram. Under a ring, a bar or a dome it changes a handful of grey levels, while the printed logo beside it changes hundreds. Brightness-based lighting cannot separate the two. Phase shift stripe lighting can: it ignores how much light each point reflects and measures where the surface bends a projected stripe pattern. This guide explains the method, derives the stripe pitch, pixel budget and eight-frame timing with the arithmetic shown, and maps each part of a station to hardware you can order. It is written for anyone specifying inspection on painted, plated, molded or transparent parts.
What is phase shift stripe lighting?
Phase shift stripe lighting captures a short series of stripe images and converts the local displacement of those stripes into a map of surface slope. The set is four phase steps of 0°, 90°, 180° and 270° for each of two perpendicular stripe directions — eight images in all.
The eight frames come in two sets. Vertical stripes step through 0°, 90°, 180° and 270°, then horizontal stripes step through the same four steps. The camera sees the stripes reflected off the part in a specular arrangement, or through the part in a transmissive one.
Wherever the surface tilts, the stripes tilt and shift with it; where it is flat, they keep their nominal spacing. A phase calculation turns each pixel's displacement into a phase value, and the gradient of that phase is the slope map: positive where the surface rises toward the source, negative where it falls away.
Painted molded panels, plated metal, packaging seals, lenses and clear molded cases are the classic workpieces. Published work in this family, phase measuring deflectometry, applies the method specifically to detect visually perceptible sink marks on high-gloss injection-molded surfaces. The decision rule: if the part offers a specular or transmissive path and can hold still for eight frames, the method applies.
Why do stripes see a sink mark that uniform light misses?
A dent does not change how much light a point reflects. It tilts the surface: a tilt of just 0.1° swings the specular reflection by twice that, 0.2°, moving reflected stripes about 0.5 mm at a 150 mm standoff while printed ink stays put.
Uniform illumination ranks surfaces by brightness, and brightness belongs to the material: paint, ink, plastic. A dent changes none of those; it changes direction. The law of reflection — the same rule that separates bright field from dark field lighting — says a ray striking a surface tilted by θ deviates by twice that angle, 2θ. So a 0.1° local slope swings the reflected pattern a full 0.2°.
At a 150 mm standoff between part and pattern source, the stripes move s = 2 × (0.1° × π/180) × 150 mm = 0.52 mm on the part. The logo under them has not moved a pixel. Sink marks are exactly this: local shrinkage that dips a glossy surface tens of micrometres and rims it with slopes of tenths of a degree.
Scattering-based methods read scratches and pits instead. The low-angle approach to dents and pits covers that mechanism. Decision: when the defect is pure geometry, image the geometry.
How does the phase math erase the printed pattern?
The phase is computed from the eight frames pixel by pixel. Any content that does not change from pattern to pattern — ink, paint grain, plastic sparkle — sums toward zero in the sine and cosine terms, and only stripe-dependent displacement reaches the slope map.
Everything that repeats identically in all eight frames — ink, paint grain, sparkle, a static shadow — lands in the constant A and cancels in the sine and cosine sums. Only displacement that follows the stripes enters φ. Nothing is ratioed or subtracted by hand: the per-pixel atan2 performs the rejection.
Unwrap φ, take its gradient in x from the vertical set and in y from the horizontal set, then threshold the combined slope magnitude. The last two lines are the sizing chain used through the rest of this guide: a full period of phase (2π) displaces the image pattern by one pitch P, and the slope that produces a shift s at standoff D is s/(2D) radians.
Decision: a stationary appearance is deleted by the arithmetic; genuine pattern distortion survives.
How many images does one part cost, and how fast can the line run?
Plan on eight captures per part: four vertical and four horizontal patterns. At a 60 frames per second camera and a 2 ms strobed exposure this costs 133 ms of image time and caps the station near six parts per second once processing is added.
Worked example, carried through the rest of the guide — a painted panel in a 120 × 90 mm field on a 2448 × 2048 global-shutter camera:
- Pixel scale: 120 mm ÷ 2448 px = 0.049 mm/px.
- Capture: 8 patterns with a 2 ms strobed exposure each at 60 fps → 8 ÷ 60 = 133 ms of image time.
- Processing: phase, unwrap and slope maps in about 30 ms on a current vision processor.
- Station rate: 1 ÷ (0.133 + 0.030) s = 6.1 → about 6 parts per second.
- A 10 parts-per-second line needs a 100 ms cycle: a 120 fps camera captures the same set in 8 ÷ 120 = 67 ms, leaving 33 ms — it fits with margin.
Strobe duty stays light: 8 pulses × 2 ms = 16 ms of LED-on time per part, roughly 10% at 6 parts per second. Pulsing at the controller's rated strobe current therefore costs little LED life. The pulse-versus-exposure trade is the same one behind pulsed IR capture timing: fire each pulse inside its exposure window, keep both locked, and keep auto-exposure out of the sequence.
Decision: budget camera frame rate as 8 ÷ (cycle time − compute margin). At 60 fps this is a 6 parts-per-second method; at 120 fps, about 10.
Do you shoot the part in reflection or transmission?
Glossy opaque parts — painted or molded plastic, plated or polished metal — reflect the stripes and are shot in a specular arrangement. Transparent parts such as lenses, glass and clear molded plastics refract them and are shot transmissively through the part.
In reflection, source and camera sit on the same side and the stripes bounce off the surface into the lens. For small parts a coaxial beam splitter folds the path so the pattern appears to come from the camera axis.
In transmission, the pattern source sits behind the part and the camera in front. Molding stresses, scratches, seeds and moisture films bend the stripes on the way through. The 2θ doubling belongs to reflection only — in transmission the deviation follows refraction geometry, so calibrate against a flat master instead of trusting a formula.
Both arrangements run the identical eight-frame sequence, and the placement logic in light geometry selection by surface and feature still decides where the source sits. Decision: glossy opaque part, reflection; clear part or lens, transmission; a part with both defect classes needs two stations or a repositionable source.
What stripe pitch and camera resolution does the job?
Size stripes to the defect: keep six pixels or more across one stripe and let each defect span a quarter of the pitch. For 3 mm sink marks in a 120 mm field on a 2448-pixel sensor that is a 12 mm pitch and 245 pixels per stripe.
Two rules set the pitch. Sampling: at least six pixels per stripe gives P ≥ 6 × 0.049 mm = 0.29 mm as a floor. Feature coverage: the narrowest defect should span a quarter pitch, so P ≤ 4 × 3 mm = 12 mm.
Choosing P = 12 mm puts 120 ÷ 12 = 10 stripes across the field, each 12 ÷ 0.049 = 245 px wide. The 3 mm mark spans 3 ÷ 0.049 = 61 px, about a quarter period — enough for the atan2 to localize its edge shift.
Sensitivity follows the formula block above. One pixel of shift equals θ = 0.049 ÷ (2 × 150) rad = 1.6 × 10⁻⁴ rad, about 0.01°, and a practical phase-noise floor sits near 0.05°. A 6 mm-wide, 30 µm-deep mark averages 0.03 ÷ 3 = 0.01 rad = 0.57° of slope across its width, an order of magnitude above that floor.
Check the pixel budget against your own sensor with the camera resolution calculator:
Decision: set pitch at four times the narrowest defect width and six or more pixels per pitch. When the defect is smaller, shrink the field or lengthen the standoff before paying for a higher-resolution camera.
What do you actually buy to build a station?
You buy three synchronized pieces: a pattern-switching light, a strobing controller and a triggered global-shutter camera. The dense two-axis eight-pattern source is normally a custom build, commissioned through the special lights line rather than picked off a shelf.
The source decides fidelity. For coarse zone and bar patterns, the nearest standard device is a 150 mm, 24 V sequencing unit with 8 zones:
- Size
- 100-200 mm
- Voltage
- 24 V
- Type
- AOI
Fixed projected patterns come from the structured line. This unit covers a 200 x 150 mm field on 24 V:
- Size
- 200-400 mm
- Voltage
- 24 V
- Type
- Structured
The dense two-axis source — eight stripe patterns at 90° steps switched within the exposure window — is a custom build. The special lights line covers AOI, structured, light boxes and custom-built designs, and the LS-SPC-S12 custom series alone lists 41 models. Bring the vendor the field size, pitch, the eight-pattern list and the switching rate, and get the pattern set in writing.
Firing zones in a timed order under camera trigger is the same discipline as multi-channel controller sequencing for photometric stereo.
Switching lives in the controller: a digital 4-channel, 200 W box with RS232 and strobe output runs a zone source:
- Output
- Constant voltage 24 V
- Channels
- 4 channels
- Control
- Digital
- Feature
- Strobe
Larger panels or shorter pulses need current headroom. This high-power 2-channel, 400 W digital controller with a network option is the series to shortlist:
- Channels
- 2 channels
- Control
- Digital
- Feature
- High power
- Output
- Constant voltage 24 V
The light controllers category and its finder filter by channels, control and output. The camera needs only global shutter, external trigger and a strobe output; the general light finder filters size and wavelength for the trial source. Decision: prototype the geometry on standard zoned or structured hardware, then commission the custom eight-pattern source for production.
When is a dome or low-angle light the better answer?
Reach for stripes when texture is the enemy — printed, painted or sparkling surfaces whose appearance you must see through. Where the surface is bare and defects shade or scatter strongly, a dome or a low-angle rig is the simpler answer for the same defect.
| Situation | Better tool | Why it wins there |
|---|---|---|
| Painted or printed glossy panel, shallow sink marks | phase shift stripes | artwork is phase-static; only slope moves the stripes |
| Curved, shiny housing with visible shading | dome light | uniform diffuse wraps curvature without hot spots |
| Scratches and pits on matte or brushed metal | low-angle dark field | edge scattering gives contrast with no compute at all |
| Full depth map of a larger freeform surface | photometric stereo | several directional frames yield per-pixel normals |
The illumination types round-up places any of these on the brightness-versus-geometry axis. Photometric stereo with a ring light is the nearest cousin — it also fires several directions per part but reads intensity rather than phase. Dome lights for shiny, curved and wrinkled parts details the diffuse option, with the dome lights category sorting by emitting diameter. These geometries gather on the lighting topic page. Where diffuse genuinely fits, it fits outright:
- Size
- 100-200 mm
- Voltage
- 24 V
Decision: bare surfaces, dome or dark field; printed, painted or transparent surfaces hiding shallow geometry, phase shift stripes.
What should you check before ordering?
Before you order anything, prove the physics on your part with a borrowed pattern source. Capture the eight frames on a known-good and a known-bad panel, run the phase math, and confirm the phase noise floor sits at least ten times below the shift your smallest defect produces.
Layout and masters
- Standoff D fixed at its design value — 150 mm in the example — and repeatable; every slope number scales with it.
- A shroud or enclosure over the station: the phase math rejects a static background, not flicker.
- A flat master and a known dented part in hand before the trial.
Source and controller
- The eight-pattern list in writing: pitch 12 mm, step order, switching rate, trigger input.
- Exposure and LED pulse matched at 2 ms and locked; auto-exposure excluded.
- Controller channels and output power checked against the source's rating.
Camera and software
- Global shutter, external trigger, frame rate at 8 ÷ (cycle time − compute margin) or better.
- Phase, unwrap, gradient, threshold — with the threshold at ten times the flat master's noise floor.
Start the build at the light configurator or the controller configurator, or describe the part through the contact page; free expert support comes with every order. Decision: run the trial with a borrowed or bench-built pattern source the same afternoon, and let the measured phase noise floor — not a catalog claim — make the buy decision.
Questions
Can phase shift stripe lighting inspect parts on a moving conveyor?
Only in stop-and-shoot at low line rates, or with a line scan camera that switches pattern state line by line under encoder trigger. The sequential eight-frame set assumes the part holds still for the full 133 ms capture at 60 frames per second; any travel in that window smears phase exactly like motion blur. Above roughly six parts per second, either move to a 120 fps camera or redesign around line scan.
Three, four or more images per direction — how many phase steps do you need?
Three steps at 120° spacing is the mathematical minimum and saves two frames per part, but the result is more sensitive to LED nonlinearity and drift. Four steps at 90° is the working standard and what the two-axis method uses. Eight or sixteen steps per direction average sensor noise further and help on dark, low-contrast surfaces, at a direct cost in cycle time of about 17 ms per extra frame at a 60 frames per second camera.
Does the method work on matte, brushed or rough surfaces?
Not as a reflection method. The phase calculation needs the stripes to return in a known direction, which means a specular or near-specular surface, or a transparent part shot in transmission. On matte and heavily brushed surfaces the reflected pattern scatters into a blur that carries no usable phase; low-angle dark-field lighting, which feeds on that same scattering, is the better tool for scratches and pits there.
Which wavelength should the stripe source use?
White suits painted and molded parts when you also want a natural-looking review image. Infrared at 850 nm or 940 nm is the usual pick for dark production halls and for ambient rejection, and the strobe runs invisibly to operators. Catalog lines list white, red, blue, green, infrared and UV variants; choose infrared when ambient light, not appearance, sets the detection limit.
How shallow a sink mark can this realistically detect?
Work it from the slope floor. With a 150 mm standoff and 0.049 mm pixels, one pixel of stripe shift equals a 0.01° surface slope, and a practical noise floor sits near 0.05°. A typical sink mark 6 mm wide and 30 µm deep carries an average slope of about 0.6° — ten times that floor — which is why the method finds marks uniform lighting misses. Smaller marks need finer pixels or a longer standoff, not a brighter light.