A dot-peen or laser-etched Data Matrix on bare metal is one of the hardest reads in plant traceability. There is no ink: the code is a pattern of craters or grooves in a surface that behaves like a mirror, and the decoder still needs dark-on-light cells with clean edges. Turning up light power does not fix that; geometry does. This guide gives the decision for each marking process, the numbers that drive it (pixels per module, light size, working distance, exposure), a worked lens example for a 5.6 mm dot-peen code, and the checks to run before ordering. The rules hold whatever decodes the image: a smart camera, a PC pipeline, or a code-reader app on an edge controller.
The decision by marking process
Match the light to how the mark was made, not to the camera. The marking process decides whether you want the flat surface bright, so the marks scatter away from the lens, or dark, so the grooves scatter toward it. A guide to laser, dot-peen and inkjet marking on metal covers the processes themselves; here is the lighting version of that decision:
| Mark and surface | Light geometry | What you see |
|---|---|---|
| Dot-peen on flat machined, ground or polished metal | Coaxial light on the lens axis | Dark craters on a bright field |
| Laser-engraved or scribed grooves, flat part | Two bar lights at 10-20° above the surface | Bright grooves on a dark field |
| Laser anneal (black oxide), flat part | Coaxial or dome, any glare-free bright field | Dark mark on bright metal |
| Any mark on a curved or stamped surface | Dome light over the feature | Even cells across the curvature |
| Mark under an oil film, or polished and curved | Dome, or cross-polarized coaxial | Glare suppressed, cells readable |
| Inkjet or printed code on metal | Diffuse ring or dome | Ink contrast without specular washout |
Three rules carry most of the weight: flat and peened means coaxial; flat and engraved means dark field; curved means dome.
Why geometry beats power on metal
Machined metal is specular: it reflects light like a flat mirror, so nearly all the light leaving the surface goes in one direction. A dot-peen crater or an etched groove breaks that mirror, because its walls are rough and angled and scatter light everywhere. Every DPM scheme is one of two choices. Bright field sends light down the lens axis so the flat background returns to the sensor: white field, dark cells. Dark field sends light at a low angle so the flat background reflects away: black field, bright cells.
Design for at least 40 percent contrast between cell and quiet zone; 50-80 percent is normal when the geometry matches the mark. Below about 20 percent, still frames may decode, but read rates on a moving line collapse. If the vocabulary is new, the illumination geometry overview walks through all the standard setups.
Coaxial light for dot-peen on flat metal
A coaxial unit folds the light 90° through a beam splitter so it travels on the lens axis, and the camera shoots through the same splitter. The flat background returns to the sensor; crater walls do not. Peened codes read as dark cells on a bright field, which makes this the default answer for dot-peen on ground, turned or milled faces.
Size by field of view, not by code: the emitting face should cover at least twice your FOV width, so code and quiet zone sit in the even centre of the field. Set the working distance roughly equal to the light's own width: closer and the centre runs hot, farther and the corners fall off. Check with a white card in place of the part: the histogram should form one tight band, not a centre peak with dark corners.
The LS-CXL-0067 is a 102 × 100 mm coaxial source.
- Size
- 100-200 mm
- Voltage
- 24 V
For larger fields or longer standoffs, step the face size up:
| Specification | LS-CXL-0067 | LS-CXL-0192 |
|---|---|---|
| Price | $499.00 | $739.00 |
All 2 share: Cable length (mm) 500. Values as printed in each product's data sheet; open a part for its drawing and full specifications.
For the trade-off between on-axis and dark field on Data Matrix specifically, the dark field versus on-axis DOAL lighting selection guide is worth ten minutes before you buy. Know the failure mode too: on brushed, sandblasted or cast finishes the background scatters as well, the whole field turns grey, and no exposure recovers the cells. That surface needs dark field.
Low-angle bar lights for engraved marks
Laser ablation, scribing and electro-etch marks leave grooves; read them with two bars aimed at each other across the code, mounted at 10-20° above the surface plane. The flat metal mirrors that light away from the lens, so the background goes near-black, while groove walls scatter into the lens and the cells light up. Start at 15°, watch the live histogram, and sweep a few degrees either way: the optimum is sharp, and 5° of tilt can double the contrast.
Bars also suit rows of codes along a conveyor or web, where one pair covers every part in the lane; sizing them for full-width coverage is its own exercise, covered under large field of view lighting. The LS-BAR-1047 is a 290 × 30 mm LED bar light. Fit one per side, aimed so the beams cross at the code.
- Size
- 200-400 mm
- Type
- Standard
- Voltage
- 24 V
Dome light for curved and stamped parts
On a cylinder, a shoulder or a stamped bracket, the surface normal sweeps tens of degrees across the code. Coaxial gives a glare band; dark field gives one bright stripe. A dome fires light from every direction at once, so every cell sees the same intensity whatever its tilt, which is also why domes are the default for reading codes on reflective and curved surfaces of any kind. Sizing and finish-dependent cases are covered in the dome light guide.
Size to the part: the emitting aperture should be at least twice the diameter of the curved region you read, and the part must fit inside the dome's height. The LS-DOM-0085 has a 166 mm outside diameter and a 116 mm emitting area.
- Size
- 100-200 mm
- Voltage
- 24 V
When the real problem is a thin oil film or a mirror polish rather than curvature, cross polarization on a coaxial or dome light removes the specular washout without changing the geometry.
The numbers that decide the read
The light sets the contrast; the optics decide whether the decoder can use it. Five numbers govern a DPM station, and they are pure arithmetic:
Sampling rule: never go below 3 pixels across the smallest cell, design at 4, and use 5 when you grade the code and want margin. A 3-pixel cell is a 3 × 3 block; one pixel of blur consumes a third of it, and process variation will find that weakness.
Worked example: 5.6 mm dot-peen code on steel
Mark: dot-peen Data Matrix, 14 × 14 modules at 0.4 mm pitch, so the code is 14 × 0.4 = 5.6 mm wide. Part placement repeats within ±2 mm, so the FOV is 12 mm.
Sampling at 4 px per module: GSD = 0.4 / 4 = 0.1 mm per pixel. Pixels needed = 12 / 0.1 = 1200 across the field, so a 1600 × 1200 sensor covers it with a third of the frame to spare.
Lens: a 1/1.8 in sensor is about 7.0 mm wide, so m = 7.0 / 12 = 0.58. With the coaxial body setting the working distance near 95 mm, f = 95 × 0.58 / 1.58 ≈ 35 mm: use a 35 mm macro lens. Depth of field at 0.58× and f/11, allowing a 2-pixel blur (8.8 µm on a 4.4 µm sensor), is 2 × 11 × 0.0088 × 1.58 / 0.34 ≈ 0.9 mm total, about ±0.4 mm: enough for a machined face located within ±0.3 mm.
Contrast: coaxial on ground steel typically renders the background near 200 of 256 grey levels and the craters near 70, so (200 - 70) / 200 = 65 percent, comfortably above the 40 percent floor.
Motion: at 0.5 m/s on a conveyor, one-pixel blur caps the exposure at 0.1 mm / 500 mm/s = 0.2 ms. Strobe the light with a 100-200 µs pulse from a controller trigger and let the pulse freeze the part; high speed barcode verification: size the strobe pulse works the same arithmetic for line-rate codes.
Run your own FOV and cell pitch through the calculator before ordering the camera:
Checks before you order
- Coaxial face at least 2× the FOV width; dome aperture at least 2× the curved region; bars longer than the row of codes.
- 4 px per module at the worst part position, measured on a captured image: divide the code's width in pixels by its module count.
- Contrast at least 40 percent cell-to-quiet-zone, read from the histogram on parts from the process spread, not on golden samples.
- Strobe pulse no longer than GSD / line speed, with camera and light on one trigger.
- ISO/IEC 29158 grade of 3.0 or better across 30 parts; log the cell-contrast sub-grade so a future failure shows whether the marker or the light drifted.
- Shroud the station: overhead plant light on a mirror-like part is stray light that flattens a dark field.
- Colour: monochrome camera with red on etched steel, blue on dark or anodized substrates, white only if the same station checks colour; the LED colour contrast guide has the full table.
When the usual answer is wrong
Brushed or cast surface with a coaxial light: the background scatters grey and the cells vanish. Switch to low-angle bars, and go lower with rougher finishes, 10° rather than 20°, to keep the background dark.
Matte cast iron with dark field: nothing on the surface is mirror-like, so field and cells both stay black. Move to a dome or a flat diffuse panel.
Adding power when contrast is low: on a mirror the direct return is already tens of times brighter than the scatter, so more power saturates the background first and clips the cells. Change the angle instead.
Laser anneal that reads poorly: if the mark re-melted to shiny grey instead of black oxide, it behaves like a specular defect, and dark field often flips it readable when coaxial fails.
Anodized aluminium: the oxide tint shifts contrast by colour more than by angle. Read up on the marking methods for aluminium and metal components and test red against blue before committing to a light.
If the only sightline bounces the code off another surface, the decode arrives mirrored; detecting mirrored Data Matrix codes reliably solves that on the decoder side without new hardware.
What to order next
Pick the geometry from the table, size it from the rules above, then run it through the light finder or the system configurator starting at the light step, with FOV and working distance to hand. More lighting set-ups are on the lighting topic page. In-stock items ship within one business day, every part carries a 3-year warranty minimum, and free expert support comes with each order; if the mark is unusual, send the finish, mark type and part speed through the contact page and have the choice reviewed before you buy.
Questions
Why does a coaxial light beat a ring light for dot-peen codes?
A ring light strikes the surface as a cone, so on a mirror-flat part only one annulus of the background reflects back to the lens and you get a bright ring with a dark centre. Coaxial light travels on the lens axis, so the whole flat background returns to the sensor at once while the craters stay dark. A ring light works again on slightly matte surfaces, because the finish scatters the cone evenly.
How many pixels per module does a DPM Data Matrix need?
Three is the floor, four is the design point, five if you grade the code rather than just decode it. At three pixels a cell is rendered by a 3 × 3 block, so one pixel of edge blur consumes a third of the cell, and process spread in mark size and part position starts failing cells quickly. Decoder demos read below this on still images, but a moving line will not hold the read rate.
Do I need a telecentric lens for DPM reading?
Only when part presentation forces it. A telecentric lens holds magnification constant as height varies, so it earns its place when the part moves more than about ±1 mm on a small FOV, or when you measure the mark as well as read it. On a fixture or machined face located within ±0.4 mm, a 35 mm macro at f/11 keeps the code in focus and saves the telecentric cost.
Which light colour reads etched metal Data Matrix best?
With a monochrome camera, red is the default on steel: the marks are geometry rather than pigment, and red LEDs hold output and lifetime well. Blue raises contrast on dark or anodized substrates because the shorter wavelength lifts scatter off fine grooves. White is only worth fitting when the same station verifies colour prints; on a mono sensor it costs contrast. Do not mix colours inside one strobe set.
Can one station read both DPM and printed labels?
Yes, and the economical way is sequencing, not compromise lighting. Fit the coaxial and a diffuse bar or ring on a two-channel controller, strobe them as separate exposures of the same trigger, and let the decoder try both frames. Use separate exposures and gains rather than a continuous overlap, because bright-field and dark-field images want opposite settings.
What grade should a DPM code pass?
Grade DPM codes to ISO/IEC 29158, not ISO/IEC 15415: 29158 adds the reference-decoding and lighting corrections that direct marks need. Pass at 3.0 overall unless the customer contract specifies higher. When a batch fails, re-grade the images with the sub-metrics on screen: low modulation points at the lighting, low cell contrast points at the marker.


