A laser marker engraves a batch code into a coated surface, and the next question from quality is: how deep did it actually cut? When the drawing calls for up to 100 μm of engraving depth at 1 μm granularity, resin embedding and cross-sectioning cost hours per part, and a 2D photograph cannot answer the question at all. The optical route is to scan the surface with a 3D laser profile sensor, build a height map, and read the groove floor against the surrounding coating. Here you get the physics behind the 1 μm figure, two reference methods, a sensor comparison, a worked example with the arithmetic shown, and the honest limits of the method. It is written for process and quality engineers who mark coatings and must prove depth to a customer.
What a laser mark is, in height units
The marker ablates or reforms the coating along hatch lines and leaves a shallow groove: typically 50-300 μm wide and, on the parts discussed here, up to 100 μm deep. Two facts follow. First, depth lives on a z axis. A 2D camera records the contrast where the coating changed colour or gloss; it records no height, so image processing alone cannot deliver a micrometre number. Second, the groove is tiny against the part, so the measurement is always depth relative to a local reference surface, never an absolute distance to a fixture.
The classical answer, resin embedding plus cross-sectioning, does work. It costs hours per part, destroys the sample, and a tilted section biases the reading. Optical profiling measures the same groove in seconds, without contact, and the same sensor then runs 3D laser line surface defect detection on other products between mark audits.
Two ways to set the reference surface
Method A, the island scan. The mark sits inside intact coating. One scan is enough: mask the groove pixels, fit a plane (a low-order polynomial on curved parts) to the unmarked ring around it, and read the floor against that plane. The mask comes from the depth residual itself or from the profiler's intensity image, where ablated coating is usually clearly darker.
Method B, before and after. Scan the panel before marking, scan again after, register the two height maps, and subtract. Registration must be tighter than the point spacing; fixture pins or two fiducial holes get you there. The subtraction is the same operation described in measuring surface height difference from a 3D point cloud.
Both methods end in one number:
The median, not the mean, keeps a few edge pixels or a dust speck from moving the result.
Where the 1 μm granularity comes from
A laser profile sensor triangulates: the camera sees the laser line displaced sideways in proportion to height. To first order, the height noise of one point is the image noise divided by the sine of the triangulation angle θ, and the depth noise after averaging N floor points and M reference points is:
Work the chain for a mid-range geometry. The camera resolves σ_image = 2 μm of line displacement at the surface and θ = 30°, so sin θ = 0.5 and the single-point height noise is σ_h ≈ 2 / 0.5 = 4 μm. Published repeatability for industrial profilers lands in a 0.1-5 μm band from close-range to long-range models; exposure, surface finish and standoff move a given sensor inside that band.
Now average, with N = 400 floor points and M = 400 reference points:
σ_d = 4 μm × sqrt(1/400 + 1/400) = 4 μm × sqrt(0.005) ≈ 4 μm × 0.071 = 0.28 μm
Even at the top of the band, σ_h = 5 μm gives 5 × 0.071 = 0.35 μm. Random noise, in other words, is not what limits a 1 μm measurement of a 100 μm groove. Tilt removal, curvature and edge definition are, and the reference fit plus the error budget below handle them.
Sensor choice at the 100 μm scale
| Sensor type | Height noise, single point | Output | Verdict for 100 μm at 1 μm |
|---|---|---|---|
| Laser profile sensor | 0.1-5 μm | Height map, micrometre grid | Right tool for opaque coatings |
| Line confocal sensor | Sub-μm to about 1 μm | Surface plus layer interfaces | Right tool when the coating transmits |
| Structured-light area camera | 10 μm to 1 mm | Fast full-area cloud | Noise at or above the groove depth |
| ToF camera | 1-50 mm | Range map | Orders of magnitude off scale |
| 2D camera and image processing | None | Brightness only | No z axis exists to read |
Both recommended picks are laser profile sensors. The MIS-DP3850-01P sits on the DP3000 line and the MV-DP4180-03P on the DP4000 line, and the 3D sensors category holds the full range with different standoffs, field widths and line rates.
- Type
- Laser profiler
- Type
- Laser profiler
For translucent coatings, steep groove walls finer than the triangulation spot, or sub-micrometre targets, a 3D line confocal sensor reads each layer interface separately from one side. The trade is a shorter standoff and slower scanning, so reserve it for the coatings a profiler cannot see into.
When the coating fights the laser line
Three coating behaviours cause trouble.
A black or rubber-loaded coating swallows the line. Raise exposure and gain first; if the line stays dim, a shorter wavelength is the next lever, since many dark organic coatings scatter violet light far more strongly than red. Persistent weak or broken lines are usually a signal-processing problem, and laser line extraction noise: get a clean height line covers the fixes.
A translucent coating lets the beam through. The triangulation then returns from the substrate or the metal beneath, and the depth reads wrong by the missing layer. Either switch to a wavelength the coating itself returns, or move to confocal, which separates the interfaces.
The visible band fails altogether. Dark coatings that reflect poorly at visible wavelengths often return strongly in the short-wave infrared, so the triangulation moves to SWIR. For the camera side, the LS-LCM-0038 is a 1K GigE line scan camera with 12.5 μm pixels, a 29.4 kHz line rate and sensitivity from 900 to 1700 nm. If the patch is wider, the line scan cameras category also holds SWIR models up to 16K line length. Triangulate with a SWIR line projector, keep the same plane-fit subtraction, and nothing else in the workflow changes.
- Lens mount
- M42
- Interface
- GigE
- Spectrum
- SWIR
- Line length
- 1K
Worked example: a 2D code engraved on a powder-coated enclosure
Spec: engraving depth 20-100 μm, reported at 1 μm granularity. Assumed inputs: scan patch 20 mm × 15 mm, point spacing 10 μm across, line step 5 μm, single-point height noise σ_h = 4 μm from the estimate above, profile rate 100 lines per second.
Scan effort: 20 mm / 10 μm = 2,000 points per line; 15 mm / 5 μm = 3,000 lines; 3,000 / 100 = 30 s per patch.
Reference ring: patch area 20 × 15 = 300 mm²; the code floor patch is 8 × 4 = 32 mm²; the ring of intact coating is 268 mm². A plane fit over that ring leaves a residual of 0.3 μm RMS, which is the tilt and curvature of the coating removed.
Depth: after excluding a guard band of about two laser-spot widths inside each wall, the floor median sits 88 μm below the ring plane, so d = 88 μm.
Random error: with N = M = 400, σ_d = 4 × sqrt(0.005) = 0.28 μm.
Systematic budget: plane residual 0.3 μm, registration 0.2 μm, edge definition 0.5 μm. Root-sum-square: sqrt(0.3² + 0.2² + 0.5²) = sqrt(0.38) ≈ 0.62 μm. Combined with random noise: sqrt(0.62² + 0.28²) ≈ 0.68 μm.
Report: 88 μm ± 1 μm, quoted conservatively. The part passes the 20-100 μm window with room to spare, the 1 μm granularity requirement is met with margin, and no resin was poured.
Mounting, motion and the error budget
- Stillness during capture. Machine vibration writes itself into the height map as fake depth. The mounting rules live in 3D camera mounting: stillness during capture decides image quality, and if the line shakes, vibration correction in 3D height scans shows the corrections.
- Moving parts. For panels on a conveyor, drive the profile rate from an encoder so every line lands at the same spacing; the set-up arithmetic is in laser line profile height measurement on a moving conveyor.
- The 10:1 rule. If the depth tolerance is ±10 μm, the measurement uncertainty should sit at or below 1 μm; the reasoning is in vision metrology: the 10 to 1 accuracy and repeatability rule. The example above lands at 0.68 μm combined, which clears it.
- Points across the floor. At least 10 points across the groove floor keep the median stable: a 0.2 mm floor wants a point spacing of 20 μm or less. The 10 μm spacing in the example gives 20.
- Z range. Size the sensor's measuring range to the flatness of the patch, not to the 100 μm of depth; the groove is a small ripple on a surface whose warps set the real span.
- Trueness check. Measure a step gauge of known height on the same fixture, and once per coating type reconcile the optical number against a cross-section; the routine is the one in 3D camera infield correction: verify dimension trueness on the line.
What profilometry cannot tell you
Depth is not thickness. When the mark burns through to the substrate, the groove floor is the metal, and the number you measured is how far the mark reaches, not how much coating remains beside it. How much coating is left is a thickness question: cross-section for opaque stacks, confocal for ones that transmit.
Very narrow, steep grooves. Below roughly 50 μm of floor width the triangulation spot straddles the walls and the floor median drifts. Confocal heads with spots in the low-micrometre range take over here.
Disputes. When a customer rejects an optical number, one destructive cross-section on one part settles the offset, and the profiler stands in from then on. For an occasional audit, an optical metrology lab or a public industrial institute will run the scan on parts you send in; anything recurring belongs on the line.
The decision and the next step
The decision, case by case:
- Opaque coating, mark inside intact surface: laser profile sensor, island scan with a ring plane fit, 400-point averages. A 100 μm groove reads to about 0.68 μm combined uncertainty, as the worked example shows.
- Translucent coating or grooves narrower than about 50 μm: line confocal sensor, which separates the coating interfaces from one side.
- Black coating that kills the visible line: SWIR triangulation with a SWIR line scan camera, same subtraction math.
- Full-face marks or strongly curved panels: archive a pre-mark height map and subtract; the ring fit has nothing to stand on.
- Structured-light area cameras and ToF: their depth noise sits at or above the groove depth itself, so they are out of scale for this job.
To pick the sensor, start at the cameras finder, or configure a vision set-up from the camera step. For a related metrology problem measured an order of magnitude larger, see how to measure 500 micron laser-drilled holes, and the 3D vision topic page collects the rest of the profiling guides. If you send the part drawing, coating type and tolerance band through the contact page, the applications team will size the standoff and field width for your marks.
Questions
Can image processing on a 2D photo measure the engraving depth?
No. A 2D image records brightness only, so the contrast where the coating changed colour carries no height information. Depth needs an optical z axis: a triangulated laser profile, a confocal scan or an interferometric sensor. Keep the 2D camera for locating and masking the mark, and read the depth from the profiler's height map.
Do I have to scan the part before marking?
Only when no intact coating surrounds the mark. A groove inside an unmarked surface carries its own reference: fit a plane to the surrounding ring and read the floor against it. Archive a pre-mark height map when marks cover a whole face or the coating is strongly curved, then register and subtract the two scans.
What works on a black coating that swallows the visible laser line?
Raise exposure and gain first, then try a shorter wavelength: many dark organic coatings scatter 405 nm violet light far better than red. If the visible band still fails, move the triangulation to SWIR, around 1300-1550 nm, with a SWIR line scan camera and a SWIR line projector, or switch to confocal with a small spot.
Is the profiler's laser safe for operators?
Profile sensors are low-power laser products, most of them class 2 or 2M, and nothing like the marking laser in power. Treat them as the warning below sets out, mount the sensor so the beam cannot sweep at eye height during set-up, and follow the laser-class instructions that ship with it.
How do I defend the optical number to a customer who trusts cross-sections?
Validate once, in parallel. Cross-section one marked part, measure the same groove both ways, and log the offset between the profiler and the section. If the offset sits inside the systematic budget of about 1 μm, the profiler stands in for sectioning on production parts, and every measurement after that stops costing a destroyed sample.


