3D line confocal sensors: measurement through transparent layers

Clear films, coated foils, glass covers and steep shiny walls break the usual 3D toolkit. Where the physical stack has three or four surfaces, a laser triangulation profiler reports one height value — and it is the wrong one. A 3D line confocal sensor was built for exactly this case. It scans a focused line across the part, judges focus independently for every pixel on that line, and reports every refractive-index transition as its own surface: top, coating, substrate, back face. A stack that defeats triangulation becomes three or four clean measurements. This guide explains the principle, gives the arithmetic that sizes a scan (lateral pitch, line rate, data rate), and compares confocal imaging with triangulation and offline microscopy. It closes with a decision for each case that reaches our bench most often: electronics, printed electronics and medical packaging.

What a 3D line confocal sensor is

A 3D line confocal sensor projects a focused, illuminated line onto the part and images the returned light through confocal optics. These optics are built around a spatial pinhole that rejects light from planes above and below focus — see how confocal imaging rejects out-of-focus light for the principle. A triangulation profiler reads one height profile. A confocal line returns three aligned data sets in the same pass: 3D topography of the outer surface, tomography of the interfaces inside transparent material, and a 2D intensity image. It acquires while the part or web moves and yields millions of data points per pass. That puts it in a class of its own: an inline microscope, doing micrometre-class surface measurement at production speed.

How the depth response separates layers

Every point on the line produces an axial response: returned intensity as a function of depth. Light from the focal plane passes back through the aperture efficiently. Light from any other plane is suppressed, and the evaluated peak of the response is that point's height. A transparent stack simply repeats the process at each interface, so one scan reports coating top, coating-to-substrate and substrate back face as separate surfaces in a single height image.

Chromatic designs reach the same place differently: depth is encoded in the wavelength that comes back in focus, so the returned spectrum reads out position directly. The class also holds its readings where triangulation starves. Transparent, glossy and dark glossy surfaces return usable light through the aperture, which is why these sensors appear on glass, film and polished plastic rather than despite them.

Sizing the scan: pitch, line rate and data rate

These three relations size the whole scan, and they are plain geometry rather than vendor figures. Worked example: a printed-electronics web, inspected by a sensor that views a 20 mm line with 2,000 points per line. Lateral pitch: 20 mm ÷ 2,000 = 0.010 mm, so 10 µm across the web. The web runs at 150 mm/s. To sample along the web at the same 10 µm, the line rate must reach 150 mm/s ÷ 0.010 mm = 15,000 lines per second. The sensor then emits 2,000 × 15,000 = 30 million points per second.

Now the reality check. The same web on a sensor capped at 5,000 lines per second is sampled every 150 mm/s ÷ 5,000 = 0.030 mm = 30 µm along-track — three times coarser than across-track. A 50 µm printed feature is covered by under two rows of points. That mismatch, not the optics, is what usually sinks a scan plan. The decision falls straight out of the arithmetic: line rate must be at least feed speed ÷ target along-track pitch, so buy the line rate first and set the web speed to match it.

The line-scan calculator solves the same division for 2D line-scan cameras; the geometry is shared, so it doubles as a sanity check for these numbers.

Why laser triangulation splits on glass

Triangulation is pure geometry: a laser line hits the surface, a camera views the scattered line from an angle, and height follows from where the line lands in the image. The method assumes one dominant scattering surface, and clear material breaks that assumption. The beam reflects from the top face, again from the bottom face, and refracts on each entry and exit. The camera sees two overlapping lines or one blur, and the computed height lands between the true interfaces.

Change the thickness or the refractive index and the reported height changes with it — a systematic bias, not noise, so averaging frames will not remove it. Steep shiny curves fail a second way: the reflected line walks out of the camera aperture and the profile drops out entirely. The opaque version of that problem is treated in improving shiny metal 3D point cloud capture, and split-line noise in laser line extraction noise: get a clean height line. If the part forces you to stay with a laser sensor on a reflective surface, one reflective-surface laser measurement workaround walks through a print-head case.

Reading a transparent stack: topography plus tomography

Because every pixel evaluates its own response curve, a transparent stack returns a peak per interface: coating top, coating-to-substrate, substrate back face. Layer thickness is the peak-to-peak distance converted with the layer's refractive index — the arithmetic is below. The same mechanism reads air gaps, bubbles and delamination inside or behind clear material. Defects buried in the transparent substrate itself show up as breaks in the expected peak pattern.

On opaque regions of the same part the sensor still delivers plain topography and intensity, so one instrument covers mixed assemblies. Glued components on glass or PET foil come out with both the glue line and the part outline in one pass. For the set-up side of this — reference surfaces, layer selection and exposure when you configure a confocal sensor on transparent material — this transparent-material confocal walkthrough follows a real measurement case.

The refractive-index conversion

The sensor reports optical path; physical thickness divides it by the refractive index:

A foil stack reporting 750 µm of optical path through PET (n ≈ 1.6) is 750 µm ÷ 1.6 ≈ 469 µm physical. The same optical path through cover glass (n ≈ 1.5) would be 500 µm. A 6% error in the index becomes a 6% error in the thickness, so measure n once per material with a refractometer and store it with the part recipe.

Where line confocal earns its keep

Electronics and assembly: solder fillet height, glue bead profile, conductive track contour, component coplanarity, step height, gap and offset, and flatness or roughness of small surfaces — measured inline, where a microscope would need staging and fixturing. The confocal premium pays when profiles are steep, shiny or sit under clear material. A flat opaque board remains a triangulation job.

Printed, hybrid and flexible electronics: substrate roughness on glass and PET foil, step height and width of screen-printed, inkjet-printed and dispensed structures, and coating defects — scratches, impurities, delamination — including defects inside the transparent substrate. For print-quality faults that are purely 2D, illumination decides the outcome before any sensor does; near-infrared lighting for printed packaging inspection covers that choice.

Medical manufacturing: parting-line flash on molded parts, burr height on precision-stamped and micro-machined parts, texture on extruded and web products, and the classic tomographic case — heat-seal completeness and integrity in sterile medical packaging, read through the transparent film without cutting the pouch. A single clear coating on an opaque core is a simpler sibling; the coated rod inspection guide shows what triangulation-based optics already achieve there.

Confocal, triangulation or microscopy: the comparison

Three instrument families compete for these measurements, and the surface stack decides:

The resolution rows are typical classes: interferometric microscopes work in the sub-micrometre band, triangulation profilers and line confocal sensors in the micrometre band, with point-sensor confocal variants reaching finer figures. Accuracy is also always governed by what the sensor is looking at, not its label — stereo calibration accuracy: the measurement volume decides makes the same argument for stereo systems.

Limits to plan around

Plan around four limits. Steep walls and deep narrow grooves return no light to any optical sensor, confocal included: slopes beyond the acceptance angle drop out of the profile. Heavily absorbing matte material starves the return path — glossy black measures, black foam does not. The line covers a fixed width, so wide parts need motion, stitching or multiple sensors, and stitching adds its own alignment budget.

Through-stack thickness inherits the refractive-index error from the conversion above. Data volume is the last practical limit: at 30 million points per second you are filling storage in real time, so plan the acquisition pipeline before the pilot. Recording 3D camera data to H5 files for offline processing shows one workable file pattern.

What we stock for opaque surfaces

Line confocal sensors are specialist instruments outside our catalogue. For the opaque majority — metal and plastic parts with one interesting surface — a triangulation profiler is the right and more economical tool. We stock laser profilers such as the MIS-DP3850-01P and the MV-DP4180-03P, plus full-field structured-light cameras such as the MV-DPS200P-02 when area coverage matters more than a single profile line. Logistics dimensioning and parcel volume sit further down the same ladder and need no confocal at all; box dimensioning: depth camera or RGB camera walks that decision.

The decision, case by case

• Opaque part, single surface, inline, micrometre-class tolerances: a triangulation profiler. Choose from the 3D sensors category or the camera finder.
• Clear single coating on an opaque core, outer profile wanted: the profiler still works; the coating's own thickness pushes you to confocal tomography.
• Multilayer transparent stack — glass, PET foil, coatings, glue lines, bubbles, delamination, heat seals: a 3D line confocal sensor in tomographic mode.
• Sub-micrometre metrology on small areas, offline: coherence-scanning or focus-variation microscopy.
• Moving web: compute the line rate as feed speed ÷ target pitch, buy that line rate, then set the web speed.

For the opaque cases, the system configurator steps through camera, lens, trigger and mounting choices. For the transparent cases, send us the stack — layers, thickness ranges, refractive indices if known, and the tolerance you must hold — through the contact page, and we will help you specify the measurement before you commit to an instrument. More measurement guides sit on the 3D vision topic page.

Checks before you commit

• Write the stack down: every layer, opaque or transparent, its thickness range, and its refractive index if known.
• Compute the required line rate (feed speed ÷ target along-track pitch) and hold it against the sensor's maximum with margin for speed variation.
• Match lateral pitch to the smallest feature: at least three points across anything you must resolve.
• Compare the steepest wall angle on the part with the sensor's acceptance angle; that is where dropouts concentrate.
• Pilot on calibrated artefacts: a step-height standard for topography, a film stack of known thickness for tomography.
• Webs inherit machine vibration; vibration correction in 3D height scans shows the diagnostic pattern that carries over.

Run those six checks and the technology choice usually makes itself: opaque and single-surface stays with the profilers we stock, and anything with a clear layer in the measurement path belongs to line confocal.