← Engineering notebook
Applications

Comparing laser-triangulation sensors

Anonymized lessons from evaluating several 3D profilers. The best datasheet rarely wins.

Anonymized comparison. No products are identified.

I have worked with several laser-triangulation systems, and one lesson appears again and again.

The best datasheet rarely wins.
How laser triangulation turns height into a measurable shift
A height change moves where the angled camera sees the laser spot. Repeatability matters more than the headline numbers.

The evaluation protocol

Every candidate sensor got the same treatment, on real parts, in this order:

  1. Static repeatability - measure the same part 100 times without touching anything. The spread is the floor of everything else; a sensor that cannot repeat cannot measure.
  2. Reposition repeatability - remove and replace the part 30 times. This adds fixturing and surface-angle effects and is always worse than the static number. The ratio between the two tells you whether the sensor or the fixture is the problem.
  3. Material sensitivity - run the full range of surface finishes from production: shiny, matte, dark, multi-material transitions. Triangulation sensors differ wildly here; the laser-line behavior on a shiny-to-dark transition produces artifacts that headline accuracy never mentions.
  4. Drift over a shift - power on cold and log a fixed reference for 8 hours. Several sensors needed 45+ minutes of warm-up the datasheet did not mention.
  5. Integration friction - how long from unboxing to first usable scan in our software? Days versus weeks of engineering time dwarfs most price differences.

Metrics that actually predicted success

  • Repeatability
  • Integration complexity
  • Throughput
  • Reliability over long runs

Metrics that mattered less than expected

  • Headline resolution
  • Headline accuracy

A typical surprise

The sensor with the best resolution spec produced the worst results on the actual parts: its narrow laser line and high sensitivity made it respond strongly to the surface texture, producing noisy profiles that needed heavy filtering - which threw away the resolution advantage. A "worse" sensor with a slightly wider line averaged over the texture naturally and delivered cleaner, more repeatable heights.

The spec sheet was not wrong. It was just answering a question nobody asked.

Why

A sensor with stellar specs that is fragile, slow to integrate, or drifts over a shift will lose to a slightly less precise sensor that simply runs reliably for months. Evaluate the whole system in production conditions, not the spec sheet.