I have supported 3D laser line-scan inspection deployments across Vietnam, Guadalajara and
Mexicali in Mexico, Houston and the broader US, Taiwan, and Thailand. Every one of those programs
passed its bench validation before it shipped. Not one of them behaved identically once it hit the
customer's floor. The failures are predictable — which means they are designable-against.
What actually changes in the field
- Lighting and ambient conditions. Laser triangulation is robust, but the 2D tools
running alongside it are not. Skylights, shift changes, and neighboring equipment all show up
in the data.
- Calibration and pitch mismatch. A head configured for one Y-pitch on the bench meets
a different transport speed in production. Height images stretch; every downstream measurement
inherits the error.
- Fixture and handling differences. The customer's board support is never quite the
demo fixture. Warp, tilt, and seating differences appear directly in coplanarity data.
- Output-format mismatches. The customer's traceability system expects columns in an
order nobody wrote down. This gets discovered at 11 pm during ramp.
- Operator workflow. If the review screen is confusing, operators bypass it, and the
inspection silently stops mattering.
The deployment path that works
The full technical path I run — and the one I would hand any team deploying depth-based
inspection:
- Pre-deployment engineering: baseline validation on representative samples, including
deliberately bad ones. A system that has only seen good boards is untested.
- Onsite machine qualification: component training and testing, then integrated
multi-component runs — interactions surface only under combined load.
- Repeatability review before tolerance debate: measure the same board thirty times
before arguing about limits. If the spread eats the tolerance, no threshold will save you.
- Customer training and defect review: engineers and operators need different
sessions. The operator needs to trust the review screen; the engineer needs to interrogate the
CSV.
- Output-format confirmation, in writing, against a real exported file — not a
screenshot of one.
- Action-list closure and final handoff: every open item owned and dated. An
inspection system without a closed action list is still a project, not a production asset.
How this changes the way I build
After enough regional deployments, you stop designing for the demo. Runtime diagnostics go in
first, not last. Every measurement ships with evidence. Pitch and calibration are parameters, not
constants. And the review workflow is designed for the operator who will actually stand in front of
it — in whatever language, shift, and production pressure that factory runs on.
← Previous: CSV to acceptance · All notes · 3D line-scan case study →