Working notes on measurement, fitting, and inspection automation. Written so you do not need a math background to follow them. Written by the consultancy behind these services.
What an FAI legally is and what AS9102 actually requires — forms, ballooned characteristics, full versus partial, and the seven-step process end to end — then where 3D scanning genuinely earns its place (profile callouts, model-based definition, one setup for hundreds of characteristics) and where the CMM stays. Closes with how FAI compares to automotive PPAP and what does not transfer to ISO 13485. With rotatable 3D models of a turbine blade and a compressor blisk.
What a CT scan actually produces and why surface determination carries the uncertainty, how industrial CT differs from the medical scanner everyone pictures, what CT is trusted for today — porosity, wall thickness, internal channels, failure analysis, reverse engineering — the honest state of its dimensional standards, what a capability really costs, and where evaluation time, not scan time, is the cost that automation attacks. With rotatable 3D models of a closed impeller and a turbo housing.
How the instrument works — light patterns, Gray codes, phase shifting, calibration, and why point spacing is not accuracy — then where industrial structured light grew up. What actually scans well (rough beats shiny), the spray debate with measured coating thicknesses, the forging defects a scanner can and cannot see, who leads and why the free viewer mattered, datum targets and forging tolerances, acceptance standards after ISO 10360-13 — anchored by the connecting rod, a part with two metrological lives. With rotatable 3D models throughout.
The naming story in one paragraph — GOM Inspect became ZEISS INSPECT Optical 3D with Release 2023 — and then the part that matters: a rename cannot move a measurement, but the settings underneath can. Six measured points on one flat face, worked by hand two ways, reporting 0.010 mm under a best fit and 0.040 mm under datum targets. Plus what to carry across a migration, and the acceptance test that proves you did.
Every position callout is measured from somewhere, and that somewhere is built out of three imperfect surfaces. Six degrees of freedom and the 3-2-1 rule, datum feature versus datum versus simulator, and a complete alignment worked from measured points — ending with the same hole passing at Ø0.1000 and failing at Ø0.3042 because two letters swapped places.
The study your customer demands, rebuilt from the arithmetic up: repeatability and reproducibility defined by what changed, a complete two-operator worked example you can check by hand, what %GRR and ndc actually gate, and what a "gage" even is when the instrument is a scanner plus software.
What U = 3 µm (k = 2) actually claims, and how the number is built: Type A from repeated readings, Type B from resolution, certificates, and temperature, combined in quadrature and expanded — a complete uncertainty budget worked with checkable numbers, and what it means at the tolerance limit.
Every measured feature ends as a fit, and the number depends on which rule did the fitting. What least squares actually does, what minimax actually does, why the same points give two different answers, and which rule belongs to which job — built up from a five-number example anyone can check by hand.