The qualified acceptance of an industrial vision system is based on statistical measurement system analysis (MSA) procedures. Measuring systems are validated via MSA Procedure 1 (capability study) and Procedure 3 (overall system without an operator). Purely visual good/bad decisions require mathematical proof of attribute inspection process suitability according to VDA Volume 5.
Why a simple function test is not enough for quality managers
The fact that a camera system sorts a handful of sample parts flawlessly in the lab has no statistical significance for industrial series production. In real three-shift operation, disturbance variables continuously act on the system: vibrations from the injection moulding machine, mechanical play in guide nests, temperature drift in electronic components, and slowly ageing ambient light. Acceptance according to automotive and mechanical engineering standards therefore requires mathematical proof of repeatability and reproducibility precision. For measuring systems with tight tolerances, lens selection also plays a decisive role – see our article on telecentric lenses.
The three pillars of industrial system validation
- MSA Procedure 1 (Cg/Cgk study): A certified, precisely measured reference part is captured at least 25 times (better: 50 times) in uninterrupted succession within the measurement window, without physically removing the part. This isolates the pure sensor and software scatter.
- MSA Procedure 3 (R&R study for automated systems): Since no human operator intervenes at fully automated inspection stations, the focus is on mechanical handling. Typically, 10 to 20 different series parts are measured across multiple runs, including real part feeding and mechanical reloading.
- Attribute inspection process suitability (kappa value): If the camera system makes a purely classification decision (e.g. presence of an O-ring, yes/no decision), metric distributions do not apply. For acceptance, a calibrated part set of clear good parts, clear defects, and critical borderline sample parts (“borderline pass” and “borderline fail”) is inspected blind multiple times.
The following are considered acceptance thresholds in industrial practice:
- MSA Procedure 1: Cg and Cgk ≥ 1.33
- MSA Procedure 3: %GRR < 10 % (max. 30 % in exceptional cases)
- Attribute inspection: Cohen's/Fleiss' kappa ≥ 0.80
The THOPinspect software architecture has integrated statistical export functions (CSV/data logger). This lets you generate raw data for MSA1, MSA3 and kappa calculations at the push of a button and present it in full for audits.