Coating Thickness Measurement System: Gage R&R and Attribute Agreement
An automotive supplier measures primer thickness on stamped panels with a handheld eddy-current gauge. The specification is 18-22 micrometers. A Gage R&R study uses 10 panels, 3 operators, and 3 trials; the reported %GRR is 31% of study variation and the number of distinct categories is 2. The process capability study from the same week reports Cp = 1.45 and Cpk = 1.38, but customer complaints about thin paint have risen. At the same time, a visual inspection audit for paint scratches uses three auditors and 50 parts; Fleiss kappa is 0.41. Operations wants to adjust the primer robot recipe immediately to raise the mean thickness by 1.5 micrometers. Quality wants to add a second gauge and average the readings. The gauge has current calibration stickers, but operators place the probe by hand at different angles. The panels include five families with different geometry. The customer requires 100% visual inspection until containment is lifted.
1. Interpret the %GRR and ndc results. Should the capability study be trusted, and what measurement actions should precede any process adjustment?
Read the worked answer
Interpretation: %GRR of 31% exceeds the common 30% acceptance threshold, and ndc = 2 means the gauge can distinguish only about two part categories, far too few for a 4-micrometer specification width. The capability numbers Cp = 1.45 and Cpk = 1.38 are therefore not trustworthy; measurement error can inflate or deflate observed capability, and the true Cpk may be lower. Before adjusting the robot: Improve the measurement system: fixture the probe, control angle and contact force, mark and measure the same location, train operators, and assess bias against a standard. Because panels have within-part thickness variation and five geometries, use a nested or crossed study with geometry as a factor. Then repeat Gage R&R and capability. Reasoning: Adjusting the process to a gauge with poor discrimination risks overcorrecting and creating thick paint defects. The first question is whether the data can see the process.
2. Interpret the kappa value and design an attribute measurement-system study for the visual scratch inspection.
Read the worked answer
Interpretation: Fleiss kappa 0.41 is only fair agreement; auditors disagree on what counts as a scratch, so 100% inspection is not reliable. Attribute MSA design: Select 50-100 parts spanning clean, borderline, and defective; include boundary samples. Have each auditor inspect all parts twice, in random order, under controlled lighting, with a written operational definition covering length, depth, location, and visibility. Calculate within-auditor repeatability and between-auditor agreement, plus effectiveness and miss or false-alarm rates against an expert standard. Actions: Retrain on boundary samples, recertify, and re-run the study; if kappa remains below about 0.7, do not release inspectors to independent judgment. Use gauges, limit samples, or automated inspection. Reasoning: Kappa 0.41 means the inspection system itself is a major source of variation. Containment is still required, but the inspection method must be fixed before its data are used for decisions.
3. Evaluate operations' proposal to raise the mean by 1.5 micrometers and quality's proposal to add a second gauge and average readings. What do you recommend?
Read the worked answer
Operations' proposal: Raising the mean by 1.5 micrometers based on a 31% GRR capability study is premature and could push thickness above the 22-micrometer upper limit or waste primer. It treats a possibly biased measurement as process truth. Quality's proposal: Adding a second gauge and averaging reduces random error but not operator angle, fixturing, or bias; averaging two bad readings does not create a valid measurement. Recommendation: Contain with the current customer-approved method, improve the gauge and method first, re-establish capability, then adjust the recipe only if the improved study shows Cpk below target. Use a small confirmation run with the fixed gauge. Reasoning: Measurement variation and process variation must be separated before process adjustment. Averaging is not a substitute for a capable measurement system.