Process Producing Defects: A Capability Diagnosis

Defects are being produced — is this a centering problem, a spread problem, a stability problem, or a measurement problem, and how do I diagnose which without disturbing the evidence?

1. Collect at least 50 in-control measurements from a stable window before any adjustment.
2. Verify gauge repeatability and reproducibility against the tolerance.
3. Plot the histogram against USL/LSL; visually classify offset vs spread.
4. Compute Cp, Cpk, Pp, Ppk with a consistent σ estimator.
5. Cross-check with the SPC chart to confirm the study window is stable.
6. Split by candidate streams (cavity, spindle, operator, lot) and recompute per stream.
7. Assign the diagnosis: centering, spread, stability, measurement, or spec.
8. Prescribe the matching action pattern; do not apply a generic "improve capability" plan.
9. Confirm with a follow-up study on a fresh window.
10. Update Control Plan, PFMEA, and reaction rules per the confirmed cause class.

• Mixed cavities or heads pooled — Cp collapses, Cpk misleads.
• One-sided characteristic (e.g. flatness) treated as two-sided.
• Attribute defect (visual) mapped onto a variable capability study.
• Specification revision mismatch between shop floor and CMM.
• Operator adjustment masking drift, then failing at the shift end.

• The same PPM can arise from three completely different processes — treating them the same is the most common capability-analysis mistake.
• If Cp and Cpk both look acceptable yet defects appear, the data almost always comes from more than one stream reported as one.
• Non-normal distributions (flatness, roundness, one-sided characteristics) will report acceptable Cpk while still generating defects at the tail.
• A "sudden defect problem" that turns out to be a slow drift over weeks was invisible only because control limits were too wide.

• Do not compute capability on the excursion window; separate stable from unstable segments.
• Do not use variable-data capability indices on attribute or ordinal data.
• Do not diagnose from pooled multi-stream data — split first.

• Cp vs Cpk: Diagnostic pair — the gap between them classifies the problem type.
• Cpk vs Ppk: A Cpk − Ppk gap points to between-subgroup drift over time.
• Cpk vs defect Pareto: Variable capability explains one mode; attribute Pareto explains the rest.

1. Collect 50–125 in-control measurements.
2. Calculate Cp and Cpk.
3. If Cp acceptable but Cpk low → centering fix.
4. If Cp low → variation reduction.
5. Use the free Process Capability Calculator to compute Cp, Cpk, Pp, Ppk, sigma level, and predicted PPM from your measurement data — no sign-up required.

• Never diagnose before capturing the pre-adjustment dataset — the evidence is destroyed by re-centering.
• Never assume normality on geometric characteristics without a normality test.
• Never close a defect investigation on a passing sample alone; require a fresh capability study under the corrected state.

• contractual critical-characteristic evidence — Critical-characteristic evidence governs production response.

• ✓Pre-adjustment dataset captured and retained
• ✓MSA acceptable
• ✓Stable-window segment isolated for the study
• ✓Data split by candidate streams and recomputed per stream
• ✓Diagnosis assigned to one class: centering / spread / stability / measurement / spec
• ✓Confirmation study run under the corrective action
• ✓Control Plan and PFMEA updated with the confirmed cause class

• Process condition: statistical stability is required.
• Distribution assumption: use a distribution model justified for the data.
• Confirm process stability and measurement-system adequacy before interpreting a capability index.
• Use a justified distribution model or non-normal method when the normal model is unsuitable.