Skip to content

Quality-Control Limit Adjustment

Quality-management rule — instantiates Adaptive Threshold Recalibration

Recomputes control and action limits on a process chart when the process's own capability or measurement noise has genuinely changed.

Quality-Control Limit Adjustment revises the control, warning, and action limits on a process monitoring chart. Its one defining principle separates it sharply from every other threshold in the set: these limits are derived from the process's own variation, not chosen against an external target or spec. A control limit is the voice of the process — it says how much this process naturally varies when nothing special is wrong. So the only legitimate reason to move it is that the process itself has changed: a machine upgrade tightened variation, a new gauge changed the measurement noise, a material change shifted common-cause spread. Recalibration is re-deriving the limits from the current capability, and the discipline is knowing when a shift is real rather than reacting to ordinary noise.

Example

A machined-parts line runs a Shewhart control chart on shaft diameter with 3-sigma control limits set at commissioning. After a spindle upgrade, a capability study shows the process now varies far less: points cluster tightly well inside the old limits, so a genuine excursion could drift halfway to the edge before the chart ever signals. The old limits have become too wide to catch real problems. Quality-Control Limit Adjustment re-derives them from the post-upgrade common-cause variation, tightening the band so the chart regains its sensitivity. The revision is licensed by the capability study, not by anyone's wish for a quieter or busier chart — and the control limits are kept firmly distinct from the customer's specification limits, which did not change.[1]

How it works

The distinctive method is that a change must first be justified as a real, sustained shift in capability — confirmed by a capability study over enough production, not inferred from a run of lucky or unlucky points. Only then are limits recomputed from the current common-cause variation. Throughout, the mechanism holds the line between control limits (the voice of the process, computed from data) and specification limits (the voice of the customer, set by requirements) — conflating them is the deepest error in the domain. The confirmed capability shift is itself the drift signal that licenses the recomputation.

Tuning parameters

  • Sigma multiplier — how many standard deviations the limits sit at (e.g. 2 for warning, 3 for action); tighter catches more but raises false alarms.
  • Recompute window / sample size — how much recent production feeds the new limits; more data stabilizes, less tracks a fresh shift.
  • Warning vs action bands — whether to carry an inner warning zone and an outer action zone, and where each sits.
  • Chart type — X-bar/R, individuals, or attributes charts; the choice shapes what variation the limits describe.
  • Control-vs-spec separation — enforcing that limits derive from process data, never from tolerance requirements.

When it helps, and when it misleads

Its strength is keeping a monitoring chart's limits faithful to the process as it actually behaves now, so signals mean something. Its classic and corrosive misuse is tampering — widening the limits simply to stop nuisance alarms, or nudging them to match a spec the process can't hold. Widening to quiet the chart blinds it to the very defects it exists to catch, and adjusting limits toward a target rather than the process is a textbook way to make things worse. The discipline is to move limits only on evidence of a real capability change, and to keep control limits derived from the process rather than from what anyone wishes were true.

How it implements the components

  • monitored_variable_or_score — it operates on the process statistic being charted, ensuring it is defined and measured consistently before limits move.
  • recalibration_rule — it defines how a capability study recomputes the limits from current common-cause variation.
  • drift_detection_signal — the confirmed, sustained capability shift is the signal that licenses recomputation (and distinguishes a real change from noise).

It does not weigh who bears the cost of an escaped defect (consequence_weighting_policy) — Policy Threshold Update and Eligibility Threshold Review carry consequence weighting — nor does it stage the change across lines (rollout_and_guardrail_plan), which Staged Threshold Rollout owns.

Notes

The recurring trap is confusing control limits with specification limits. Spec limits express what the customer requires; control limits express what the process delivers. Only the latter are recomputed here, and only when the process changes — a spec change does not license moving control limits, and a process improvement does not relax the spec.

References

[1] In statistical process control, control limits are calculated from the process's own variation (commonly at 3 sigma), distinct from specification limits set by requirements. Adjusting limits to reduce alarms or to match a spec, rather than to reflect a demonstrated change in the process, is tampering and degrades the chart's diagnostic value. ↩