How often to check: setting inspection frequency against what a bad interval costs

Start from the quantity at risk between checks

The exposure of an interval is everything produced since the last conforming result, multiplied by what happens to that quantity if it is bad. On a slow process making a high-value item, the quantity is small and the interval can be generous. On a fast press feeding directly into an assembly that ships the same day, a long interval risks a substantial quantity already built into product. Doing this arithmetic explicitly, per characteristic, changes intervals more than any general policy, and it usually reveals that a few features deserve much tighter checking while many deserve less.

How the process actually fails determines the pattern

Gradual mechanisms such as tool wear, bath depletion or thermal drift move predictably, so periodic checking catches them before the limit is reached and the interval can be set from the observed rate of movement. Sudden mechanisms — a tool chipping, a sensor failing, a fixture shifting, material from a different lot — arrive without warning, and no achievable frequency detects them early. For those, the answer is detection at the machine or an error-proofing device rather than a tighter sampling interval, because increasing the sample rate on a step change only reduces the average loss, not the risk.

Setup checks earn more than steady-state ones

A large share of dimensional defects originate at a changeover, so verification before the run is released is usually worth more than the same effort spread through it. First-off approval by a second person, with the result recorded against the setup, catches wrong tooling, wrong programme, wrong material and mis-set offsets while nothing has been produced. A last-off check at the end of the run serves the opposite purpose: it bounds the batch, so that if the last piece conforms, the production between two good results can be released with far more confidence.

Frequencies should move, in both directions

An interval set at launch reflects what was known then. Once the process has run, the data says whether it was right: if no check has ever found a problem across a long production history and the process is demonstrably stable, the interval is a candidate for relaxation. If checks regularly find drift, or defects escape between them, it is too loose. Tighten automatically after a change of tooling, material or operator, after any escape, and following a repair. Recording why each interval was changed prevents the ratchet where frequencies only ever increase after incidents.

The cost of checking is not only the inspector's time

Every in-process check interrupts the operator, may require the machine to stop, occupies gauging that other jobs need, and adds handling that can damage the part. Where a check is destructive, it consumes product. These costs are real and they are why intervals cannot simply be tightened everywhere. They are also the argument for moving measurement to the machine or the fixture where possible, since a check that costs nothing to perform can be done often, which changes the entire calculation and removes the trade-off that the interval was managing.

Frequently asked questions

Should every characteristic on the drawing be checked at the same frequency?

No, and doing so wastes effort on features that never move while under-protecting the ones that do. Sort characteristics by consequence of failure and by how likely the process is to shift them, then assign intervals accordingly. Features held by tooling geometry rarely change unless the tool is damaged, whereas features controlled by a setting, a temperature or a wearing edge need watching. The control plan should show that reasoning, not a uniform column.

What should happen when a check fails?

The reaction has to be written before it is needed, because deciding under pressure produces inconsistency. It should cover stopping or continuing, identifying and quarantining everything made back to the last conforming result, who is called, what is done to the machine, and what evidence releases production again. Without the retrospective boundary defined, the natural instinct is to fix the setting and carry on, leaving an uninspected quantity of suspect product already moving down the line.

Can we reduce checking if the process shows good capability?

That is the right basis for reducing it, provided capability was established on data from normal production rather than a favourable trial, and provided monitoring continues at a level that would detect a shift. The reduction should be a documented decision with a stated reinstatement condition, so that a tooling change, a material substitution or an unexpected defect restores the original frequency automatically instead of requiring someone to notice and argue for it.