How to create a quality inspection process
How to design a quality inspection process around the sampling plan and the reaction it triggers: what a passing sample proves about the lot, the hold and retest route, and the loop that corrects drift.
A worked example, stage by stage
Plan the checks before the run
Measurement waits on "Confirm control plan inspection points", the row that names the characteristic, the tolerance, the gauge, the sample size and the frequency. Only then are readings recorded, a sample drawn per the plan, and verified against the specification.
A fail is a hold, not a verdict
"Within specification?" sends "Out-of-spec" units to a hold, then to quarantine and a lab retest. The retest gate can answer "Invalid result" and return the sample for re-verification — gauges, methods and operators go wrong often enough that one reading should not condemn a batch.
An accepted unit is not an in-spec unit
The three routes separate at "Disposition of held units?": "Rework" reaches "Verify sample against the specification" again by way of its own row, "Scrap" ends at a write-off, and "Accept" arrives where the "In spec" branch does — so the record has to say which it was.
Feed it back or it is a sieve
A drift leaves "SPC trend alert?" for "Adjust the process before drift", which points back at "Record in-process check measurements": the correction lands on the operation, not the parts. "Final inspection passed?" then sends "Failed" units back to the hold, not into a second disposition route.
How it works
Start from the control plan, not the checks
For each characteristic, write the specification and tolerance, the gauge, the sample size, the frequency, and what happens when a reading falls outside limits. That last column is the process you are about to draw. Inspection points chosen without it collect wherever the last complaint landed.
Fix the sample size and who chose it
Sampling accepts a known escape rate, so the plan needs a name against it, not just a number: lot size, sample size, the accept number, and the scheme the figures were taken from. Decide too which characteristics are sampled and which are checked on every unit — safety features usually earn the second — and write that split on the step.
Build the rows: check, gate, hold
Type each check into Box text as its own row, and let Line to carry the row number the units move to next. Keep the hold and the quarantine as separate rows from the measurement. Physically moving units out of the flow is the step most often left implicit in a document, and the one most often skipped on the floor.
Put the criteria on the gates themselves
Where a reading meets a limit, the Shape column reads Decision and the label asks the question out loud; Line text carries the outcomes in the same order. Then put the reaction on the gate itself: which run rule counts as an alert, who is called, and whether the run stops. A gate with no stated reaction is answered differently on every shift.
Draw the routes back into production
Point rework at the re-inspection row rather than at the next phase, and point a trend alert at the upstream check it exists to correct. Those backward arrows are what make the chart a control loop. Without them it describes a sieve that sorts good units from bad and learns nothing from either.
Lane the owners, then prove the gauges
The lane columns carry owner and phase: Production operator, Line QC inspector, QC lab, Production supervisor, Warehouse down one side, the phases across the other. Then prove the gauges before the chart reaches the line: every characteristic needs an instrument in calibration whose resolution is finer than the tolerance it judges.
Frequently asked questions
What is the difference between in-process and final inspection?
In-process inspection checks characteristics while the run can still be adjusted; final inspection checks conformity before the batch leaves. The difference that matters is the reaction plan: an in-process reading outside limits should stop and correct the operation, whereas a final failure can only sort, rework or reject what already exists. A process relying on final inspection alone finds the same defects at the point where every remaining option is expensive.
What is a sampling plan in quality inspection?
A rule fixing what is inspected, how many units, how often, and how many defects accept or reject the lot — normally taken from a published scheme such as ISO 2859-1 at an agreed acceptance quality limit, rather than invented per line. Every plan carries an escape rate, and the published tables are worth using precisely because they state it: the probability that a lot at a given defect level still passes. Put the numbers on the sampling step, with the standard they came from.
How should quality inspection results be recorded for an audit?
Against the batch, with the measured value, the gauge or method used, the specification compared against, the inspector, and the revision of the control plan in force. Auditors test traceability in both directions: from a despatched batch back to its readings, and from a specification change forward to the first batch inspected under it. Initials against a pass stamp satisfy neither. Keep the disposition decision on the same record, including who authorised an acceptance.
What happens to units that fail final inspection?
They go on hold and rejoin the disposition route rather than entering a second process of their own. The failed branch of final inspection should return units to the same hold step, so held material is dispositioned by the same authority against the same criteria wherever the failure was found. The usual alternative — a rework loop bolted onto final inspection — is how one factory ends up with two standards of acceptance, one of them undocumented.