The Validation That Silently Expires
The sample test proves a configuration on your parts - the images and OK/NG decision log come back, and the configuration is validated on your parts before the vision inspection machine ships. What the validation does not say is how long it stays true. Parts change under a configuration that does not: molds wear and start producing flash the tested parts never showed, a new compound lot can look different to the camera (surface reflectance shifts), handling scuffs a face the tested lots kept pristine. None of this is the machine degrading - it is the part drifting away from the proof. This article maps the drift sources, the warning signals that show up in the NG statistics, and what an honest re-confirmation - revalidation, in audit language - looks like. The tuning side of this - thresholds and the two error directions - is in false rejects and false accepts (/posts/false-rejects-false-accepts-inspection); this article is about the parts, not the thresholds.
The Three Drift Sources
• Mold and tooling drift. The molding-defects glossary (/posts/rubber-molding-defects-glossary) attributes flash to mold wear or misalignment opening the parting line, and notes burrs come from trimming and handling as often as from molding. A mold in month six produces different flash than the mold the configuration was validated on in month one: same family, new geometry, possibly new severity. Tooling changes - a re-cut cavity, a cleaned vent, a repaired gate - are validation events even when nobody labels them that way.
• Material drift. Compound lots vary in color, surface texture and flow. The multi-lot good-parts samples that the sampling-plans guide (/posts/sampling-plans-vs-full-inspection) and the false-reject guide (/posts/false-rejects-false-accepts-inspection) recommend for the test package exist precisely because next month's good parts will not look identical to this month's - and a material supplier's process change can move the whole appearance baseline far enough that a tuned threshold starts rejecting honest parts or passing marginal ones.
• Handling and environment drift. A new tote, a slower cooling rack, a seasonal humidity shift - handling changes scuff faces and shift how parts rest and present. These are the least dramatic drift source and the most commonly missed, because nobody associates a packing-material change with an inspection result.
The Warning Signals in the NG Statistics
Drift announces itself in the machine's own records before it announces itself in a claim - which is why reading the NG statistics weekly (/posts/ng-statistics-process-improvement) is the cheap early-warning system:
• A known family starts climbing: mold wear announcing itself as rising flash or burr counts.
• A previously clean family appears: a new compound lot showing flow marks the old compound never produced.
• The reject rate rises with no defect-family explanation - good parts being rejected more often: the parts moved, the thresholds did not. This is the quiet one, and it shows up as operator workarounds unless someone is watching the rate.
What Re-Confirmation Looks Like
Re-confirmation is a mini sample test, not a full one: current-lot good parts, a handful of the worst-case defective parts for the families that drifted, and the limit samples already on file - if a genuinely new family appears, its first borderline parts are the worst-case parts you have; send those. The published deliverables stay the same - parts run, images and the OK/NG decision log come back - and the results are read against your acceptance thresholds and the borderline cases that defined them. Three triggers, all yours to define:
• After tooling events: mold maintenance, re-cuts, vent cleaning, gate repair - anything that touches the parting line or the cavity surface. The same discipline the changeover guide (/posts/vision-inspection-machine-changeover) applies to a new SKU applies to a changed mold of the old SKU.
• On material lot changes that are visible: a new compound lot that looks different at the bench deserves a confirmation run before the line runs unattended.
• On statistics signals: a family climbing week over week for a window you define - two weeks is a common example choice, yours to set - or an unexplained reject-rate rise; either is a re-confirmation trigger regardless of the calendar.
Kept together, these records are also the change-control story an auditor asks for: what triggered the check, what was run, and what came back.
The Boundaries
• No vision inspection revalidation interval is published or advisable to copy: how much your process drifts is your number, read from your own NG statistics. A plant with stable tooling and one compound supplier drifts slower than one cycling molds and compounds - the interval is a quality-system decision, not a brochure figure.
• Re-confirmation has the same blind spot as the original validation: it measures surface-visible families. Material-level and internal drift needs the lab methods that see it - the purchase-boundary (/posts/when-not-to-buy-vision-inspection-machine) logic applies to time as well as to defect types.
• Drift works in both error directions: material drift can push marginal good parts over the line (more false rejects) or pull marginal defective parts under it (more escapes (/posts/re-screening-rejected-lot-containment)). Watch the reject rate and the claim feed together.
Frequently Asked Questions
How often should we re-confirm the configuration?
There is no honest universal interval - it depends on your mold wear rate, compound lot variation and handling stability. Define triggers (tooling events, visible lot changes, statistics signals) rather than a calendar, and let your NG statistics set the rhythm.
Is re-confirmation the same as the original sample test?
It is a scaled-down version: current-lot good parts, drifted-family worst-case parts, and the limit samples on file. The full package matters at the start and at major changes; the mini version keeps the decision line honest between them.
The machine passed a customer audit last month - does that prove the configuration is still valid?
It proves the configuration was valid for the parts tested at that time. Validation is a statement about parts, not about the machine - and the parts keep moving.
Can drift make the machine reject good parts?
Yes - material or handling drift moves good-part appearance toward the threshold, and the reject rate climbs without any defect-family explanation. That signal is the prompt for a re-confirmation run.
Next Step
Pull your NG statistics and check the trend lines for the three warning signals - a climbing known family, a newly appearing family, an unexplained reject-rate rise. If any is present, assemble the mini package (current good parts, drifted-family worst cases, limit samples) and send it; where that run happens for an installed machine - at your floor or at the factory - is part of the service scope confirmed in your contract. The images and OK/NG decision log will tell you whether the decision line still fits today's parts. Contact the team (/contact-us) - Monica (+86 138 5726 5705, WhatsApp) or Arya (+86 173 2430 9735), or email monica@aivisioncheck.com.