Inspection Record Retention and Retrieval

Retention is normally decided by storage capacity: keep images until the disk fills, then reduce the period, then compress. That is the one criterion with no relationship to what the record will be needed for. The question that actually determines retention is the one somebody will ask months later — did this part pass, and on what evidence — and the answer determines not only how long to keep things but which things, in what form, and how they will be found. Answered that way, the archive is usually smaller than the one that grows by default, and considerably more useful.

Three tiers, three requirements

The tiers have different clocks and, importantly, images are the tier that dominates both the cost and the temptation. The workable arrangement for most stations is that every part produces a verdict and a set of values, while images are kept routinely for a short window and are retained selectively and indefinitely for parts that were flagged, that were disputed, or that belong to a batch later placed under investigation. This keeps the mandatory record complete while keeping storage proportional to the number of occasions that actually matter. What ties the tiers together is the reference: a value with no part reference, or an image with no verdict alongside it, is not evidence of anything.

Retrieval is part of the requirement

An archive that cannot be searched quickly has not really been kept. Investigations have deadlines — a customer complaint, an audit, a containment decision that has to be made today — and a folder of files named by timestamp is unusable under that pressure. The retrieval question has to be answered at design time, and it is usually the same question in every case: given a part serial number, a batch number or a time window, produce the records. That means the station must be able to state the reference it was inspecting, which in turn means the reference has to arrive with the part, either from the line control system or from a code read at the station.

The other prerequisite is time. A record is findable by timestamp only if the clocks agree, and stations, line controllers and plant systems drift apart by seconds or minutes when they are not synchronised. When an investigation asks what happened between two o'clock and three, unsynchronised clocks turn a simple query into an argument. Synchronising every station to a common time source is a small piece of work that makes every later investigation materially easier, and it is worth treating as part of the station rather than as an infrastructure concern for somebody else. The granularity question — whether the record attaches to an order, a batch or an individual part — is a decision in its own right and is covered in traceability granularity.

Deciding the period

• Start from the obligation. Customer contracts, industry standards and internal quality policy may each state a minimum, and the longest of them sets the floor. Establish what those say before choosing anything.
• Then add the investigation horizon. How long after shipment could a defect plausibly be traced back to this station? A product with a multi-year warranty needs a longer horizon than one consumed within weeks.
• Then check the reference survives. There is no point keeping a record for five years if the part serial numbers are only legible in the plant system for two. Retention is limited by the shortest-lived link in the chain, not the longest.
• Then verify retrieval, not just storage. Take a part from a batch six months old and try to produce its record within the time an investigation would allow. The first attempt usually reveals that something essential was never captured.