In-process inspection: catching drift while the material is still cheap

Check before value is added, not after

The natural place for an in-process check is immediately before an operation that is expensive, slow or irreversible. Verifying a bore before a part goes for coating, or a weld before an assembly is closed up, protects everything spent afterwards. Checking straight after the operation that creates the feature has its own logic, because feedback reaches the machine that caused the problem while the settings are still known. Most plants need both kinds and confuse them, ending up with checks positioned where records are convenient to collect rather than where a fault costs least to catch.

Setup approval carries more weight than anything that follows

A large share of batch losses trace back to the first piece after a changeover: wrong offset, wrong programme, wrong material, fixture not seated. A formal first-off approval before the run is released stops that class of failure outright, and it is the highest-return check in most machine shops. It has to be a genuine stop, though. Where the operator sets, checks and starts within the same minute while the machine is already cycling, the approval has become a signature rather than a gate. Independent confirmation for the first piece off a changeover repays itself on the first save.

Everything since the last good reading is at risk

The quantity exposed by any checking interval is simply what the process makes between looks. On a slow machine producing large components that may be a handful of pieces. On a press running quickly it can be a full container before anyone notices. Since containment cost scales with that number, the interval is really a decision about how much material you are willing to put in question at once. Where output is high and separation poor, the practical answer is often to add an automatic in-cycle check rather than to ask a person to look more often, though the equipment side of that belongs to the automation discipline.

Patrol checking and its blind spot

Roving inspection covers several machines by sampling each in turn, which suits low-volume shops and mixed lines where dedicated checking cannot be justified. Its weakness is that the timing is uncorrelated with the events that matter. Patrols arrive after changeovers rather than at them, miss the moment a tool starts to wear, and provide reassurance based on samples taken at arbitrary points. Used well, patrols audit whether the operator checks are being done properly and pick up problems nobody reported. Used as the primary defence, they leave long unobserved intervals that nobody has quantified.

Recording enough to reconstruct the run later

In-process results earn their keep twice: once when they trigger an adjustment, and again months later when a customer complaint has to be bounded. The second use requires the readings to be tied to something identifiable, which usually means the container, the shift and the tool. Values written on a sheet that lists neither the batch nor the machine are almost useless for containment, however diligently they were collected. Deciding what identifier goes on the record is a five-minute conversation at the time and the difference between recalling a container and recalling a month of production.

Frequently asked questions

Should the operator or a separate inspector do in-process checks?

Operators for the routine measurements, because the value lies in correcting the process immediately and nobody else is standing at the machine. Reserve independent checking for the first piece after a changeover, for judgement-based characteristics, and for occasional audits of the operator checks themselves. That mix keeps the cost sensible while removing the two situations where self-checking is weakest: the moment of highest risk, and the readings where the operator has an interest in the answer.

What is a first-off check and how does it differ from first article approval?

A first-off confirms that this particular setup, on this shift, is producing acceptable parts before the run is released, and it happens at every changeover. First article approval is a one-time exercise proving the part and process as configured can meet the drawing, done at introduction or after a significant change. One protects a batch, the other qualifies a process. Plants that treat them as the same activity usually do the more expensive one badly and the cheap one not at all.

How do we stop in-process records being completed after the fact?

Design the record so backfilling is obvious and pointless: entries tied to a container or a timestamp, readings that vary rather than repeat, and periodic comparison of the sheet against machine data. Beyond that, look at whether the plan asks for more checking than the shift can perform, which is the usual reason people write results from memory. Fixing the workload almost always achieves more than a disciplinary conversation about honesty.