Defect classification: grading faults so the response matches the consequence

Grade by consequence, not by how obvious the fault looks

The natural instinct is to rank faults by how noticeable they are, which puts scratches at the top and an unmarked internal condition at the bottom. Consequence inverts that list. The questions worth asking are whether the fault can cause harm, whether it stops the product working, whether it stops the customer's assembly, and whether it is merely cosmetic. A blemish on a visible face of a consumer product may genuinely outrank a dimension slightly outside tolerance on a hidden feature, so the grading has to reflect the application rather than an abstract severity.

The safety-related grade behaves differently from the rest

Faults that could cause injury or fire, or that defeat a protective function, sit in a category with its own rules rather than at the top of a scale. They are not candidates for sampling, not candidates for acceptance under concession, and they oblige a notification path that runs beyond the quality department. Where such faults are possible, the classification exercise is the point at which they get identified and the special handling gets written down. Regulatory notification duties for unsafe product are a separate matter dealt with under compliance rather than here.

Inspectors have to grade consistently or the data is noise

Classification only works if the same fault gets the same grade regardless of who saw it and when. In practice grades drift with delivery pressure, with which customer is affected, and with how much argument a grade invites. Periodic exercises where several inspectors classify the same set of faults expose the inconsistency and, more usefully, show which definitions are unclear. Photographic examples of each grade for the specific product family do more than a written definition. So does the rule that nobody may downgrade a fault on their own once it has been raised.

The grade decides the response, which is the whole point

A classification scheme is useless unless something follows from it. Grades should map explicitly to how much material gets checked, whether a fault may be accepted under concession, who authorises the disposition, how quickly a customer is informed, and whether a formal investigation is opened or a note is enough. Making that mapping explicit is what stops trivial faults consuming engineering weeks while a serious one waits its turn in the same queue. It also removes the arguments about severity that otherwise happen at every review. Publish the mapping on a single page and keep it where dispositions are made, so nobody has to interpret it from memory during a busy shift.

Counting by grade tells you where to spend improvement effort

Once faults are graded consistently, the tally becomes a management instrument rather than a scrap report. A rising count in the least severe grade may be worth ignoring, while a single occurrence in the most severe grade justifies stopping and investigating. Weighting the count by consequence produces a very different priority list from a simple frequency ranking, and it is usually a better guide to where money should go. Plants that report only total defect numbers reliably end up working on the commonest problem rather than the most expensive one.

Frequently asked questions

Who decides the classification of a particular fault?

The grading rules should be agreed in advance between the customer's engineer and your quality function, because only the customer knows what the feature does in the assembly and in service. Applying those rules to an individual occurrence then falls to the inspector, with a review for anything at the severe end. What causes trouble is grading being decided case by case at the moment of discovery, since the answer then reflects the delivery situation rather than the fault.

Should cosmetic faults be classified at all?

Yes, and for consumer-facing products they often deserve a higher grade than engineers assume, because appearance drives returns and complaints even where function is untouched. The classification should distinguish faults on surfaces the user sees from those on hidden faces, and it should reference physical boundary samples rather than adjectives. Treating all appearance faults as minor is one of the more common ways a technically sound plant ends up with an unhappy consumer brand as a customer.

How many grades should a scheme have?

Few enough that people can apply them without hesitating, which for most plants means a small handful with clearly different consequences attached. Elaborate schemes with many levels produce hesitation, inconsistency and long arguments about boundaries, and the extra resolution is rarely used for anything. The test is whether each grade leads to a genuinely different response. If two grades produce the same actions, they are one grade with extra paperwork.