City Of Articles

Repeatability Is a Different Claim from Capability

Capability is the ability to make one part exactly right. Repeatability is the ability to make ten thousand parts acceptably right without anyone intervening. These are different claims resting on different evidence, and a supplier is judged on the second while the first is what gets photographed. The part that came off on the first attempt, dead on nominal, with a printout taped to the box, is real. It is also the least informative thing in the building, because every process produces at least one such part and most produce several per week by accident.

The failure modes are asymmetric. A shop with high capability and low repeatability ships an excellent sample followed by a distribution. A shop with low capability and high repeatability ships a consistent mediocre part and finds out immediately, which is a commercial problem rather than an engineering one. The first case is the dangerous one, because the sample is then offered as evidence of a capability the process does not have, and the discrepancy surfaces downstream, at volume, where it is expensive.

So when tolerance capability comes up, the useful question is not about the machine. A machine specification tells you what a machine can do once, in a laboratory, with ideal workholding and unlimited setup time. What a customer buys is a process that holds the same distribution on a Tuesday in the eighth week of an order as it did on the first piece. That is a property of the process, not of the asset, and for cnc milling work it is the only part of the specification that stays true.

What Actually Has to Be Controlled

Fixture setup has to survive across jobs, not merely within one. A fixture adjusted by eye to suit the part in front of it will produce that part and need rework for the next one, and the rework will repeat indefinitely. Repeatable setups are located off the part's features where possible, use hard stops rather than clamps that squeeze the workpiece, and are documented well enough that a second operator can mount the same part without guessing.

Tool life is managed as a number, not as a feeling. The decision point is not when a tool breaks, because a broken tool usually means a scrapped part and possibly a damaged fixture; it is when wear has pushed the feature out of the centre of its band. On a face or a bore that means a stated number of parts or a stated length of cut, and a tool changed at that point rather than when somebody notices the finish going off. The difference is between a dimension sitting mid-tolerance and one that walks toward an edge and stays there.

Calibration needs an interval and a record, for the measuring equipment obviously, and for the machines too, because a spindle that drifts out of position over a month will keep producing parts that are individually acceptable and collectively out of position. Datum strategy has to be independent of how the part happened to be loaded yesterday; if inspection locates a feature from a face the operator picked because it looked clean, the measurement reports fixture luck rather than part geometry.

Operator-to-operator consistency is mostly a documentation problem, and it matters more in cnc milling work than the machine count suggests. The steps that get skipped are always the ones that feel obvious to the person who knows the job, which is why they vary. A route sheet naming the tools, the offsets and the inspection points does more for repeatability than a training session, because it survives the departure of the person who gave the training.

Sequencing is the least visible control and one of the most effective. Jobs that share a setup or a workholding arrangement should be grouped, so a setup is built once and amortised across a family of parts rather than rebuilt for every part number. This reduces the number of setups that can be subtly different from one another, and it gives the operator a repeatable series instead of a stream of unrelated one-offs. A shop that sequences deliberately has fewer processes to control, and each gets exercised often enough for drift to show up before it becomes scrap.

Why the First Article Off a New Machine Is Often the Best

It is common for the first part off a fresh setup to be the best part of the run. Fresh offsets, a clean setup, an attentive operator, a warm spindle and a workpiece not yet carrying enough heat to have shifted anything all line up in the same direction. The measurement is taken under favourable conditions and the result reflects them.

Treating that as evidence of capability is the error. The first article tells you what the process can do under ideal conditions; it says nothing about part four hundred, on the second shift, after three tool changes. The informative number is not how good the first part was, but how far the hundredth part drifted from it, and that number exists only if somebody measured the hundredth part too.

The Question Worth Asking

What is your tolerance capability invites a general answer, and general answers are easy to give. What is your Cpk on this feature, in this material, at this tolerance, over how many pieces, measured how, invites a specific one. Cpk describes how much room a process has between its natural spread and the tolerance limits, so it moves when material, tool, fixture, operator or environment moves. It cannot be quoted as a company figure, only for a feature on a part, and asking for that forces the conversation onto the process rather than the brochure. It also forces the supplier to state the measurement conditions, because an index taken on hot parts in the fixture is a different number from one taken cold.

A good answer states the sample size and the measurement method, and it often comes with a figure that is not flattering. A supplier willing to quote a marginal index on a difficult feature is telling you they have measured the process rather than described it. One who answers with a large number for every feature, without a sample size, has measured one setup on one day.

The same logic applies to a shop advertising a wide process range. Listing milling, turning, grinding, casting, forging and sheet metal work is not a catalogue claim; it is a claim about how many processes the organisation can hold under control at once. That means carrying a tolerance consistently across parts that arrived as castings, as forgings, as bar stock and as assemblies, each with its own residual stress and machinability, rather than only across parts that were convenient to machine.

Owning more machines does not deliver that. Breadth of capability is inventory; breadth of control is competence applied repeatedly across inventory, and the second is harder. A shop running twenty machines on one part has one process to control. A shop running twenty different parts has twenty, with the setup, datum scheme and inspection plan varying between them. That is why parts sent out on an ordinary Tuesday are more informative than those produced when a visitor is watching.

Capability is what a supplier can show you once. Repeatability is what they can show you every time, with documentation. For a buyer, only the second matters, and the reliable way to tell them apart is to ask for evidence rather than an impression: the Cpk on the feature, the sample size behind it, the tool change interval in the route sheet, and the conditions under which a part is measured. On cnc milling jobs, those answers separate a controlled process from a lucky setup faster than any capability claim.