Tolerance, Repeatability and the Difference Between a Good Part and a Good Process

A machined part that arrives within tolerance and a machined process that produces the same part twice are not the same achievement, and only one of them survives contact with production. The distinction is worth making early, because a supplier who can only promise the first will quote you a price for sampling rather than for a production run.

Consider a flange. The finished part is a disc with a flat sealing face, a central bore, and a circle of bolt holes. The drawing states the face flatness, the bore tolerance, the bolt circle diameter and the positional relationship of each hole to the bore. Every one of those numbers describes the part. None of them describes the process that produced it, and the process is what determines whether the hundredth part matches the first.

Accuracy is a measurement, repeatability is a property

Accuracy is how close a single part sits to the nominal dimension. A first article can be accurate by luck — a favourable temperature, a fresh tool, a datum that happened to seat well. Repeatability is how little the process varies from part to part. It is the property that makes a batch interchangeable, and it is the one a production buyer is actually purchasing.

The two come apart in predictable ways. A machine that has lost a degree of rigidity will still produce a good part on a short warm-up, then drift as it heats through the run. A spindle bearing with play will produce an accurate bore at one depth and a different bore at another, because the error varies with tool extension. A fixture locating on a rough surface will hold the first part correctly and shift it slightly on the two-hundredth as the surface wears.

This is why a first-article inspection report proves very little. It proves the process can reach tolerance once. Production needs the process to reach tolerance every time, and the two are established by different evidence: the first by measurement, the second by the design of the process itself.

Where tolerance comes from, and what makes it move

Three contributors dominate the total error budget on a milled part, and they do not respond to correction in the same way.
























Contributor What it is What reduces it
Machine geometry and rigidity Deflection of the structure under cutting load, and geometric error in the axis alignment Rigid machine, short overhang, appropriate spindle torque, avoiding a cut that exceeds the machine's working envelope
Thermal growth The machine and the part change size as heat is generated and as ambient temperature moves Coolant management, allowing for warm-up before cutting, keeping the cutting load steady
Tool wear The cutting edge breaks down, and the part it cuts slowly changes Correct tool grade and geometry for the material, and a tool-change interval set on measured wear rather than on hours

Of these, thermal growth is the one most often underestimated. A spindle that has been running cuts a slightly different part from a spindle that has not, because the shaft has grown. A cutting process that removes a lot of heat can drift through a tolerance band over a long run even when nothing is wrong with the machine. The practical implication is that the first few parts of a run are often the least representative of the batch, and that a process specification which does not mention a warm-up requirement is incomplete.

Tool life is a process parameter, not a consumable cost

Tool cost per part is rarely the number that determines whether a milling operation is viable. What determines viability is how quickly the tool wears and how much that wear moves the dimension.

Consider a stainless steel pocket. The material is abrasive and work-hardening. A tool that is cutting well generates a small chip and stays sharp. A tool that has dulled slightly generates a rubbing action, work-hardens the wall it is cutting, and then has to cut the hardened layer. The dulling accelerates. What was a tool-life issue becomes a tolerance issue within a few parts, because the wall of the pocket follows the tool as it wears.

The practical rule is that tool-change intervals should be set on the material and the tolerance, not on a calendar. A production estimate built on a fixed tool-change interval will be accurate for aluminium and badly wrong for stainless — and wrong in the direction that costs money after the parts have shipped.

Rigid setups are a tolerance decision

Clamping is often treated as a shop-floor concern rather than an engineering one, which is a mistake, because the clamping error lands directly in the part.

Wherever a part is located during a cut, the machining accuracy of that surface is bounded by the accuracy of the location. A part held in a vise with the stock gripped over 40 mm of length is more rigid than the same part held over 15 mm, and the deflection under a wide finishing cut is proportionally smaller. A part located by a spigot is more repeatable than the same part located by a stop pin against a machined edge, because the spigot defines the position in two directions rather than one.

This is the argument for combining operations in a single setup. Each additional setup is a re-establishment of the datum, and each re-establishment consumes part of the tolerance budget. A flange faced, bored and drilled in one clamping holds the relationship between the sealing face and the bolt circle directly. The same flange faced on one machine, turned on a second and drilled on a third consumes three datum transfers, and the positional error at the bolt holes is the sum of what happened at each.

The inspection record is part of the deliverable

For a part going into a rail transit or an automotive application, the machining is a small part of the compliance story, and the record that accompanies it matters as much as the part.

A production inspection record should be able to answer three questions without further enquiry: which features were measured, what the actual values were, and which machine and tooling produced them. Measurements against the actual drawing values, not against a summary, and identified by the machine that took them — because a process that cannot be traced back to a machine and a tool cannot be corrected when it drifts.

Where a customer requires it, this extends to material traceability: the grade and heat number of the stock, tied to the cast or forged blank it came from. A machining supplier that also operates casting, forging and grinding is in a position to carry that chain, and to say exactly where within it a given defect would have originated. That is a materially different conversation from one where the finished part has no upstream history.

Choosing between processes for the same part

Several parts can be made by more than one process, and the choice is rarely decided by capability alone. It is usually decided by which operation will hold the tolerance on the feature that matters, and how many times the part has to be re-referenced to get there.

A bore that must run at a close tolerance with good surface finish is a grinding operation on most parts, regardless of how well the mill or the lathe might hold it. A face that must be flat across a large diameter is a milling operation, because a single face-milling pass achieves a flatness that repeated turning passes approach only with diminishing returns. A pattern of holes related to a sealing face belongs on the same machine as the face, so the relationship is established once.

The correct sequence for a typical component follows from this. Casting or forging establishes the near-net shape and the material history. Turning produces the rotational features while the part is still gripped on its axis. Milling then produces the faces, pockets and hole patterns referenced to those turned features. Grinding takes the features whose tolerance is tighter than the milling machine holds. cnc milling occupies the middle of that chain, and its job is to hold the relationships between the features that the previous operations created.

Choosing that middle step well is what separates a part that passes inspection from a part that keeps passing inspection. cnc milling on a rigid machine, run with tool intervals set on the material rather than on the calendar, is a repeatable operation. The same operation run with intervals set on hours is not.

What to ask before placing the order

Four questions separate a process that will produce the same part in April as it did in January from one that will not.

Which machine and which tooling will run this part? Not a class of machine — the machine. A process specified against a machine has a known envelope; a process specified against a capability has a guess in it.

What is the tool-change interval, and what is it based on? An interval stated in pieces for a specific material is a process decision. An interval stated in hours is a purchasing note.

What is the warm-up requirement before the first part is cut? The answer indicates whether thermal growth has been considered as a contributor or ignored.

What will the inspection record contain? Feature-by-feature actual measurements against the drawing, with machine identification, is the minimum that makes a production run auditable.

None of this is exotic. It is the ordinary content of a specification written by a supplier who intends to deliver the same part twice. The reason to ask for it at the enquiry stage is simply that it is far more expensive to establish after the first batch has been shipped.

The same four answers define the process behind cnc milling work generally. A machine named, an interval justified by material, a warm-up stated and an inspection record specified — that is the whole difference between a quote for a sample and a quote for production.


Google AdSense Ad (Box)

Comments