"Tight tolerance" is not a number, it is a set of conditions
Almost every machining supplier's capability page contains a sentence about producing complex parts with tight tolerances. It is the least informative sentence in the industry, and it sits on the pages of excellent shops right beside the pages of shops that are guessing. A tolerance is a number, but a usable tolerance is that number plus the conditions that decide what it is measured from, under what material condition, against what surface finish, and by what method the result is judged. Strip those conditions away and the number becomes an adjective.
That requirement usually reaches a shop as a line in an email body rather than as a feature on a drawing: the part is complex, the tolerances are tight, please quote. At that point a responsible cnc milling supplier cannot price anything, because the drawing carries the requirements and the email carries only an impression of them. The first useful move is to ask for the drawing, and then to ask which features are the tight ones. The answer is almost never "all of them."
Five conditions that make a tolerance mean something
The control type and its band. A band is only interpretable once you know whether it applies to a size feature, a form feature, a profile, or an orientation relationship. A linear band on a diameter and a parallelism control on a face both constrain material removal, but they constrain different things and are verified differently. Quoting a band without knowing which control is meant is estimating, not quoting.
The datum scheme. This is the largest source of ambiguity in machining drawings, because a tolerance is always measured from something. If the drawing does not state which features establish the datum, in what order, and whether they are simultaneous or accumulated, the same nominal part can pass in one setup and fail in another. Establish the bore from the mounting face, or the mounting face from the bore, and you have two tolerance stacks with two costs and two yield patterns. Two suppliers can quote the same print in good faith and still describe different parts.
The material condition. Size features can be controlled at maximum material condition, at least material condition, or as truly independent features. The choice decides who carries the risk of accumulated stack-up. Under a worst-case independent scheme the part is guaranteed within limits at its worst combination of features. Under a maximum material condition scheme the guarantee comes with bonus material in the feature, and that bonus pays for the accumulation. This changes yield, and yield is a large part of the price.
The surface finish requirement. A dimensional tolerance says nothing about the texture of the surface it applies to. For a sealing face, a bearing seat or a sliding fit, texture is frequently what decides whether the part functions at all. Finish is specified as a roughness parameter and verified by a trace across the surface, not by the instrument that checks the dimension. That is a second inspection, usually a second setup.
The inspection method. Who measures the part, with what instrument, and to what measurement uncertainty decides whether it ships. A feature accepted on a shop-floor comparator, on a coordinate measuring machine, or by a functional test are not equivalent claims. As a band approaches instrument uncertainty, the method becomes part of the requirement rather than an implementation detail.
The same nominal dimension, two entirely different parts
This is where the vagueness turns into money. Consider one nominal bore in two unrelated assemblies. In a non-critical housing it is a clearance fit for a sliding or rolling element: a modest band, a modest finish, and a bore gauge settles the argument. In a hydraulic circuit the same nominal bore sits inside the sealing path, so its size, its roundness, its position relative to the mating face, and the texture of its wall all influence whether the joint holds pressure. Those features must be inspected as a set, because a bore correct in diameter but tilted to the face will still leak.
Nothing about the number changed. One instance is a fit and the other is a seal, and the seal version is not the same feature with a tighter band; it is a different feature that happens to share a nominal diameter. This is the trap in treating cnc milling as a capability statement: what a process holds depends on feature type, size, material condition and setup rigidity, not the process name.
A flange face is three requirements wearing one label
Flange facing is where the vagueness becomes almost embarrassing, because everyone calls a flange face "flat." A flange is a critical component in piping, mechanical systems and industrial equipment, existing to create a secure, leak-proof connection that withstands high pressures and temperatures. The face doing that work carries at least three independent requirements, and any one can go out while the face still looks flat.
| Requirement on the flange face | What it controls | What fails when it drifts | Usual evidence |
|---|---|---|---|
| Flatness across the sealing area | Even contact under bolt preload | Over-compression one side, a leak path the other | Optical flat or a height scan |
| Surface texture of the sealing area | How the gasket or metal seat beds in | Uneven compression and weeping below working pressure | Profilometer trace |
| Squareness of the face to the bore axis | Concentric distribution of bolt load | Bolts pull unevenly, seal misaligned to the bore | Face position against the datum axis |
| Bore roundness and concentricity to the face | Even seating around the bolt circle | Cross-bolting and edge loading of the gasket | Roundness scan, bore-to-face position |
Note what that table does not contain: a single number called "precision." Each row is a different control, verified differently, and each can be satisfied while the others fail. A drawing that says "flat flange face, good tolerance" has specified none of them. That is why two quotations for the same flange can differ widely without either shop being careless, and why the cheaper one is often quoting a face that is flat, square and sealed only mid-range.
The same decomposition applies elsewhere: a bore is a fit and a seal at once, a mounting face is a datum and a thermal path at once. Collapsing either into one callout makes the shop judge it twice from the same information.
What a buyer should be able to state so the quotes match
For a flange face, a comparable quotation requires five items on the drawing: the flatness control and how it applies across the sealing area, the surface texture requirement expressed so a trace can be judged against it, the datum features and their priority, the size and geometric controls on the bore together with their position relative to the face, and the method by which conformity will be demonstrated.
It is also worth putting one more question in writing before the price comes back: which operation establishes the face, and is it cut in the same setup as the finished bore or referenced from a separate support. The answer decides whether the face is controlled against the bore directly or through an indirect chain, and whether squareness is achievable in one clamping or must be recovered by hand fitting.
None of this makes the trade harder. It makes the trade legible. The industry sentence is not dishonest, only incomplete, and incomplete specifications get resolved by whoever asks a precise question first. The datum scheme, the material condition, the finish, the inspection method and the setup logic are the conditions, and those conditions are what actually are the machining. A cnc milling capability stated in those terms is one an engineer can compare against another; one stated as tight tolerances is a claim taken on faith.
Comments