City Of Articles

Metrology Is a Process Step, Not a Final Inspection

The most common mental model of a machined part is chronological rather than causal. You cut it, and then you measure it. The tool makes the part; the inspection reports the part. That sequence feels like common sense, and it survives for years, because nothing about it looks wrong until a batch that passed inspection on the shop floor fails on the customer's assembly line and nobody can reconstruct what the real geometry was.

The flaw is a timing error, and it is not subtle. Between the final cut and the reading on the gauge, the part is still moving. Heat from the cut zone is still working outward through the section. Residual stress inherited from the bar stock, and rebalanced by the material that has been removed, is still redistributing. And the part, which has been held in a shape by a fixture for the last several operations, has just been unclamped and is now finding the geometry its internal forces actually want. The reading taken in that window is a real number. It is simply not the number that describes the part the customer will receive.

Shops doing serious cnc milling work discover this through rejection rather than through theory. A part clears inspection at 09:40, sits in a tray, is boxed at 11:15, and arrives at 14:00 with a bore two tenths of a millimetre smaller than the report states. Nothing failed. The report is not falsified; it is simply true about a moment that nobody thought to specify.

A Part Cut Without a Relaxation Plan Was Never Measured Correctly

Consider a thin flange face, the kind used in piping, mechanical assemblies and industrial equipment where the flange has to seal against a mating face through a defined gasket line. A thin face is exactly the geometry where stress relief moves dimensions. It is stiff out of plane, because plate stiffness scales sharply with thickness, but essentially free within the plane, because nothing restrains it there. Machine one side heavily and finish the other lightly and the section acquires a bending moment that wants to dish the face. Clamped, it stays flat. Unclamped, it relaxes.

Flatness measured while the part is still located in its machining fixture is a measurement of a constrained plate, not a free plate. The two readings are related, but the relation is not constant. It depends on how much material came off each side, in how many passes, and in which direction the tool travelled. None of that appears on the drawing, and all of it is decided at the setup. Which means the drawing cannot tell the inspector when to look, so the setup has to.

A bore is the same problem in different clothing. On an electric motor stator housing, the bore accepts a rotor and usually also locates it, so its size and its relationship to the mounting face are both live dimensions. Cut that bore hot, with heavy cuts and a full spindle load, and the material at the bore is expanded. The gauge reads large. The same bore twenty minutes later, cleaned and at ambient, reads smaller. If the housing is then matched to a shaft selected against the hot reading, the fit that looked correct on the bench becomes a press fit on the floor.

A Wrong-Moment Measurement Is Random, Not Conservative

The usual assumption is that measuring early is the safe choice: look while it is still hot and you will catch it while it is still bad. The assumption is false, because thermal change and stress relief do not have a single direction. A face can dish on release and a bore can shrink on cooling, but a length can grow, and a feature's position can move toward or away from the datum depending on how the cutting forces were distributed through the section. When the sign is not predictable, an early measurement is not conservative. It is a sample from an unknown distribution at an unknown offset, and the error it carries is not bounded in a direction you can rely on.

That distinction has a direct commercial consequence. A part that is out of tolerance on the bench and in tolerance in service wastes material. A part that is in tolerance on the bench and out of tolerance in service stops a line. Both come from the same mistake, and the second one is the expensive one.

cnc milling operations that handle this treat measurement as a process step with its own inputs. There is a defined dwell between the last cut and the first reading. There is a defined cleaning step, because a part still carrying cutting fluid is a part whose thermal state is still moving and whose surface is being measured through a film. There is a defined ambient condition. And there is a defined restraint condition, which normally means unclamped, supported so as not to load the feature being measured, and located on the datum scheme the drawing specifies rather than the one that happened to be convenient at the machine.

What Changes When You Measure the Part Again

Measure a part at the machine, then clean it, leave it at ambient for an hour, and measure it again on the same equipment against the same datum, and the two sets of numbers will usually differ. On a stable process the difference is small and consistent in direction, and that consistency is itself the useful finding: the relaxation is predictable and can be allowed for. On an unstable process the difference is large and inconsistent in direction, which means the process is not under control regardless of what either reading says.

This is why the phrase the parts passed inspection is a statement about timing rather than about the parts. It carries no information about the condition the parts were in when they were read. Two parts from the same lot, one measured hot in the fixture and one measured cold and unclamped, can both pass and still be materially different. The report is valid for both. The conclusion drawn from it is not.

The practical consequence for anyone specifying work is to ask about the process rather than about the report. Ask what the dwell is before measurement. Ask whether the part is measured in the machining setup or after removal. Ask whether the inspection equipment is calibrated on an interval and whether the records exist. Ask whether the feature with the tightest tolerance is verified in the condition in which it will function. A supplier with direct answers is describing a process. A supplier with vague answers is describing a habit.

The same logic applies to the whole route, not just the last operation. Deburring, washing, plating, passivation and handling each add their own timing and thermal effects, and a part verified correctly after milling can still move before it ships. For cnc milling work carrying tight flatness on a thin section or a tight bore-to-face relationship on a housing, the inspection plan belongs at the end of the full process, and the conditions at that point have to be specified rather than assumed.