MILLWRIGHTQC
← Back to Millwright QC
Last updated 12 August 2026
Four things that cost money on precision-maintenance jobs, and are not obvious until the second time they happen to you. Each one produces a record that looks fine.
A machine that runs hot does not stay where you left it. Aligned to zero cold, a hot pump is misaligned every hour it runs — and the sheet says it passed.
Steel grows about 0.0000063 inches per inch per degree Fahrenheit. That sounds like nothing until you put real numbers through it. A pump with its shaft centreline fourteen inches above the baseplate, running product at 300 °F against an 80 °F shop ambient, lifts roughly twenty thousandths by the time it is at temperature. The motor beside it, running cooler and sitting lower, barely moves.
So the alignment you set to a perfect zero on a cold Saturday is twenty thousandths of vertical offset on Monday afternoon, and it stays there for the life of the run. The coupling absorbs it, the bearings absorb what the coupling does not, and the machine is consuming its own service life on every rotation.
What makes this expensive rather than merely wrong is that the paperwork is immaculate. As-left reads 0.000. The tolerance box says Excellent. Nobody looks at it again until a coupling fails at eight months, and by then the alignment sheet is the strongest evidence that alignment was not the problem.
Align to a cold target rather than to zero, so the machine walks into alignment as it heats. Record everything the target depends on:
A hot check settles the argument permanently. Run the machine to temperature, shut down, and shoot it again inside the few minutes before it cools. Tedious, and it turns every future alignment on that machine into arithmetic instead of estimation.
The practice worth arguing about: “We always align to zero.” That is correct for a machine at ambient and wrong for anything that runs hot, and the two are not distinguishable from the alignment sheet afterwards. If the targets are zero, say so and say why.
Four thin shims and one thick shim are the same total thickness and are not the same machine. One holds; the other relaxes, and the alignment walks off within weeks.
Under load, a stack of thin shims does not act like solid steel. Each interface has surface roughness, a trace of oil, and sometimes a burr, and every one of those compresses slightly when the bolt is torqued and creeps for days afterwards. Four interfaces creep four times as much as one.
The result is a machine that was genuinely aligned when you left, and is genuinely out three weeks later, with nothing in between to explain it. The alignment record says the job was done properly. It was. The shim record — if it recorded only the total — cannot show why it did not hold.
Rust and paint make it worse in the same way, and dirt makes it worse instantly rather than slowly. A shim pack assembled on a foot that was not cleaned is a pack with grit acting as a fifth interface.
The practice worth arguing about: “It totals right, so it is right.” Thickness is one of two variables and the record usually captures only that one. The count is free to write down and impossible to reconstruct later.
Fit and internal clearance are two different numbers and they interact. Get the fit right and the clearance class wrong and the bearing is preloaded from the moment it is mounted.
An interference fit works by stretching the inner ring onto the shaft. That stretch does not disappear — it comes out of the bearing’s internal clearance. Press a C3 bearing onto a shaft machined for a heavy interference and you may land near the clearance a CN bearing would have had. Do the same to a CN bearing and you can arrive at zero, or at preload, which is a bearing running hot from the first minute with no fault anyone can point to.
Heat compounds it. A bearing mounted with an induction heater and then run immediately under load, before the inner and outer rings equalise, is a bearing whose clearance is temporarily wrong in the same direction. It survives; it just does not last.
What makes this hard to catch is that every individual step passes inspection. The shaft is on print. The housing is on print. The bearing is the specified part number. The failure is in the combination, and no single measurement on the sheet is out of tolerance.
The practice worth arguing about: “Same part number as the one that came out.” If the one that came out failed early, matching it reproduces the failure. The old bearing’s clearance class is data about a machine that broke, not a specification.
Not whether the numbers are good. Whether the record can be trusted — which is settled before anyone reads a value.
Auditors work in a fixed order, because the cheap checks eliminate most packages before the expensive ones are needed:
Notice that none of these is about workmanship. A crew that did excellent work and recorded it loosely fails all five; a crew that did ordinary work and recorded it completely passes. That is not unfair — the auditor cannot see the work, only the record of it.
The same machine under three names is three machines to anyone who was not there.
Tag drift is the quiet one. P-1201A on the alignment sheet, P1201A on the vibration record, 1201-A on the timesheet. Every one is legible to a human and none of them matches, so nothing can be pulled together without someone reading all three and deciding they are the same. On a job with two hundred records that decision has to be made two hundred times, and it will be made wrong at least once.
Trains are the other one. A motor, a gearbox and a pump are three tags with two couplings between them. An alignment recorded against “the pump” leaves the reader guessing which coupling was shot — and since a train is aligned one coupling at a time in the field anyway, the record should say which.
Both cost the same thing: not rework, but hours of somebody senior reconciling paperwork at the end of a job, usually under time pressure, usually while the client waits.
These are field observations, not standards. Where a number matters — a tolerance, a clearance class, a mounting temperature — take it from the OEM manual for the machine in front of you and the edition of the standard your job is bought to. The machine’s own documentation outranks everything general.
Field notes — millwrightqc.com/field-notes