How to Allow for Thermal Growth When You Align Cold

Why this matters

Nobody aligns a machine hot. Everybody runs it hot. So the alignment you set cold has to be deliberately wrong, by a known amount, in a known direction, so that it becomes right at temperature. The arithmetic for the amount is short and a sibling article covers it. Everything that actually goes wrong on these jobs is bookkeeping: the offset applied to the wrong machine, the sign inverted, the correction entered twice, or a correct offset stripped out six months later by the next tech who found a cold reading that was not zero and helpfully fixed it. This is the bookkeeping.

Isolate, and treat the startup and the hot check as separate hazards

Open and lock the motor disconnect and prove the terminals dead with a live-dead-live check on a known live source (29 CFR 1910.333(b)(2); NFPA 70E-2021, 120.5). Then under 29 CFR 1910.147: confirm zero rotation, block the rotor, and isolate, lock, relieve and drain any hot or pressurized line before a foot bolt or a flange moves. Turn shafts by hand with a bar on a barring point with the disconnect still locked, and re-block the rotor before any hand enters the coupling gap.

This procedure ends with a run, and the run has its own requirements. Refit and secure the coupling guard before power is restored, because it is a required guard on power-transmission apparatus under 29 CFR 1910.219 in general industry and 29 CFR 1926.300(b) on construction work, and stand out of the plane of the coupling on the first start. The hot check afterwards goes back through the full sequence: shut down, isolate, lock, prove dead, confirm zero rotation, then read. Machine surfaces will still be above the roughly 140 F contact-burn threshold long after shutdown, so read them with a non-contact infrared thermometer and wear gloves rated for the contact temperature if you must touch metal.

The one rule, written so the sign cannot slip

Cold offset equals the negative of the differential growth of the movable machine relative to the stationary machine, measured in the same direction convention, applied as a target reading at the coupling stations.

Every word in that sentence is load-bearing, and the four sections below are the four places it goes wrong.

Sign point one: declare the datum before you compute anything

Pick the stationary machine and write it down. In a field-service shop it is almost always the hard-piped one, because moving it means moving pipe.

Then compute growth for both machines and take the difference in one direction only: movable minus stationary. If the movable machine grows less than the stationary one, that difference is negative, and the offset is positive. Doing this consistently is more important than doing it intuitively, because intuition on this is unreliable and gets reversed roughly half the time.

Skip the declaration and you get the error that produces a doubled misalignment: two people on the same job, each mentally treating the other machine as fixed, and the offset applied in opposite directions on the vertical and horizontal passes.

Sign point two: declare positive, once, on the sheet

Write "up is positive at the coupling stations" at the top of the record before any number goes on it. Growth figures, cold targets, hot targets and measured readings all live in that one convention. Mixing a growth figure quoted as "the pump rises 5 mils" with a reading convention where a high shaft reads negative on the indicator is exactly how a correctly computed offset lands upside down.

Sign point three: the offset is a target reading, not a shim quantity

This is the step that eliminates double-application, and it is why the offset is expressed at the coupling stations rather than at the feet.

Do the alignment exactly as you would for a cold machine, with one change: instead of driving both station readings toward zero, drive them toward the cold target. The shim arithmetic, the foot projection and the validity checks are unchanged. Nothing extra is added anywhere.

The failure this prevents is real and common: a tech computes a 3 mil offset, adds 3 mils of shim to both feet to "put it in", then aligns the set to zero at the coupling, which removes the 3 mils they just added. The same failure appears with laser tools when the offset is entered into the instrument and also shimmed for, applying it twice. The offset lives in exactly one place, and that place is the target reading.

Sign point four: know when growth is a translation and when it is a tilt

The clean case, and the one the worked example below uses, is a machine whose feet all sit the same distance below its shaft centerline. That machine's growth is a pure vertical translation of its centerline, so the differential between two such machines is a pure parallel offset with no angular component, and the cold target is the same number at both stations.

That condition fails when a machine's supports are bolted at different elevations - a fan or a between-bearings machine with one pedestal on a plinth and the other on the floor. Then each support has its own height above its own shim plane, each grows by its own amount, and the shaft tilts. Run each support separately: a drive-end pedestal 6.0 inches above its shim plane and an outboard pedestal 14.0 inches above its own, both steel at 6.5 microinches per inch per degree F and both rising 30 F, grow 1.17 mils and 2.73 mils. The difference is 1.56 mils across the pedestal spacing, which over a 30 inch spacing is 0.052 mil per inch of tilt. Small, but it is angular, so it goes into the cold target as a different number at each station rather than as one number at both.

Worked example: the ledger for a hot-water pump set

Pump is hard-piped and stationary. Motor is movable. Both machines have all feet at equal height below their own shaft centerline, so both grow as pure translations. Predicted growth, taken from the mechanism article's method with measured support-metal rises: pump plus 5.0 mils, motor plus 2.1 mils. Convention: up is positive at the coupling stations, differences taken as movable minus stationary. Coupling stations 6.0 inches apart; motor front foot 9.0 inches and back foot 24.0 inches from station 1.

Differential growth of the movable relative to the stationary: 2.1 minus 5.0 equals minus 2.9 mils. The motor falls 2.9 mils relative to the pump as the set heats.

Cold offset: the negative of that, so plus 2.9 mils. The motor is set high cold.

Sanity-check it forward rather than trusting the rule. Cold, the motor sits 2.9 mils above the pump line. Hot, the motor has risen 2.1 to sit at 5.0 absolute, and the pump has risen 5.0 to sit at 5.0 absolute. Relative offset hot: zero. The rule holds and the direction is confirmed, which is worth the ten seconds because this is the step people get backwards.

Cold target at the stations: plus 2.9 mils at station 1 and plus 2.9 mils at station 2, because the differential is a pure translation and carries no angle. Round to available shim resolution and call it plus 3 mils at both stations.

Foot projection: the target line is y equals 3 mils, flat, so the projection to the feet is 3 mils at the front foot and 3 mils at the back foot. Both feet move the same way because the target has no slope. If the target had carried the 0.052 mil per inch tilt from the pedestal case above, the two feet would differ by 0.052 times the 15.0 inches between them, which is 0.8 mil, and rounding both to the same value would throw that away.

Axial: the pump grows away from its anchor along the shaft as well, which changes the distance between shaft ends. Set the cold gap so that the hot gap lands on the coupling's specified between-shaft-ends dimension, and on a sleeve-bearing motor treat that as a hard requirement rather than a preference, because the end-float window is what keeps the rotor off its thrust face.

What the record must say. Not "aligned." The record says: up is positive at the coupling stations, pump is the stationary datum, predicted differential growth minus 2.9 mils, cold target plus 3 mils at both stations, achieved cold reading, station spacing, foot distances, bolt torque, and machine condition at the reading.

The failure mode this record prevents is the most common one in the whole subject. Six months on, a tech re-reads the set cold, finds plus 3 mils at both stations, sees a number that is not zero, and shims it to zero. The set is now correctly aligned cold and 2.9 mils out hot, permanently, and there is nothing in the file to tell anyone why the bearings started failing. A thermal offset that is not written down and explained on the equipment record has a short life expectancy, and the person who removes it will be doing careful work.

What changes the answer

A centerline-mounted casing removes most of the term. Hot pumps are often built with their feet at the shaft centerline elevation rather than at the bottom of the casing, precisely so that the height from the shim plane to the shaft centerline is near zero and vertical growth with it. If the machine you are working on is centerline mounted, do not carry over a growth allowance computed for a foot-mounted one; recompute with the actual height, which may be small enough to ignore.

A cold service reverses the sign. On chilled water or refrigeration duty, the process machine's support contracts while the motor still grows from its own losses, so the differential and therefore the cold offset both flip. Compute it, do not pattern-match from the last hot job.

An intermittent machine has no steady state. A set that runs in short cycles never reaches the temperature the calculation assumes, so the true growth sits somewhere between zero and the computed figure. On those, either take the manufacturer's published growth or run a hot check, and do not apply a full-value calculated offset to a machine that never gets there.

Verifying the offset actually worked

Run the machine to steady operating condition, then do a hot check under the safety sequence above and record the elapsed minutes from shutdown to reading alongside the numbers, because the machine starts cooling immediately and that elapsed time is part of the measurement.

Compare the hot reading to zero, not to the cold target. If the hot reading is inside tolerance, the offset was right and the record should say so with both numbers on it. If the hot reading is out in the same direction as the original growth, the offset was too small. If it is out in the opposite direction, the offset was too large, which usually means the support-metal temperature rise used in the prediction was the process rise rather than the measured metal rise. Correct the cold target by the measured hot error and re-align; one iteration of measured correction beats any second attempt at predicting.

References

• 29 CFR 1910.147 for mechanical isolation, stored energy and rotor blocking; 29 CFR 1910.333(b)(2) with NFPA 70E-2021, 120.5, for the motor circuit; 29 CFR 1910.219 with 29 CFR 1926.300(b) for guarding the coupling before restoring power.
• Machine manufacturer documentation for published thermal growth, designed anchor points, mounting arrangement (foot or centerline), and the between-shaft-ends dimension and end-float window.
• See related: "What Thermal Growth Does to an Aligned Machine"; "How to Set Up a Shaft Alignment and Know It Held"; "The Alignment That Was Perfect Cold and Wrong Hot".