The Alignment That Was Perfect Cold and Wrong Hot

Why this matters

This is a case where the first diagnosis was right, the correction was right, the verification passed, and the bearings kept failing anyway. That combination is worth more than a case that was simply misdiagnosed, because the trap here is not ignorance. It is a verification method that was correct in every respect except one, and the one it missed happened to be the whole fault. A shop that can find this fault stops treating "we did the thermal offset" as the end of the conversation about a set that runs hot.

The safety envelope this case ran inside

Every reading taken with the machine stopped went through the full sequence: motor disconnect open and locked, terminals proved dead with a live-dead-live check on a known live source (29 CFR 1910.333(b)(2); NFPA 70E-2021, 120.5), rotor stopped and blocked, and the suction and discharge isolated, locked, relieved and drained to a container before any flange or foot bolt moved (29 CFR 1910.147).

The one measurement in this case that had to be taken with the machine running was set up so nobody was near a rotating part. The indicators were mounted on an independent stand anchored to the floor, reading the pump casing at a point well outside the guard line and out of the plane of the coupling, with the coupling guard fitted and secured before power was restored (29 CFR 1910.219 in general industry; 29 CFR 1926.300(b) on construction work). Readings were taken visually from outside the guard line. Nothing was mounted, moved or read by reaching past a guard on a running machine.

Two more hazards belong to this job specifically. Casing and pipe surfaces on a 200 F system are far above the roughly 140 F contact-burn threshold, so every temperature in this case was taken with a non-contact infrared thermometer, and the pipe insulation was left alone rather than opened to reach metal, because thermal system insulation installed before 1981 is presumed asbestos-containing material under 29 CFR 1926.1101, with general industry work covered by 29 CFR 1910.1001. And adjusting a pipe hanger on a hot full line means working with a loaded spring under stored energy: the line was temporarily supported before any hanger was released, nobody worked under the run while it was unrestrained, and the riser hangers were reached from a properly set and secured ladder with fall protection where the exposure met the trigger in 29 CFR 1910.28 for general industry work, or 29 CFR 1926.501 on construction work.

The signal

A 40 hp motor driving a hot-water circulating pump on a 200 F loop. The pump inboard bearing had been replaced at 11 months, 14 months and 12 months. The bearing's basic rating life at the design load and this speed works out to about 60 months.

Two years earlier the shop had diagnosed this correctly. No thermal offset had ever been applied to the set. They computed one: pump shaft centerline 13.0 inches above the shim plane, cast iron at 6.0 microinches per inch per degree F, with a measured support-metal rise of 65 F, giving 5.07 mils. Motor centerline 9.0 inches above its shim plane, steel at 6.5, with a 45 F rise from its own losses, giving 2.63 mils. Differential of 2.4 mils, so the cold target was set to plus 2 mils at both coupling stations with the motor high, that being the nearest shim resolution, leaving under half a mil of residual. The record was written up properly, with the convention, the datum machine and the reasoning on it.

Then they hot-checked it and it read near zero. Everything a good shop is supposed to do had been done, and the bearing interval did not move.

What the failed bearing said

The third failed bearing was bagged rather than binned, cut open, and read. The raceway carried a band that wandered diagonally across the ring and ran hard onto a shoulder. That is the tilt signature: the rings were being held at an angle to each other in service, and the contact was edge loading against the raceway shoulder.

That single observation is what made the case tractable, because it flatly contradicted the hot check. The bearing was reporting real misalignment during running hours. The hot check was reporting an aligned machine.

Running the load back out of the life history gives a floor on the severity: median observed life of 12 months against a 60 month rating is a ratio of 0.20, so the load ratio is the cube root of 5, about 1.71 times the design load. That figure is a floor for two reasons stated in the bearing article and worth repeating here: rating life is the life 90 percent of a population reaches, so a population with a 12 month median has an L10 shorter than 12 months, and the presence of edge loading means the rating life model's own condition - the contact fully supported inside the raceway - was violated, so the calculation describes a bearing the model does not actually cover. The number's job was only to say whether this was a small problem or a large one. It said large.

Three candidates the records eliminated in an afternoon

Base or grout settlement. If the base had moved, the cold reading would have drifted off the recorded plus 2 mils. It was re-measured under the recorded cold conditions, same stations, same spacing, same brackets and sag, same bolt torque, and it read plus 2 mils at both stations. Nothing had moved cold.

Soft foot. Re-checked at specification bolt torque, one bolt at a time. All four feet under 2 mils. The frame was sitting the way it was machined.

Coupling capacity. Guard pulled on a locked-out machine: element clean, no fretting powder, no crumbs, no cracking. The coupling was not being worked past its own limit, which is a fact about the coupling and, as a sibling article makes clear, not by itself a statement about the alignment.

Three plausible explanations gone, and the contradiction between the bearing and the hot check still standing.

The contradiction that made the case

The question that broke it open was not about the machine. It was about the hot check.

Every hot check on this set had followed the shop's lockout procedure, which meant that before anybody put a dial on the coupling, the isolation valves were closed and the casing was drained. That is the correct procedure and nobody was going to change it. But it means the hot check measured the machine hot with the piping isolated, unpressurized and already starting to cool.

Set that against the cold pipe-strain check done at installation, where the suction flange bolts were loosened with indicators watching and nothing moved. That check passed with the pipe cold.

So the set had been verified in two states: cold with the pipe connected and cold, and hot with the pipe isolated and draining. It had never once been measured in the state it actually runs in, which is hot with a hot pipe connected and pressurized. Any strain that only exists when the line itself is hot was invisible to both checks by construction.

The measurement that separated pipe from machine

The test was arranged to distinguish two things that both happen when you start a machine: the machine's own metal heating, and the connected system heating. They happen on very different clocks, and that is the discriminator.

Indicators on an independent floor-mounted stand were set to read the pump casing vertically and horizontally, outside the guard line. The set was started cold with the loop cold, and the casing position and the pump support-metal temperature were logged as the system came up.

The pump's own support metal reached steady temperature 35 minutes in. The casing kept moving for the next four hours, tracking the loop temperature rather than its own, and settled at 9 mils vertical and 6 mils horizontal from its cold position. Predicted thermal growth for that support was 5.07 mils vertical and effectively nothing horizontal.

That is the finding in one line: the vertical movement was almost double the machine's own growth and the horizontal movement had no thermal explanation at all, and both of them continued long after the machine's own metal had stopped changing temperature. Machine growth follows machine metal temperature. Pipe strain follows system temperature. The clocks separated them.

What the supports were doing

Walking the hot loop with that number in hand made the rest short. The discharge riser was carried on rigid rod hangers where the design called for variable spring hangers, and the guide that was supposed to take the run's axial growth was missing. With nothing else to absorb it, the line's expansion was being delivered into the pump discharge nozzle, lifting and twisting the casing as the loop heated.

None of that is visible cold, because a cold line has not grown. None of it is visible on an isolated, draining hot check, because a line that is no longer pushing on a nozzle is not straining anything.

The repair was piping work, not machine work: variable spring hangers on the riser sized and set by the piping designer, the missing guide installed, and the anchor confirmed where the design intended it. Then the same heat-up test was repeated with the same stand and the same indicators. Casing movement over the full loop heat-up dropped from 9 mils vertical to 1.5 mils, and the horizontal movement went to under a mil.

Re-verifying the offset once the strain was gone

With the strain removed, the thermal offset became the only remaining term, and it had to be re-verified rather than assumed still correct. It was: cold reading back to the recorded plus 2 mils at both stations, and a hot check under the same lockout procedure reading inside tolerance. The difference is that the hot check now means something, because with the pipe supports doing their job there is no longer a strain that the isolation step hides.

The record was updated with both the original offset reasoning and the heat-up test result, including the 35 minute machine-metal figure and the four hour loop figure, so that the next person who sees a hot problem on this set knows which clock to look at.

Reading the outcome honestly

The bearing has not been replaced since. That is not proof yet. Against an interval history of 11, 14 and 12 months, a set that has run past those without a failure is evidence the load came down, and the honest statement is that the mechanism was identified, measured before and after, and reduced by a factor of about six on the vertical axis. The bearing life will confirm or refute it on its own schedule.

The portable lesson is the discriminator, not the piping fix. When a machine reads good hot and its bearings say otherwise, ask what state the hot check was actually taken in, and then log the casing movement against both temperatures at once. If the movement stops when the machine's own metal stops changing, it is growth and the arithmetic in the thermal articles covers it. If the movement keeps going with the system, the machine is being pushed by something bolted to it, and no amount of correct alignment work at the feet will hold it.

References

• 29 CFR 1910.147 for mechanical isolation, stored energy including a loaded spring hanger, 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 coupling guarding; 29 CFR 1910.28 and 29 CFR 1926.501 for fall protection duties in general industry and construction; 29 CFR 1926.1101 and 29 CFR 1910.1001 for presumed asbestos-containing thermal system insulation.
• ISO 281 rolling bearing rating life, for the load-versus-life relationship used to bound the severity, and its stated conditions.
• Piping designer or the system's support drawings for hanger type, spring sizing, guide and anchor locations.
• See related: "What Misalignment Does to a Bearing in Mechanical Terms"; "How to Allow for Thermal Growth When You Align Cold"; "Why a Flexible Coupling Does Not Fix Misalignment"; "How to Set Up a Shaft Alignment and Know It Held".