The New Motor That Shook and the Old One That Did Not
Why this matters
A motor is replaced with one that matches the nameplate, and the machine that ran acceptably for years now shakes hard enough that the customer calls before the invoice does. The instinct is that the new motor is defective, and the usual next move is a warranty claim or a rotor balance. Both were wrong on this job. The interesting part is that by the time anyone thought to investigate, the fault had already been made to disappear by putting the old motor back, so there was nothing left to measure in the failed state. The diagnosis had to be reconstructed from paperwork, two nameplates, and one test that could be repeated in both configurations. What the paperwork was missing turned out to be as instructive as what it held.
Before anyone goes back on the roof
The reconstruction below involves an impact test, a motor swap and readings on a running machine, on a rooftop unit. Lock and tag the disconnect and confirm the fan wheel has stopped rather than slowed before a hand or a hammer goes near the assembly, per 29 CFR 1910.147 and its verification-of-isolation step, because a large wheel coasts for minutes on good bearings. Rewiring or metering at the motor terminal box or the starter is electrical work under 29 CFR 1910.333(b)(2), not 1910.147, which excludes electric utilization installations at 1910.147(a)(1)(ii)(C); prove the conductors dead using the live-dead-live sequence in NFPA 70E-2021, 120.5. Nobody positions under a motor while it is rigged, and the sheave key is removed or positively retained before any shaft is turned under power, since an unrestrained key leaving a shaft at speed is a projectile. Fall protection on the roof is a duty, under 29 CFR 1910.28(b)(1) for service work on an existing building or 29 CFR 1926.501 where the job falls under construction, and the two Parts trigger at different heights. Running readings come from outside an intact guard with a magnetic-base sensor on a stationary housing, per 29 CFR 1910.219.
What the file actually contained
The unit is a belt-driven rooftop supply fan. The service history had five useful items.
- The old motor's nameplate, photographed at removal: four-pole, full-load speed 1,740 rpm.
- The replacement motor's nameplate: same horsepower, same voltage, same enclosure, premium-efficiency, full-load speed 1,775 rpm.
- An invoice line for a shop-fabricated adapter plate, because the replacement's frame did not match the existing rail hole pattern.
- A vibration reading from a PM visit fourteen months before the swap, logged as elevated but inside the shop's pass band, with no note and no follow-up.
- The customer's own timeline: no complaint before the swap, complaint the same week after it, complaint gone when the rebuilt original went back on.
Nothing in that file names a defect. The nameplates match on every field a purchaser compares. Full-load speed is the only one that differs, and 35 rpm out of 1,740 is 2 percent, which nobody treats as a difference.
What the file did not contain
Three gaps, and each one is the reason a question could not be answered from the record.
No weight for either motor. Neither the removal line nor the install line recorded it. A premium-efficiency motor of the same rating commonly carries more active material and weighs more than the older design it replaces, and that is a mechanical change to the assembly which no paperwork in this file captured.
No description or weight for the adapter plate. It appears as a fabrication charge with no drawing, so its mass and its stiffness are both unknown from the record.
No natural-frequency baseline of any kind. No coast-down, no impact test, ever. There was therefore no way to answer "did the assembly change" from documents alone, only by re-creating both states physically.
That third gap is the expensive one, and it interacts with the fourth item on the list above. A reading logged as elevated but passing, on a pass/fail band with no trend and no frequency information, is exactly the record that hides a machine already sitting on the shoulder of a resonance. The number was reported honestly and it told nobody anything.
The one test that could be run in both states
The old motor was still on the shelf, so both configurations could be rebuilt and the same measurement taken twice: a soft-face hammer on the motor pedestal and rails, machine locked out and stopped, reading where the assembly rings.
- With the replacement motor and the adapter plate installed: 1,798 cpm.
- With the original motor back on its original rails: 2,050 cpm.
That is the finding, and it is measured rather than inferred. The assembly's natural frequency dropped by 252 cpm, about 12 percent, when the motor was changed.
Working backwards through the frequency relationship gives the size of the change that caused it. Frequency falls with the square root of mass, so the participating mass ratio is (2,050 / 1,798) squared, which is about 1.30. Roughly 30 percent more mass moving in that mode, consistent with a heavier motor plus a fabricated plate, and the two cannot be separated from this measurement alone. The mode in question is the motor pedestal and rail assembly, in which the motor is the dominant participating mass, so a heavier motor moves it directly.
What that did to the machine
Now put running speed against each frequency.
Original configuration. Motor at 1,740 rpm against a 2,050 cpm mode: ratio 0.849. At an assumed 5 percent of critical damping, typical for a bolted steel assembly and not measured here, the amplification works out near 3.4.
Replacement configuration. Motor at 1,775 rpm against a 1,798 cpm mode: ratio 0.987, essentially on the peak. Amplification near 9.8.
The ratio between them is about 2.9. The residual unbalance in the two rotors is not identical, but neither is out of tolerance, so treat the force as comparable and add the 2 percent speed increase, which raises unbalance force by about 4 percent since force goes with the square of speed. The predicted change in motion is therefore around three times, with the same rotor quality, the same bearings, and the same drive.
That number leans on the damping assumption, so it is worth bounding. Across the damping range plausible for this assembly, roughly 3 to 8 percent of critical, the predicted increase runs between about two times and four and a half times. The direction and the order of magnitude hold across the whole range, which is what the conclusion actually rests on. The single figure of three does not.
Note the second half of the finding, which is easy to miss: the original configuration was already amplifying its own unbalance more than threefold. It was never a quiet machine on a stiff base. It was a machine sitting on the shoulder of a resonance and passing, which is precisely what the elevated PM reading from fourteen months earlier was reporting to a record that had no way to say so.
Two corrections that were considered and rejected
Balance the new motor's rotor. This reduces the force but does nothing to the amplification, and the arithmetic settles it. Take the original machine's motion as 1.0. Halving the replacement's residual unbalance gives 1.04, times 0.5, times 9.8, divided by 3.4, which is about 1.5. Even a rotor balanced twice as well as the one installed leaves the machine shaking about half again as much as it did before the swap, at real expense, and every future disturbance to that rotor lands on a ninefold multiplier.
Add isolators under the unit. The mode that moved is the motor pedestal and rail assembly inside the unit. Isolators sit between the unit and the curb, in a different part of the load path, and they do not change an internal pedestal mode. Adding them would also lower the mount system's frequency, which is the correct direction for a mount system and irrelevant to this fault. The companion article on what a flexible mount cannot do covers why these two get confused.
The correction that was made, and the number it targeted
Put the pedestal mode back above running speed with a proper margin. Clearing 1,775 rpm by 15 percent requires the mode to reach 2,041 cpm. From the as-installed 1,798 cpm that is a 13.5 percent rise in frequency, and since frequency rises with the square root of stiffness, it needs stiffness up by about 29 percent.
Two things got that: the fabricated adapter plate was replaced with the motor manufacturer's rail kit for the new frame, removing an unengineered element from the load path, and gussets were welded at the ends of the motor rails to shorten the unsupported span. That welding carries its own hazard on a route the rest of this job does not: zinc-coated or galvanized rail releases zinc oxide fume and stainless releases hexavalent chromium when welded, so the work was done with local exhaust and respiratory protection under a written 29 CFR 1910.134 program, with 29 CFR 1910.1026 governing the chromium exposure assessment, plus a hot-work fire watch because it happened on a roof. The span change is the one that does the mechanical work. Adding a member alongside an existing rail adds far less stiffness than shortening the span it bridges, and a 29 percent stiffness increase is not something a bolt-on brace reliably delivers.
Note where that target landed: 2,041 cpm is within half a percent of the 2,050 cpm the assembly had with the original motor. The design intent was never written down anywhere, but the machine had been telling anyone who measured it what the number should be.
How they confirmed it held
Three checks, in this order, because each one can pass while the next one fails.
Re-run the impact test on the locked-out, stopped machine and confirm the pedestal now rings above 2,041 cpm. This is the only measurement that proves the structure changed. A vibration reading alone could drop for half a dozen unrelated reasons, including the belt being re-tensioned during the same visit.
Re-run a coast-down from outside the guard and confirm no amplitude peak appears between the new mode and running speed. Stiffening one member occasionally exposes a second mode that was previously masked, and a coast-down is the only thing that finds it.
Re-read at the same sensor points on the running machine and record the numbers with the frequency, not just a pass or fail. That last part is the change the shop made permanently: the PM sheet for rotating equipment now carries a natural-frequency line alongside the amplitude line, and any motor or major-component swap records the weight of what came off and what went on. Both fields cost a minute on the job and both were the fields whose absence made this diagnosis a reconstruction instead of a lookup. The library's article on capturing a healthy baseline covers how to set that up before you need it.
References
- 29 CFR 1910.147 for isolation, lock and tag, and stopped-rotor verification before impact testing or handling the assembly
- 29 CFR 1910.333(b)(2) and NFPA 70E-2021, 120.5 for terminal-box and starter work during a motor change
- 29 CFR 1910.28(b)(1) and 29 CFR 1926.501 for rooftop fall protection under general industry and construction respectively
- 29 CFR 1910.219 for guarding during running and coast-down measurements
- 29 CFR 1910.134 and 29 CFR 1910.1026 for respiratory protection and hexavalent chromium exposure during the welded rail stiffening
- See related: What Resonance Is and Why It Finds Your Machine; How to Work Out Whether You Are Near a Critical Speed; What a Flexible Mount Does and What It Cannot Do; Capturing a Healthy Baseline So a Later Complaint Has Something to Compare Against