The Mount That Held Until the Equipment Ran

A mounting that is fully adequate for the machine's weight can still shake itself apart in a month, and the reason has nothing to do with the fasteners. This is the case where the first diagnosis was right, the first repair was done properly, and the joint failed again anyway. That combination is worth walking through in full, because the instinct after a repeat failure is to doubt the workmanship, and doubting the workmanship is what kept this one going for two visits.

Stop the machine before you touch the mount

• Nothing gets touched on a mounting while the machine can start. Isolate and lock out under 29 CFR 1910.147 for general-industry mechanical and stored energy, including any electrical supply at its disconnect under 29 CFR 1910.333(b)(2) with the live-dead-live proving sequence in NFPA 70E-2021, 120.5. A machine on loose mounts that starts while somebody is under it moves in a direction nobody predicted.
• Rotating parts coast. Confirm the shaft has stopped, not that the contactor has dropped out.
• A machine with failed mounting anchors does not get run for a diagnostic reading until the mounting is at least temporarily secured against walking. Secure it or do not run it; there is no third option that ends well.

The call and the first repair

A piece of rotating equipment sits on a fabricated steel stand, bolted down to a concrete pad through four anchors. Six weeks after commissioning it is noisy and the customer can see the stand moving. On site, two of the four hold-down bolts are finger loose and one anchor turns in its hole.

The tech's diagnosis is vibration-induced loosening, which is correct. The repair matches it: the anchors are replaced with the correct family for the pad, set to the manufacturer's installation torque with the embedment measured, the hold-down bolts are cleaned and torqued to a verified value with a calibrated wrench, thread-locking compound applied where the equipment manufacturer allows it, and a witness line marked across every fastener, washer and joint face. A 30-day re-torque is put on the schedule.

That is a good visit. Every element of it is defensible on its own.

Five weeks later, the same joint

The same two bolts are loose. The witness lines have rotated, which is the useful part: the marks prove the fasteners moved rather than that somebody under-torqued them. The anchors are still set. The bolts came loose against a correct, verified preload.

That result eliminates the entire fastener-workmanship branch in one observation, and it is why the witness lines were worth the sixty seconds they cost.

What preload can and cannot hold

A bolted joint resists sideways movement by friction, and the friction comes from the clamp force the bolt's preload generates. A fastener backs off when the joint slips transversely under it, because slip lets the thread walk. That is the mechanism a sibling article covers in full.

The consequence that matters here: correct preload raises the alternating force the joint can take before it slips, but it raises it by a bounded amount. If the alternating force at that joint is several times what the mounting was expected to see, no achievable preload holds it, and the joint slips on every cycle at any torque you can put in it.

So the question stopped being "was the bolt tight enough" and became "why is the force at that bolt so much bigger than anyone expected."

Separating a fastener problem from a structure problem

Two measurements, and neither requires taking anything apart.

A bump test, machine stopped and locked out. Strike the stand near the top with a soft mallet and record the ring-down with a vibration instrument. The frequency it rings at is the stand's natural frequency in that direction. This is a stopped-machine test, done under the isolation described above, with the striking hand clear of anything the stand can pinch against as it moves.

A running reading, machine secured first. With guards in place and the mounting temporarily braced so it cannot walk, take vibration readings at three points: the machine's bearing housings, the top of the stand, and the concrete pad beside the anchors. Use a magnetic-base accelerometer placed on accessible housing surfaces. Do not reach past a guard, do not hold a probe against a rotating component, and keep sleeves and lanyards clear.

The comparison across those three points is the whole diagnosis. If the amplitude at the machine's housings and at the pad are both modest and the amplitude at the top of the stand is far higher, the stand is not transmitting the machine's vibration, it is amplifying it.

The two numbers that named it

The machine runs at 3500 rpm, which is 58.3 Hz of once-per-revolution forcing.

The bump test rang the stand at about 57 Hz.

The ratio of forcing frequency to natural frequency is 58.3 divided by 57, which is 1.02. The machine is running within about 2 percent of the stand's natural frequency, which is the definition of resonance, and at resonance a support amplifies rather than resists. A small residual unbalance in the rotor, one that would be unremarkable on a stiff base, arrives at the anchors multiplied.

Two facts about the site history closed it. The stand was built for a machine that ran at 1750 rpm and was reused when the equipment was replaced with a 3500 rpm unit. Unbalance force rises with the square of speed, so at double the speed the same residual unbalance produces four times the force before any amplification. And the doubled running speed walked the forcing frequency straight onto the stand's natural frequency, which the original 1750 rpm machine, at 29.2 Hz, was nowhere near.

The direction holds at both ends of the range, which is how you know the mechanism is the right way round. Well below the stand's natural frequency, the stand simply deflects with the force and there is no amplification. Well above it, the stand cannot follow the force and the machine's motion stops being transmitted. It is only in the narrow band around the natural frequency that a small force becomes a large one, and that band is precisely where this machine parked itself.

Stiffen or soften: the fork, and how it was decided

Natural frequency rises with the square root of stiffness and falls with the square root of mass, for a support idealized as a single spring carrying the machine's mass in one direction. A real fabricated stand has several modes; the bump test tells you which ones are close enough to matter. Two ways out:

Stiffen it, moving the natural frequency above the running speed. Getting from 57 Hz to 84 Hz puts the frequency ratio at 58.3 divided by 84, which is 0.69. That is off resonance but it is not the same as unloaded: at 0.69 an undamped support still amplifies about 1.9 times. The two branches need mirrored margins, so below resonance aim for 0.5 or less, which for this machine means 117 Hz or above and brings the amplification down near 1.3. Treat 0.69 as the minimum acceptable rather than the target. Because frequency goes with the square root of stiffness, that requires about (84/57) squared, or 2.2 times the stiffness in that direction.

Soften it, moving the natural frequency well below the running speed. Isolation only begins above a ratio of about 1.41, and a working design target is a ratio of 2.5 or better, so the natural frequency would need to come down to 58.3 divided by 2.5, about 23 Hz. For a linear spring carrying the machine's weight, natural frequency in hertz is roughly 3.13 divided by the square root of the static deflection in inches, so 23 Hz corresponds to about 0.018 in of static deflection. Getting there from 57 Hz means dropping the stiffness to about (23/57) squared, roughly a sixth of what it is.

Stiffening won here, for two reasons specific to this job. Softening a stand that is already carrying a machine means fitting isolators between the machine and the stand, which changes the alignment of the connected piping and needs flexible connections that were not in the original scope. Stiffening was cheap: a rectangular fabricated frame resists this kind of load in bending, and adding diagonal bracing changes the load path from bending to axial, which is a large stiffness gain for very little added mass. Two diagonals and a gusset at the base got there.

Doing the work without creating a second problem

Welding braces onto an existing stand introduces hazards the diagnosis did not have:

• Hot work needs a fire watch and a cleared area before the arc strikes, per 29 CFR 1910.252 in general industry and 29 CFR 1926.352 in construction, and where the site runs a hot-work permit program, that program and the NFPA 51B edition it was written against govern the permit.
• Coated steel puts metal in the air when it is welded or ground. Galvanized steel releases zinc oxide fume. A chromate-containing primer releases hexavalent chromium, which is an inhalation carcinogen and is not addressed by gloves or a face shield: it needs local exhaust or respiratory protection under a written program per 29 CFR 1910.134, with exposure governed by 29 CFR 1910.1026 in general industry and 29 CFR 1926.1126 in construction. Grinding the coating back before welding creates the same airborne exposure and needs the same control, so removing the coating is not a way around it.
• Nothing gets welded to the machine itself, to a pressure boundary, or to any component serving a relief or protective function. Braces go between members of the stand.
• The machine stays isolated and locked out for the whole of it, and welding return current is routed so it does not pass through the machine's bearings.

Confirming the fix, and keeping it confirmed

• Repeat the bump test after the work. The stand rang at 84 Hz, putting the frequency ratio at 0.69. That single number is the acceptance criterion, and it is the only one that would have distinguished a real fix from a hopeful one.
• Repeat the three-point running reading. The once-per-revolution amplitude at the top of the stand fell to roughly a sixth of its previous value, while the readings at the machine's bearing housings barely moved. That split is the confirmation that the stand was the amplifier and the machine's own balance was never the issue.
• Re-mark the witness lines and re-check at 30 days. With the amplification gone, the fasteners should hold their marks. If they still walk, the diagnosis was incomplete and there is a second mode in play, which the bump test can be repeated in the other two axes to find.
• Make the bump test an acceptance step on any fabricated stand you build or reuse, before the machine is commissioned on it, and compare the result against the running speed. It costs a mallet, an instrument and five minutes, and it is the only check that catches a stand that is fine for the weight and wrong for the speed. A stand reused for a faster machine is the specific case to never skip it on.

References

• 29 CFR 1910.147 for mechanical and stored-energy isolation; 29 CFR 1910.333(b)(2) with NFPA 70E-2021, 120.5, in the edition adopted by your employer's electrical safety program, for electrical isolation and proving dead
• 29 CFR 1910.252 and 29 CFR 1926.352 for hot-work fire prevention; NFPA 51B in the edition your site's hot-work permit program or your authority having jurisdiction has adopted, which binds through that program
• 29 CFR 1910.1026 and 29 CFR 1926.1126 for hexavalent chromium; 29 CFR 1910.134 for respiratory protection programs
• Equipment manufacturer documentation for permissible mounting arrangements, isolator selection and whether thread-locking compound is allowed on hold-down fasteners
• See related: Why Things Vibrate Loose; How to Choose an Anchor for What You Are Hanging