Loose Foundation Bolt Resonance Shift Mistaken for Sensor Fault

A 2.4 MW SAG mill gearbox pinion bearing shows a rising 1× running-speed amplitude on its radial accelerometer. The vibration analyst raises a work order against the sensor. The sensor is replaced. The amplitude returns to baseline for eleven days, then climbs again. The second sensor is also replaced. On the third cycle, someone inspects the foundation and finds two anchor bolts on the drive-end pedestal have lost preload. The resonance shift from the loose foundation bolt was being read as a sensor fault for six weeks.

This is a diagnostic-coverage problem, not a hardware problem. It matters because condition-monitoring channels increasingly feed proof-test credit under IEC 61511 safety instrumented function (SIF) verification. If a channel’s diagnostics cannot separate a mechanical resonance shift from an instrument fault, the proof-test interval you have claimed is not the interval you are running.

The mechanism: preload loss and stiffness change

A foundation bolt holds a machine pedestal to a concrete plinth or steel baseplate through a preloaded joint. Correct preload clamps the joint so the bolt sees a fraction of the dynamic load. When preload is lost — through vibration-induced relaxation, thermal cycling, grout degradation, or corrosion under the nut — the joint’s effective stiffness drops.

The machine’s mounted assembly is a mass-spring system. Its natural frequencies scale with the square root of stiffness over mass. Mass is unchanged. Stiffness falls. The first bending mode of the pedestal, and often a coupled mode involving the gearbox housing, moves down in frequency. If that mode was previously well clear of a running-speed harmonic or a gear mesh frequency, it can migrate onto one.

The result is a resonance shift. Amplitude at the affected frequency rises, phase relationships change, and the spectrum develops a peak that was not there before. To a single-channel alarm with a fixed band threshold, this looks exactly like a sensor going bad: an amplitude excursion with no other correlated change.

Why conventional alarms misclassify it

Most condition-monitoring alarm logic is built on per-channel band thresholds. A channel trips when its narrowband amplitude exceeds a limit. That logic has no model of the machine, so it cannot ask the question that distinguishes the two cases: did the change occur at a frequency consistent with a structural mode, or is it broadband and uncorrelated with running speed?

Three signatures separate the cases, and conventional alarms typically evaluate none of them.

**Frequency selectivity.** A loose foundation bolt produces a narrow, running-speed-related or mode-related peak. A failing accelerometer, a cracked cable shield, or a loose connector produces broadband noise, 50/60 Hz mains pickup, or a signal that drops out intermittently.

**Cross-channel coherence.** A structural change appears simultaneously on multiple channels — drive end and non-drive end, horizontal and vertical — with a consistent phase relationship. An instrument fault appears on one channel only, or on channels sharing a junction box or cable route.

**Persistence across sensor replacement.** This is the diagnostic that catches the failure mode. If amplitude returns to baseline after sensor replacement and then climbs again over days to weeks, the sensor was not the cause. A genuinely failed sensor does not reproduce its own signature on a new transducer.

A per-channel threshold alarm cannot perform any of these three tests. It reports an excursion. The human then guesses, and the guess is biased toward the cheapest intervention — swap the sensor.

IEC 61511 framing: diagnostic coverage and proof-test credit

Under IEC 61511, the verification of a safety instrumented function depends on the claimed failure rates of each subsystem and the diagnostic coverage (DC) applied to dangerous detected failures. A condition-monitoring channel used to support proof-testing of a final element — a SAG mill clutch, a conveyor drive brake, a control valve actuator — is contributing evidence to that verification. It is not itself a safety device, and no accelerometer carries a SIL rating. The question is narrower: does the diagnostic architecture detect the failure modes you have claimed it detects, within the proof-test interval you have claimed?

A loose foundation bolt is a dangerous undetected failure mode of the mechanical assembly. If your monitoring channel reports it as an instrument fault, the diagnostic has not covered it. The proof-test interval for the mechanical subsystem is then unverified, because the mechanism that would have revealed the defect has been misattributing it.

The corrective is methodological, not a product feature:

• **Model the machine, not just the channel.** Store baseline natural frequencies and mode shapes for each monitored asset — SAG mill, gearbox, conveyor drive, pump, control valve — and alarm on deviation from the structural model, not only on absolute amplitude.
• **Require cross-channel correlation before raising an instrument fault.** A single-channel excursion with no corroborating channel is a sensor suspect. A multi-channel excursion with consistent phase is a structural suspect.
• **Track post-replacement recurrence.** Log every sensor replacement against the amplitude history. Recurrence within a defined window should route to mechanical inspection, not a third transducer.
• **Feed the result into proof-test scheduling.** When resonance analysis identifies a loose bolt, the finding is a proof-test result for the mechanical subsystem. It resets the interval with evidence, and it triggers torque verification before the next scheduled test.

None of this requires the monitoring channel to be a safety device. It requires the diagnostic to be honest about which failure mode it has detected.

Getting started

Pick one rotating asset with a history of repeated sensor replacements. Pull its amplitude trend and overlay every replacement date. If amplitude recovers after replacement and then degrades again, you have a structural suspect, not an instrument fault. Inspect the foundation bolts and baseplate grout, record the preload, and re-baseline the natural frequencies. Then write the diagnostic rule that would have caught it: cross-channel correlation plus post-replacement recurrence. That rule is what converts a misclassified sensor fault into a proof-test result you can defend.

Source: internal-backlog — grounded in S5 engineering design notes, not a current market event.

Talk to S5 System about your site

S5 System designs and manufactures ultra-low-power industrial IoT sensors for mining, agriculture, infrastructure and utilities.

Discuss your IoT monitoring project — book a call.