A Horrible Noise That Disappeared: Diagnosing a Loose Motor Cooling-Fan Part

A strange mechanical noise does not have to remain present for the fault to be real.

In a Bently Nevada success story from an offshore oil and gas platform, a technician heard a “horrible noise” from a dissolved-salts pump motor during routine vibration data collection. The noise disappeared shortly afterward, but the measurements taken during the event preserved the evidence. What followed is a valuable lesson in combining human observation, spectral analysis, waveform interpretation and focused inspection.

A transient sound may disappear before an inspection begins, but correctly captured vibration data can preserve the mechanical signature of the event.

The operating context

The machine was considered moderately critical and was monitored periodically using a portable Scout 220 data collector. Offshore technicians collected the readings, uploaded them to a System 1 server and requested remote diagnostic support from Bently Nevada Machinery Diagnostic Services engineers.

This arrangement matters. It shows how a condition-monitoring program can connect three important capabilities:

• People near the machine who notice abnormal sound, smell, heat or movement.
• Portable measurements that preserve the spectrum and time waveform.
• Diagnostic expertise that can interpret the data and guide a targeted inspection.

What the vibration data showed

The measurement taken while the noise was present contained a single prominent peak at approximately 1210 Hz, accompanied by 1X and 2X running-speed sidebands. The unusual vibration appeared only at the motor non-drive end and was absent from the later measurement taken after the noise stopped.

The 1210 Hz component did not match an expected forcing frequency for the machine. This prevented the analyst from simply assigning it to a normal rotating or electrical source. Instead, the spectrum and waveform suggested that repeated impacts were exciting a local resonance near 1210 Hz. The 1X and 2X sideband spacing showed that the high-frequency response was being modulated at running-speed-related intervals.

Measured observation: A 1210 Hz peak with 1X and 2X sidebands appeared at the motor non-drive end during the audible event.

Interpretation: Periodic impact or rubbing was likely exciting a structural resonance.

Location hypothesis: A loose rotating part or contact between rotating and stationary parts near the overhung cooling fan.

Discriminating action: Open the fan cover and inspect the cooling-fan assembly.

Why sidebands were important

A sideband is a spectral component that appears on either side of a carrier frequency. In this case, the carrier was the resonance near 1210 Hz. Regular variation in the amplitude of that vibration produced sidebands separated by the modulating frequencies.

The key point is not merely that sidebands existed. Their spacing connected the high-frequency resonance to a once-per-revolution and twice-per-revolution mechanical event. That relationship supported a hypothesis involving a rotating component repeatedly contacting, striking or changing load as the shaft turned.

Sidebands should always be interpreted with the waveform, machine speed, measurement location and equipment geometry. They describe modulation; they do not identify the damaged component by themselves.

Why the noise was so noticeable

The abnormal frequency was high enough to be heard clearly by the technician. Baker Hughes noted that it was near the frequency range where human hearing is particularly sensitive. This helps explain why the event sounded severe even though it was temporary.

Human senses remain useful screening tools in maintenance. An experienced technician may recognize a change before an alarm is triggered. However, the safest workflow is to treat sound as an observation, capture objective data and avoid approaching or opening moving equipment until it is isolated under the approved procedure.

The inspection confirmed the diagnosis

Because the abnormal response was localized at the non-drive end, the diagnostic team suspected the overhung cooling fan. They recommended removing the fan cover and looking for abnormalities.

The inspection found part of a broken retaining ring loose inside the cover. The ring belonged to the fan assembly and had separated. The technician replaced it, and the machine returned to service less than 48 hours after the abnormal measurement.

According to the source case study, failure to detect the problem could have allowed the fan to separate from the rotor and destroy the motor. The estimated replacement cost of the motor was approximately €40,000. The larger value, however, also included avoided secondary damage, unplanned downtime and operational risk.

A practical diagnostic workflow

1. Record the human observation. Note what was heard, where it was strongest, when it began and whether operating conditions changed.
2. Preserve event data. Save the spectrum, time waveform, speed, load, measurement direction and timestamp. Do not overwrite an abnormal reading with a later normal one.
3. Compare locations. A response isolated to one end or one direction can greatly reduce the search area.
4. Identify measured facts. List the dominant frequency, sideband spacing, waveform features and differences from the baseline.
5. Develop more than one hypothesis. Consider loose parts, rubbing, impacts, bearing faults, aerodynamic effects, electrical sources and structural resonance as appropriate.
6. Use machine geometry. Ask which component near the measurement point could physically produce the observed periodic event.
7. Choose a focused inspection. Inspect the suspected area under proper isolation instead of dismantling the entire machine.
8. Verify after repair. Repeat comparable measurements and confirm that the abnormal sound and vibration signature are gone.

Lessons for a condition-monitoring program

• Intermittent faults deserve immediate attention. Disappearance of the symptom is not proof that the defect corrected itself.
• Keep the abnormal dataset. Event measurements may contain evidence that routine follow-up readings no longer show.
• Use both spectrum and waveform. Frequency-domain sidebands reveal modulation, while the time waveform helps show the underlying impact pattern.
• Measurement location matters. Localization to the motor non-drive end directed attention toward the cooling fan.
• Combine technology with technician experience. The technician’s report of abnormal noise gave essential context to the data.
• Convert diagnosis into a specific action. The recommendation was not simply “monitor closely”; it identified the fan cover as the next safe inspection point.
• Moderately critical assets can still create major losses. Periodic portable monitoring can prevent expensive failures outside the permanently monitored machine population.

Final takeaway

This case is a strong example of evidence-based vibration analysis. A temporary noise was captured as a localized high-frequency resonance with running-speed sidebands. The pattern suggested impact or rubbing near the motor cooling fan, and a focused inspection revealed a broken retaining ring.

The successful diagnosis did not come from one spectral rule. It came from linking sound, timing, location, spectrum, waveform, machine geometry and physical inspection. That is the habit that turns condition-monitoring data into reliable maintenance decisions.

Source and further reading

This educational analysis is based on the Baker Hughes/Bently Nevada success story Loose Part on Motor Cooling Fan, authored by Håkon Myklestad of Norway MDS. Refer to the original case study for its spectrum, waveform and damaged-component images.

Educational note: This article summarizes a published case and expands on its diagnostic lessons. Actual machinery decisions must follow approved isolation procedures, manufacturer guidance, applicable standards and qualified engineering judgement.