Vibration condition monitoring: from sensor placement to fault frequencies

Vibration monitoring is the most proven predictive maintenance technology in rotating equipment: decades of physics, standardized severity zones, and fault signatures precise enough to name the failing component — inner race, outer race, misalignment, unbalance — weeks before functional failure. Programs fail not for lack of theory but for lack of discipline in sensing, baselines, and response. This guide covers the chain from accelerometer to work order.

Sensing: placement and mounting decide everything

Measure velocity or acceleration at each bearing, in three axes where possible (radial horizontal, radial vertical, axial), with the sensor mounted on a solid part of the bearing housing — stud-mounted for permanent installations, never on thin covers, guards, or painted-over rust. Magnet mounts serve route-based collection; adhesive and stud mounts serve online systems. Beyond about 1 kHz of interest, mounting stiffness dominates: a hand-held probe rolls off where early bearing-defect energy lives, which is why route readings miss faults that online systems catch.

Wireless vibration sensors have matured into a legitimate option for balance-of-plant equipment — pumps, fans, and motors where cable runs cost more than the measurement. Size the tradeoff honestly: wireless nodes sample intermittently (minutes to hours between captures) and report derived features plus occasional waveforms, while wired online systems capture continuously with full spectra. Critical machines get wired continuous monitoring; the long tail of supporting equipment gets wireless screening. Either way, confirm the sensor's frequency range, sampling rate, and temperature rating against the machine's actual fault frequencies before buying.

Reading the data: overall levels, spectra, envelope

Overall levels (velocity RMS per ISO 10816/20816 zones A–D) answer "how bad" with one number per machine — ideal for alarming and dashboards. Spectra (FFT) answer "what": 1× rotational speed indicates unbalance, 2× misalignment, blade-pass and gear-mesh frequencies point at their components, and bearing defect frequencies (BPFO, BPFI, BSF, FTF — calculable from bearing geometry) identify the exact failing element.

Envelope analysis (demodulation) answers "how early": it extracts the low-frequency repetition of high-frequency impacts, revealing bearing and gear defects months before overall vibration rises. If your program skips envelope spectra, it detects failures late by construction.

Set alarms in two stages: alert (investigate, trend more closely) and danger (plan the intervention), with baselines captured per machine after commissioning or overhaul — identical machines vibrate differently, and fleet-average limits false-alarm the rough ones while missing drift in the smooth ones.

Closing the loop to maintenance

Monitoring without a response workflow is expensive telemetry. Every alert needs an owner, a diagnosis step (analyst review of spectra, not just the overall number), and a path to a planned work order in the CMMS — with feedback when the diagnosis was wrong, so alarm limits and rules improve. Feed validated data upward: clean vibration features are among the highest-value inputs to the predictive maintenance data stack, and the signal quality guide covers the aliasing, grounding, and timestamping pitfalls that corrupt analysis silently. Start with ten critical machines done well rather than two hundred done thinly; a program that catches three failures in year one gets funded for year two.