industrial//maintenance//condition monitoring//vibration analysis

Vibration analysis is the condition-monitoring technique that reads the health of a rotating machine from the vibration measured on its casing, and it is the most widely used tool for catching imbalance, misalignment, looseness and bearing defects weeks before they become failures. It works because each defect leaves its own signature: causes that are tangled together in the time signal separate cleanly into lines of the Fourier spectrum. A line at the shaft's rotation frequency (1×) suggests imbalance; a strong 2× suggests misalignment or looseness; a bearing defect produces impacts at a rate fixed by its geometry (bearing defect frequency).

Vibration analysis is the condition-monitoring technique that reads the health of a rotating machine from the vibration measured on its casing, and it is the most widely used tool for catching imbalance, misalignment, looseness and bearing defects weeks before they become failures. It works because each defect leaves its own signature: causes that are tangled together in the time signal separate cleanly into lines of the Fourier spectrum. A line at the shaft's rotation frequency (1×) suggests imbalance; a strong 2× suggests misalignment or looseness; a bearing defect produces impacts at a rate fixed by its geometry (bearing defect frequency).

The chain starts with an accelerometer bolted to the bearing housing, sampled at 10 to 25 kHz behind an anti-aliasing filter, because bearing impacts ring structural resonances of several kilohertz (aliasing). From the waveform come a few overall numbers. The RMS value of the vibration velocity, measured between 10 and 1,000 Hz, is the overall level that severity standards rate: ISO 20816 (the former ISO 10816) sorts it into four zones, from newly commissioned (A) through fit for unrestricted operation (B) and restricted operation (C) to damaging (D), with limits that depend on machine class and mounting. The crest factor, peak over RMS, and the kurtosis, the fourth standardized moment (3 for Gaussian vibration), rise when the signal becomes impulsive, which is how an early bearing defect announces itself before the RMS moves.

An accelerometer and two trended limits come before any neural network.

RMS velocity against the ISO 20816 zones, an envelope spectrum for the bearings and two thresholds with trend, alert and alarm, catch most imbalance, misalignment and bearing defects weeks ahead at little cost; many plants gain more from that, well installed, than from any model (flyswatter rule).

Order normalization expresses frequencies as multiples of the shaft speed, the orders, so a defect's signature stays at the same place when a variable-speed drive changes the rpm. Without it, a machine that speeds up looks like a machine whose spectrum changed, and a classifier trained on spectrograms learns the speed instead of the fault.

The plain spectrum hides early bearing defects, because each impact spreads its energy over a wide band; demodulating around a resonance recovers them (envelope analysis).

A spectrum assumes the signal keeps its character over the window; run-ups, coast-downs and intermittent rubs are read in a spectrogram or in the time trace.

The physics behind the readings (modes, natural frequencies, the resonance that amplifies an impact) is in vibration; a loosely mounted sensor adds its own resonance and turns a healthy machine into an alarm.