HUM

Monitoring & Diagnostics

Vibration 102 · Listening on purpose

The 101 lessons taught you to read a spectrum. This module is about using it: deciding how bad a machine is, watching it change, naming the fault from its signature, and knowing what to do about it. The machine throughout is a 1780 rpm motor driving a five-vane pump through a coupling, the commonest machine train in any plant.

Why listen at all

Faults announce themselves long before they fail. Quietly.

The P–F curve.

Most machine faults develop over time. From the moment a flaw first exists (P, for potential failure) to the moment the machine stops (F), condition declines along a curve that starts flat and ends steep. Different senses notice it at different points. Vibration notices it first, with months to spare. Ears notice it with weeks; a hand on the housing, with days.

The value of a monitoring programme is the length of warning it buys: time to order the part and plan the outage. Measure often enough to catch the fault well before F. For bearings on ordinary machines, that usually means monthly readings, and weekly once something has been seen.

How bad is it?

One number, four zones.

ISO zones.

For a first judgement of a machine you have never seen, the standards offer a single number: the overall velocity, in mm/s RMS, between 10 and 1000 Hz, measured on the bearing housings. ISO 20816-3 (the successor of 10816-3) divides it into four zones by machine size and foundation. Velocity is the unit that gives low and high frequencies a fair weighting (lesson 5).

Zone boundaries are guidance for machines in general, not a verdict on yours. A fan on a tall steel structure may live happily in zone C; a precision spindle may be scrap long before it leaves zone A. The overall level also says nothing about what is wrong: that needs the spectrum.

Trends and alarms

A machine is best compared with itself.

Watch the change.

The same overall value, measured month after month, tells a story the standards cannot. A machine that has read 1.8 mm/s for two years and now reads 3.6 has doubled, and something has changed, even though 3.6 mm/s is still "acceptable" on the chart for a large machine. Alarms are therefore set relative to each machine's own baseline: an alert at two or three times baseline, a danger level above that.

Lower the alert multiplier and you catch the fault earlier but risk false alarms from normal scatter; raise it and you buy certainty with lead time. Try a few machines. Notice how the curve bends upward near the end: faults accelerate, so a reading that has started to climb deserves a shorter interval, not a longer one.

Overall alarms catch a fault late, because a bearing tone at 0.3 mm/s hides under a 2 mm/s 1X. Modern programmes add band alarms on parts of the spectrum: the bearing band, the 2X band, the gear-mesh band, each trended against its own baseline. The spectrum of a healthy machine, saved on day one, is the most useful measurement in the programme.

The fault simulator

Nine faults, one healthy machine, your ears.

Name that fault.

Pick a fault, a measurement point and a direction. The simulator synthesises the time waveform and the spectrum you would see there, computes the overall value and its ISO zone, and lets you hear it. The signature notes under the plots are the field rules of thumb; the point of the exercise is to see them in the picture before you read them.

Bearings: four stages

From a whisper above hearing to a roar you cannot miss.

How a bearing fails.

A rolling-element bearing almost never fails suddenly. It goes through four recognisable stages, each with its own place in the spectrum: ultrasonic first, then the ringing of the bearing's own natural frequencies, then the defect frequencies in the ordinary spectrum, and finally a broadband roar as the discrete peaks melt away. In stage 4 the clearest signs disappear just as the end approaches. Slide through the stages.

Rule of thumb for lead times on ordinary machines: stage 1, months; stage 2, weeks to months; stage 3, weeks; stage 4, days to hours. The numbers vary enormously with load, speed and lubrication, which is exactly why you trend rather than predict.

Enveloping

Hearing the rhythm inside the ring.

Demodulation, step by step.

Early bearing impacts are tiny compared with the shaft vibration, and they live mostly in the high-frequency ring they excite, not at the defect frequency itself. Enveloping (also called demodulation) throws the shaft away and keeps the rhythm of the ringing: band-pass around the ring, rectify, low-pass, and take the spectrum of what remains. The defect frequency pops out of the floor, often a full stage earlier than in the ordinary spectrum.

An AM radio and a tremolo pedal work the same way: a fast carrier (the ring, or the radio station) whose loudness rises and falls at a slow rate (the impacts, or the music). Only the pattern of the carrier's loudness carries the information.

Sidebands

The spacing names the shaft.

A tone that wobbles grows a family.

When a tone's loudness rises and falls at some slow rate, its spectrum sprouts pairs of lines on either side of it, spaced by that rate: sidebands. Borrow the 23:57 gearbox of lesson 4 for a moment, its pinion on our 29.7 Hz motor shaft and its gear turning at 12 Hz. A gear mesh modulated once per turn of the input shaft grows sidebands 29.7 Hz apart; modulated by the output shaft, 12 Hz apart. The carrier tells you where the trouble is (the mesh); the spacing tells you which gear carries it. The same logic reads the 1X sidebands around an inner-race bearing frequency.

Resonance

When the structure sings along.

The run-up.

Every structure has natural frequencies at which it amplifies whatever shakes it. When a machine's running speed sits near one, a small unbalance becomes a large vibration, and no amount of balancing cures it. The classic test is a run-up or coast-down with a waterfall plot: a spectrum taken at each speed, stacked. The 1X peak walks diagonally across the plot and swells where it crosses the natural frequency, which stays put.

With the machine stopped, a bump test finds the same thing in seconds: hit the structure with a soft hammer (or a rubber mallet), record the ring-down, and read the natural frequencies from the spectrum. The 90° phase shift at resonance (Balancing, section 3) is the other fingerprint. Fix a resonance by stiffening, adding mass, bracing, or changing speed. Never by balancing harder.

A diagnostic workflow

The same seven questions, every time.

Is it worse? Compare the overall and the band values with the machine's own baseline and trend, then with the ISO zone.
Which frequencies? Convert the spectrum to orders. Synchronous (1X, 2X, 3X…), non-synchronous (bearing tones, twice line frequency, other shafts), or sub-synchronous (whirl, cage, looseness)?
Which direction? Radial points to unbalance or looseness; axial points to misalignment or a bent shaft; one direction only hints at resonance or a loose part.
Which location? One bearing: a bearing fault. Both sides of the coupling: misalignment. Motor bearings only, at line-related frequencies: electrical.
What does the waveform say? Clean sine: unbalance. M or W shape: misalignment. Clipping: looseness. Sharp repeating impacts with ringing: a bearing. Bursts once per turn: an inner-race flaw. Hiss like gravel: cavitation.
What does phase say? Steady, with horizontal and vertical about 90° apart: unbalance (both bearings in phase if static, 180° apart if a couple). 180° across the coupling: misalignment. 180° axial across one bearing: a bend. Unsteady: looseness or resonance.
What changes when you change something? Cut the power: electrical faults vanish at once. Change speed: a resonance stays at its frequency while everything else moves. Change load, temperature or flow and watch what follows.

Back to the 101 lessons, or on to Balancing.