Why Vibration Is the Earliest Honest Signal a Pump Gives

Why this matters

A pump's head, its flow and its power draw are aggregate numbers. They describe what the whole machine is doing, so a single part quietly coming apart barely moves them - a failing bearing changes discharge pressure by nothing at all and motor current by about one percent, which is under the reading error of a clamp meter. Vibration is not an aggregate. It is a direct read of the one element that is degrading, and it climbs while every other gauge on the machine still says fine.

That is only true of the mechanical half: bearings, alignment, imbalance, looseness, the coupling, the mounting. The suction side announces itself differently and a sibling card covers that. And it is only honest under conditions you have to hold deliberately, because the same pump measured a different way on a different day produces a number that means nothing. This card is about that discipline and about what the numbers can legitimately carry, using eleven months of one pump's record.

The library already has a card mapping vibration frequencies to specific faults. This one is about why the signal leads, and how to keep it trustworthy enough to act on.

Taking the reading without creating a hazard

The coupling guard stays on and the reading is taken on the bearing housing outside it, never through a guard opening and never with a hand inside the guard line. No loose sleeves, no lanyards, no dangling meter strap.

A magnet base snatches. Bring it to the housing at an angle and roll it onto the surface rather than presenting it flat, and keep fingers off the contact face; a rare-earth base landing on steel takes a fingertip with it.

Do not test a bearing housing's temperature with your hand. A housing at 180 F burns on contact in under a second. Read it with a non-contact thermometer, and treat a hot housing as its own finding rather than as a nuisance.

Marking permanent measurement points means putting a punch or a paint mark on a machine, which is stopped work. Lock and tag the energy isolating device under 29 CFR 1910.147 first; a level control or a building automation command will start a pump that someone only pressed stop on.

Why the mechanical signal arrives first

Think about what each instrument is actually integrating.

A discharge gauge sums the work of every impeller vane against the whole system. A bearing that has lost its clearance does not change that sum in any measurable way, because the energy the bearing wastes is a small share of shaft power on a machine of this size. So the gauge cannot see it, and it is not the gauge's fault.

A clamp meter sums motor load. If bearing drag doubles, the shaft power moves by around one percent, which is inside the noise of a current reading taken on a different day at a slightly different supply voltage. A sibling card on locating a pump on its curve makes the same point from the other direction: motor current is an ordinal indicator, not a power measurement.

An accelerometer on the bearing housing is not summing anything. It is measuring the force that specific bearing is transmitting into that specific housing, and the increase from a degrading rolling element is a large fraction of the small signal it started with. That is the whole reason vibration leads: the fault represents a fraction of a percent of the machine's output and a large percentage of the quantity being measured.

Six conditions that make two readings comparable

A vibration number is a comparison or it is nothing. These are the conditions that let two readings be compared, and every one of them has produced a false alarm somewhere when dropped.

1. The same physical point, marked on the housing, not "about there." Half an inch on a housing changes an overall reading measurably.
2. The same direction, with horizontal, vertical and axial recorded separately rather than blended into one figure. The direction that rises is itself diagnostic.
3. The same mounting method. A hand-held probe, a magnet base on bare metal and a magnet base on paint give three different numbers from the same bearing. Pick one and record which.
4. The same shaft speed. On a variable speed machine, amplitude is not comparable across speeds, so the speed is part of the reading and a reading without it is discarded.
5. The same operating point. A pump running at a third of its best efficiency flow has internal recirculation that raises vibration for reasons that have nothing to do with wear. Record the flow or the discharge pressure alongside.
6. The same measured quantity and band. Overall velocity in one frequency band is not comparable to overall velocity in another, and neither is comparable to an acceleration or displacement figure.

The record: eleven months on one circulating pump

Overall velocity, outboard bearing housing, horizontal, magnet base on a bare marked spot, fixed speed, taken each time at the same discharge pressure with the same meter.

Over the same eleven months, discharge pressure at the same flow went from 40.0 psi to 39.4 psi, a change of 1.5 percent, which is inside the gauge's own error and tells you nothing. Motor current was recorded three times and moved by less than the difference between two readings taken an hour apart. Both hydraulic legs were silent for the whole period. The mechanical leg had already more than doubled.

What those numbers can carry, and what they cannot

Before any of the steps above are called significant, two error figures need their basis and their character stated, because the two behave completely differently.

Instrument scale error: 5 percent of reading, systematic. Percent of reading is the basis, so it shrinks as the reading shrinks. Systematic is the character: it is the same meter with the same calibration on every visit, so the error has the same sign and the same proportion each time. That has a consequence people usually get backwards. In a ratio of two readings from the same instrument, it cancels completely. The month-11 to baseline ratio of 0.24 / 0.10 = 2.4 is not degraded by the 5 percent at all. If instead you quote the difference, 0.24 - 0.10 = 0.14 in/s, the same systematic error leaves 5 percent of that difference, which is 0.007 in/s, not 5 percent of the larger reading.

Repeatability spread: about 0.01 in/s per reading, independent and random. This is the scatter from re-seating the magnet, small speed variation and ambient conditions. Independent random terms combine in quadrature, so the spread on a difference between two readings is 0.01 x the square root of 2 = 0.014 in/s, not 0.02.

Now the steps can be read honestly:

• Month 4's step of 0.01 is inside the 0.014 spread. It is not a signal.
• Month 8's step of 0.02 is marginally outside it. Worth noting, not worth acting on alone.
• Month 10's 0.06 and month 11's 0.05 are each more than three times the spread. Those are real.

The one comparison that gets no cancellation. Comparing your reading to a published severity limit is not a ratio of two readings from your instrument, so the 5 percent systematic error applies in full and does not disappear. Severity zones in a standard such as ISO 20816, in the edition your customer's specification or your service contract references, bind through that contract rather than on their own, and their boundaries depend on the machine class and on whether the mounting is rigid or flexible. For a shop without that classification in hand, a ratio against the machine's own baseline is the stronger number precisely because the instrument's systematic error cancels out of it.

What was found, and when it was found

At month 10 the reading was flagged and the axial component was checked, since only the horizontal had been trended. Axial had risen further in proportion than horizontal, which pointed away from imbalance and toward the coupling and thrust end. Housing temperature at the outboard bearing read 24 F above the inboard, measured with a non-contact thermometer from outside the guard.

The pump was scheduled for the following weekend rather than run to failure. The outboard bearing was found with spalling on the outer race and the grease degraded. The coupling was inside tolerance, which retired the second hypothesis. Nothing about the hydraulic performance had changed at any point, and the pump would have carried on producing rated pressure right up to the moment the bearing seized.

What running it to failure would have cost, in the units that matter. A planned weekend bearing replacement is a short job on a stopped machine with the parts on the truck. The same bearing failing in service takes the shaft, usually the seal, sometimes the impeller and the wear rings, and it happens at the worst time on the worst day; the labour multiple on that job runs several times the planned one, and the outage is unscheduled. That multiple, not the vibration meter, is what the trend record is buying.

What the shop changed after this one

A baseline is taken at commissioning or it does not exist. The single most common reason a trend program produces nothing is that the first reading was taken when someone already suspected a problem, which makes every subsequent comparison a comparison against an already-degraded machine. Any new or rebuilt pump gets its six-condition reading recorded on the start-up ticket before it leaves.

The trigger is a ratio and it is written down, not judged. A rise to twice baseline at the same point and conditions converts the machine from monitored to investigated, meaning axial and vertical get added and the housing temperature gets read. A rise to four times baseline converts it to scheduled repair with a date. Both are starting points to tune to your own machines and your own duty; what matters is that the number is agreed in advance, because a threshold decided while looking at a rising reading is decided by whoever wants the weekend off.

Readings that fail a condition get discarded rather than recorded. A number taken at a different speed, a different flow or with a different mounting is not a data point with a caveat, it is noise that will be averaged in later by somebody who never saw the caveat. The record is worth more with four honest readings in it than with nine of mixed provenance.

References

• ISO 20816 vibration evaluation series, in the edition referenced by your customer's specification or your service contract, which is what gives it force and which also fixes the machine classes and mounting conditions its severity zones depend on
• Pump and bearing manufacturer documentation for the specific unit, which owns the acceptable housing temperature rise and the bearing's expected life at the applied load
• 29 CFR 1910.147 for isolation before marking permanent measurement points or working near a coupling
• See related: The Vibration Signature Reference; How to Tell Whether a Pump Is Running Off Its Curve; The Pump That Was Replaced Twice and Failed the Same Way Each Time