How to Read a Pump Vibration Complaint Without an Analyzer

Why this matters

A vibration analyzer measures a level and names a frequency. You do not have one, and the customer is standing next to a pump that "sounds wrong." You can still split the complaint into the three families that matter, because each family responds differently to three things you can change at the machine: whether the rotor is turning, how fast it is turning, and whether liquid is flowing through it. The coast-down is the sharpest of the three and it costs one press of the stop button. Getting the split right decides whether the next visit brings a coupling insert, an alignment kit, or a suction-side investigation, and those are three different trucks.

Before you touch anything

Coupling and belt guards stay on and stay closed while the machine runs. Every observation below is made through or around an intact guard, or with the machine locked out. Do not wear gloves within reach of a rotating shaft; a glove is a snag path, and the protection here is bare hands kept outside the guard, not better gloves.

Where a step requires the rotor stopped - the uncouple test, applying reflective tape for an optical tach, touching a bearing housing with a rod - open the motor disconnect, apply your lock and tag, and try the start command to confirm it will not run. The hazard in that step is mechanical, so it falls under 29 CFR 1910.147, and 1910.147(d)(5)(i) is the part that also requires you to relieve residual and stored energy, which on a pump means bleeding the casing and closing the suction and discharge valves before you crack anything. If your work goes further and you open the starter enclosure to meter conductors, that act is electrical and 1910.147 expressly excludes it at (a)(1)(ii)(C); the rules for it are 29 CFR 1910.333(b)(2), and on a construction site the counterpart is 29 CFR 1926.417.

The pump casing and the liquid in it may be hot. Read casing temperature with a non-contact infrared thermometer at standoff rather than a palm, and do not crack a vent or a gauge port on a hot pumped-liquid line until it has been isolated and allowed to cool, because a flashing hot-water release scalds.

Step 1: Turn the complaint into a comparison

Vibration is never read as an absolute number in the field. It is read as a change. So the first question is not "how much is it vibrating" but "compared to what," and you need to manufacture that comparator before you interpret anything.

Three comparators, in the order you should prefer them:

1. The machine's own commissioning baseline. If somebody recorded the as-installed sound, the amps at a known valve position and the coast-down behaviour, you are done arguing. This is the single best reason to keep an installation record.
2. An identical sibling on the same duty. Two pumps in a lead-lag pair, measured with the same instrument in the same spots within a few minutes, cancel most of what you do not know.
3. The machine against itself under a changed condition. Speed changed, flow changed, coupling removed. This is always available and it is what the rest of this procedure uses.

If a customer says "it started last Tuesday," ask what else happened last Tuesday. A strainer cleaned, a valve repositioned, a tank level dropped, a second pump brought online. A vibration complaint with a date on it is usually a system change wearing a mechanical costume.

Step 2: Rule out amplification before you interpret anything

A soft or resonant support turns a small force into a large motion, and you will chase a rotor that is fine. Put a hand flat on the baseplate, then on the floor beside it, then on the discharge pipe two feet out. If the pipe or the baseplate edge moves more than the bearing housing does, the structure is amplifying and you are reading the building, not the pump.

The press-and-hold test covers this and is described in the sibling article on isolating a vibration with a simple tool; do not re-derive it here. What matters for the split is that amplification does not change what is generating the force, so if you skip this step you will correctly identify a large motion and incorrectly assign it to the rotor.

Step 3: The coast-down, which is the test that pays

Stand where you can hear and see the machine, have someone press stop, and watch what happens in the first few seconds while the shaft is still turning. Three outcomes, and they map cleanly onto the three families:

• The noise stops almost immediately, well before rotation does. That is hydraulic. Cavitation and suction recirculation depend on flow and on the margin between available and required suction head, and required suction head falls roughly with the square of speed, so the moment the pump decelerates the noise has nothing to feed it. A rotor fault cannot behave this way, because the rotor is still spinning.
• The noise fades in proportion as the machine slows and disappears only when the shaft stops. That is mechanical, driven by rotation. Imbalance force rises with the square of speed, so it decays visibly on the way down. This is the fork the sibling article on telling imbalance from misalignment picks up; hand it off there rather than guessing between them here.
• The noise gets briefly worse at one particular speed on the way down, then goes quiet. That is resonance. The machine passed through a natural frequency. Nothing is wrong with the rotor at all, and balancing it will not help. This is the outcome that makes the coast-down worth more than a stop: it is a continuous sweep through every speed between running and zero, which is the speed change you would otherwise need a variable-speed drive to produce.

This is the whole reason the step exists: no other analyzer-free test separates a hydraulic source from a mechanical one, because both are running-speed phenomena while the machine is at speed and they feel identical through a hand.

Step 4: Change the flow and watch the roughness

If the coast-down said hydraulic, confirm it by changing flow deliberately. On a centrifugal pump, throttle the discharge valve part way and listen. Cavitation and recirculation noise change with operating point, usually loudest well off the best efficiency point; a mechanical fault does not care what the valve is doing.

Never close the discharge valve fully, and never throttle a positive-displacement pump on its discharge at all. A closed discharge on a centrifugal pump turns all of the shaft power into heat in a trapped volume of liquid and can flash it; on a positive-displacement pump the flow has nowhere to go and the relief valve, or the weakest joint in the line, decides what fails. If you cannot identify the pump type from the nameplate or the casing, do not perform this step.

Step 5: Check speed and current against the nameplate

Two readings you can take without an analyzer, and they cross-check each other.

Speed. With reflective tape applied to the shaft while the machine is locked out, then the lock removed and the machine restarted, an optical tach reads running speed from a standoff. Nameplate speed is the speed at rated load. Slip rises with load, so a motor pulling less than rated load runs faster than nameplate, approaching synchronous speed at no load, and a motor reading below nameplate speed is pulling more than rated load. That direction is the one people get backwards.

Current. A clamp meter on one motor lead, compared with nameplate full-load amps. On a centrifugal pump, current tracks flow: less flow means less current.

Worked example: the pump that was "vibrating"

A lead-lag pair on the same header. The lead pump has been rough for a week; the lag pump is quiet. Same model, same duty, so comparator number two is available.

Coast-down. Stop is pressed. The crackle disappears inside a second while the coupling is still visibly turning, and the machine then coasts down silently. Hydraulic, not mechanical.

Flow change. Discharge valve throttled to roughly half open, not closed, on a centrifugal pump. The crackle drops noticeably. Confirms hydraulic.

Speed. Nameplate reads 1,760 rpm, illustrative for this machine. The optical tach reads 1,780 rpm. Above nameplate speed means below rated load, so the motor is doing less work than it was designed for. On a centrifugal pump that means less flow, not more.

Current, with the comparator built properly. Same clamp meter, same lead position, both pumps within four minutes.

• Lag pump (quiet): 11.0 A
• Lead pump (rough): 9.4 A
• Raw difference: 1.6 A

Now the correction the general rule demands. The meter's accuracy is quoted as plus or minus 2 percent of reading, which is a percent-of-reading basis, so it shrinks as the reading shrinks, and within one instrument over four minutes it behaves as a fixed proportional offset rather than an independent spread. A fixed proportional offset largely cancels in a difference, leaving about 2 percent of the difference itself: 2 percent of 1.6 A is 0.03 A. If instead you treat the two readings as carrying independent random spreads, they combine in quadrature, not linearly: 2 percent of 11.0 A is 0.22 A, 2 percent of 9.4 A is 0.19 A, and the root-sum-square is 0.29 A. Either way the 1.6 A gap is five times the worst of the two error terms, so the difference is real.

Reading the whole set. The rough pump is running faster than nameplate and drawing 14 percent less current than its identical sibling on the same header, while making a noise that stops the instant flow stops. Every one of those says the same thing: it is not moving the liquid it should be moving, and the restriction is upstream of the impeller. The lag pump shares the header, so the difference is in the branch each one owns.

What was found. A partially closed suction isolation valve on the lead pump's branch, left that way after a strainer service the previous Tuesday. Which is the date the customer gave in the first thirty seconds.

The failure mode of getting this wrong. Every one of those symptoms is also consistent with a worn coupling insert to a tech who skipped the coast-down, and the pump would have been uncoupled, aligned, re-coupled and returned still cavitating, with the real damage - impeller vane erosion and a mechanical seal running on flashing liquid - still accumulating. The seal would have been the callback, four to eight weeks out, and it would have been read as a bad seal.

How to verify you got the split right

• Undo the change and see the symptom return. Restore the suction valve to its wrong position for a few seconds with the machine running and confirm the crackle comes back. A fix you cannot switch on and off is a coincidence until proven otherwise.
• Re-run the coast-down after the correction. The noise should be gone from the running machine entirely, not just quieter on the way down.
• Re-read current against the sibling. The two should land within the same margin your meter's percent-of-reading spec allows on the difference, which for this pair is a few hundredths of an amp on a 1.6 A gap that should now be near zero.
• Write down what the machine sounds like when it is right. That is next year's comparator, and it costs one line.

References

• 29 CFR 1910.147, the control of hazardous energy, for locking out the mechanical hazard before uncoupling or touching a shaft, including the relief of stored and residual energy at (d)(5)(i)
• 29 CFR 1910.147(a)(1)(ii)(C), which excludes exposure to electrical hazards from work on conductors and equipment in electric utilization installations, sending that work to 29 CFR 1910.333(b)(2) in general industry and 29 CFR 1926.417 in construction
• Pump manufacturer documentation for the allowable operating range, the minimum continuous flow and the required suction head at the duty point, which own the numbers this method only points at
• See related: How to Tell Imbalance From Misalignment Mechanically; Using a Simple Tool to Isolate Where a Vibration Is Coming From; What Cavitation Is and How It Announces Itself; How to Record a Pump Installation So the Next Visit Is Shorter