What Cavitation Is and How It Announces Itself
Why this matters
Cavitation is one of the few faults that tells you exactly what it is if you know the sound, and one of the few that will destroy a good pump in a season if you do not. Techs lose the call two ways: they hear gravel and replace the pump, which cavitates again within weeks because the pump was never the problem, or they hear it, decide it is "just noisy," and come back to a wire-drawn impeller and a chewed seal face. This card is about recognizing it and separating it from its look-alikes. A sibling card owns the suction-side budget that decides whether it happens at all.
Before you go near a running pump
Rotating equipment is the immediate hazard and the action is to leave the coupling guard on. Take amp readings with a clamp meter on the lead outside the guard, never by reaching past it, and keep sleeves, lanyards and meter leads out of the coupling plane. If the guard is already off when you arrive, the pump does not run until it is back on.
Hot fluid at a gauge port sprays at face height. Where you are opening an isolation cock to fit a gauge on a loop above scald temperature, isolate and let the section cool if the schedule allows, and if it does not, stand out of the discharge path, crack the cock a fraction of a turn, and wear a face shield rather than safety glasses, because a gauge port aims sideways.
If the job moves from reading the pump to opening it, that is mechanical isolation and stored energy under 29 CFR 1910.147, so lock and tag the disconnect rather than switching it off, and vent the trapped pressure before you break the first joint. Work at the starter or disconnect itself is electrical and falls under 29 CFR 1910.333(b)(2), with the live-dead-live proving sequence taken from NFPA 70E-2021, 120.5.
The mechanism, stated once
A liquid boils whenever its local static pressure falls to its vapor pressure at its actual temperature. That is the whole of it. Heat is one way to get there and dropping the pressure is the other, and a pump inlet does the second one for a living: the fluid accelerates into the eye of the impeller, and accelerating fluid trades static pressure for velocity.
If the static pressure at that point reaches vapor pressure, vapor bubbles form. A few thousandths of a second later the impeller vane has raised the pressure again, and the bubbles collapse. They do not deflate gently, they implode, and each implosion is a microscopic jet against whatever surface it is touching.
Vapor pressure is the term people forget, and it is temperature-dependent, not fixed. Water at 212 F has a vapor pressure equal to standard atmospheric pressure, 14.7 psia, which is exactly why it boils at that temperature at sea level. At 180 F it is roughly half that. At domestic cold-water temperatures it is under one psia. That range matters in the field: on a hot loop the vapor-pressure term dominates the suction budget, and on a cold domestic line it is nearly negligible, so the same noise in the two systems usually has different causes.
The call, as it went
A booster set in a mid-rise, complaint of a rattling in the pump room, reported as worse on summer afternoons. The building engineer had already been told by two people that the water was too hot for the pump.
What was ruled out, and on what evidence. Entrained air from a riser that had been drained the month before was the first candidate; the risers were vented and the noise did not change, and the character was wrong anyway, because air slugs gulp intermittently while this was continuous. A failing bearing or a loose impeller was second; that noise tracks shaft speed and is present at every flow, and this pump was silent at low demand. A solenoid slamming somewhere upstairs was third; water hammer is a single bang tied to a fixture event, not a sustained rattle.
What the gauges said. A compound gauge at the pump inlet read a steady positive value at low demand. At high demand it fell close to zero and the needle flickered a needle-width, which is the tell, and the noise arrived in the same second. Discharge pressure fell at the same time and the clamp meter showed motor current dropping by roughly a tenth, because a pump passing a vapor-and-liquid mixture is moving less mass and doing less work.
The number that lied. The seasonal story was real and the explanation was wrong. Differential across the suction strainer measured several psi where a clean strainer of that size should sit at a small fraction of that, and strainer loss rises with roughly the square of flow, so it only bit hard at peak demand. Peak demand in that building is a summer afternoon. The supply water temperature had moved a little, but at cold-water temperatures the vapor-pressure term was too small to account for it. The correlation was with demand, and demand happened to correlate with the season.
The fix was the strainer basket, which had construction debris in it. What made the diagnosis defensible was that the inlet gauge went back to a steady positive value at peak demand afterwards, not that the noise stopped.
Where the damage sits, and what that tells you
Cavitation damage is not distributed. It marks the place where pressure recovered, which is downstream of the place where pressure was lost, and reading that gets you to the right cause.
| Damage location | What it points to |
|---|---|
| Pitting on the leading edge of the impeller vanes, low-pressure side | Classic suction cavitation, bubbles forming in the eye |
| Pitting near the vane trailing edge or at the volute cutwater | Discharge recirculation, from running far below best efficiency, not a suction problem |
| Pitting just downstream of a control valve seat or an orifice | The valve is the low-pressure point, not the pump |
| Frosted or matte patch with no pitting yet | Early stage; you have caught it before metal loss |
The surface texture is distinctive: cavitation eats a spongy, honeycombed cavity into the metal rather than thinning it evenly. Erosion from grit polishes and grooves in the direction of flow. Corrosion attacks the whole wetted surface. If the damage is a sharp-edged pocket in an otherwise clean vane, you are looking at collapse energy, not chemistry.
Three things that sound like it and are not
- Entrained air. Gulping, intermittent, often clears after a purge and comes back over days if the leak is on the negative-pressure side of the loop. Air also compresses, so it damps the system rather than hammering it, and it leaves no pitting.
- Debris in the volute. A metallic tick or scrape that tracks shaft speed exactly. Cavitation noise does not track speed, it tracks flow.
- Flashing across a throttled valve. Same physics, different location, and the fix is at the valve rather than the pump. If the noise moves when you open the valve and the pump inlet gauge never dipped, the valve owns it.
What changes the reading
Two conditions genuinely invert how you interpret the same evidence.
Fluid other than plain water. Glycol mixes, refrigerant blends, fuel oil and process fluids all have their own vapor-pressure curves, and some are far more volatile than water at the same temperature. Do not carry a water intuition into them; go to the fluid's property data before you decide there is margin.
Altitude and open versus closed systems. An open tank at elevation starts with less atmospheric pressure pushing on the surface, so the same pump and the same piping that never cavitated at low elevation can cavitate on the identical install higher up. A closed loop with a working expansion tank and correct fill pressure does not have this exposure in the same way, which is why the same symptom in a sealed hydronic loop usually points at fill pressure or a waterlogged tank instead.
When you find it and cannot fix it the same day
You will sometimes confirm cavitation on a Friday afternoon on a pump nobody can shut down. There is one legitimate interim move and one that makes it worse, and they are easy to confuse because both involve a valve.
The legitimate move is to throttle the discharge. Adding resistance downstream walks the pump left along its own curve to a lower flow, and the suction head a pump requires falls as flow falls, so less flow means more margin. Two limits on it: do not take the pump below the manufacturer's stated minimum continuous flow, because a centrifugal pump run near shutoff overheats its own contents and loads its shaft radially, and do not do this at all on a positive-displacement pump, where throttling the discharge raises pressure instead of dropping flow. What throttling costs is covered by the valve card in this group.
The move that makes it worse is throttling the suction. It looks equivalent and it is the opposite: it drops the pressure arriving at the impeller eye, which is the exact quantity that was already too low. Any suction-side isolation valve on a running pump belongs full open, and a partly closed one is a finding rather than a fix.
Then log the interim setting on the equipment record with the date and the reason. An interim throttle that nobody wrote down becomes a permanent mystery restriction, and the next tech will find it as a balance defect.
How to verify you got this right
Do not accept "the noise stopped" as proof. Noise stops when flow drops for any reason, including a new restriction you just created.
- Read the inlet gauge at the worst condition you can create, not at the condition you found. If the system cavitates at peak demand, open enough terminals to reach peak demand and watch the needle there.
- Confirm the inlet stays positive and steady with no needle flicker at that condition. A steady reading at low flow proves nothing.
- Confirm discharge pressure and motor current are both back to their expected values together. A pump that is passing vapor shows low discharge and low current at the same time, so both recovering together is the signature of a real fix.
- Log the strainer or filter differential at a known flow, so the next tech has a clean baseline to compare against rather than a subjective one.
- If you replaced a pump, open the old one before it leaves the site and look at the vane leading edges. If they are clean and the volute cutwater is pitted, the replacement will do the same thing, because the cause was operating point and not suction.
References
- 29 CFR 1910.147, control of hazardous energy, for lock and tag plus pressure relief before opening a pump
- 29 CFR 1910.333(b)(2) for work at the starter or disconnect, with NFPA 70E-2021, 120.5 for live-dead-live proving
- Hydraulic Institute standards for cavitation, recirculation and best-efficiency terminology
- Published steam tables or the fluid manufacturer's property data for vapor pressure at the actual operating temperature
- See related: How Suction Conditions Fail in Practice; Why Throttling a Valve Changes More Than Flow; Reading Pump Curves Reference