Cavitation

Water boils at 100°C at atmospheric pressure. Drop the pressure and it boils colder, and at every temperature there is a pressure at which it will boil with no heat going into it at all. That pressure is the medium's vapour pressure. Cavitation is what happens when the pressure inside a pump drops below it.

Brennen's Hydrodynamics of Pumps defines it as the formation of vapour bubbles in low-pressure regions within a flow. The bubbles need somewhere to start. A genuinely clean liquid holds together in tension and has to be pulled well below its vapour pressure before anything forms, but real media are not clean: they carry suspended nuclei, microbubbles and particles, and those decide when cavitation actually begins. Gas bubbles already in the flow can also swell as they pass through the low-pressure region while the pressure stays above vapour pressure throughout. That is pseudo-cavitation, and it is a different fault.

The damage is in the collapse, not in the bubble. Past the impeller inlet the pressure recovers, the vapour has nowhere to go, and the cavity implodes. Next to a solid surface it cannot stay spherical. One side accelerates inwards faster than the other, and drives a high-speed re-entrant microjet through the far wall of the bubble. Shock waves off the collapse itself add to that, with impulsive pressures Brennen puts at two to three times those of the microjet. Repeat the loading at one spot on one vane and the metal fails by fatigue and flakes away, which is the pitted surface that identifies cavitation on a stripped impeller.

Vapour pressure climbs steeply with temperature, which is why a duty that has run for years can start cavitating after a process change. Put into head of the liquid being pumped, the form that goes into an NPSHa sum, water looks like this:

Take a duty from 20°C to 80°C and you have removed close to five metres from NPSHa without touching a single pipe. On a hygienic line the cleaning cycle can do it on its own, because clean-in-place runs hotter than the product does. So check the temperature the process actually reaches, including the wash, rather than the one on the specification.

The NPSHr curve a manufacturer publishes is a head-breakdown figure, not a line you can safely sit on. This is the part that gets missed, and it decides whether a pump has a service life or a replacement cycle.

ISO 9906, the acceptance-test standard for rotodynamic pumps, defines required NPSH loosely enough to cover several criteria, from visible cavitation through to a limit on cavitation erosion. It then names one of them as the standard basis for performance curves: NPSH3, the NPSH at which the total head of the first stage has already dropped by 3%. That is what almost every published NPSHr curve is. It is a repeatable, agreed test point, reached by lowering the suction pressure on test until the head falls away.

It is not the point at which cavitation starts. Brennen separates three critical values as you lower the inlet pressure on a running pump. First the inception value, where cavitation appears at all. Then a critical value fixed by a set percentage loss of head, typically 2, 3 or 5%. Then the breakdown value, where performance collapses. Between the first and the second it takes a further substantial drop in pressure before any loss of performance shows up at all, and the gap between them is not a fine margin: the inception value can be an order of magnitude larger than either the critical or the breakdown one.

So a duty point sitting exactly on the published NPSHr curve is a pump running with cavitation already established. A bare pass is not a margin. There is a whole Hydraulic Institute guideline for the gap you should leave, ANSI/HI 9.6.1, and its 2024 revision dropped the 3% metric in favour of a manufacturer-declared NPSHr that must sit at or above the tested NPSH3. It sets recommended minimum margins for ten market segments, water and wastewater, chemical process, slurry and building services among them, because those duties do not fail the same way as each other.

Work the margin at the worst case rather than the typical one: the lowest tank level the process reaches, the hottest the medium gets, and the highest flow the system could ever pull the pump out to. NPSHr rises steeply to the right on the performance curve, so a margin that looks comfortable at the duty point can be gone by run-out.

Treat the noise as a warning rather than a measurement, because a good deal of cavitation is pitched above what you can hear. The frequencies cavitation typically produces run from 10 to 100 kHz, set by the resonant frequency of the nuclei taking part. Human hearing gives out somewhere around 20 kHz. The gravel rattle from a badly cavitating wet end is the audible tail of something largely ultrasonic.

Noise does track damage, loosely. Brennen cites work by Lush and Angell in which the rate of weight loss correlated with the noise as flow velocity was varied, and notes that noise magnitude is often used as a crude measure of erosion rate. Crude is the operative word. It is a proxy, not a reading, and a pump can be well into damaging cavitation before it sounds dramatic.

Erosion also depends on where the pump is sitting on its curve, not only on how far the pressure has dropped. Erosion rate measured in a centrifugal pump against flow relative to design flow, reproduced from Grist, rises on both sides of the design point, because off-design incidence at the blade leading edge increases the cavitation. Weight loss also tends to begin only after an incubation period, so a pump can be accumulating damage for some time before there is anything to weigh.

What is reliable is the metal. Strip the pump and look at it. Cavitation damage is localised, because the cloud collapses in a particular place, and the surface opposite the leading edge of the next blade is a common one. It also has a crystalline, jagged look to it, which is the appearance of fatigue failure. Abrasion is the other thing that eats a wet end, usually grit on a works where the grit removal is not doing its job, and it does not look the same: an abraded surface is smoothly worn, with scratches from the larger particles.

A pump run well below its design flow cavitates too, and the suction pipework can be perfectly sound while it happens. Most people meet cavitation as a run-out problem: the pump pushed too far right on its curve, NPSHr climbing faster than NPSHa can follow, and the wet end starting to rattle. That is the common case and it is not the only one.

Below the design flow a loaded pump can drive its tip clearance flow upstream and set up an annular backflow ahead of the inlet plane, reaching many diameters up the suction line. Cavitating bubbles and vortices get swept into it. Brennen gives that its own name, backflow cavitation, and notes it is often the most visible form of cavitation an observer sees. Nothing about the suction line caused it. The operating point did.

Which is why an oversized pump can cavitate whichever way you run it. The mechanism is margin stacking, and nobody in the chain is being careless. The customer states a flow. The consultant adds margin for future duty. The contractor adds margin. The installer adds a little more head. What comes out of that is too big for the job. Run it as installed and it goes off the end of its curve hunting back pressure the system will not give it. Throttle it back to the real duty and you have parked it in the part-flow region where backflow cavitation lives. Neither is a pipework fault. Both were settled at selection, years before anyone heard a noise.

The impeller is the visible damage. It is rarely the expensive part.

A cavitating pump shakes, and on most machines the mechanical seal goes before anything else does. A seal needs a stable film of liquid between its faces to survive; a medium that is part vapour and vibrating does not give it one. Once the faces have run dry in patches they are finished, and you replace them alongside whatever the vibration has taken out of the bearings. That is the same ending as dry running, reached by a different road.

Flow and head go unstable too, which matters more on a process line than it sounds. A dosing duty that swings, a transfer pump that will not hold rate, a booster set hunting: all of them get chased as control faults for a while before anyone thinks to look at the suction side. The noise and vibration also travel out into the pipework and the structure, and in a food factory or a plant room that is a complaint in its own right.

None of it stops when you fit a new wet end. Put a fresh impeller into the same suction conditions and it goes the same way as the last one. The installation is the fault, and the pump has been telling you so.

A positive displacement pump cavitates as well, and survives it far better. Cavitation gets written up as a centrifugal problem because that is where it does the most damage. But a rotary lobe or progressing cavity pump runs the same mechanism: the medium cannot fill the cavity as fast as the pump opens it, and vapour appears in the gap.

What differs is the consequence. With the seal selection right, a positive displacement pump tolerates a degree of cavitation. You get noise and vibration rather than a ruined wet end, and it holds its rate, because a PD pump's flow barely moves as pressure changes. That tolerance is part of why PD gets specified on the awkward media to begin with.

The causes shift too. On a thick medium the limit is usually the inlet rather than the vapour pressure: a bore too small, a run too long, or a pump turning faster than the medium can follow it in. The answer is nearly always to slow the pump down and open out the inlet, and slowing down is the right move on those media anyway, since abrasive and shear-sensitive products both want pumping slowly. Viscosity decides how far you have to go with it, and the figure to design against is the worst the medium reaches, not the one it sits at most of the time.

Work the suction side before you touch the pump. Clean the strainer, open the inlet fully, check the tank still delivers the flow at its lowest working level, and find out what the medium's temperature is doing. Those cost nothing, and one of them is usually the whole answer. Preventing cavitation in a centrifugal pump runs the full sequence, including how to separate cavitation from air entrainment before you spend money curing the wrong one.

If the suction side is as good as the installation will allow and the margin still is not there, move the duty point next and change the pump last. A pump with a lower NPSHr is usually a larger, slower machine. You pay for it in capital and often in efficiency, and it does nothing whatever about a tank that cannot supply the flow.

Tell us your duty and we will size it. Send the flow, the head, the medium and the connection, and send the suction side with them: the lift, the pipe run and bore, the tank, and the temperature the process really runs at. Where the answer is a change to the pipework rather than a pump, we will say so, because it is usually the cheaper fix and it is the one that holds.