NPSHa vs NPSHr: Why Pumps Cavitate and How to Fix It

A pump that cavitates does not usually announce itself with a calculation error. It announces itself with a noise like gravel in the casing, a head curve that has quietly collapsed, and an impeller that is being eaten from the inside. By the time anyone measures it, the damage is weeks old.

The quantity that governs whether that happens is NPSH — net positive suction head. It is not difficult arithmetic. What makes it catch people out is that the term which usually does the damage is the one that has nothing to do with the pipework at all.

Here is the result worked through at the end of this article. Take one suction line — same pipe, same length, same fittings, same flow — and change only the water temperature:

The friction loss moves by four centimetres across that entire range. The vapour-pressure term moves by seven metres. The pipework is not what killed this pump; the temperature is.

Note that the density column is doing something helpful and small: as the water thins, the same atmospheric pressure supports a taller column, so that term gains about 0.35 m over the range. It is nowhere near enough to offset a 7 m loss.

The two quantities, and who owns each

NPSH comes in a pair, and half the confusion in the field is people comparing one against something that is not the other.

• NPSHa — available. How much suction head your system delivers to the pump inlet, above the liquid's boiling point at that temperature. It is a property of the installation: the source pressure, the height, the suction pipework, the temperature. You calculate this.
• NPSHr — required. How much the pump needs at that flow to avoid cavitating. It is a property of the machine, measured on a test rig, and it rises with flow. The manufacturer gives you this, as a curve against capacity.

The design rule is NPSHa ≥ NPSHr + margin. Neither number means anything alone — quoting "the pump needs 3 metres" without knowing what the system delivers is not a check.

One definitional trap worth clearing immediately: NPSHr is not the suction pressure the pump needs. It is measured as the point at which the pump's developed head has already dropped by 3% due to cavitation ($\text{NPSH}_3$). At exactly NPSHr the pump is cavitating, mildly. That is why margin is not optional padding — it is the difference between the catalogue's failure threshold and actual clean operation.

Calculating NPSHa

For a pump drawing from a tank or vessel:

• $P_{abs}$ — absolute pressure on the liquid surface (Pa) — atmospheric for an open tank, the vessel pressure for a closed one
• $P_v$ — the liquid's vapour pressure at the pumping temperature (Pa, absolute)
• $\rho$ — liquid density at that temperature (kg/m³)
• $g$ — gravitational acceleration (9.81 m/s²)
• $z$ — static height of the liquid surface above the pump centreline (m) — negative for a suction lift
• $h_f$ — total friction and fitting loss in the suction line (m)

Why there is no velocity-head term here, when NPSH is defined on total head at the suction: there is one, and it cancels. Starting from a stationary liquid surface, the fluid must be accelerated to pipe velocity, and that costs exactly $V^2/2g$ of static pressure — which is then handed straight back as the velocity head NPSH counts. Measured from the tank surface the two are the same number with opposite signs. Adding it (or subtracting it) is a common slip worth about 0.16 m in the example below.

You do need the term if you start from a measured static pressure at the suction nozzle rather than from the tank, because then the acceleration has already happened and is not in your starting number. That is the form the solver uses internally.

Everything is in metres of the liquid being pumped, and everything is absolute pressure. Working in gauge here is the single most common way to get NPSH wrong, because atmospheric pressure is the term doing most of the lifting — about 10.35 m of water at sea level.

The $P_v$ term is subtracted because that is the head the liquid needs simply to stay liquid. Whatever is left over is what the pump has to work with.

Why temperature dominates

Vapour pressure is not linear in temperature — it climbs steeply. Water goes from 2.34 kPa at 20 °C to 70.18 kPa at 90 °C: a factor of thirty. Expressed as head, that is 0.24 m against 7.41 m, and it comes straight off your available suction head.

Meanwhile the friction term in the table above barely moved. That is the practical lesson, and it inverts most people's instinct: on a hot line the suction pipework is almost never the problem, and on a cold line it almost never is either. Temperature and static height are the terms that decide.

This is why boiler feed, condensate return, hot-oil circulation and anything drawing from a flash vessel are the classic NPSH casualties. A liquid at or near its saturation temperature has, by definition, nearly zero vapour-pressure margin — which is why those services are almost always designed with a flooded suction and a tall static leg.

The worked example

An open tank, water at 20 °C, feeding a pump through 12 m of DN80 Schedule 40 (inside diameter 77.9 mm) at 30 m³/h. The pump sits 2 m above the water level — a suction lift. Suction fittings: a sharp-edged entrance, two 90° elbows and a full-open gate valve (ΣK = 1.72). Assume a datasheet NPSHr of 3.0 m.

Stated assumptions, so the result is reproducible: absolute roughness 0.045 mm (clean commercial steel), friction factor from Colebrook-White, atmospheric pressure 101.325 kPa at sea level, and density, viscosity and vapour pressure taken from IAPWS-IF97 at each temperature rather than held at their 20 °C values. Fittings are Crane TP-410 at $f_T = 0.018$ for 3-inch pipe: sharp-edged entrance 0.5, two 90° elbows at $30 f_T$, one gate valve at $8 f_T$, giving ΣK = 1.72.

At 30 m³/h the velocity is 1.748 m/s, Reynolds number is 135,700 — firmly turbulent — the Colebrook friction factor is 0.0199, and pipe plus fittings come to 0.747 m of head. So:

• 10.35 — atmospheric head at sea level, in metres of 20 °C water
• 2.00 — the suction lift, subtracted because the pump is above the source
• 0.747 — friction and fittings in the suction line
• 0.24 — vapour-pressure head of water at 20 °C

7.37 m against a 3.0 m requirement is a comfortable installation. Now heat the same water to 80 °C and nothing else changes — same pipe, same flow, same geometry:

2.95 m against a 3.0 m requirement. The pump cavitates — and it fails by five centimetres. Nothing was done to it. The same installation that had 4.4 m of spare margin at 20 °C is now under water, purely because the liquid got hotter.

At 90 °C it reaches 0.58 m and the pump is gas-bound.

What actually fixes it, ranked

This is the part worth internalising, because the instinctive fix is close to the worst one. Starting from that 80 °C case at NPSHa = 2.95 m:

Lowering the pump buys about seven times what upsizing the pipe does. Every metre of static height is a metre of NPSHa, directly and for free, while attacking the friction term can only ever recover a fraction of 0.747 m — you cannot win back head that was never being lost.

So the order to think in is: geometry first, temperature second, pipework last. Reach for a bigger suction line only once the static arrangement is fixed and you are chasing the final few tenths.

Two more levers exist when those are exhausted: raise the pressure on the source vessel (which adds to $P_{abs}$ directly), or specify a pump with a lower NPSHr — an inducer, a lower-speed machine, or a double-suction impeller, which halves the flow per eye.

How much margin is enough

Because NPSHr is the 3%-head-drop point rather than the onset of cavitation, you need real margin over it. Common practice:

• A fixed minimum of 0.5 to 1 m over NPSHr for ordinary cold-water duty.
• A ratio of 1.1 to 1.3 × NPSHr for larger or more critical machines.
• More for hot or volatile liquids, where a small temperature excursion moves the vapour-pressure term a long way — as the table above shows.

Whatever margin you choose, check it at the worst operating point, not the design point. NPSHr rises with flow, so a pump running out to the right of its curve — a system with less resistance than predicted, a parallel pump tripping, a control valve opening fully — demands more NPSH exactly when the extra flow is also increasing suction friction.

Cavitation and flashing are not the same failure

These get used interchangeably and they are different mechanisms with different consequences.

• Cavitation — the liquid drops below its vapour pressure at the impeller eye, forms bubbles, and those bubbles collapse violently as pressure recovers across the impeller. The collapse is what does the damage: pitted impellers, eroded vanes, bearing and seal loads, the characteristic gravel noise. The liquid is still liquid when it leaves.
• Flashing — the liquid drops below its vapour pressure and stays vapour, because the downstream pressure never recovers. There is no collapse and usually no pitting; instead the pump gas-binds and stops pumping altogether, and the vapour carries on downstream.

Both start the same way — pressure below $P_v$ — and the difference is only whether pressure recovers afterwards. Cavitation destroys the pump slowly; flashing stops the process immediately. Neither is acceptable, but they present completely differently, so knowing which one you are looking at tells you where to look.

Common mistakes, quickly

• Working in gauge pressure. NPSH is an absolute-pressure calculation, and atmospheric is the largest single term.
• Using vapour pressure at the wrong temperature. Design for the hottest the liquid will actually be — summer conditions, a warm start, recirculation heating a closed loop.
• Comparing NPSHa at design flow to NPSHr at rated flow. Both change with flow, and they must be compared at the same point.
• Treating NPSHr as a safe operating value. It is where the pump has already lost 3% of head.
• Forgetting the strainer. A suction strainer blinding off is a friction term that grows over time — the classic "it used to be fine" cavitation call.
• Assuming a bigger suction pipe will fix it. As the table above shows, it is usually the smallest lever available.
• Ignoring altitude. At 1,500 m the atmospheric term is roughly 1.5 m lower, straight off your margin.
• Forgetting dissolved gas. Liquid drawn from an open tank carries dissolved air, which comes out of solution as pressure falls and eats into the effective margin before the vapour pressure itself is reached. It is a second-order effect next to temperature, but it is one reason a cold-water pump can grumble at a margin the calculation says is adequate.

Where a network simulator fits

None of this needs software for a single pump on a known suction line — the equation above and a vapour-pressure table will do it.

What changes in a real system is that $h_f$ and the flow are not known in advance. The suction friction depends on the flow, the flow depends on where the pump sits on its curve, and the pump's position on its curve depends on the whole network downstream. Change a valve position and the flow moves, which moves both the suction friction and the NPSHr you need to clear.

SimuPipe computes NPSH available at every liquid pump on every run, from the solved suction pressure and velocity rather than an assumed flow. If you enter the pump's datasheet NPSHr it will warn you twice: once when the margin falls under half a metre, and again when the available head drops below the requirement outright. The check is deliberately opt-in — the available value is always shown, but nothing warns until you supply a requirement, because we cannot know your pump.

For the suction friction half of the calculation, see how to calculate pressure drop with Darcy-Weisbach and minor losses in pipes; vapour pressures for 38 fluids are in the fluid properties table, and there is a solved booster-pump example in the worked examples gallery.