What Air Entrainment Does That Cavitation Does Not
Why this matters
Both faults put gas in the impeller eye and both make a pump noisy and short of flow, which is why they get confused in the first minute of a call. Everything after that first minute is different. One of them eats the impeller and leaves the seal alone until the vibration gets there; the other leaves the metal untouched and kills the seal directly by taking the liquid film out from between its faces. One is fixed by finding feet of suction head; the other is fixed by finding a joint. And the quantity involved is not intuitive: in the case worked below, under one cubic foot per minute of air ingress, on a pump moving 180 gpm, is enough to take a real bite out of the head, and because the whole section is under vacuum there is nothing to see and nothing to hear at the leak. Getting the two mixed up costs a season of the wrong work.
Before you chase air on a running pump
A pump that has gas-bound is doing shaft work on liquid it cannot move, and the liquid in the casing heats fast. Do not open a vent, a drain or a gauge plug on a pump that has been running gas-bound. Stop it, isolate it, let it cool with a gauge watched, and drain to a routed hose rather than to the floor. On hot service, opening it hot is a flash-to-steam event, not a spill.
Wetting joints to find air ingress is done with the pump running and the coupling guard on. Sleeves buttoned, no rings, and nothing but a squeeze bottle goes near the machine.
If the section is to be pressurized to find the leak, the weakest component sets the pressure, and on a suction line that is usually a flexible connector, a sight glass or the strainer cover rather than the pipe. Use water rather than compressed air wherever the geometry allows: a gas-pressurized volume stores energy, and a failure is a projectile event rather than a squirt. Never pressurize with air a system that contains flammable vapor.
Two ways gas gets into the eye, and only one of them implodes
Cavitation makes its own gas. When local pressure at the eye falls below the liquid's vapor pressure at its actual temperature, the liquid boils in place. The bubbles are vapor of the liquid itself. As the impeller does work on them and pressure recovers a fraction of an inch downstream, they condense back to liquid in an instant, and the surrounding liquid rushes into the void. That collapse is what removes metal. A sibling card owns the mechanism and how it announces itself; this one is about telling it apart from its neighbour.
Entrained air is delivered. It is free gas that came from outside the liquid: through a joint on a section that is under vacuum, through a vortex over a tank outlet, from a return line splashing into the source, or from a partly drained line. Because it is a permanent gas rather than vapor, recompression does not make it disappear. It compresses, it passes through, or it accumulates in the low-pressure region at the eye and stays there. Nothing implodes, so nothing pits.
There is a third, quieter mechanism worth naming so you do not misfile it: gas that was dissolved in the liquid coming out of solution as pressure falls. Air and carbon dioxide are far more soluble at high pressure than at low, so a liquid that arrives saturated releases free gas at the eye. It behaves like entrained air once it is out, because it is a permanent gas, and it redissolves slowly rather than collapsing.
The field key
| What you are looking at | Cavitation | Free gas at the eye |
|---|---|---|
| Where the gas came from | the liquid itself, boiling in place | outside the liquid, or out of solution |
| What happens as pressure recovers | violent collapse | compression, or accumulation |
| Sound | sharp and random, like gravel through the casing | lower, gurgling or burbling, often rhythmic |
| Discharge gauge | rough, but a broadly steady mean | swings you can watch, sometimes several psi |
| Motor current | down with the lost flow, unsteady | down and swinging with the gauge |
| Metal condition when you open it | pitting on the low-pressure side of the vanes near the eye, working back into the passage | no pitting from the gas |
| What fails instead | impeller first, then seal and bearings via vibration | seal faces first, from lost film and dry running |
| End state if ignored | material loss, then breakthrough | total loss of pumping, pump mechanically intact |
| Where the fix lives | in the suction head budget | at joints, tank level, and submergence |
The row that decides most arguments is the metal. Pull a pump that has been noisy for a season: pitting near the eye is cavitation and no amount of joint-chasing will fix it, and clean vane surfaces under a wrecked seal is free gas and no amount of suction-head work will fix that.
How little air it takes
A centrifugal pump commonly handles a couple of percent gas by volume at the eye with a small head penalty, loses substantial head somewhere in the mid single digits, and gas-binds around ten percent. Those bands are impeller-geometry-specific and the pump manufacturer owns the real numbers; use them as a screen, not as a specification. Two things about them get skipped.
The percentage is by volume, at the pressure at the eye, not at atmospheric. Air drawn in through a joint enters at atmospheric pressure and then expands on its way to the lowest-pressure point in the system. That expansion is the correction, and it always runs against you.
A suction line under vacuum has no visible leak. The pressure difference across the joint points inward. Nothing drips, nothing hisses outward, and every visual inspection comes back clean.
Worked determination. Suction-lift transfer pump, water at 85 F, specific gravity 0.9965, so 2.318 ft per psi. Station barometric 14.3 psia. Suction line 4 in schedule 40. Measured flow 180 gpm. Suction gauge, 0.9 ft above the shaft centerline, reads minus 4.2 psig. Published required suction head at 180 gpm is 8.0 ft.
First, rule cavitation in or out arithmetically rather than by ear:
| Line | Value |
|---|---|
| Gauge, static | -4.2 psig |
| Absolute, using station barometric 14.3, not 14.7 | 10.10 psia |
| Minus vapor pressure at the measured 85 F, 0.596 psia | 9.50 psi |
| To feet at 0.9965, so 2.318 ft per psi | 22.03 ft |
| Plus velocity head, 4.54 ft/s at the measured 180 gpm | +0.32 ft |
| Plus datum, gauge 0.9 ft above the eye | +0.90 ft |
| Available | 23.25 ft |
| Required, read at the flow the pump should be delivering, not at the degraded flow | 8.0 ft |
A ratio of 2.9 is not a marginal suction condition. Cavitation is out, and it took one gauge, one thermometer and four minutes.
Now the gas. The eye sits at roughly the inlet absolute pressure of 10.10 psia, and air enters the leak at 14.3 psia, so it expands by 14.3 divided by 10.10, a factor of 1.42 on its way in. That expansion is the correction the band needs: to reach 4 percent by volume at the eye you only need about 2.8 percent measured at atmosphere.
Put that in units you can picture. Liquid flow is 180 gpm. The 2.8 percent at atmospheric conditions that the expansion factor just gave us is 5.04 gpm of air, which is 0.67 cubic feet per minute. Under one cubic foot per minute of air, into a pump moving 180 gpm, is a real head penalty, and two thirds of a cfm through a threaded joint or a packed gland is a leak nobody would find by looking, listening or feeling, because the pressure gradient carries it inward.
The failure mode, and it is the one that produces repeat visits: the tech hears noise, calls it cavitation, cleans the strainer and finds it fairly clean, then quotes a pump with a lower required suction head. The new pump does the same thing, because the required value was never the constraint. The arithmetic above closes that door in the first hour.
Where the air comes in
- A packed gland on a suction lift. Packing is a designed leak path and the pressure gradient decides which way it leaks. Below atmospheric, it draws air in, and the gland stays bone dry while doing it. A packed box on a lift needs a lantern ring with a flush at a pressure above box pressure precisely so this cannot happen.
- Threaded joints and flange gaskets on the vacuum section. Any joint upstream of the pump on a lift is a candidate, and a joint that has never dripped is not evidence of anything.
- Valve stem packing and instrument connections on the same section, for the same reason.
- A vortex over the tank or sump outlet when submergence falls at low level. That path is worked through in a sibling troubleshooting article.
- A return line that splashes into the source and carries entrained air back down to the outlet, particularly where the return discharges above the liquid surface.
- An air-release or vacuum-breaker device set for the wrong pressure, or stuck.
- A partly drained line refilling, which is a transient and clears, so it belongs on the list only when the complaint is at startup.
Finding it when nothing leaks out. Three methods, all generic. With the pump running, wet each suspect joint with a soap solution or a heavy-bodied liquid; the joint stops drawing air for a moment and the discharge gauge swing and the noise change audibly. Or isolate the section, put it under a low positive pressure limited by the weakest component, and look for the drip in the normal way. Or hold a vacuum on the isolated section with a vacuum pump and watch the decay: note that a decay measurement reads the deficit, meaning the volume that leaked in, so what you get is an inward leak rate directly, not something you have to convert.
Where the two overlap, and what to do then
They can run together, and on a hot system they usually do, because the same low pressure that boils the liquid also drags air through every joint on the section. When both are present, the order is fixed: fix the air path first. Free gas at the eye lowers the effective suction condition further, so an air leak can push a pump that had adequate margin over its own onset, and you cannot judge the suction budget honestly until it is out.
The one situation that inverts this: if the metal already shows pitting, the suction budget was inadequate at some point regardless of what the joints are doing, and sealing every joint will leave a pump that is quieter and still losing material.
How to verify you got this right
After the repair, take the discharge gauge reading over a full minute and describe its behaviour in writing, not its value. Free gas produces a swing; the useful evidence is that the swing has gone, and a single spot reading cannot show that.
Then re-take the inlet reading at the same flow and temperature and compare with the pre-repair figure. Sealing an air path does not change available suction head, so that number should be essentially unchanged. If available suction head moved a lot, you did not fix an air leak, you changed something in the friction or the level, and the diagnosis needs re-running.
Finally, put the flow and the liquid temperature on the record beside both readings. Without them the next tech cannot tell whether the swing came back or the conditions changed.
References
- 29 CFR 1910.147, control of hazardous energy, for lock and tag plus pressure relief before opening a pump or its suction piping
- Pump manufacturer's data for the gas-handling capability of the specific impeller and for the required suction head at the running flow
- Hydraulic Institute standards for net positive suction head terminology, as adopted by the specifying engineer or by contract
- See related: What Cavitation Is and How It Announces Itself; The Pump That Lost Flow Every Afternoon With Nothing Changed; What Packing Does That a Mechanical Seal Cannot