The Pump That Was Replaced Twice and Failed the Same Way Each Time

Why this matters

Every diagnosis in this case was correct. The seal had failed. The faces were heat-checked. The replacement seal was the right one, fitted right, and the third time around the whole pump was replaced with an identical unit by a competent crew. It came back anyway, on the same schedule, with the same wear pattern. Being right about the component was not enough, because the component was not the fault - it was the only part of the machine reporting a suction condition that nobody had measured. The evidence that finally named it was already in the shop's own records: the interval between failures. Three failures spaced within a hundred run hours of each other is not three coincidences, it is a measurement of something that is still there.

Stop the exposure before anyone diagnoses anything

The fluid in this case is water at 170 F in an open process tank, and every step below happens after these.

Isolate the pump at its disconnect and apply a lock and a tag under 29 CFR 1910.147 rather than pressing stop, because a level control can restart this pump on its own. Close and lock the suction and discharge valves, then bleed at the casing vent to a hose routed to a floor drain, standing to the side, wearing a face shield and heat-rated gloves. Water at 170 F scalds on contact and this line is under static head from the tank above it. Do not break a casing joint on hot fluid; let the casing drop below 120 F first if the schedule permits, and if it does not, drain it fully to a drain and confirm the vent runs dry before any bolt comes out.

Do not clean or read the strainer basket while the line is hot and under head. Same isolation, same drain, same face shield.

If the tank or the pit below it has to be entered to read a level or a temperature, stop. That is a permit-required confined space under 29 CFR 1910.146 in general industry, or 29 CFR 1926 Subpart AA on a construction site, and it needs the permit, the atmospheric testing and the attendant before anyone's shoulders go past the opening. A tank temperature can be taken with a probe through the fill opening or read from the process control, and it was in this case.

What the record already said

The shop had three tickets on the same pump position over about two years.

Three things in that table are doing work. The failure mode is identical every time, which rules out a random parts problem. The alignment was checked and was good, which takes the mechanical drivetrain off the table. And the intervals are not just similar, they are falling: 2,050, then 1,980, then 1,910, each one about 70 hours shorter than the one before, roughly three and a half percent each time. A component that fails at a repeatable run-hour count is being consumed by a condition that is present every time the machine runs, and a count that keeps shrinking says the condition is getting worse.

The pitting found on the third teardown is the piece that named it. Erosion on the low-pressure side of the vanes near the impeller eye is where vapor bubbles collapse. That is a suction-side signature, and it appears on the impeller long before it costs the pump its head.

Why the first two repairs were right and still failed

A mechanical seal runs on a very thin film of the pumped fluid between two flat faces. That film is doing two jobs at once: it separates the faces and it carries the friction heat away. When the fluid at the seal chamber is close to its own boiling point for the pressure there, the film flashes to vapor between the faces, the faces touch dry, and the heat has nowhere to go. What you find afterwards is exactly what was found here: heat-checked faces, fine radial cracks from thermal shock.

So the seal report was accurate and the repair was appropriate. What it could not tell anyone is why the film kept flashing. That question lives on the suction side, and it is answered with arithmetic rather than with a teardown.

Running the suction arithmetic at the machine

Net positive suction head available is the margin, in feet of the fluid being pumped, between the absolute pressure at the pump suction and the pressure at which that fluid boils at its current temperature. Four terms, each measured or read on site, each printed on its own line so the corrections are visible.

As designed, years earlier: open tank, water at 120 F, level 3 ft above the pump centerline, 1.5 ft of suction friction, site barometric pressure 14.2 psia.

• Surface pressure in feet of this fluid: 14.2 psia x 2.31 / specific gravity 0.990 = 33.1 ft
• Static level above pump centerline: +3.0 ft
• Suction friction loss: -1.5 ft
• Vapor pressure at 120 F, 1.69 psia, in feet of this fluid: 1.69 x 2.31 / 0.990 = -3.94 ft
• NPSH available = 30.7 ft

As found: heat recovery had been added to the process two and a half years ago, taking the tank to 170 F. A strainer had been added after the first failure. The level control lets the tank draw down to 0.5 ft above the pump centerline at the end of a batch.

• Surface pressure, now at specific gravity 0.976 because hot water is less dense, so the same barometric pressure is worth more feet: 14.2 x 2.31 / 0.976 = 33.6 ft
• Static level at end of batch: +0.5 ft
• Suction friction with a partly loaded strainer: -4.5 ft
• Vapor pressure at 170 F, 6.0 psia, in feet of this fluid: 6.0 x 2.31 / 0.976 = -14.2 ft
• NPSH available = 15.4 ft

The suction condition lost roughly half its margin, and the split is worth stating because it decides the fix. Of the 15.3 ft lost, the temperature rise accounts for about 10.3 ft, or two thirds. The strainer accounts for 3.0 ft, about a fifth. The lower operating level accounts for 2.5 ft, about a sixth. The density change gave back half a foot, which is the one term that moved in the shop's favour and is also the term most likely to be dropped by someone doing this in their head.

What that margin has to beat, and why the curve's number is not it

The curve sheet gives NPSH required as 14 ft at this duty flow. Compare:

• Normal tank level with a clean strainer: 33.6 + 3.0 - 1.5 - 14.2 = 20.9 ft available, a margin ratio of 20.9 / 14 = 1.49
• End of batch with a loading strainer: 15.4 ft available, a margin ratio of 15.4 / 14 = 1.10

NPSH required on a published curve is defined at the flow where the pump's head has already dropped by three percent from cavitation. It is a damage-in-progress threshold, not a safe operating point, so running at a ratio of 1.10 is not running with ten percent to spare. Common design practice is a margin ratio of at least 1.2 to 1.5, or a few feet, whichever is greater, with more required on high-energy services; the Hydraulic Institute's margin guidance and the pump manufacturer's own recommendation for the specific casing own that number, not a rule of thumb.

That is the shape of the failure. For most of each batch the pump has adequate margin and does no damage. During the last part of every batch, with the level down and the strainer loading, it drops under margin and erodes. The damage is dose-based, the dose per batch is nearly constant, and the run-hour count at failure is therefore nearly constant - which is precisely what the ticket history showed, and what got shorter as the strainer fouled faster each cycle.

The correction and how it was confirmed

The process needed 170 F, so the largest term was not available to fix. The two that were - static level and suction friction - were both moved, in three changes.

• The pump was lowered on a new base so the minimum static level became +3.5 ft instead of +0.5 ft
• The suction line was upsized one pipe size and the strainer moved to a location and mesh where its loaded loss stays near 1.5 ft, with a differential gauge across it so loading is visible rather than inferred
• The low-level cutout was raised so the tank cannot draw below the new minimum

Recomputed at the worst condition: 33.6 + 3.5 - 1.5 - 14.2 = 21.4 ft available, a margin ratio of 21.4 / 14 = 1.53, which now clears the practice range at the worst point of the cycle rather than only at the best.

Confirmation was not "it has not failed yet." The fourth pump passed 2,050 run hours, the point where the first one died, with no seal weep and with the overall vibration reading flat against the baseline taken at start-up. The strainer differential has stayed under the value that corresponds to 1.5 ft of loss. At the first scheduled inspection the impeller showed no new pitting, which is the direct evidence that the collapse stopped rather than the indirect evidence that a seal survived.

What the shop changed about how it books repeat work

The ticket history contained the answer for two years and nobody could see it, because each visit was opened as a new job with no field for the last one. Two changes fixed that, and they are cheap:

Record run hours at every failure, not calendar dates. A pump that runs eight hours a day and one that runs sixteen fail at the same run count and at completely different calendar intervals, so calendar dates hide the pattern that run hours expose.

On the second identical failure at the same position, stop replacing and start measuring the system. The rule the shop wrote is a count, not a judgment: the second occurrence converts the job from a repair to an investigation, and the investigation starts on the suction side, because that is where a centrifugal pump is usually killed even though the complaint always arrives from the discharge side.

References

• Hydraulic Institute standards for NPSH margin and centrifugal pump application, in the edition your specification or service contract references, which defines NPSH required at the three percent head-drop point
• Pump manufacturer curve sheet and seal manufacturer documentation for the specific unit, which own the NPSH required value and the seal chamber conditions
• 29 CFR 1910.147 for isolation and stored energy before opening the pump, and 29 CFR 1910.146 in general industry with 29 CFR 1926 Subpart AA on construction sites for any tank or pit entry
• See related: What Cavitation Is and How It Announces Itself; How Seals and Gaskets Fail; How to Prime a Pump Without Damaging It