How a Mechanical Seal Fails and What You See

Why this matters

A seal that comes out of a pump is a written record of what the system did to it, and it stays readable for about as long as it takes to wipe it on a rag. Read it and you find out whether the seal reached the end of its life or whether something in the installation killed it, which are the only two answers and they lead to completely different work. Skip it and you fit a new one, which will fail on the same schedule, because the schedule was never the seal's. The case at the end of this card is a pump that ate three seals in fourteen months, roughly one every five months, and the evidence on the faces pointed at a control sequence that nobody had ever connected to the seal at all.

Before you pull a seal

The casing may still be at pressure with the motor off, if the discharge check valve seats. Lock and tag the disconnecting means under 29 CFR 1910.147 rather than switching it off, relieve at the casing vent, and confirm it on a gauge before a gland bolt moves.

If the pump has been running against a closed discharge, it is a scald hazard and not an ordinary drain. A centrifugal at shutoff puts nearly all its shaft power into the small volume of liquid in the casing, so that liquid can be well above its boiling point at atmospheric pressure and will flash the moment anything is opened. Stop the pump at the starter, isolate it, let it cool with a gauge watched, and drain to a routed hose. This applies to the diagnostic in this article too: watching a start to see what the discharge valve does means standing at a gauge, not at a vent, and if the valve stays shut longer than the sequence claims, the pump gets stopped at the starter rather than relieved at a plug.

Clean the faces to inspect them with a lint-free cloth and a solvent the SDS says is compatible, used with the ventilation that SDS calls for, never with compressed air and never with anything abrasive. Compressed air drives grit across the surface you are trying to read and puts the residue of whatever the pump was handling into the air you are standing in.

Do not grind or dress a face. Dry-machining silicon carbide or tungsten carbide is a respirable-dust exposure, and on a cobalt-bound carbide, cobalt is the health concern, so the control is respiratory under 29 CFR 1910.134 rather than a glove. Lapping is a wet shop operation.

What normal looks like

You cannot call a failure without knowing what success looks like, and most techs have never deliberately looked at a seal that simply wore out.

Normal end of life is an even, narrow, polished wear band that runs the full circumference at a constant width, on faces that are still flat, with a soft carbon face that has lost measurable thickness and a hard face that has not. The secondary elastomer is still soft and springs back when you press it. The shaft or sleeve under it is clean and unmarked. A seal in this condition leaked gradually and increasingly for some weeks before it was pulled.

Everything that is not that description is the system talking.

Three things a failed seal tells you

The wear band names the mechanics. Its width, its evenness around the circumference, and whether the two faces still sit flat on each other tell you whether the seal was running square and steady.

A deposit names the fluid condition. Anything baked, crusted or plated onto the faces happened because the liquid at the faces was hotter than the liquid in the pipe, or because it was carrying something.

The elastomer names chemistry and temperature, and the two look different. Over-temperature hardens it, takes a permanent set, and kills its resilience without changing its size much. Chemical attack changes its size: swelling, softening, or shrinking with surface cracking. Do not lump them together, because one sends you to the flush and the other to the compound.

The field key

Ten rows, and the last one is the one people do not check because it looks like a good part. A seal that comes out unworn did not fail; it was never asked to work.

Direction check: what heat checking can and cannot mean

Heat checking is fine radial cracking on the hard face, caused by rapid heating of a thin surface layer that the bulk of the ring does not follow. It means the faces got hot, which requires either that there was no liquid between them or that they were generating more heat than the flush could remove.

Now run it the other way, because that is where the misreads live. A seal that leaked steadily and ran cool cannot produce heat checking, because a leaking seal has liquid between its faces by definition. So when the complaint is "it has been weeping slowly for months" and the hard face comes out cracked, the weeping was not the failure developing slowly. It was a late symptom of periodic dry running, and the periods are what you go looking for. At the other end of that range, a seal that genuinely wore out over months shows a widened, polished, uncracked band and leaks increasingly rather than in episodes.

Worked case: three seals in fourteen months

A transfer pump, 25 hp motor, water service starting at about 95 F. Three seal replacements in fourteen months, roughly one every five months. Each replacement was a correct repair of a genuinely failed seal.

What came out on the third pull. Hard face heat checked across the full width of the wear band. Carbon face worn evenly, full circumference, constant width, no edge chipping. Shaft elastomer hard and set, dimensions unchanged, no swelling. No particles embedded, no scoring. Sleeve clean, no fretting.

Cross-referenced against the field key, that combination rules out four whole families in one pass: no edge chipping means no shock loading, no one-sided band means the seal was running square, no embedded particles means abrasives never reached it, and an elastomer that hardened without changing size means heat rather than chemistry. Everything left points at the faces getting hot with an otherwise correctly installed seal.

Where the heat came from. Watching a start with a gauge on the discharge showed the control sequence holding the discharge valve closed for 4 minutes after the motor starts, twice a day.

Put a number on what those 4 minutes do. The pump manufacturer's power curve gives about 12 hp at shutoff for this impeller, which is roughly half of the motor rating and is the kind of proportion a radial-flow impeller shows; the curve owns the figure, not the rule of thumb. Casing volume is about 1.2 gallons, which is about 10.0 lb of water.

So each start spends no more than about 1.5 minutes of its 4-minute dead-head period with the liquid at the faces at or past its boiling point at chamber pressure, and possibly none if the casing metal absorbs more than assumed. Twice a day for the roughly five months a seal was lasting is on the order of 280 starts, so somewhere up to about 7 hours of accumulated dry-face running per seal life. The sibling card on what a seal is sealing against works the chamber-margin arithmetic in full; the point here is that the evidence on the face and the number from the sequence agree, and neither one alone would have been convincing.

The fix was the sequence, not the seal. The discharge valve was re-sequenced to be open before the motor is called, and a minimum-flow arrangement was quoted separately for the case where the system genuinely cannot take full flow at start, which is its own subject.

The failure mode if you skip the read: the third seal gets replaced with a different face pair on the theory that the last one was a bad part. It fails on the same five-month clock, and now there is a fourth invoice and a customer who has stopped believing anybody.

How to verify you got this right

Photograph both faces and the elastomer before anything is cleaned, and again after, with the pump identified in the frame. The pattern is evidence and cleaning destroys some of it; two pulls compared side by side is what turns "it failed again" into "it failed the same way again," which is a different and much more useful sentence.

Then pair the face evidence with one running observation that would be absent if your explanation were wrong. Here that was the discharge gauge at start: if the sequence had opened the valve immediately, the heat-checking explanation would have needed another source and the diagnosis would have gone back to chamber pressure and flush flow. Evidence that only confirms is not evidence.

Finally, put an expiry on the conclusion. Write the date and the running hours on the pump record when the new seal goes in. If the fix was right, the next failure is the wear-out pattern from the top of the field key, arriving much later. If it arrives on the old clock with the old pattern, the sequence change did not take, and the record is what lets you see that in one look instead of three visits.

References

• 29 CFR 1910.147, control of hazardous energy, for lock and tag plus pressure relief before pulling a pump or a seal
• 29 CFR 1910.134, respiratory protection, where a face is machined or an SDS names an inhalation route for a solvent or a pumped-fluid residue
• Pump manufacturer's power curve for shaft power at shutoff, and the seal manufacturer's data for the face pair and its temperature limits
• Published steam tables or the fluid supplier's property data, for saturation temperature at the measured seal chamber pressure
• See related: What a Mechanical Seal Is Actually Sealing Against; How Seals and Gaskets Fail; What Packing Does That a Mechanical Seal Cannot