The Joint That Failed Every Summer and Held Every Winter

A seasonal fault is a gift, because the season is a variable you did not have to introduce. The trap is assuming the season is the mechanism. This one was reconstructed almost entirely from the service history, three years after the first ticket, by a tech who never once saw the joint leaking.

Depressurize the bowl before you touch it

The joint in question is the seal on a filter bowl in a compressed air line, and the bowl is a pressure vessel with stored energy in it. Before any part of this diagnosis:

• Close the isolation valve upstream, then open a bleed or drain downstream of the filter and confirm zero on a gauge at the bowl itself. A closed valve upstream with trapped air between it and the bowl is still a loaded vessel.
• Do not crack the bowl to test for pressure. Verify on the gauge, then relieve, then open. 29 CFR 1910.147 requires stored energy to be relieved, disconnected, or restrained before servicing begins, and a pressurized bowl is exactly that.
• Treat a clear plastic bowl as suspect if there is any crazing, clouding, or fine cracking in it. Certain synthetic lubricants and most solvent cleaners attack transparent bowl plastics, and an attacked bowl can let go at working pressure rather than split quietly. If you see crazing, isolate and replace the bowl, and never clean one with a solvent.

That last point turns out to be the same failure mechanism as the fault in this case, arriving on a different part.

The call nobody could reproduce

A small manufacturing customer, one compressed air drop feeding a spray booth. The complaint each time was oil and water on the floor under the filter housing and a hissing leak at the bowl seal. Each time a tech went out, replaced the bowl seal, pressure tested it, and left. Each time it held.

Nobody could reproduce it on a return visit because by the time anyone went back for a second look, the seal was new. The fault was being erased by the repair on every single visit, which is why three years went by without a cause.

What the record actually said

The tech who caught the fifth call did not go to the site first. He pulled every ticket on that address and laid the dates out.

• Five bowl-seal replacements across three cooling seasons: two in the first year, in July and September, two in the second year, in June and August, and one in the third year, in July.
• Zero failures between October and May in any of those three years. Not one.
• The unit ran year round. The customer's production schedule did not change seasonally, and the run hours logged on the compressor were flat across the year.

That last line is what made the pattern worth something. If the shop had run harder in summer, the seasonality would have been a duty-cycle story. Flat run hours across the year with all five failures in the warm half is a temperature story or a nothing story.

The gaps in the record were as informative as the entries. No ticket recorded a part number for the seal that went in. No failed seal was kept. Two of the five tickets said only "replaced o-ring, tested ok" with no cause noted, and none of the five named a cause at all. That is how a repeat fault stays invisible: five separate closures, spread across three years and several techs, never add up to a pattern in anyone's head, only on a page.

Two changes landed in the same year

Working backwards past the first failure, two things changed at this site in the spring of year one, four and three months before that first July ticket.

• In March, the customer switched compressor lubricant from a mineral oil to a synthetic.
• In April, the shop switched to a generic seal assortment for bowl and filter kits instead of ordering the kit specified for the housing.

Both are strong candidates. A lubricant change is a classic seal-attack trigger, because the seal was chosen for the old fluid. A seal source change is an equally classic one, because a bin of assorted rings gives you the right dimensions and tells you nothing about the compound.

Dates alone could not separate them. They were four weeks apart and both preceded the failures.

The control case that separated them

The same customer runs a second building with an identical air setup. Same compressor model line, same filter housing, same drop.

The second building got the same lubricant change on the same day, from the same drum. It did not get the assortment seals, because it had been serviced out of a different van that still carried the specified kit, and its bowl seal had never been replaced.

Three cooling seasons, zero failures at the second building.

That is the whole answer, and it came from the record rather than from a meter. The lubricant is common to both buildings and only one building fails, so the lubricant is not sufficient to cause the failure. The assortment seal is present only in the failing building. Two candidates, one control, one survivor.

A caution on how far that carries: the control shows the lubricant alone does not cause the failure, not that the lubricant is irrelevant. The seal and the fluid are a pair, and the finding is about the pair.

Why the season mattered, and what it was not

The seasonality is not a separate mechanism. It is the same mechanism running at two different speeds.

Chemical attack on an elastomer is a rate, and rates rise with temperature. The working rule of thumb is that reaction rate roughly doubles for every 18 degrees F of rise. It is an approximation, not a specification, but it is close enough to explain a service history.

The mechanical room in the failing building was measured at 104 degrees F on an August afternoon and 58 degrees F in January, with no ventilation of its own. That is a 46 degree F spread, which is about two and a half doublings, so the attack on the seal runs roughly six times faster in August than in January.

Run that against a seal that would take years to fail at winter room temperature. At six times the rate, years becomes months, and months lands inside a single cooling season. Then a new seal goes in, the weather turns, the clock effectively stops for the winter, and the following summer it starts again from a fresh part. The pattern of the seasons was never the cause; it was the throttle.

This is also why "it only leaks in summer" sent three techs looking for thermal expansion in the housing. Expansion is reversible and it happens on the first hot day. This fault took weeks of heat each year and never once appeared in the first hot week of the season.

Measuring the failed seal

The fifth replacement was done differently: the old seal was kept and measured against a new one out of the same assortment bag.

• Nominal cross-section of the ring, new and unused: 0.139 inch, one of the standard sizes.
• The seal pulled out of the housing measured 0.155 inch across the same cross-section, and that reading is net of any compression set the ring took in the gland, so the swell is if anything understated.

That is 0.016 inch of growth on a 0.139 inch section, which is about 11.5 percent linear swell. Swell is a volume effect, and volume goes as the cube of the linear change, so 11.5 percent on each dimension is roughly a 39 percent volume increase.

For context on whether that number is bad: a small amount of swell is helpful in a static seal, and manufacturers commonly treat volume swell up to somewhere around 15 to 20 percent as tolerable for a static application. Nearly 40 percent is not in that band. A seal that swollen no longer fits the groove it was designed for. It gets squeezed out of the gland, it goes soft, and it extrudes under pressure - which is precisely the hissing leak the customer kept reporting.

The failed ring was also noticeably softer than the new one and left a smear on a rag. A ring that stains a rag and gives under a thumbnail against a new ring's springback is a compound being dissolved, not one being worn.

The fix, and the part of it that was not a part

The replacement seal was ordered as the housing kit rather than pulled from the assortment, so its compound was specified against the fluid rather than matched on dimensions. In broad classes, the nitrile group resists petroleum oils. Do not extend that to synthetics: synthetic lubricants are not one class, and the ester and glycol-based ones attack seals that mineral oil never touched, which is exactly the failure this case describes. After any lubricant change, read the compound off the lubricant manufacturer's seal-compatibility guidance rather than inferring it from the oil that came out. The family that resists water, glycol and steam, and that is poor in petroleum oils, is the ethylene propylene group; the family that is excellent in water, glycol, and steam and poor in petroleum oils is the ethylene propylene group. An assortment bag will happily hand you the wrong one in exactly the right size.

The part of the fix that was not a part: the assortment bin came off the van for pressure-side seals entirely, and the ticket template got a required field for the seal part number. That second change is what makes the next occurrence diagnosable in one visit rather than five.

How they confirmed it. Not by a pressure test on the day, which had passed every previous time and proved nothing. They scheduled a look at the seal in the following August, at the hottest point of the year, and measured its cross-section against a new ring from the same kit. It read within a couple of thousandths of nominal and sprang back under a thumbnail. That is a confirmation with the mechanism's own clock accounted for; a pressure test in the week after the repair is a confirmation that the tech can install an o-ring.

What the record could not tell them

Two things stayed unresolved, and they are worth naming rather than papering over.

The record could not say whether the original seals would have survived the synthetic lubricant indefinitely, because nobody kept an original seal from either building to measure. The second building is still running its originals, which is evidence but not proof.

It also could not say whether the room temperature had always run that hot. The 104 degree F reading was taken in year three. If the room had been cooler in earlier years, part of the failure timing belongs to a ventilation change nobody logged. That is the general lesson from the gaps: a service record that captures the part and the condition it was working in turns a three-year pattern into a one-visit diagnosis, and a record of one-line closures cannot.

References

• 29 CFR 1910.147, energy control, for the requirement to relieve or restrain stored energy before servicing a pressurized component
• Manufacturer elastomer compatibility charts for the specific fluid, including lubricant manufacturer guidance on seal materials after a lubricant change
• Manufacturer documentation for transparent filter bowls, including solvent and lubricant restrictions and replacement criteria for crazed bowls
• See related: Gasket + O-Ring Material Reference; What Temperature Does to a Material's Behaviour; Reading Rust and Corrosion Patterns