Why a Gasket Fails

Why this matters

The gasket is the part you hold in your hand when a flange leaks, so it is the part that gets blamed and the part that gets replaced. Then the same joint leaks again in the same season with a brand new gasket in it, and the shop starts arguing about materials.

Almost none of those are material failures. They are load failures: the joint had enough clamping force on the face when it was built and did not have enough later. That is a different problem with different fixes, and the useful thing about it is that load loss is largely predictable from the shape of the joint. Two flanges on the same skid, carrying the same fluid at the same pressure with the same gasket cut from the same sheet, can behave completely differently, and the difference is usually visible before either one leaks.

The gate

One rule decides both cases below:

A joint holds when the seating stress remaining on the gasket face stays above the material's minimum for the entire service cycle, including the coldest start and the hottest steady run. Not at assembly. Throughout.

Assembly stress is the opening balance. Everything after assembly is withdrawals. Four of them, and they are worth knowing separately because they have different fixes.

Gasket relaxation. Compressed gasket materials creep under sustained load, quickly at first and then slowly, and faster when hot. Thicker gasket, more material to creep, more total movement. This is why a thick gasket used to bridge a bad flange face is a trade, not a free choice.

Embedment. The bolt head, the nut face, the washers, and the flange faces all have surface roughness, and that roughness flattens under load. Each interface gives up a few ten-thousandths of an inch. It is a small absolute number and it happens once, mostly in the first hours.

Differential thermal movement in the bolt-flange stack. If the bolt runs hotter than the flange it grows more, the stack gets shorter relative to the bolt, and preload drops. Run it the other way and the direction reverses: a flange alloy with a higher expansion rate than the bolts, brought up to temperature, grows the stack faster than the bolts grow and preload climbs, which is how a joint that never leaks arrives at an outage with a gasket crushed thin and no recovery left. Both ends of that range are real, and which one you have depends on the alloy pairing and where the heat is.

Cycling. Pressure and temperature swings work the joint back and forth. Where the transverse movement is enough to let the threads slip, nuts rotate loose, which is a genuinely different mechanism from relaxation and leaves a different signature: relaxation loses load with the nut exactly where you left it.

Why loss is a fraction, not an amount

Here is the part that turns this from a list into a prediction. The bolt is a spring. When you preload it, it stretches. Everything in the list above is a small permanent shortening of the stack, and the bolt gives back load in proportion to how much of its stretch that shortening consumes.

So the number that matters is not how much the joint relaxes. It is how much it relaxes divided by how far the bolt was stretched in the first place. A bolt with a lot of stretch stored in it barely notices a thousandth of an inch. A short, stubby bolt through two thin flanges has almost no stretch stored, so the same thousandth is most of what it had.

Bolt stretch at a given load is proportional to grip length. Triple the grip, triple the stretch, and the same absolute relaxation costs a third as much of your preload. That single relationship explains most of the "identical joints, different behavior" cases in the field.

Before you touch either joint

Both cases below involve opening a flange that has been in service.

• Isolate and lock the mechanical and stored energy sources under 29 CFR 1910.147, then prove zero pressure by cracking a vent or drain at the joint itself with the isolation locked. A gauge upstream of a closed valve tells you about the gauge.
• Where any conductor, heat trace, or motor lead is on the equipment, that half is electrical work: de-energize and verify absence of voltage under 29 CFR 1910.333(b)(2), proving your tester live-dead-live per NFPA 70E-2021, 120.5, which reaches you through your employer's electrical safety program or your contract, not on its own.
• Let a hot line cool before loosening anything, and stand out of the split plane while backing bolts off in the crossing pattern. Retained load springs the last few bolts even on an empty line.
• Do not re-tighten a hot or pressurized joint to chase a leak. Hot bolting is a written, engineered procedure with its own controls, not a wrench decision, and a stud at temperature has lost yield margin you cannot see.

Two joints, one skid, one shutdown

Both flanges carry the same fluid, both were built the same day by the same crew with the same gasket material and the same target load, both are 8-bolt joints using 1/2-13 bolts. Illustrative values throughout; the method is what travels.

The stretch numbers come from load times grip divided by (stress area x modulus): 6,000 x 1.5 divided by (0.1419 in2 x 29,000,000 psi) is about 0.0022 in, and the 4.5 in grip gives three times that, about 0.0066 in.

Now the withdrawals, and here is the point: they are nearly the same absolute movement on both joints, because they come from the gasket and the contact interfaces, which are identical. Call the total 0.0008 in from relaxation and embedment together.

• Joint A loses 0.0008 of 0.0022 in of stretch, about 36 percent of preload.
• Joint B loses 0.0008 of 0.0066 in, about 12 percent.

Assembly stress was the same on both. After a week in service, Joint A is sitting at roughly 64 percent of the seating stress it started with and Joint B at 88 percent. If the material's minimum happens to sit at 70 percent of what they were built to, Joint A is now below it and Joint B is comfortably above, and the crew that built both to the same spec on the same day did nothing wrong.

Add a thermal cycle. Say the process runs the bolts about 30 F hotter than the flanges on both joints during a hot start, and steel expands at roughly 6.5 millionths per inch per degree F near these temperatures. Joint A's 1.5 in of grip gives a differential growth around 0.0003 in; Joint B's 4.5 in gives about 0.0009 in. Bigger absolute number on B, but as a fraction of stored stretch it is about 13 percent on both, because grip length scaled both terms. That is the tell that separates a thermal problem from a relaxation problem: relaxation punishes short joints disproportionately, differential thermal growth punishes both equally. If Joint A leaks and Joint B does not, look at stack stiffness. If both leak together, and only after a temperature swing, look at the thermal pairing.

What each finding actually changes

If the diagnosis is a short, stiff stack losing too large a fraction:

• More stretch for the same load. A longer stud, or a smaller-diameter bolt at the same load if the joint design allows, stores more stretch per unit of preload. Both change the joint, so both belong to whoever engineered it.
• Less movement to lose. A thinner gasket, or a material with less creep at temperature, cuts the withdrawal instead of raising the balance. This is often the cheapest real fix and it is a gasket manufacturer question: minimum seating stress and creep relaxation at your service temperature are properties they publish for their material and thickness.
• A live spring in the stack. Conical spring washers of an appropriate rating hold load through small movements. Rated by the manufacturer for the load, and the wrong washer is a shim.

If the diagnosis is nuts rotating loose under cycling, none of the above helps, because you are not losing stretch, you are losing turns. That is a transverse-motion problem: kill the movement at the pipe support or the equipment mount, or use a locking method rated for the joint.

The failure mode to avoid: replacing the gasket with a thicker one because the joint would not seal. Thicker gives you more conformability at assembly and more creep afterward, so it can seal on the day and leak by the next season, which reads to the shop as a bad batch of gaskets. If a thicker gasket was needed to seal a flange face at all, the face is the finding, and a sibling article covers reading witness marks to tell a face problem from a load problem.

How to verify you got this right

• Record grip length and gasket thickness on the joint record, not just the torque value. These two numbers predict the behavior of a joint better than the torque figure does, and neither one is recoverable later without taking the joint apart.
• Measure compressed gasket thickness at four points on assembly and again at the next opening. Uniform loss all around is relaxation. Loss on one side is geometry.
• Check nut position, not just tightness, on a joint you suspect is cycling loose. Mark nut to flange at assembly with a paint line so the next visit reads rotation directly. A joint that lost load with the paint lines intact did not vibrate loose.
• When a joint has failed twice with different gaskets, stop replacing the gasket. Two materials failing the same way is evidence about the joint, not the material.

References

• 29 CFR 1910.147, control of hazardous energy, for mechanical isolation and stored energy before opening a bolted joint
• 29 CFR 1910.333(b)(2) for electrical isolation, with NFPA 70E-2021, 120.5 for live-dead-live proving where your employer's electrical safety program adopts it
• Gasket manufacturer documentation for minimum seating stress, maximum stress, and creep relaxation at service temperature for the specific material and thickness
• ASME PCC-1, in the edition your facility or project specification has adopted, for bolted joint assembly practice including the treatment of relaxation and re-tightening; it binds through that specification, not on its own
• See related: How a Gasket Actually Seals; The Gasket That Was Crushed and the One That Was Not; What Torque Actually Controls and What It Does Not