Preventing Compression Set Failures in Industrial Silicone Mold Design

The gasket still looks intact. The joint leaks. That is compression set, not a missing bolt. The rubber took the flange shape in service and did not push back when temperature or load changed.

Compression set is not a mystery property and it is not a single catalog number. ASTM D395 (Method B is the usual rubber callout) compresses a specimen a defined amount, holds it at a defined temperature for a defined time, then measures how much of the squeeze never comes back. A low number at 70 °C for 22 hours does not describe a cover that lives at 180 °C. Do not copy a percentage off a general HCR bulletin into a hot joint.

What the test is telling you

Method B is constant deflection: buttons (or a cut O-ring) squeezed, typically 25 percent, against spacers, in an oven, then released and measured after a recovery period. Set is the unrecovered fraction of that deflection.

Method A is constant force. Most silicone seal drawings that mention D395 mean Method B. ISO 815 is the cousin. If the drawing says “compression set < 20 percent” with no method, time, or temperature, it is not a requirement yet.

Post-cure of the part before the test changes the number. A ring tested as-molded and a ring tested after a 200 °C post-cure are different articles. Specify the state.

Do not invent a passing value here. Ask the gum supplier for D395 Method B at your temperature and time, on your hardness and post-cure. Then design to that.

The three design knobs that cause shop failures

Squeeze. Static silicone faces often work in a moderate squeeze band (many glands are designed around the same neighborhood as the 25 percent used in Method B, commonly discussed as roughly 15-30 percent for O-rings and similar static seals). Too little squeeze, no line pressure, leak at rest. Too much squeeze, you have already spent the recovery the rubber owns, and heat finishes the job. A gasket that is “a bit thicker so it seals” at the same bolt length is a set generator.

Gland fill for O-rings: an overfilled gland at high temperature has nowhere to expand. The rubber takes a permanent square shape. An underfilled gland can pump. Follow a gland design (AS568 / ISO 3601 type tables or your corporate equivalent). Do not freehand a groove from the ring’s ID only.

Temperature. Heat accelerates set. A silicone that is honest at 150 °C can still be the wrong grade at 200 °C continuous, or the right grade if you post-cure and keep squeeze moderate. Thermal cycling that goes through a high soak and then a cold start is when the leak appears: the flange opened a few hundredths and the rubber stayed put.

Time. D395’s 22 hours is a standard, not your warranty. A joint that is clamped for years at temperature will set more than a 22-hour button. Use the test to compare grades, then add margin in squeeze and in retighten policy if the joint is serviceable. For a non-serviceable cover, pick the low-set grade and the post-cure, do not hope.

Those three knobs sit on top of grade: polymer (VMQ vs fluorosilicone when fuel is involved), filler, and cure. Platinum LSR and peroxide HCR both can be low-set if formulated and post-cured. A random 50A shop HCR is not “silicone, therefore low set.”

Mold-design mistakes that look like compound failure

• Thickness variation. A compression tool that dishes in the middle makes a gasket thick at the bolt circle and thin in the center, or the reverse. The thick region is over-squeezed. Fix plate stiffness and charge, not the datasheet.
• Knit lines on the sealing bead. LSR weld lines on a thin lip can take a local set or crack. Gate so the lip is a continuous flow.
• Oval O-rings from post-cure. Rings dumped in a hot pile. Mandrel or tray them.
• Flash that is sanded into the sealing line. You just changed the squeeze locally.
• Wrong hardness for the flange roughness. Hard rubber on a coarse as-cast cover needs more squeeze to fill the valleys, which then sets. Either finish the metal or use a softer grade with a low-set formulation, not unlimited crush.

The mold cavity should hold thickness the way the gland expects. Call DIN ISO 3302-1 (or your class) on the sealing thickness, and inspect it with a method that does not squash the part (see the CMM article).

Process: cure and post-cure are part of set

Undercured HCR and LSR have more unset chains to rearrange under load. They set. A “faster cycle” that still looks filled is a set problem six weeks later.

Post-cure (often a hot-air oven after demold, per the gum) drives the network and volatiles toward the state the datasheet assumed. Skip it on a low-set seal and you will not match the bulletin. Run it on a fixture if the geometry can sag.

How to stop the leak before you cut steel

1. Write the joint: media, temperature profile, clamp load or gland, service life, whether the joint can be retorqued.
2. Choose squeeze from the gland or flange gap, not from leftover rubber thickness.
3. Call ASTM D395 Method B at a temperature at least as hot as the soak, for 22 hours or the longer time your spec uses, on the post-cured state.
4. Pick the grade that has data there. Fluorosilicone if the media demands it; do not use it just because it sounds tougher.
5. Tool thickness and bead to that squeeze. Stiffen the compression plate. Gate away from the bead.
6. First-article: measure thickness without crushing, then a heat-soak of the assembled joint, then a leak check after cool-down. That last check is the one that catches set.

If the part already leaked in the field, do not first “mold it thicker.” Measure remaining thickness versus original, look at the flange gap, and compare to the D395 condition you never wrote down. The rubber is showing you the squeeze, temperature, and time you designed by accident.

FAQ

Why does a gasket that still looks intact leak after a heat cycle?

The rubber took the flange shape and did not push back when temperature or load changed. That is compression set, not a missing bolt. ASTM D395 Method B (ISO 815 is the cousin) names the test: defined squeeze, temperature, time, then unrecovered deflection. A low number at 70 C for 22 hours does not describe a 180 C cover.

Is "compression set less than 20 percent" a requirement by itself?

Not without method, time, temperature, and as-molded versus post-cured state. Ask the gum supplier for Method B at your soak and time, on your hardness and post-cure. A random 50A shop HCR is not "silicone, therefore low set."

Why is "make the gasket thicker" a set generator?

At the same bolt length, extra thickness is extra squeeze. You have already spent the recovery the rubber owns; heat finishes the job. Too little squeeze leaks at rest. Follow a gland design for O-rings; an overfilled gland at high temperature squares the ring permanently.

How do mold mistakes impersonate a bad compound?

A dished compression plate makes thickness vary so one region is over-squeezed. Knit lines on a thin LSR lip, oval rings from a hot pile, flash sanded into the sealing line, and hard rubber on a coarse as-cast cover all look like datasheet failure. Call thickness class and inspect without crushing the part.

What first-article check catches set the oven button test can miss?

Measure thickness without crush, heat-soak the assembled joint, then leak-check after cool-down. That last check is when the flange opened a few hundredths and the rubber stayed put. If a field part already leaked, measure remaining thickness versus original before molding it thicker.