CMM vs Optical Measurement for Soft RTV Cavity Dims

A CMM stylus on a Shore 40A seal measures how far the probe sank, not the part. Silicone is an elastomer. It yields under a touch probe, under a clip, and under its own weight. A QA lab that treats an O-ring like a steel bushing will pass a thick, squashed ring and fail a correct one that was standing free.

That is a parts-lab problem. On this site you are often measuring the cavity: a 10A–20A platinum skin, a master, a dump from the tray. The probe still sinks. A lab that treats a chocolate pocket like a steel bushing will recut a good master and leave a floppy tray that was never jacketed. Optical methods avoid that dent. Measure free-state or in the jacket, not with fake Gage R&R.

If your drawing says “CMM all dims” and the quote returns a cold-runner, you are on the wrong factory.

Free-state is a method, not a mood

Decide, in writing, the state of the part — or the skin — at measurement:

• Free-state, no squeeze. Most O-rings, gaskets, and flexible housings. The part sits, supported so it does not sag into a banana, but is not clamped to a drawing that assumed metal. Most RTV skins too, *unless* they ship in a mother. A 12-cavity tray laid floppy on granite is a sag study. Loaded-tray edge sag without a jacket is often about 0.5–1.5 mm. Outer-row dumps often run 3–6% heavy versus center. That is jacket, not shrink.
• Defined compression. Only if the drawing is a gland or a compressed thickness. Then the fixture must reproduce that squeeze with stops, not with a technician’s fingers. RTV analog: in the jacket, on stops, the way it dumps. A skin post-cured floppy will not match one post-cured in the jacket.
• On a mandrel or in a nest that copies the use orientation (a tube ID, a strap on a wrist form, a tray in its mother). Say so. A strap laid flat and a strap on a form are different lengths. A cavity on the table and a cavity in the box are different pockets.

Temperature matters. Silicone expands more than steel. A lab at 23 °C and a press-side check on a 40 °C part will disagree. Condition parts, then measure. Do not argue about 0.1 mm on a 50 mm LSR part across a 15 °C swing without doing the arithmetic. Same arithmetic on a warm dump and a cold master: platinum RTV shrink is about 0.1–0.3% after full cure. LSR *part* shrink is about 2–3%. Do not paste the LSR number onto the master.

Contact CMM: where it still belongs, and where it lies

A tactile CMM (touch probe on a moving bridge) is excellent on rigid datums: overmolded metal frames, plastic carriers, the steel mold itself, and the CNC or SLA master. On the rubber:

• Probe force dishes the surface. Low-force probes and large-radius styli reduce the dish. They do not make it zero. A 10A–20A food skin dishes more than a 40A seal.
• A scanning probe dragged on a soft lip will plow. Dragged on a chocolate highlight, it matts a witness you then call “wear.”
• Fixtures that “hold datum A” on a gasket often are the measurement. Fixtures that tape a tray flat are the measurement.

Use contact CMM on silicone when you need a 3D point that vision cannot see (a deep undercut, a hole axis in a stiff-enough boss) and when you have a fixture that supports the part the way the drawing meant. Record probe force and tip radius. If two labs disagree, those two numbers are the first suspects, not the molder’s shrink factor — and not the RTV kit lot.

Do not accept a CMM report on an O-ring CS that was taken by touching the torus at the top of a ring lying on a granite plate. You measured squash plus gravity. Do not accept a CMM report on pocket depth taken by touching the floor of a 15A bear while the tray was hanging off the granite. You measured squash plus sag. Measure the master on CMM if you need steel-like numbers; measure the skin optically or in the jacket.

Optical and vision: the default for flexible parts — and flexible cavities

Vision measuring machines, optical CMMs, profile projectors, and calibrated camera systems take edges without a dent. They are the right first tool for:

• O-ring ID/OD and cross-section on a glass stage, ring lying naturally, no clip.
• Gasket outlines and hole positions in free-state.
• Flash width, gate vestige, pin dishes (the defect is optical). RTV analog: split flash, pour scar, key witness, peel stretch.
• Thin membranes and slits that a probe would open.
• Pocket outlines, logo depth as a silhouette, dump outline versus the master.

Lighting is the new fixture. Backlight for a silhouette. Raking light for a step. A shiny LSR lip can fake an edge. A glossy chocolate highlight can too. Telecentric optics help on height-sensitive silhouettes. A human with a steel caliper is also an optical method with an uncontrolled squeeze; it is not “more real” than a vision system. Calipers on a 12A pocket are a dent with extra steps.

Laser or structured-light 3D scans can map a surface, but the part still must not sag, and shiny or translucent silicone can drop data. A light dusting that the customer forbids is not a method. If you scan a torn jacket to “save the master,” prove the setup on a retained dump — then admit you are reverse-engineering, not inspecting. Scanning a healthy master is the honest 3D path.

Fixturing that does not become the dimension

Support long, floppy parts on a bed of pins or a printed nest that matches the free-state curve, or measure in sections. Do not tape a gasket flat unless the drawing says the part is used flat and the tape does not pull. Do not tape a tray flat unless dumps happen taped flat, which they do not.

For compressed measurements (installed O-ring height, a pad under a lid), use hard stops at the designed gap and measure the remaining gap or the load. Do not “press until it looks seated.” For a cavity, seat the jacket on its stops and measure. Never use a vise.

ISO 3302-1 is a rubber tolerance *class* on the drawing of a molded article, often M2 as a common default. It is not the word “precision,” and it is not a CMM program. A dump from an RTV cavity is not automatically an ISO 3302-1 article. Put the class on the *part* drawing if you are buying parts. Put pocket depth, wall, and jacket state on the *cavity* drawing.

What a competent silicone FAIR — or first-dump pack — looks like

• Drawing with datums that make sense on rubber (a sealing face, an ID, not six points on a floppy wall). Cavity analog: pocket floor, split, jacket seat — not six points on a 12A wall.
• Measurement state: free or compressed, temperature, orientation. Cavity: free vs in-jacket, post-cure yes/no.
• Method per dimension: vision / optical, low-force contact, mandrel, go/no-go of a defined design. Master: contact CMM is fair. Skin: optical first.
• Shrink applied to the steel (or to the master), not corrected in QA by “the rubber always measures small so we add 0.2.” Platinum RTV ~0.1–0.3%. Do not add LSR 2–3% in the lab.
• Thickness of seals by a non-crush method (optical, or a drop-gauge with a large foot and a specified force, used with a correction you have demonstrated). Pocket depth the same way.
• Flash called as a maximum, measured, not ground off before FAIR unless the process includes that destage. RTV: not trimmed off the skin before the first-dump pack unless destage is the process.

If the lab only reports a CMM printout of 40 points on a 20A gasket — or a 15A tray — ask how the part was held. If they cannot answer, the numbers are a fixture study.

Do not fake capability theater

Gage R&R on silicone is sensitive to operator squeeze and to part relaxation between trials. It can be done, with a fixture that removes those knobs, on a defined dimension. Publishing a table of made-up %GRR values does not make a lab. Run the method, on the actual part or skin, with the actual fixture, or do not claim it. Do not invent a named lot or a client.

A shop-floor check can be a nest and a vision snapshot, or a mandrel and a go/no-go, aligned to the same state as the lab. Two methods in two states will never match. Pick one state and put it on the drawing. For trays: pick jacketed vs free and put it on the RFQ.

When a dimension is chronically “off,” look at fixturing and probe force before you recut steel — and before you sand the master “to clean it up.” A surprising number of silicone tools are innocent. The CMM was measuring rubber as if it were a block. A surprising number of RTV masters are innocent too. The tray was sagging.

Common mix-up: CMM the skin, then recut the master

A parts FAIR on a 50A LSR lip does not approve a chocolate pocket. Probe force that was “fine” on a stiff boss is a crater on 12A. Recasting from the unsanded master is cheaper than chasing a dent with a toolpath.

What to ask when you're buying

• Object: RTV cavity / dump, or LSR/HCR part?
• Measurement state: free, jacketed, compressed with stops? Temperature?
• Method per critical dim: optical vs low-force contact vs CMM on the master?
• Probe force and tip radius recorded if contact is used?
• ISO 3302-1 class on the *article* (M2 common) — not as a slogan on the jacket?

Worked example: 12-cavity bear vs LSR O-ring

12-cavity 18 mm bear, platinum 10A–20A, jacket. First-dump pack: optical pocket outline in the jacket, after post-cure, named illuminant if color is in play. CMM the CNC master, not the skin floor. If pocket 12 is “deep,” check jacket sag before you recut. Sister PO: 64-cavity infant O-rings, free-state optical CS, no clip on granite — different quote. Same “CMM the silicone” email. The bridge CMM does not deposit pectin.

When this article is the wrong quote

If you already know you need a parts FAIR on LSR/HCR *articles*, send the parts OEM the measurement state and method list. A tiny open-back ice tray can skip optical-theater. Geometry can override: a deep undercut still needs a support nest, not a smaller stylus. Do not paste a Gage R&R spreadsheet onto a floppy glove.

Related reading: shrinkage allowance, scanning damaged masters, micro-tolerances, custom RTV molds.

RFQ language that gets you a method, not a printout

“Object: platinum RTV cavity. Measure dumps in the jacket after post-cure. Optical (or non-crush) on pocket depth and outline. CMM the master, not the 10A–20A floor. State: free vs jacketed, temperature. Probe force and tip radius if any contact. No invented %GRR. Master kept; who owns it.” If the quote returns 40 CMM points on a taped tray, that is a fixture study. Ask for the fixture photo, not a capability slogan.

Send the ballooned 2D, what is cavity vs article, and the state you will inspect. We will measure the master and the jacketed dump — and tell you when the CMM belongs on a parts FAIR.

We would not recommend CMM-probing a 15A pocket floor on granite

We would not recommend recutting the master from a dent. We would not paste 2–3% LSR shrink onto the skin in QA. We would not tape a tray flat and call it free-state. We would not raise Shore A first so “the probe works.” We would not publish a fake Gage R&R table.

RUUIPON Mold Structure Decision: if you dump or pour into silicone, stay RTV. Flat back? open-back — still measure in the box if that is how it dumps. Deep undercuts that cannot peel? two-part / plug / core; datums on the split, not on a floppy wall. Soft 10A–20A or wet dump? mother mold; inspect jacketed. If the silicone *leaves* a hot or compression tool, stop — LSR/HCR on a parts OEM, optical default, free-state written.

Typical shop values

Reference values only. Confirm material-specific limits against the supplier’s technical data and agree project acceptance criteria before production.

| Item | Typical shop value | | --- | --- | | RTV mix (this site) | Platinum 1:1 or 10:1; vacuum degas | | RTV cavity Shore | 10A–20A skin with jacket — dishes under a stylus | | Platinum RTV shrink | About 0.1–0.3% after full cure | | Contrast: LSR *part* shrink | About 2–3% — do not add this in QA | | Contrast: HCR *part* shrink | About 1.5–2.5% | | Lab condition | Often 23 °C; do not fight 0.1 mm across a 15 °C swing without the arithmetic | | ISO 3302-1 | M2 common on a rubber *article*; not a cavity slogan | | Outer-row vs center / edge sag | 3–6% heavy / about 0.5–1.5 mm — jacket, not a CMM miss |

FAQ

Why does a contact CMM on a Shore 40A seal — or a 15A cavity — measure the fixture and the dent? The elastomer yields under the stylus, under a clip, and under its own weight. A lab that treats an O-ring like a steel bushing will pass a squashed thick ring and fail a correct free one. A lab that treats a chocolate pocket the same way will recut a good master. Record probe force and tip radius; those are the first suspects when two labs disagree, not the molder's shrink factor.

What does "free-state" have to mean on the drawing? A written measurement state: no squeeze for most rings and gaskets, or a defined compression with stops if the drawing is a gland height, or a mandrel that copies use. RTV: free vs in the jacket, the way it dumps. A strap laid flat and a strap on a form are different lengths. A tray on granite and a tray in the mother are different pockets. Condition temperature; silicone expands more than steel.

When does a tactile CMM still belong on a silicone job? On rigid datums: overmolded metal frames, plastic carriers, the steel mold, and the CNC or SLA master, or a deep undercut vision cannot see in a stiff-enough boss. Do not touch-probe O-ring cross-section on a ring lying on granite. Do not touch-probe a 15A pocket floor while the tray sags. You measured squash plus gravity.

Why is lighting the fixture on optical measurement? Vision avoids the dent and is the default for ID/OD, outlines, flash, vestiges, membranes a probe would open, and RTV pocket silhouettes. Backlight for silhouette, raking light for a step; a shiny LSR lip or chocolate highlight can fake an edge. A steel caliper is also an optical method with uncontrolled squeeze. Never use a vise.

What should I check before recutting steel — or sanding the master — for a chronic "off" dimension? Fixturing and probe force. A surprising number of silicone tools are innocent; the CMM was measuring rubber as if it were a block. A surprising number of RTV masters are innocent; the tray was sagging. A competent FAIR or first-dump pack names state, method per dimension, and non-crush thickness. Shrink is applied to the steel or the master, not corrected in QA by adding a constant. Two methods in two states will never match.