DEV Community

Mingchun Huang
Mingchun Huang

Posted on

Why Seal Surfaces Leak While the Ra Number Looks Fine

A buyer sent us a leak test report last quarter. Twelve fittings, all VCR-type glands in 316L, all machined from certified bar, all showing Ra 0.4 µm on the inspection sheet. Six of them leaked at helium. The Ra number could not explain it, because Ra was never the problem.

This post is about what Ra does not measure, and why metal-to-metal seal surfaces fail in ways that show up only at leak test — long after the dimensional report said everything was good.

Ra is an average, and averages hide grooves

Ra integrates the absolute deviation of the profile from its mean line. A fine, uniform tool mark and a surface with one deep spiral scratch can share the same Ra. On a sealing land, that spiral scratch is the leak path: helium finds it, the average never did.

The second thing Ra ignores is lay — the direction of the dominant marks. A turned seal face has concentric marks, which sit crossways to the leak path and actually help sealing. A face that was turned and then lightly polished in a circular motion can end up with random lay and fine scratches radiating outward, worse for sealing than the turned surface despite a better Ra. I have seen that exact swap fail. I would not claim it always fails — the depths matter too — but it fails often enough that we stopped trusting polish as a finish step on seal lands.

Waviness is the third blind spot. Ra reads short wavelengths. A gland face that is slightly concave across its whole width — a few tenths of a micron of crown error from a tool deflection or a worn guide bushing — can read a beautiful Ra while the contact pressure line sits off the land. At flange dimensions under 1/2 inch, the sealing band is narrow, and a waviness error eats most of it.

What actually predicts a seal

In practice, three things decide whether a VCR-style joint seals:

First, flatness of the lands, checked on a granite surface plate or with an optical flat, not inferred from Ra. Second, the absence of directional scratches crossing the land — this is a handling and chucking problem as much as a machining one; a part dropped into a plastic bin with chips can pick up a line no cutter ever made. Third, surface texture fine enough and isotropic enough that the two metal faces conform under load. That last one is where electropolishing earns its cost: it removes the smeared, micro-cracked layer left by machining and reduces the asperity height, so the joint needs less seat load to close the remaining gaps.

What electropolishing does not do is fix geometry. It removes material roughly uniformly (slightly more on peaks, which helps), so if the face is not flat before the bath, it is not flat after. And it will soften an edge radius you may have wanted sharp. Sequence matters: machine, deburr, measure, polish, then re-measure the critical diameters, because the polish pull-down on a small gland OD is real — on the order of a couple of microns, which on some drawings is a tolerance you do not have to spend.

The inspection sequence we ask for

When a drawing says "Ra 0.4 on seal land" and nothing else, we quote asking for three additions: a flatness callout on the lands, a leak-based acceptance if the assembly is going to be tested (because a leak test at the fitting level correlates better than any surface metric), and a note on which features are measured before versus after polishing. None of this is exotic. It is mostly about deciding, before the first chip is cut, which measurement owns acceptance.

The failure reports we see in this product family almost never trace back to the cutting process itself. They trace back to handling between operations, to polishing done as a cosmetic step rather than a controlled one, and to drawings that specified Ra on a feature where flatness was the real requirement. One of those, the bin with chips, is embarrassingly cheap to fix and still shows up. I have no elegant explanation for why; shops are shops.

If you are buying fittings or any machined part with a sealing function, ask the supplier how they measure flatness, not just roughness. The answer tells you whether the leak test is going to be boring, which is exactly what you want a leak test to be.

CHX Precision Metal machines stainless, aluminium, titanium and engineering plastics in Shajing, Bao'an, Shenzhen, working to +/-0.005 mm where the drawing calls for it. 220 machines, MOQ 1 piece, prototypes 3-7 working days, batch production 7-15 days, EXW Shenzhen. chxmetal.com

Top comments (0)