A supplier that runs a 4.5 MPa hydrostatic pressure test on every liquid cooling connector will still ship you parts that leak in the field — because micro-leaks (10⁻² Pa·m³/s class) hide inside O-ring grooves and micro-channels, and they only open up after thermal cycling. This page walks through what each test actually catches, why pressure testing alone misses the field failures that keep your field engineer on a plane, and what certificate your supplier must hand you for the leak-critical parts.
For a data center liquid cooling connector, "zero-leak" is a business term, not a physical term. The seal is the last line of defense between expensive GPU silicon and a puddle of glycol on the floor of a colocation hall. A leak that drips at 0.05 ml/min is invisible on the test bench, invisible at room temperature, and catastrophic on day 400 of a 5-year service contract when the dielectric coolant has crept along a micro-gap in an O-ring groove and shorted a board 14 rows deep in the rack.
The buyer-side definition of zero-leak for a cooling connector is therefore not "the part held 4.5 MPa for 30 minutes in your shop." It is:
A pressure test verifies the first point. A helium leak test verifies the second. A thermal-cycling helium test verifies the third. None of the three is optional for a connector going into a hard-to-reach installation; the question is which combination matches your risk and your budget.
The hydrostatic pressure test is the baseline gate for any cooling connector. The standard sequence is: fill the part with water (or a water-glycol mix matching the field fluid), pressurize to the design test pressure (typically 1.5× the maximum working pressure — for a 3 bar / 45 psi working pressure, that is 4.5 bar / 67 psi), hold for 30 minutes, and inspect for visible drips, pressure decay on a calibrated gauge, or pressure decay on a mass-flow leak detector. The 30-minute dwell is chosen because most gross leaks (porosity, missing weld, scratched sealing face) will show a pressure drop within the first 10 minutes, and the remaining 20 minutes catches slow leaks from under-torqued fittings or hairline cracks.
| Test parameter | Typical value (cooling connector) | Why this value |
|---|---|---|
| Test medium | Deionized water or 25–30% glycol mix | Matches field fluid density for realistic leak paths; water is cheaper and faster to dry |
| Test pressure | 1.5× working pressure (e.g. 4.5 bar for 3 bar working) | Catches the part that just barely passed the working pressure but has no margin |
| Dwell time | 30 min (10 min minimum for production) | Slow leaks from under-torqued joints need >10 min to show on a gauge |
| Pass criterion | No visible drip AND pressure decay < 1% over the dwell | Visible drip is a fail; gauge decay > 1% is a fail (larger leaks would fail in <10 min anyway) |
| Coverage | 100% production (every part) | Cost per test is low enough to run on every part; risk of shipping a gross leak is too high not to |
What pressure testing catches well:
What pressure testing does not catch:
Helium leak testing uses a helium mass spectrometer to detect helium gas escaping from a pressurized (or evacuated) part. The detection floor of a modern helium mass spectrometer is around 5×10⁻⁹ Pa·m³/s in vacuum mode and 1×10⁶ Pa·m³/s in sniffer mode — that is six to nine orders of magnitude more sensitive than a hydrostatic pressure decay test on the same part volume. The trade-off is that helium leak testing requires a sealed test fixture, a helium charge (or vacuum), and 30–120 seconds per part — which is why it is not a 100% production test for every cooling part on the market.
| Test parameter | Vacuum-mode (outside-in) | Sniffer-mode (inside-out) |
|---|---|---|
| Part state | Sealed inside a vacuum chamber, helium sprayed on outside | Pressurized with helium inside, sniffer probe on outside |
| Sensitivity | 5×10⁻⁹ to 1×10⁻⁴ Pa·m³/s | 1×10⁶ to 1×10⁶² Pa·m³/s |
| Cycle time | 30–60 s per part (chamber pump-down dominates) | 15–30 s per part (faster, less fixturing) |
| Best for | Small sealed parts, leak-critical (medical, semiconductor, cooling connectors) | Larger or hard-to-fixture parts, field troubleshooting |
| Cost per part | Higher (vacuum chamber, helium recovery) | Lower (sniffer probe only) |
| Typical pass criterion for cooling connector | ≤1×10⁻² Pa·m³/s total leak rate | ≤1×10⁶ Pa·m³/s at any single point |
The ≤1×10⁻² Pa·m³/s threshold is the number to write on your drawing or PO. It is a working compromise: tight enough to catch an O-ring groove that is at the high end of compression tolerance and is weeping under thermal cycling, but loose enough to be achievable in a 30–60 second production test on every part. Tighter (10⁻⁶ or lower) is feasible for aerospace and semiconductor applications but adds cycle time and cost that the cooling connector market is not yet willing to pay.
A cooling connector that passes a 4.5 MPa hydrostatic test for 30 minutes and shows no visible drip can still fail in the field. The failure mode is a thermal-cycle-induced micro-leak: at room temperature, the O-ring and the groove are within tolerance and the seal is good; after 500–1000 thermal cycles, the differential expansion between the O-ring (elastomer, CTE ~200×10⁻⁶/K) and the stainless groove (CTE ~17×10⁻⁶/K) walks the O-ring out of its optimal compression range, and a leak path opens up that was not there at room temperature.
The walk-out mechanism is well documented in Parker O-Ring Handbook and Trelleborg sealing guides:
A 30-minute pressure test at room temperature will not see this. The squeeze is in tolerance, the surface finish is in tolerance, the part holds pressure. The leak appears only after the part has been in service long enough for the compression set to accumulate — which in a data center running 24/7 with daily thermal swings is typically 6–18 months. The fix is one of two tests:
A "we leak-tested every part" claim without a certificate is not a quality record. The certificate is what makes the test auditable, what allows you to trace a field failure back to a specific production lot, and what your downstream customer will ask for during their own incoming inspection. A complete cooling-connector leak-test certificate has these fields:
| Certificate field | What it must say | Why it matters |
|---|---|---|
| Test method | Hydrostatic per ASME B31.3 / Helium leak per ASTM E499 or internal SOP number | Method must match what was quoted; "leak test" alone is ambiguous |
| Test pressure and dwell | e.g. 4.5 bar, 30 min, room temperature | Verifies the part was actually tested at the spec, not at a lower pressure |
| Pass/fail criterion | e.g. ≤1% pressure decay (hydrostatic) or ≤1×10⁻² Pa·m³/s (helium) | Threshold must be on the cert, not just in the supplier’s head |
| Part identification | Part number, lot number, serial number range | Links the cert to the parts you received; lets you trace a field failure back to a production run |
| Material/heat lot | Material heat number(s) used to make the parts in this lot | For 316L stainless parts, the heat number links to the MTC and the PMI record |
| Equipment used | Mass spectrometer model/ID, calibration date, leak standard used | Calibration must be current; results must be traceable to a leak standard (e.g. NIST-traceable reference leak) |
| Operator | Name or ID of the test operator | Accountability; the same person signs off on the FAI |
| Date and quantity | Test date, total quantity tested, quantity passed, quantity failed | Quantity failed > 0 is a process signal; should be reported even if the failed parts were reworked and re-tested |
| Disposition of failures | Rework, scrap, or concession (with engineering approval) | Concession without engineering approval is a red flag |
A common audit failure: the supplier provides a single line item — "leak test passed” — on a Certificate of Conformance, with no test method, no threshold, no quantity breakdown, and no link to a specific lot. This is not a cert; it is a sticker. Reject it and ask for the full certificate per the table above.
Helium leak testing adds cost and cycle time. The buy-side decision is whether the added cost is justified by the risk reduction. The numbers below are the typical cost-and-time impact for a small stainless cooling connector (50–200 g finished weight, 1–3 critical sealing interfaces) at production volumes of 5,000–50,000 pieces per month, based on Sinbo experience and helium mass spectrometer vendor cycle-time data.
| Cost driver | Pressure test only | Pressure test + helium leak test | Delta |
|---|---|---|---|
| Equipment cost (amortized) | Pressure gauge, water tank, manual fixture — low five figures | Helium mass spectrometer ($80k–$200k), vacuum chamber or sniff fixture, helium recovery — mid six figures | Most of the helium cost is the equipment, not the per-part consumable |
| Per-part cycle time | 30–45 s (manual) or 15–20 s (automated) | 45–90 s (single-station) or 30–45 s (parallel two-station) | +30–100% cycle time, mitigable with parallel fixturing |
| Per-part consumable cost | Tap water, electricity for pump | Helium (small fraction of cost; recovery system can capture 80%+) | +<1% of part cost |
| Per-part labor | Operator loads/unloads | Operator loads/unloads, equipment runs automated cycle | Minimal delta if automated |
| Net per-part cost impact | Baseline | +5–8% on a $20–$50 connector (typical Sinbo ai-liquid-cooling-connector case) | +5–8% |
| Failure mode it catches | Gross leaks, missing welds, porosity | All of the above + micro-leaks at the O-ring groove + thermal-cycle walk-out | The leak you cannot see at room temperature |
The decision rule that comes out of this is straightforward:
No — for the leak-critical parts, pressure test alone is not enough. A 4.5 MPa hydrostatic test for 30 minutes catches gross leaks (porosity, missing welds, visible scratches on the sealing face), but it does not catch micro-leaks of 10⁻³ mbar·L/s class, and it does not predict the leak that opens up after thermal cycling. For connectors going into hard-to-access data center installations, add a helium leak test at ≤1×10⁻² Pa·m³/s on 100% of parts.
≤1×10⁻² Pa·m³/s total leak rate is the working threshold for OCP/UQD-class liquid cooling connectors. It is a compromise: tight enough to catch O-ring grooves at the high end of compression tolerance (the parts that would walk out of seal after thermal cycling), loose enough to be achievable in a 30–60 second production test on every part. For leak-critical semiconductor or aerospace applications, you can tighten to 10⁻⁶ or lower, but expect higher per-part cost and longer cycle time.
For a small stainless cooling connector, helium leak testing adds +5–8% to per-part cost and +30–100% to per-part cycle time at the equipment level. The cycle time impact is the bigger operational issue: a single-station helium test takes 45–90 seconds per part vs 15–20 seconds for an automated hydrostatic test. Most production lines mitigate this with a parallel two-station fixture that brings cycle time back to 30–45 seconds per part. The equipment cost (helium mass spectrometer + vacuum chamber + helium recovery) is mid-six-figures, so it is a capital decision for the machine shop, not a per-part cost.
Thermal cycling walks the O-ring out of its compression window. The O-ring (elastomer, CTE ~200×10⁻⁶/K) and the stainless groove (CTE ~17×10⁻⁶/K) expand and contract at very different rates. On the hot half of the cycle, the metal expands more than the O-ring, so squeeze drops. On the cold half, the metal contracts more, so squeeze increases and the O-ring takes a compression set. After 500–1000 cycles, the compression set does not fully recover, the squeeze window drifts, and the seal starts to weep at the worst-case temperature. A 30-minute room-temperature pressure test will not see this — the squeeze is in tolerance and the seal is good. The fix is a thermal-cycle helium leak test for qualification, plus a 100% production helium leak test at ≤1×10⁻² Pa·m³/s to catch the parts whose tolerance stack is at the bad end.
A complete leak-test certificate has: (1) test method (e.g. ASTM E499 sniffer or E493 vacuum), (2) test pressure and dwell (e.g. 4.5 bar / 30 min for hydrostatic, or charge pressure for helium), (3) pass/fail criterion (e.g. ≤1% pressure decay, or ≤1×10⁻² Pa·m³/s), (4) part number, lot number, and serial range, (5) material/heat lot for traceability, (6) test equipment model/ID and calibration date, (7) operator name/ID, (8) test date, quantity tested, quantity passed, quantity failed, and (9) disposition of any failures (rework, scrap, or concession with engineering sign-off). A single-line “leak test passed” on a C of C is not a cert — it is a sticker. Reject and ask for the full certificate.
Yes — for prototypes, hydrostatic pressure test is usually enough. The prototype exists to find gross design errors (wrong port size, mislocated O-ring groove, missing weld), not to verify micro-leak rate. A 4.5 MPa / 30 min hydrostatic test catches all of those. Add the helium leak test for prototype only if you are qualifying a new supplier and need a baseline number to compare against their production process. For the production run itself, helium leak test on 100% of parts is the buyer-side default for leak-critical cooling connectors.
For production cooling connectors, sniffer mode (inside-out, ASTM E499) is the better fit. The part is pressurized with helium inside, and a sniffer probe on the outside scans for escaping helium. Cycle time is 15–30 s per part, fixture is simpler, and the sensitivity (1×10⁶ to 1×10⁶² Pa·m³/s) is more than enough to catch the O-ring groove micro-leaks at issue. Vacuum mode (outside-in, ASTM E493) gives better sensitivity (down to 10⁻⁹ to 10⁻⁻ Pa·m³/s) but requires a vacuum chamber, a pump-down cycle, and helium spraying on the outside — which adds 30–60 s of cycle time. Vacuum mode is reserved for qualification testing of new designs, or for leak-critical applications like semiconductor or aerospace where the 10⁻⁶ floor is the spec.
Send your drawing and leak-rate spec. We will quote the test combination, cycle time, and per-part cost in 24 hours.
Request a Quote