Vacuum Chamber Helium Leak Testing Machine 1.0E-9 Mbar.L/S Two Chambers For Liquid Cooling Cold Plates

A cold plate that weeps does not fail in your factory. It fails in a rack, a year later, when the coolant has been quietly topping itself up and the leak finally reaches something expensive. This two-chamber helium system is built to catch that defect at the line, at a takt the line can actually hold.

What actually fails on a cold plate

Almost never a hole you could see. The failures we are called in to explain come from the joining process, and they share one property: they are small enough to pass a water test and large enough to matter over a service life.

• Friction stir weld defects. A worm hole or a lack of penetration along the weld toe. The joint looks continuous, the water test holds, and the leak opens once the plate has been through a few hundred thermal cycles.
• Braze voids in the channel joint. Same story in a vacuum-brazed plate: a void that is bridged by a thin film of filler and fails in fatigue.
• Port threads cut too deep. The fitting bottoms out on the thread instead of the seal face, and it weeps only under pressure and at temperature.
• Header and manifold welds. The heat cycle leaves a crack that starts at the toe and grows.
• Handling damage. A plate that has been dropped often looks perfect and leaks at the port.

Why water and air tests are not enough

A hydrostatic test proves the plate will not burst today. It does not prove it will stay sealed for eight years, because the defects that cause returns are two to four orders of magnitude below what a water test resolves.

Air decay is cleaner and faster, but a cold plate has the worst possible geometry for it: a small internal volume, a large metal mass that fights temperature stability, and a criterion that is usually tighter than air can resolve inside a takt. You end up either accepting a coarse limit or waiting three minutes per part.

Helium inside the plate, vacuum outside it, and a mass spectrometer reading the rise. The plate is never exposed to internal vacuum, so a thin plate stack is not deformed by the test itself — which is a real risk on a 1 mm channel wall.

A number worth remembering

Leak rates confuse people because the units look academic. One useful conversion: 1×10^-6 mbar·L/s is roughly 0.6 mL per minute of gas at atmospheric pressure — about 300 litres a year. For a coolant circuit that is a visible loss, not a theoretical one.

Which is why the acceptance criterion deserves a proper conversation. A plate specified at 1×10^-5 mbar·L/s is allowed to lose litres of coolant over a service life. A plate at 1×10^-7 is not. Both are "leak tested"; only one of them will keep a data hall dry.

Design choices and why we made them

Gross stage first, fine stage second

Helium enters at up to 3.0 MPa and the pressure is held. A plate with a bridged weld is rejected in seconds, before the analyser is asked to work. This is what turns a three-minute test into a one-minute test.

Two chambers, out of phase

While one chamber measures, the other is unloaded and reloaded. One operator, about a hundred plates an hour. A single chamber with the same sensitivity needs a second person to reach that output.

Recovery sized for a high cycle count

A hundred plates an hour, two shifts a day, is a lot of helium down the vent if you do not recover it. The loop is integrated with purification after the buffer, so what goes back into the next plate is still within specification.

A drying option, because plates arrive wet

If a plate has been through a water circuit, that moisture outgasses under vacuum and reads as a leak. We can add a dry-air purge step to the recipe, or you can test before the water circuit is filled — both work, guessing does not.

How a cycle runs

• Load and clamp — plate into the chamber, coolant ports connected, pneumatic seal on the lid, barcode scan.
• Evacuate — roots and rotary pump set pull the chamber down; the recipe waits for a stable base pressure.
• Gross stage — helium charge to the recipe pressure and hold; a pressure drop ends the cycle immediately.
• Fine stage — analyser integrates the chamber signal and converts it to a leak rate against the limit.
• Recover and release — helium pulled back, chamber vented, plate released with a verdict and a stored record.

Technical parameters

What it finds

Setting the limit, and proving the machine can hold it

• Work backwards from coolant loss. Use the conversion above, the coolant volume and the acceptable annual top-up, and derive the rate. Most server cold plates land between 1×10^-6 and 1×10^-7 mbar·L/s.
• Run ten good plates and ten with a calibrated leak. If the two populations are separated by a factor of three or better, the test is sound. If they overlap, lengthen the test rather than tightening the limit.
• Repeat after a process change. A new weld program or a new brazing cycle changes the defect population, and the limit should be revalidated, not inherited.
• Document the method with the limit. Charge pressure, dwell, chamber size, ambient band. Auditors ask for it, and so will your customer's quality engineer.

Compared with the alternatives

The comparison that usually decides the purchase is not with the alternatives but with the cost of a field failure: a rack pull, a coolant cleanup, a customer visit and a reputational hit that no test machine budget can absorb. Once a customer has had one, 100% inspection stops being a cost argument.

Delivery, installation and the first month

• Factory acceptance test on your plates before shipment, with video and calibration record.
• Fixture and port connector concept agreed against your drawing before fabrication.
• On-site commissioning: fixture prove-out, leak-rate correlation against a calibrated leak, operator training.
• Repeatability study on good plates and a calibrated-leak plate, signed off with your quality engineer.
• Recipe per plate family, selected by barcode, with the drawings handed over.
• 24-month warranty on mechanics, pneumatics and control; pumps per manufacturer terms.
• Spares in Guangzhou: lid seals, port connector seals, helium valves, pump service kits — 48-hour dispatch.
• Annual calibration with traceable certificate and remote support for recipe tuning.

Frequently asked questions

About Guangzhou Hejin Automation Control Technology

Since 2003 we have built leak detection and electrical test equipment in Guangzhou, with 50–100 employees and around half of our production exported. Our engineers commission chamber systems, tracer gas hoods and air decay testers on the same floor, so the recommendation for a cold plate comes from test data rather than from a catalogue.

Every system is assembled, pump-down tested, calibrated and recorded under its serial number before it leaves us. When a question comes up three years later, the original test data is still on file.