Aluminum FSW Cooling Cavity: Holding a 0.8 MPa Seal on 350 Parts

In the third trial batch, we pulled ten cavities off the line for a hydraulic leak test at 0.8 MPa. Three of them bubbled. Not a fast drip, a slow weep at the weld seam, the kind that takes a minute to show and a lot longer to explain to a customer. We cut sections and put the seal faces under a comparator. The contact surface was uneven. The weld had pulled the body, and the seal face had followed it.

The final production run: 350 aluminum FSW cooling cavities, 349 passed the hydraulic leak test the first time. The one that failed was reworked and retested to the same 0.8 MPa, 30-minute standard and passed. First-pass yield 99.7%. No redesign, no tolerance change, no scrap.

That trial batch is why we stopped running the standard rough, finish, weld sequence and rebuilt the process from the distortion source.

1.The part and the four requirements

The customer builds semiconductor inspection equipment in Europe. The cooling cavity moves heat across a dense inspection array, so it has to seal under pressure, hold its datum face flat enough to seat on a precision housing, and land in the same position on every part in the batch. Specs: 425 x 218 x 72 mm, Al 6061-T6, internal closed-loop cooling channels, minimum wall thickness 3.2 mm, cover plate sealed by friction stir welding. Order quantity 350 pieces.

Four hard requirements at once: cover plate seal at 0.8 MPa for 30 minutes with zero leaks; datum face flatness of 0.04 mm or better; locating hole position within ± 0.02 mm; overall body distortion after all operations of 0.06 mm or less.

Every cavity ran on our 5-axis CNC machining line, with the cover plate closed by FSW.

Engineering read: on a closed-loop cooling cavity that has to pass a hydraulic leak test, FSW is the right sealing method for a thin aluminum wall. It runs cooler than TIG or MIG and bonds without the porosity that lets water find a path. On a thick-wall housing where distortion is not a concern, TIG or MIG is faster and costs less.

2.Why the standard sequence leaked

The conventional route roughs the billet, finish machines it, then welds the cover plate by FSW. That order produces two distortion events with no recovery step between them.

• Roughing pulls a large volume of material off the billet. The rolling stress locked in the stock releases as the metal comes off. With 3.2 mm walls left standing, that release bends the part.
• FSW lays heat in a concentrated band along the weld path. On a thin-wall cavity, that thermal gradient bows the body. The datum face follows. When the sealing face is not flat, the cover plate cannot seat evenly, and the joint leaks.

Third trial batch, hydraulic test: 3 of 10 sampled cavities leaked, a slow bubble weep at the weld line. Sectioning showed uneven contact on the seal face. The weld distortion had also shifted the locating holes, and a sample assembly showed interference on the machine frame.

Engineering read: on a thin-wall FSW assembly, finishing before welding and expecting a post-weld skim cut to recover flatness does not work. FSW distortion on a 3.2 mm wall pushes the datum face past its tolerance before the final cut even starts. On a thick rigid housing, the body resists weld distortion and a post-weld skim cut is enough.

Engineer’s read on why this class of part leaks “Most shops chase FSW parameters and miss the stacked distortion. Rough a thin wall, go straight to finish, then weld it, and you have layered two deformation events with no recovery between them. By the time you hit the leak test the seal face is already gone.”

3.Five changes that held the seal

• Billet pre-treatment. Low-temperature anneal on the billet before any cutting, to release the rolling stress locked in the stock.
• Three-stage machining with stress relief built in. Rough machine, then natural age 48 hours; semi-finish the channels and outer profile; second age cycle to drain the stress the first age did not reach; leave 0.12 mm stock for finish.
• Low-stress finish cut. PCD dense-tooth cutter, 0.12 mm finish pass at reduced cutting pressure. Channel inner wall to Ra 0.8 µm or better, datum face to Ra 0.4 µm or better. No burrs, no loose chips in the flow channel, particle-free on delivery, which the semiconductor application needs.
• FSW rework. Segmented intermittent weld path, pin plunge depth and travel speed dialed down from the original single-pass parameters, symmetric weld sequence so each side of the cavity sees the same heat load. Post-weld datum face stayed within 0.035 mm, inside the 0.04 mm limit.
• Post-weld sizing, then finish last. A post-weld low-temperature sizing age, then the final finish cut on the datum face and locating holes after welding, so the weld distortion is already complete before the final dimensions get set. 100% hydraulic leak test at 0.8 MPa with a 30-minute hold on every part, plus CMM spot checks on flatness and hole position.

We built two aging cycles into the sequence, one after roughing and one after semi-finish. The timing of stress relief around machining changes how much a thin wall moves later, and a single age does not drain every layer of stress.

Engineering read: cutting the datum face and locating holes after welding beats the conventional order, because it takes post-weld distortion out of the final dimension. On a rigid casting or thick-wall housing that barely moves under weld heat, machining before welding saves time and the tolerance still holds. Deciding what to finish last comes down to which features need tighter tolerance control.

4.Results, measured on this part

350 pieces delivered. Every number below was measured on this production run, not taken from a material or process datasheet.

Flatness (datum face)	0.04 mm or better	0.020 to 0.035 mm	Sealing face seats clean, no gap for the water to find
Overall body distortion	0.06 mm or less	0.05 mm or less	Cavity drops into the housing without fighting it
Locating hole position	± 0.02 mm	All within ± 0.02 mm	No interference on machine assembly
Hydraulic leak test @ 0.8 MPa / 30 min	Zero leaks	349 passed first time; 1 reworked and passed	First-pass yield 99.7%; 1 rework, 0 scrap

Client outcome The customer received 350 cavities with no design change and no tolerance relaxation. All 350 passed the hydraulic test before shipment. The single reworked part was repaired and retested to the same 0.8 MPa, 30-minute standard and passed. First-pass yield: 99.7%.

5.Carry this into your next thin-wall FSW part

• On thin-wall aluminum FSW assemblies, anneal the billet before roughing. Rolling stress in the stock adds to cutting stress before you even touch the part.
• Build two age cycles into the sequence, one after roughing and one after semi-finish. A single age does not drain all the layers of stress.
• Run the FSW path symmetric. A single-pass linear weld pulls the body to one side. Balanced heat keeps the datum face in tolerance.
• Machine the datum face and locating holes after welding, not before. Post-weld distortion will not come back out by re-machining a pre-welded surface.
• Leak test every piece at 0.8 MPa with a 30-minute hold. Sampling is not enough on a sealed flow channel. Our third batch proved it: 3 of 10 sampled pieces had weep leaks a 100% test would have caught before shipment.

6.Client info

From	Europe
Industry	Semiconductor inspection equipment: precision thermal management, particle-free internal channels
Scope	350-piece low-volume production run of Al 6061-T6 FSW cooling cavities, 99.7% first-pass hydraulic yield
Processes	5-axis CNC milling | FSW sealing | PCD finish | Low-temperature stress relief | 100% hydraulic leak test | CMM inspection

This was a low-volume production run, 350 parts, quoted and delivered as one batch with one project manager from first drawing to shipment.

7.Questions buyers ask before quoting a sealed cooling cavity

Can friction stir welding seal a 3.2 mm aluminum wall to 0.8 MPa without leaks?

On this part, yes. FSW ran cooler than TIG or MIG and bonded the 6061-T6 cover plate without the porosity that lets water weep through a seam. 349 of 350 cavities passed the 0.8 MPa, 30-minute hold on the first test. The one that failed was reworked to the same spec and passed.

Why did the standard rough, finish, weld sequence leak on a thin-wall cooling cavity?

It stacks two distortion events with no recovery between them. Roughing releases the rolling stress locked in the billet and bends the 3.2 mm walls. FSW then lays a concentrated heat band along the weld path, bows the body, and moves the datum face off flat, so the cover plate cannot seat evenly and the joint weeps.

What datum face flatness did the cooling cavities hold after welding?

We measured 0.020 to 0.035 mm across all 350 pieces, inside the 0.04 mm limit. Cutting the datum face and locating holes after welding, once the weld distortion was already complete, is what kept the number there.

Is 100% hydraulic leak testing worth it over sampling on a sealed cooling cavity?

On this run it was. In the third trial batch, 3 of 10 sampled cavities showed a slow weep at the weld line. A sampling plan would have shipped leakers. Every one of the 350 production parts got a 0.8 MPa, 30-minute hold before it left the floor.

Does this process add cost compared with the conventional machining route?

It adds two aging cycles and moves finish machining to after welding, so machine time per part goes up. It removed the scrap, redesign, and tooling rework that a leaking seal face would have forced. First-pass yield came in at 99.7% with zero scrap.