Liquid Cooling Manifold Inspection with Vision Measuring Machines

Inside a coolant distribution unit (CDU), the manifold has a simple job in principle: it splits one coolant feed into several branches and sends each of them to a cold plate. In practice the part is a milled aluminum plate roughly 300 mm square that carries forty to fifty holes, two or three seal grooves and a set of datums. A port a few hundredths of a millimeter off center will not accept its branch fitting. A groove slightly too shallow will not compress the gasket. A plate that warps will not sit flat against its bracket. Calipers and profile projectors cannot cover this combination, which is why vision measuring machines, the workhorse of non-contact optical dimensional measurement, have moved into manifold inspection.

Four feature groups that decide whether a manifold works

The port network comes first. Inlet, outlet, branch ports, mounting holes and dowel holes are not independent features; they form a network whose relationships are fixed by the datums on the drawing. Many branch ports sit inside the cavity, and their bores have to line up with fittings on the outside of the plate. A position error is amplified along the assembly chain and shows up as a fitting that will not seat, or a gasket that sits off center.

Bore size matters next. Branches connected in parallel share one feed, so bore consistency decides how the flow splits. When bores differ, the branch nearest the inlet takes more flow and the far branches take less, and the temperature difference between cold plates follows. Manifold bores are therefore commonly held to plus or minus 0.02 mm, with roundness and burr height also on the inspection sheet.

Seal grooves are the third group. Groove width, depth, floor flatness and edge burr decide whether gasket compression lands inside its design window. Too shallow and the joint will not seal; too deep and the gasket is under-compressed; a rolled edge will cut the gasket. Calipers cannot measure these, and a projector shows only the outline, never the depth.

Plate face and thickness are the group most often skipped. Manifolds are usually milled from aluminum plate and then vacuum brazed. The heat input of brazing warps the face by amounts the eye cannot detect. Insufficient flatness leaves a gap between the manifold and its bracket or cold plate; clamping closes the gap but concentrates stress, and over time that becomes a leak path.

Why conventional tools struggle here

A caliper measures length and diameter but not the relationship between holes, and it cannot evaluate position tolerance against a datum. A profile projector projects a silhouette, so when the ports sit inside a cavity and the cavity wall blocks the light path, only the outer edge can be seen. A coordinate measuring machine is accurate enough, but point-by-point probing is slow, and a stylus pressing on a thin wall or a sealing face carries a real risk of damage. Raise the sampling rate and inspection becomes the bottleneck.

The material adds a second problem. Anodized or bead-blasted aluminum is diffuse, while unmasked areas are close to a mirror, so one plate presents large brightness differences. Edge detection on ordinary optical equipment drifts between these zones: the edge pulls in on one pass and out on the next, and the same bore reads three different values in three measurements. This is part of why manifolds have a reputation for being difficult to measure.

Case: a 46-port manifold, measured in a single setup

A coolant distribution unit manufacturer in South China sent a plate measuring 320 mm by 180 mm, with 46 holes and three seal grooves, to Sinowon for a sample run. Their existing method combined a caliper for the envelope, a projector for the profile and CMM sampling for hole positions: about 25 minutes per part, with the report filled in by hand.

The sample run used a SIG fully automatic vision measuring machine. The plate was placed on the glass stage with adjustable supports under the base, so nothing loaded the sealing face. Transmitted light from below the stage turned every hole and groove into a clean black silhouette with a sharp boundary. Alignment used one datum edge and two datum holes to build the same reference frame as the drawing; without that step, a position tolerance value means very little.

Programming started from the CAD file. Circles, slots and lines were recognized automatically and the measuring path was generated; the operator only confirmed which ports needed a position tolerance and which grooves needed a width. Illumination switched with the feature: coaxial light for deep bores so the wall and floor are evenly lit, low-angle ring light for groove edges and burrs, transmitted light for the outer contour.

After one setup, bore diameter, roundness, center distance, position tolerance to datum, groove width, groove-edge straightness and the plate envelope were all captured in a single program run. Depth items such as groove depth and cavity steps were taken with an optional laser probe or touch probe, so two-dimensional and three-dimensional data landed in the same report.

What changed after the switch

Measurement time fell from about 25 minutes to under five, most of it unattended program run time with the operator only loading and unloading. Ten re-clamps of the same plate kept position results within plus or minus 3 micrometers, which confirms that the fixturing and alignment method are repeatable. The first-article report is generated with the part, and every bore, position and groove width is judged against its tolerance as it is measured.

The more useful output was the data itself. Thirty consecutive plates, plotted on groove depth and face flatness, showed a consistent pattern: after brazing the center of the plate dished inward and the edges lifted slightly. That pattern used to be inferred from spot checks and feel; once it was quantified, process engineering adjusted the braze support points and the cooling curve, and rework caused by branch fittings that would not seat declined.

Changeover stopped depending on one experienced operator. For a new manifold of similar construction, the existing program template is recalled, the datum and port count are edited, and a working program is ready in about half an hour, ready for the line the next morning.

What to confirm when specifying a system

Manifolds range from about 200 mm to nearly 600 mm, so travel should be sized from the upper end of the part range; a gantry machine is the better fit for measuring a full plate in one setup. Specify a glass stage with transmitted illumination, which is what makes hole and groove contours readable. The light package needs ring, coaxial and bottom light with zone control, otherwise the bright and dull areas of a single plate cannot be handled together.

On the software side, three questions decide most of the value: can position tolerance be evaluated from the drawing datums, can groove width and profile be computed in one evaluation, and can the report template be exported in the customer format. If the manifold carries deep cavities and steps, add a laser or touch probe so depth dimensions join the same program.

In daily use, a glass reticle check at fixed intervals catches magnification drift before it reaches a report, and disciplined program naming and version control prevents two conflicting programs for the same model. These look like administrative habits; in practice they decide whether measurement data stays trustworthy over years.

A manifold packs many holes, thin walls, mixed reflectivity and brazing distortion into one part, and conventional gauges can reach only part of it. An optical dimensional measurement approach built around a vision measuring machine puts the port network, the seal grooves and the face topography into one non-contact program, with no stylus pressure on thin walls and results that repeat and can be traced. For manifold suppliers, that capability eventually shows up in assembly rework and in how smoothly customer audits run. For more application notes, visit www.sinowon.com.