BMW factory 3D prints sand cores for large engine cylinder heads

Binder jetting builds a sand core the way an inkjet printer lays down ink. A recoater spreads a thin layer of foundry sand, usually 0.2–0.4 mm, and a print head deposits a liquid binder only where the core geometry needs it. The build box drops, another layer goes down, and the cycle repeats until the core sits buried in a block of loose sand. No core box, no pattern, no draft angles to plan around.

That last point is the reason BMW factory 3D prints sand cores at all. A conventional core requires a metal or resin core box, and the box requires draft so the core can be pulled without tearing. Printed cores have no pull direction, so internal water jackets can take shapes a two-part box cannot release. When cooling channels need to follow a valve bridge or wrap a bore, that freedom matters more than the cost per core.

Layer thickness sets the surface you get. A 0.2 mm layer leaves a finer stair-step on internal passages than a 0.4 mm layer, but it also takes roughly twice the build time. For a water jacket that will be sand-blasted and coated, a coarser layer is usually acceptable. For a passage that stays as-cast, the finer layer saves finishing work later.

After the build, the core is depowdered and cured. Uncured binder in the loose sand is brushed or blown away, then the core goes through a thermal cure that hardens the binder. Cure temperature and time decide the core's transverse strength. Push the cure too short and the core erodes when metal hits it. Push it too long and the core becomes hard to shake out.

• 1
  Layer thickness0.2–0.4 mm; finer layers cost build time
• 2
  No draft neededInternal passages can be non-releasable
• 3
  Cure mattersShort cure erodes; long cure resists knockout

Printed cores win on low volume and geometry that a core box cannot release. A cylinder head program that builds a few hundred units, or a design that is still changing between builds, fits the process well. Change the CAD and you print a new core the next day. There is no core box to cut, no pattern to store, and no revision to scrap.

Volume is the other half of the argument. Below roughly a few thousand cores a year, the tooling cost of a conventional core box rarely pays back. Above that, a machined core box spreads its cost across enough parts that printing loses on unit price. The crossover point depends on core size and complexity, not on a fixed number, so run the comparison for your own part.

Geometry that a two-part box can release is also a poor candidate. A simple round bore or a straight passage prints fine, but it does not need the process. Save the printed route for cores with undercuts, curved jackets, or thin walls that would tear during pulling.

There are limits. Large cores take a long time to build, and a build failure near the top wastes the whole run. Sand cores also need a foundry that can handle them; not every caster has the handling and core-setting practice for printed parts. If the caster is not set up for it, the printed core is only half a solution.

• 1
  Good fitLow volume, changing design, undercuts
• 2
  Poor fitHigh volume, simple releasable geometry
• 3
  Watch outLong builds, foundry handling skill

A printed core only gets you a near-net shape. The head still needs its deck faced, cam bores line-bored, valve seats cut, and bolt holes drilled. Those features sit on datums that the casting itself defines, so the first operation is usually establishing a reliable datum from rough surfaces.

Cylinder heads are awkward to hold. The deck is large and flat, the sides are irregular, and wall thickness varies. A 5-axis setup lets us reach the deck, the ports, and the cam bore in fewer fixtures, which matters when the casting is expensive and the batch is small. On our 16 simultaneous 5-axis centers, a head-sized part up to 4,000 mm can be machined without repositioning it in a second machine.

Wall thickness is the risk on any thin-wall casting. If the printed core shifted during pouring, the wall may be thinner than the model says. We check raw castings before cutting and adjust stock allowance so the deck and bores clean up without breaking through. Valve seat and guide bores tolerate no porosity, so those areas get extra stock and closer in-process checks.

Tolerance and finish follow the function. Deck flatness and cam bore alignment are the tight ones; a machined finish of Ra 0.8–1.6 μm is typical for sealing faces, and finer work down to Ra 0.2–0.8 μm is possible where a seat or journal calls for it. A general tolerance of ±0.005 mm applies to machined features, not to the as-cast surface.

• 1
  First opEstablish datum from rough casting
• 2
  Fixturing5-axis reduces setups on irregular heads
• 3
  Stock controlCheck casting before cutting thin walls

Send the casting drawing and the machined drawing together. The gap between them is the stock allowance, and it tells the shop where the risk sits. If the casting supplier can hold ±0.5 mm on wall thickness, say so; if not, the shop has to plan for more variation.

State the datums and the critical features. Deck flatness, cam bore position, and valve seat concentricity are the usual ones. Then list the features that must not break through, and the minimum wall you will accept. That list drives the inspection plan more than any general tolerance callout.

Tell us the batch size and whether the design is frozen. A frozen design at high volume points to one process; a design still moving points to another. We quote both if the volume is ambiguous, so you can see where the crossover lands for your part.

Keep the foundry and the machine shop talking. If the printed core changes, the casting changes, and the machining stock changes with it. A shop that sees only the final drawing will quote conservatively; a shop that sees the core revision history can quote the real number.