A collaborative robot manufacturer moved a joint housing from machined prototypes to ADC12 die casting at 12,000 units a year, cutting 180 g per joint while holding ±0.02 mm concentricity between the harmonic drive seat and the encoder bore.
The housing appears three times in each arm, so mass in the structure directly reduces rated payload. The machined prototype weighed 520 g, and at three joints per arm the housing family alone was consuming enough of the mass budget that the target payload could not be met without redesigning the part.
The housing reacts 165 Nm of peak torque from the harmonic drive. Deflection at the drive seat had to stay under 0.05 mm at peak load, because any more degrades the gear mesh and shows up as backlash and position error at the tool point.
A 2 mm wall running 180 mm from the gate is where cold shuts and short fills live. Filling that reliably shot after shot, rather than on a good day, is a gating and alloy problem rather than something that can be corrected on the machine.
The Ø90 harmonic drive seat and the Ø24 encoder bore must be concentric within ±0.02 mm. Any runout between them is read by the encoder as position error, so the requirement is a control-loop specification rather than an assembly convenience.
ADC12 carries 9.6 to 12.0% silicon against A380's 7.5 to 9.5%, which raises fluidity and keeps the fill front alive over long thin flow paths. The trade-offs were accepted deliberately: 310 MPa tensile against A380's 324 MPa, and a lower machinability rating that pushed the drive seat and bore onto PCD tooling.
Flow and solidification simulation ran through five gating designs. The released tool uses a two-stage tangential fan gate at 46 m/s gate velocity with a 41 ms fill, and five overflows sized to pull the leading 6% of the metal front off the part. Predicted cold shut at the far rim went to zero and the first articles confirmed it.
Where FEA showed excess deflection, the answer was a 1.4 mm internal rib lattice at 14 mm pitch, held at 70% of the 2.0 mm wall so it does not become an isolated thermal mass. Measured deflection at the drive seat under 165 Nm came out at 0.031 mm. A plain 3.2 mm wall would have reached the same stiffness at roughly 470 g and a substantially longer cycle.
The drive seat and the encoder bore are machined in a single 5-axis setup located on three cast pads in the fixed die half, so their concentricity comes from the machine's rotary accuracy rather than fixture repeatability. Measured concentricity across the qualification lot ran 0.008 to 0.017 mm against the ±0.02 mm requirement.
The drawing arrived calling for Type III hard anodize. Type III is not viable on a 9.6 to 12% silicon alloy carrying up to 3.5% copper, and we said so before quoting rather than after sampling. A 70 µm matte black polyester powder coat in RAL 9011 held ΔE below 0.8 batch to batch, covered the parting line, and cost less. Bores and the drive seat are masked.
The customer builds collaborative robot arms in the 5 to 10 kg payload class. They had validated the arm with machined prototypes, which is the correct way to do it, and were preparing to launch at 12,000 units a year. The joint housing blocked them twice over.
Cost was the visible problem. Machined from billet the housing came out at roughly $58, and it appears three times per arm, so it carried $174 of the arm’s bill of materials on its own.
Mass was the real one. Structural mass on a robot arm cannot be traded against cost, because it comes directly off the payload the product can advertise. At 520 g per joint the housing family consumed enough of the budget that the arm could not be rated where it needed to be. Casting was attractive precisely because it makes rib geometry free, and stiffness per unit mass is what a joint housing is really buying.
A380 is the default for a reason and it would have been cheaper to machine. The deciding factor was the 2 mm nominal wall over a 180 mm flow path. ADC12 carries 9.6 to 12.0% silicon against A380’s 7.5 to 9.5%, and more silicon means better fluidity and a fill front that stays liquid longer, which is exactly what a thin wall on a long flow path needs.
Two costs came with that and both were accepted with open eyes: tensile strength drops from 324 MPa to 310 MPa, irrelevant on a stiffness-limited part, and machinability drops a grade, which moved the drive seat and encoder bore onto PCD tooling rather than carbide.
The instinct when FEA shows too much deflection is to thicken the wall. On a die casting that is close to the worst available answer, because a thicker wall adds mass, lengthens solidification and creates the isolated thermal masses that become shrinkage porosity.
The housing was stiffened instead with a 1.4 mm internal rib lattice at 14 mm pitch, held at 70% of the 2.0 mm wall so the ribs freeze before the wall rather than after. Deflection at the drive seat under 165 Nm came out at 0.031 mm against a 0.05 mm limit. A plain uniform wall would have needed roughly 3.2 mm for the same stiffness, putting the part near 470 g. That is why the casting is lighter than the machined part it replaced rather than merely cheaper.
A 2 mm wall filling 180 mm from the gate is not automatic, and it is not something to discover at first articles. Flow and solidification simulation ran through five gating designs before the tool was released.
The accepted design uses a two-stage tangential fan gate at 46 m/s gate velocity, filling the cavity in 41 ms, with five overflows sized to pull the leading 6% of the metal front off the part. That leading metal is the coldest and most likely to carry entrained air, so removing it rather than letting it freeze at the far rim is what eliminates the cold shut. Sectioned first articles confirmed the predicted clean fill at the rim. The tool went in as a two-cavity straight pull, with cable routing and connector features arranged so no slide was needed.
The encoder reads angular position through the drive seat, so any runout between the Ø90 drive seat and the Ø24 encoder bore is indistinguishable from a position error in the control loop.
The answer was to make ±0.02 mm concentricity a machine problem rather than a fixture problem. Both features are cut in a single 5-axis setup located on three cast pads in the fixed die half, so concentricity is limited by the machine’s rotary accuracy rather than by how repeatably a second fixture picks the part up. Measured concentricity across the qualification lot ran 0.008 to 0.017 mm. Splitting the work across two setups would have consumed 0.05 to 0.08 mm on fixture repeatability alone, before a single cut.
The drawing specified Type III hard anodize for scratch resistance and a dark uniform appearance. On ADC12 that is not achievable, and it is better to say so at quotation than to sample it and argue afterwards. Anodizing converts aluminum to oxide and silicon does not convert; at 9.6 to 12% silicon the film grows around coarse eutectic particles, and up to 3.5% copper dissolves preferentially in the bath, darkening and pitting the surface further.
What the customer needed was a consistent dark matte appearance. A 70 µm matte black polyester powder coat in RAL 9011 held ΔE below 0.8 against a signed master between batches, covered the parting line witness, and cost less. Bores and the drive seat are masked, since 70 µm of film would take both out of tolerance.
The casting weighs 340 g against 520 g machined, a 34.6% reduction, removing 540 g from each arm across the three joints and restoring roughly 0.4 kg of rated payload. Piece cost came in at $9.10 cast and machined against about $58 machined from billet, and the part runs on the original two-cavity tool at a 15 day production lead time.
The change that mattered most was not the process conversion. It was replacing thickness with ribs, which is the one thing casting gives you for free and machining never will.
| Part weight | 340 g (0.75 lb) |
|---|---|
| Envelope | 182 × 118 × 94 mm (7.17 × 4.65 × 3.70 in) |
| Alloy | ADC12 (A383), 310 MPa UTS, 96 W/m·K |
| Wall thickness | 2.0 mm nominal, 1.6 mm at the encoder shroud |
| Machine | 400 ton cold chamber, 41 ms fill time |
| Cavity count | 2 cavities, straight pull, no slides |
| Annual volume | 12,000 units |
| Critical tolerance | ±0.02 mm concentricity, Ø90 drive seat to Ø24 encoder bore |
| Finish | Vibratory deburr, blast, 70 µm matte black powder coat, RAL 9011 |
| Part weight | 340 g, down from 520 g machined (−34.6%) |
|---|---|
| Arm mass removed | 540 g across three joints, about 0.4 kg of restored payload |
| Piece cost | $9.10 cast and machined, from about $58 machined (−84%) |
| Bore concentricity | 0.008-0.017 mm measured against ±0.02 mm spec |
| Drive seat deflection at 165 Nm | 0.031 mm against a 0.05 mm limit |
| Production | 12,000 units a year on a 2-cavity tool, 15 day lead time |
Send your drawing and we will tell you what this geometry costs in your annual volume.
Request A Quote
A European industrial pump manufacturer converted a billet-machined housing to A360 die casting to reach 8,000 units a year, holding 6 bar pressure tightness, an H7 bearing bore and a 0.05 mm sealing face.
An LED lighting brand replaced a bonded two-piece design, an extruded heat sink adhered to a die cast housing, with a single A413 casting, removing the thermal interface and taking 11 °C off the LED junction temperature at 60,000 units a year.
Send your 2D drawing and 3D model and a manufacturing engineer will review the geometry, confirm the alloy and return pricing with a DFM report. No obligation, no sales scripts.