Machining Tolerances and Surface Finishes for Robot Joints, Harmonic Drives and Structural Parts: What to Specify and What Drives Cost

Which joint and harmonic-drive features need single-digit-micron tolerances

Tolerance on a robot is not one number for the whole part. The features that set endpoint accuracy are the ones that locate rotating elements: bearing journals, cross-roller races and harmonic reducer mounting faces are reported at ±0.005–0.008 mm under ISO H6/h6 fit classes¹².

The mechanism is cumulative — a deviation beyond ±0.01 mm at one joint compounds through the linkage, so the arm loses endpoint repeatability and gear teeth bind rather than mesh¹². Harmonic-drive housings sit at the extreme of that logic.

The flex spline, wave generator and circular spline must stay concentric, and bore eccentricity above 0.01 mm is reported to cause binding and premature wear¹² (supplier-reported figure). Which is why these housings are described as among the highest-precision custom parts in robot manufacturing¹⁵.

A housing bore that drifts past that limit does not fail at assembly; it wears in service. Away from the interfaces, the requirement drops sharply.

Arm link end interfaces are held to about ±0.01 mm on bolt patterns and pilot diameters, while mid-span structure can accept standard tolerances ¹⁵. Structural faces, pocket depths and non-interface surfaces are cited at ±0.05 mm to ±0.1 mm without affecting robot performance.

And tightening them is described as a common DFM mistake that adds machining cost without improving accuracy¹⁵. So the drawing rule is feature-by-feature: put the micron band on bearing seats, gear interfaces, servo motor pilot diameters and joint pivot holes, and leave the rest at standard.

That split also cuts inspection time, since ±0.01 mm versus ±0.1 mm callouts can double inspection time and scrap rate^{22 12} (supplier-reported figure). When you release the drawing, mark which dimensions are functional interfaces and which are default-tolerance, so the shop does not apply the tightest note to every face.

Which surfaces need a fine Ra and which can stay as-machined

Surface finish behaves like tolerance: it costs only where you call it out. Ra 0.8 µm is reported to require a secondary machining pass, while Ra 3.2 µm is as-machined^{22 15} (supplier-reported figure).

That single distinction decides whether a surface adds a finishing step or comes off the machine finished. The robot-specific case is contamination control.

Cleanroom robot end-effectors are reported to need Ra 0.8 µm or better with electropolished stainless steel to prevent particle generation¹⁵. The finish and the material work together there — the electropolish removes the surface that would shed particles, which is why the requirement is written as a pair rather than a bare Ra value.

For comparison, aerospace bearing bores are cited at Ra 0.4 µm, a finer band than the cleanroom figure³². That is an aerospace requirement under aerospace conditions, not a robotics default, and the sources give no required Ra for general robot joint mating surfaces beyond the cleanroom and aerospace examples.

Practically, you specify fine finishes on sealing faces, sliding contacts and contamination-sensitive surfaces, and leave structural faces, pocket floors and non-mating geometry as-machined. Each Ra 0.8 µm callout you add is a machining step you have chosen to buy¹⁵ (supplier-reported figure); each one you drop is a step removed from the route.

Choosing material by machinability, strength and rate premium

Material choice moves two things at once: how fast the part machines and what the stock costs.

The published ranking puts 6061-T6 at excellent machinability and low relative cost, 7075-T6 at good machinability and medium cost with the best strength-to-weight of the aluminum alloys, 4140 alloy steel at good machinability and low-medium cost with high strength after heat treatment. And Ti-6Al-4V at difficult machinability and high cost².

Titanium is the expensive end of that scale for a second reason. It is reported to add 30–100% to 5-axis machining rates, on top of the slower cutting and faster tool wear that a difficult material implies³¹.

That premium is a 5-axis rate figure under stated conditions, not a part-cost figure, so use it as a direction rather than a multiplier on your quote.

For robot structure, the common starting point is aluminum: arm links are described as aluminum 6061 with wall pockets machined into the inner faces to cut weight while keeping bending stiffness¹⁵. Where a shaft or gear needs strength after heat treatment, 4140 is rated good machinability at low-medium cost².

Titanium earns its premium where strength-to-weight or corrosion resistance is the binding constraint, not as a default for structural parts. Before you commit a material on the drawing, ask the supplier to quote the same geometry in the two candidates you are weighing.

The machinability ranking tells you the direction; only the quote tells you the size of the gap for your part.

When to stay with CNC and when to move to casting or additive

The volume effect on process choice is consistent across sources. CNC machining is positioned for prototypes, low volumes, tight tolerances and designs that may still change, because it needs no tooling and absorbs revisions¹⁹.

Die casting and injection molding become more economical at higher volumes, once the tooling investment is spread across enough parts¹⁹. The pricing mechanism behind that is setup amortization: setup cost dominates the price at low quantity, while material dominates at high quantity²².

A prototype quote is mostly the time to program, fixture and set up the machine; a production quote is mostly stock and cycle time. Die casting has a defined tolerance floor to weigh against that.

It reaches about ±0.1 mm per 25 mm depending on material and part design, with additional CNC machining used to refine features or meet tighter specs⁷. That is roughly an order of magnitude wider than the joint interfaces discussed above.

So a die-cast housing with functional bores still needs a machining pass on those bores — a hybrid route rather than a replacement. Investment casting follows the same pattern: it suits complex near-net shapes and repeat production, but critical features may still need CNC machining, so hybrid strategies are common²⁵.

The decision you are making is not CNC versus casting in the abstract; it is which features can be left as-cast and which must be machined after. No source in this set gives a numeric volume break-even, so ask your supplier where their own crossover sits for your part.

Scoping inspection and certification to the application

Inspection scope is a line item, not a formality. A full CMM FAI is reported to add $150–$500 per lot, with PPAP Level 3 adding more²².

That figure is per lot, so it repeats on every batch you order — worth knowing before you write "full CMM FAI" on a part that does not need it. Certification level gates which suppliers can bid at all.

ISO 9001 covers most industrial work, AS9100D is required for flight hardware, and A2LA accreditation validates measurement competence separately from the quality management system²³. Aerospace and defense work is reported to carry a 20–50% premium for mission-critical components, driven by traceability, quality systems and exotic materials^{31 22} (supplier-reported figure).

That premium is an aerospace/defense figure under stated conditions; it is not a robotics price adder, but it tells you the direction if you specify flight-grade documentation on an industrial robot. What you ask for should match what the part does.

For a general industrial robot joint, ISO 9001 with a defined inspection report is the baseline; for flight hardware, AS9100D is the entry requirement²³.

Two checks sharpen the choice before you commit: request a sample FAI report from a comparable part, since its format and completeness say more about quality discipline than a sales conversation²³. And require measured values rather than pass/fail marks, because a certificate with no numbers cannot confirm tolerance compliance or track process drift²⁴.

If your application is regulated, confirm the supplier's certification scope statement rather than the logo on their website²³.

Design features to change before you release the drawing

Some cost is built into the geometry and cannot be recovered by relaxing a tolerance. Deep pockets, thin walls, undercuts, small internal radii and features spread across multiple faces are consistently identified as cost drivers, because they force long-reach tooling, slower cutting paths and extra setups²⁸²⁹³⁰.

Review these before release, when a change is still free.

• ✓Replace sharp internal corners with the largest radius the function allows — small radii force smaller tools and slower paths.
• ✓Limit deep, narrow pockets and cavities that require long-reach tooling with poor rigidity.
• ✓Consolidate features onto fewer faces so the part needs two setups instead of four.
• ✓Keep tight tolerances on critical surfaces only, and let non-functional dimensions carry the default class.
• ✓Specify fine surface finishes only where function demands them, not across the whole part.
• ✓Add ribs or bosses where the part needs reliable clamping, so the shop does not build custom workholding.
• ✓Standardize hole sizes, radii and wall thicknesses across the product family so tooling, fixtures and programs can be reused.

Lead times to quote and plan around

Lead time scales with what you ask the part to be. A one-off CNC prototype is typically quoted at 2–5 business days, while metal DMLS structural parts run 7–14 business days including post-processing and inspection²⁶.

Robotics parts with full CMM inspection and material certifications are stated at 7–15 days¹⁸. Metal prototyping generally spans 1 day to 3 weeks depending on method and complexity⁸.

The pattern is that the machining itself is rarely the long pole. Tooling fabrication, material procurement, queue time, secondary finishing, documentation, shipping and rework all add time, and for molded or cast parts tooling is often the biggest driver¹⁹.

That is why a part with a fine finish, a heat treat and a full inspection package quotes longer than the same geometry as-machined. When you plan a build, add the inspection and certification steps to the machining estimate rather than assuming they run in parallel.

The sources give no single lead-time figure for robot joint or harmonic-drive parts at production quantity, so ask the supplier for their queue depth on the specific process route rather than accepting a catalog number.

Where the sources disagree on the tolerance band

Three sources place the same critical features in different bands. The disagreement is not about whether joint interfaces need tight tolerances — all three agree they do — but about how tight, and none of them cites a governing robotics standard.

Treat the tightest cited value as a requirement to verify with your supplier rather than a class to copy onto the drawing.

What the sources do not establish

• No source gives a per-feature cost curve or cost-per-part model linking tolerance, material and volume for robot joints, harmonic drives or structural components.
• No governing robotics standard or authoritative technical source is cited for tolerance classes or surface finish requirements — the figures are supplier- or article-reported values under their own stated conditions.
• No numeric volume break-even threshold between CNC and casting or molding is provided.
• No source quantifies the lead-time impact of CMM, FAI or PPAP in days, nor the cost of AS9100D certification for a robotics supplier.
• No source gives required Ra values for general robot joint mating surfaces beyond the cleanroom and aerospace examples.
• No source addresses harmonic drive gear tooth profile tolerances specifically.
• No source gives cost per part by material for a specific robot component, only relative rankings and rate premiums.
• No source addresses titanium or stainless machinability in the context of robot structural components specifically.