Generating Fasteners, Gears & Bearings Correctly (Not Just Plausibly)

Standard mechanical components — bolts, bearings, gears — should never be freely "imagined" by an AI model. They should be selected from real catalog data, because these parts only work if they match an actual manufactured product someone can buy, and the tolerance for error is essentially zero: a bolt that's 0.3 mm off on thread pitch isn't "close enough," it simply doesn't thread into its mating nut.

Why standard parts are a different problem than custom geometry

When a model generates a custom bracket, there's a reasonable range of "correct" — as long as it meets the functional dimensions and clears manufacturing constraints, minor variations in fillet size or wall routing don't matter. Standard parts don't have that flexibility. A bearing has to match a real bore/OD/width combination from an actual manufacturer's catalog, or it's not a real bearing — it's a shape that merely resembles one. This is why the right approach for standard components is fundamentally different from freeform generation: it's a matching/lookup problem ("which real part does this description refer to"), not a design problem ("what shape would work here").

Fasteners: what "correct" actually requires

A correctly specified bolt or screw needs, at minimum:

• Thread designation matching a real ISO standard (see our ISO metric thread sizes guide for the coarse/fine pitch reference) — not an approximated pitch that's close but doesn't match any cut tap or die.
• Head geometry matching a real standard (ISO 4762 socket head cap screw, ISO 4014 hex head bolt, etc.) — head height, wrench flat size, and under-head fillet all follow standardized dimensions that a "plausible-looking" generated head will usually get slightly wrong.
• Property class appropriate to the load (see the thread guide above for what 8.8/10.9/12.9 actually mean) — this is a specification decision, not a geometry decision, but it needs to be present and consistent with the thread size.
• Length and thread engagement appropriate to the joint — a bolt that's geometrically valid in isolation can still be wrong if it's too short to develop full thread engagement in the mating material.

Gears: the parameters that must be internally consistent

A spur gear's geometry is fully determined by a small set of parameters, and they have to agree with each other — you can't set them independently:

• Module (m): the fundamental size parameter, in millimeters. Module is defined as the pitch diameter divided by the number of teeth (m = PD / N). This single number, combined with tooth count, determines the entire tooth profile.
• Number of teeth (N)
• Pitch diameter: PD = m × N — this is a direct consequence of the module definition, not an independent choice. A generated gear where pitch diameter doesn't equal module × tooth count is simply wrong, not "approximately right."
• Pressure angle: almost universally 20° in modern designs (14.5° was the older standard, still found in legacy equipment but rarely specified for new designs).
• Full tooth depth: conventionally 2.25 × module for a standard full-depth tooth form.

The critical check for any generated gear: two gears only mesh correctly if they share the same module and pressure angle. A generated gear pair with mismatched modules will look fine as individual parts and simply not mesh — this is a failure mode that's invisible until assembly, which is exactly why it needs to be checked automatically rather than left to visual inspection.

Bearings: matching a real catalog part, not a shape

A bearing needs to match a real manufacturer's bore/outside-diameter/width combination — these three dimensions aren't independently chosen, they come as a fixed set for any real bearing part number. Beyond fitting the shape:

• Bearing type matters as much as size. Deep groove ball bearings handle radial loads well and limited axial load; tapered roller bearings handle combined radial and axial loads with higher stiffness but need to be used in opposed pairs and preloaded correctly. Specifying the wrong type for the load case is a functional failure even if the dimensions are perfect.
• Fit tolerances matter more than most other mechanical interfaces. The shaft and housing that a bearing sits in need specific ISO 286 fit classes (see our tolerances in CAD guide for how hole/shaft fit pairs work) because bearing performance and life depend heavily on getting the interference or clearance right — too loose and the inner/outer ring can spin in its seat; too tight and you can preload the bearing beyond its rating.
• ABEC / precision class (or the equivalent ISO tolerance class) should match the application — general machinery rarely needs anything beyond the standard class, and specifying an unnecessarily high precision class adds cost without adding function.

A practical checklist before trusting a generated standard part

1. Does the thread/gear/bearing designation correspond to an actual, purchasable standard part number — not just a plausible-looking dimension set?
2. For fasteners: does the property class match the stated or implied load requirement?
3. For gears: do module, tooth count, and pitch diameter satisfy PD = m × N, and does the mating gear share the same module and pressure angle?
4. For bearings: does the bore/OD/width combination match a real catalog entry, and is the bearing type (ball vs. roller, radial vs. angular contact) appropriate for the actual load direction?
5. If the answer to any of the above is "unclear," treat the output as a starting point for engineering review, not a final specification.

Why this shapes how we handle standard parts

This is the core reason standard components in our pipeline are pulled from real, cataloged geometry rather than generated freehand by a language model guessing at plausible dimensions — the failure mode for a "close but wrong" bolt or bearing isn't a minor cosmetic issue, it's a part that doesn't function. Custom geometry (brackets, housings, one-off features) is a different problem with a different tolerance for iteration, covered in our standard parts vs. custom geometry cost comparison.

Sources: KHK Gears — Basic Gear Terminology and Calculation · Tameson — Ball vs. Roller Bearings · National Precision Bearings — ABEC Tolerance Classes.