Tolerance Stack-Up in CNC Machined Assemblies: Bounding Cumulative Error with GD&T

Control the assembly condition, not each part dimension

A dimension that looks acceptable on its own drawing can still push the assembly out of position once it interacts with other features.

A small variation on one surface, hole or locating feature has little effect by itself; when several dimensions feed the same assembly requirement, their combined effect shows up as position, clearance or alignment error in the finished product¹⁰. That is why reviewing each dimension separately does not predict how the assembly will perform¹⁰.

Stack-up analysis exists to make that combined effect visible before production. It identifies which dimensions have the greatest influence on the assembly condition and lets you assign practical tolerances that satisfy both the assembly requirement and what the process can actually hold¹⁰.

The result is compared against the required assembly limits; if the summed variation exceeds the allowed range, you adjust the critical dimensions or rework the tolerance scheme³¹. You should resist the instinct to tighten every feature.

Control belongs on functional surfaces, locating features and the dimensions that directly affect assembly, while non-contact surfaces, outer profiles and clearance areas can carry standard machining tolerances without affecting performance⁹. A tight tolerance on a non-critical feature adds machining and inspection work and buys nothing at the assembly⁹.

Start your own review by listing the assembly conditions that matter — shaft movement, bearing location, gear spacing, hole alignment, mating clearance — and tracing which part dimensions feed each one⁹.

Worst-case or statistical summation: what the choice depends on

The three common treatments differ in what they assume about your contributors. Arithmetic worst-case summation adds the limits and guarantees 100% fit, at the price of being over-conservative²¹.

RSS is used when all parts are normally distributed and sums variances to give a realistic spread²¹.

Monte Carlo simulation is engaged for non-normal distributions and simulates thousands of assemblies²¹; applied to positional tolerances it models feature location distributions and captures bonus-tolerance behaviour and geometric interactions, typically yielding tighter, realistic predictions than linearising the callouts²².

For positional tolerances specifically, the conservative route converts the callout to a linear worst-case contribution and adds it like any other dimension — safe, but often over-conservative²².

The statistical route needs a distribution assumption: for a circular positional tolerance at MMC, one practitioner approximation treats it as a 2D normal with sigma derived from P/6, or uses supplier data instead²².

That sigma value is an assumption you are choosing, not a measured property of your process, so it is worth confirming against your own data before you rely on it. Neither source establishes which method is required for a given assembly, and neither gives a numeric criterion for choosing between them^21,22.

The decision therefore rests on evidence you hold: whether your contributors are actually normal, how much risk of a non-fitting assembly you can accept, and whether the geometric interactions in your stack are linear enough for RSS to mean anything.

If you cannot support the normality assumption, worst-case is the defensible default; if you can, RSS or Monte Carlo will let you loosen tolerances that worst-case would force tight. Ask your supplier for the distribution data behind the capability figures they quote, and state in the drawing package which method the stack was run with.

Anchor the stack to a functional datum reference frame

The datum reference frame is the coordinate system the stack is measured from, and it changes the effective path of the stack because GD&T locational controls reference the DRF rather than chaining feature to feature²⁵.

Choose datums that reflect how the part is actually assembled and fixtured for inspection and manufacture: the primary datum should be the surface assembly references first, and the secondary and tertiary datums should reflect how the part is constrained next, such as mating surfaces or locating pins²⁵.

Precedence is set by the sequence in the feature control frame, read left to right as primary, secondary and tertiary — the letters themselves do not imply order¹². The primary datum is the first feature contacted, requiring at least three points of contact; the secondary at least two; the tertiary at least one¹².

Note that form controls such as flatness or straightness carry no datum reference at all, while location controls such as position typically require one¹².

A worked comparison makes the effect concrete: two holes held by positional tolerances referenced to a robust DRF stack more predictably than two holes dimensioned with offsets that reference different local features²⁵.

The second scheme chains through local features that do not correspond to how the parts meet, so the stack no longer describes the assembly. A bad DRF gives you artificially tight or meaningless stacks²⁵, and the failure mode is quiet — the numbers add up, they just do not describe your product.

Verify datum choices with your manufacturing and fixture engineers before releasing the drawing²⁵, and ask your supplier which surface they will actually locate on for the first operation.

Which GD&T controls bound the stack, and where size alone is not enough

Size dimensions control size; they do not control the relationships between features, and most assembly problems come from location, orientation or alignment²⁸. A hole can be the correct diameter and still fail assembly if its centre position falls outside the allowed location tolerance²⁸.

GD&T closes that gap by defining feature relationships through datums and geometric tolerance zones²⁸. Compared with coordinate tolerancing, the difference is structural.

Coordinate tolerancing uses square or rectangular zones that allow independent X and Y variation, so diagonal position is controlled more loosely, and it gives no direct control over form or orientation²⁶.

GD&T uses circular, cylindrical or profile-based zones tied to datums, controls form, orientation and runout explicitly, and aligns inspection to the functional datums used in the real assembly²⁶. That is why a circular position zone better represents the actual clearance a fastener needs than a rectangular coordinate box does¹⁷.

Material condition modifiers are the lever that loosens a stack without loosening the design. Position tolerance with MMC or LMC modifiers minimises cumulative tolerance accumulation across multiple mating features, and bonus tolerance under MMC reduces the risk of the stack exceeding functional assembly limits¹⁷.

The mechanism is size-dependent: a positional tolerance at MMC grants bonus tolerance equal to the difference between the actual feature size and the MMC size, so an oversized hole earns extra positional allowance at assembly without any change to the drawing²².

LMC works the opposite way and is used where minimum material matters, such as wall thickness²². The same latitude gives the shop more room in feature size selection and reduces part rejection¹⁷.

Where a pattern is called out once as a composite feature control frame, the upper segment locates the pattern and the lower segment refines orientation, with basic dimensions between features still holding⁵.

No source in this set quantifies how much stack reduction a given bonus tolerance delivers for a specific feature, so treat the benefit as directional and confirm it in your own stack.

When you write the drawing, ask yourself for each callout whether a size dimension plus a general block tolerance would actually control the assembly condition — if not, that feature needs a geometric control with a datum reference.

Check the per-part tolerances against what the process can hold

Before you release a stack, confirm that each tolerance feeding it is achievable by the process and material you have chosen. The figures below are capability envelopes that depend on feature size, fixturing, material and tooling — starting points for a conversation with your supplier, not guarantees²⁹.

Note that the hole-sizing method changes the stack contribution materially: drilling and reaming are not interchangeable inputs to the same chain²⁹. The single ±0.01 mm reference for turn-mill parts is conditioned on geometry, material, datum relationships, machining route and inspection requirements, so it is not a blanket capability³⁰.

Neither source establishes an IT grade or that any specific tolerance is achievable for a given part without process and material qualification^29,30.

Absorb variation in the design before you tolerance it away

Some variation is cheaper to absorb than to control. Larger clearance holes, compliant elements such as gaskets, O-rings and shim spacers, and stable datums let you loosen tolerances on the rigid parts without violating the assembly requirement²⁴.

The compliant element compresses during assembly and takes up the stack error that would otherwise have to be machined out of the metal²⁴.

A worked instance from one source: increasing a clearance hole diameter from 4.2 mm to 4.5 mm allowed the position tolerance of the matching threaded hole to be loosened, saving machining time²⁴. Datum design is part of the same move.

A primary datum that is a large, flat, easily machinable surface, directly interfacing with the mating part, gives you a stable reference; secondary and tertiary datums placed close to critical features keep the tolerance chain short²⁴. Long chains accumulate more contributors, and every contributor you add widens the stack.

Setup count is the other lever. Single-setup machining lets multiple faces and features be cut relative to one primary datum, which reduces the alignment shifts that come from repeated re-fixturing, and fewer setups help control stack-up across holes, bores and mating faces²⁰.

Multi-setup handoffs are themselves a source of dimensional deviation²⁰. One machine-tool supplier reports that single-setup turn-mill processing reduces fixtures and handling steps, which lowers stack-up error³⁴ — that is the supplier's position on its own equipment, not a general rule.

When positional relationships across several faces matter more than the tolerance on any individual feature, that is a common reason to look at five-axis work²⁰.

Ask your supplier how many setups the part needs and which datum each operation locates from; if the answer changes between operations, the stack has a contributor you did not model.

Specify how the stack will be verified before the drawing is released

A stack is only as good as the inspection that confirms it. Functional gauges and CMM software give rapid pass or fail determination for position tolerance without complex manual calculations¹⁷, and where a feature directly influences alignment or fit, the machining process and inspection method are matched to that requirement³¹.

No source in this set states measurement uncertainty values or a verification acceptance criterion, so you have to set those yourself. Work through these before release:

• ✓Confirm every geometric callout on the drawing can be measured by a method the supplier actually has — a functional gauge or CMM routine, not a hand tool.
• ✓For position tolerances, decide whether a functional gauge built to the worst-case mating geometry or a CMM routine will verify the feature, and say so on the drawing.
• ✓Identify the features that directly influence alignment or fit and confirm the inspection method is matched to them.
• ✓Compare the final stack result against the required assembly limits and record the comparison, so a later change to any contributor can be re-checked.
• ✓Ask the supplier to review the drawing against the functional assembly requirements and flag tolerances that are unnecessarily restrictive.
• ✓Set your own measurement uncertainty allowance for each verification method, since no source provides one.

Where the sources disagree, and what to verify

Two of the inputs your stack depends on are not settled in this evidence set. The capability figures come in two different forms — per-process ranges versus a single conditioned reference — and the method sources describe when statistical treatment is valid without agreeing on when it is required.

Keep both sides visible and verify against your own process data rather than treating either figure as a rule.

What the sources do not establish

• No numeric assembly-level functional limits — maximum allowable clearance, interference, misalignment or float — for any specific mating interface.
• No specific per-part flatness, perpendicularity, position or profile values for a named part feeding a stack.
• No IT grade values or authoritative process-capability tables for milling, turning, drilling or boring by material.
• No measurement uncertainty values for CMM, functional gage, fixture or assembly check.
• No cost-per-tolerance-grade, tooling or fixture cost, volume threshold, lead time, or scrap and rework rate data.
• No worked stack-up example with actual numbers showing how GD&T controls reduce cumulative error in a CNC machined assembly.
• No governing standard (for example ASME Y14.5 or ISO 1101) cited in the evidence to establish mandatory GD&T practice or limits.
• No source establishes which stack method — worst-case, RSS or Monte Carlo — is required for a given assembly, or a numeric criterion for choosing between them.