Skip to content

Worst-Case Stack Calculation

An accumulation model — instantiates Tolerance Stack Management

Sums every contributor's tolerance in its most harmful direction to guarantee the fit holds even if all deviations align at their extremes — buying absolute assurance at the price of the most conservative, and often most expensive, budget.

Worst-Case Stack Calculation makes the most pessimistic assumption available: that every contributor simultaneously deviates to the edge of its tolerance, all in the direction that hurts the fit most. It simply adds the tolerances arithmetically, so the predicted stack is the largest any real assembly could ever produce. Its defining property is guarantee: if the worst-case sum fits inside the limit, then no assembly built to spec can ever fail the fit — full stop, no probability, no yield caveat. That certainty is bought at a steep price: because the aligned-extremes scenario is vanishingly rare in practice, the worst-case budget is far larger than reality demands, forcing tolerances tighter and costs higher than a statistical view would. It is the conservative pole of the accumulation choice — right exactly where a single failure is unacceptable.

Example

A cooling fan spins inside a shroud, and the blade tips must never touch the shroud wall — a contact would shed debris into the airflow, so the minimum radial clearance is a safety requirement, not a quality target. The clearance is set by a stack: shroud bore, fan hub bore, bearing runout, blade-length tolerance, and mounting eccentricity. Here a statistical estimate is the wrong tool — a "0.1% chance of contact" is unacceptable when contact is catastrophic. The Worst-Case Stack Calculation sums every contributor at its worst extreme, all pushing the blade toward the wall at once, and confirms the minimum clearance stays positive (say ≈0.15 mm) even in that impossible-but-bounding case. Because it also exposes that blade-length tolerance and bearing runout together account for most of the worst-case consumption, it points to exactly which two contributors to tighten if the guaranteed clearance needs to grow — telling the engineer not just whether it fits but which extremes to attack.

How it works

The method takes each contributor's tolerance, orients it in the direction that worsens the fit, and sums the absolute magnitudes — a pure arithmetic stack with no cancellation. The result is a hard bound: the maximum possible deviation of a conforming assembly. Comparing that bound against the fit limit yields a guarantee (fits always) or a refutation (can fail), with no probability attached. Because contributions add linearly, the arithmetic also ranks contributors directly by band size, exposing which extremes dominate the bound. What distinguishes it from every statistical sibling is that it makes no independence or distribution assumption at all — it is assumption-free precisely because it assumes the worst.

Tuning parameters

  • Directional assignment — orienting each contributor toward the harmful side of the fit. Get a sign wrong and the "worst case" isn't; this is the method's most error-prone step.
  • Contributor inclusion — whether marginal or rare contributors are folded in. Worst-case is already conservative, so padding it further with unlikely contributors compounds the over-design.
  • Bound scope — a single worst direction versus the worst over a multi-axis or multi-condition envelope (temperature, wear). Wider envelopes give a stronger guarantee but a larger, costlier bound.
  • Guarantee target — whether "fits" means zero interference or a required minimum clearance/margin held even at the extreme.
  • Selective application — applying worst-case only to the safety-critical portion of a stack while combining the rest statistically, to avoid taxing the whole assembly for one critical fit.

When it helps, and when it misleads

Its strength is unconditional assurance: it is the only accumulation model that guarantees conformance without probabilistic caveats, which is exactly what a safety-, interference-, or contract-critical fit requires when a single failure is intolerable. It also needs no independence or normality assumption to defend, so it cannot be undone by hidden correlation the way a statistical stack can — its conservatism is its robustness.[1]

It misleads when its conservatism is applied where it isn't warranted. Used on an ordinary, non-critical fit, it demands tolerances far tighter than reality needs, inflating cost and sometimes declaring "impossible" a design that would assemble fine 99.99% of the time — the over-design tax the statistical siblings exist to avoid. And a worst-case "pass" can breed complacency about the far larger typical margin, or be run backwards to justify gold-plated tolerances a supplier is already selling. The discipline is to reserve worst-case for genuinely critical fits, apply it selectively rather than blanket-wide, and use a statistical model where a rare, tolerable stack failure is an acceptable trade for realistic cost.

How it implements the components

  • accumulation_model — it is the conservative combination rule: the arithmetic sum of contributors at their harmful extremes, the maximum any conforming build can reach.
  • system_fit_limit — it compares that hard bound directly against the fit limit to certify guaranteed conformance, with no probability attached.
  • critical_contributor_map — because contributions add linearly, the sum ranks contributors by band size, exposing which extremes dominate the bound and where to tighten first.

It does not inventory the contributors or frame the requirement — that end-to-end setup is Tolerance Stack Analysis — and where a realistic (non-conservative) estimate is wanted, the statistical model is Root-Sum-Square Calculation.

Notes

Worst-case and root-sum-square are the two poles of the accumulation decision, and the choice between them is about consequences, not mathematics: worst-case where a single failure is intolerable and the cost of over-design is acceptable; statistical where a rare stack failure can be tolerated in exchange for realistic tolerances and cost. The common mistake is applying worst-case reflexively to a whole assembly when only one fit in it is truly critical — selective application is what keeps the guarantee without paying for it everywhere.

References

[1] Worst-case (or "arithmetic") tolerance stacking is the conservative bound recognized in geometric-dimensioning practice: it assumes all contributors reach their limits in the worst direction simultaneously, guaranteeing assembly at the cost of the tightest, most expensive tolerances — the deliberate opposite of a statistical stack. ↩