Blog
Tolerance Stack-Up Analysis: When Worst-Case vs RSS Actually Matters
When a designer hands you a stack of tolerances and asks “will this assemble?”, the honest answer depends on which math you use. Pick worst-case and you build a part that fits 100% of the time but costs more than it should. Pick Root-Sum-Square (RSS) and you save money — but only if you actually understand the statistics behind it.
This is the short, working engineer’s guide to deciding which stack-up method belongs on which print, with the trade-offs your shop floor will actually feel.
What a Tolerance Stack-Up Actually Solves
A stack-up answers a single question: given the individual feature tolerances on a chain of parts, what is the range of possible values for some critical dimension at assembly? That critical dimension might be a gap, a grip length, an alignment, a press depth, or the engagement of a snap fit.
Everything else — GD&T callouts, datum schemes, statistical process control — is upstream or downstream of that one question. If your stack-up math is wrong, the rest of the documentation is theater.
Worst-Case Stack-Up: The Conservative Default
The worst-case (WC) method sums all tolerances in their most adverse direction:
T_assembly_WC = T1 + T2 + T3 + ... + Tn
If every part lands on its worst-tolerated face simultaneously, this is the resulting variation in the critical dimension. It guarantees acceptance regardless of process capability.
Use worst-case when:
- The chain is short (typically 4 or fewer contributors)
- The consequence of an out-of-tolerance assembly is safety-critical, regulatory, or otherwise non-negotiable (medical devices, aerospace primary structure, pressure boundaries)
- Production volume is low enough that the cost of tighter tolerances is acceptable
- You don’t know the supplier’s process distributions yet
RSS Stack-Up: The Statistical Shortcut
Root-Sum-Square assumes each contributing tolerance is a normal distribution centered at nominal, with the tolerance equal to ±3σ. Total variation becomes:
T_assembly_RSS = √(T1² + T2² + T3² + ... + Tn²)
For a chain of five contributors each at ±0.005 in, worst-case predicts ±0.025 in. RSS predicts ±0.0112 in — less than half. That delta is real money. It is also the source of every “why is this part rejecting at incoming inspection?” phone call when the math gets misapplied.
Use RSS when:
- The chain has 5 or more contributors
- You have documented process capability (Cp / Cpk) data and parts are running centered with at least Cpk ≥ 1.33
- An occasional out-of-spec assembly can be reworked or scrapped at acceptable cost
- Volumes justify investment in SPC and supplier qualification
The Hybrid You’ll Actually Use Most Often
Real-world stacks are rarely pure. A common and defensible approach is the modified RSS, sometimes called the “benderized” method or a correction factor approach:
T_assembly = C⊂f · √(T1² + T2² + ... + Tn²)
Where C⊂f is a correction factor — typically 1.4 to 1.6 — that accounts for processes that aren’t actually centered, aren’t actually normal, and don’t actually run at 3σ. This gives you something between worst-case paranoia and RSS optimism, and it’s what most automotive and consumer electronics tolerance analyses are built on.
A Worked Example: Five-Part Press-Fit Stack
Consider a bearing pressed into a housing, with a shaft running through it. The critical dimension is the axial clearance between a snap ring groove and the bearing face. Five contributors:
| Feature | Tolerance (± in) |
|---|---|
| Housing bore depth | 0.004 |
| Bearing width | 0.002 |
| Shaft shoulder location | 0.005 |
| Snap ring groove location | 0.003 |
| Snap ring thickness | 0.002 |
Worst case: 0.004 + 0.002 + 0.005 + 0.003 + 0.002 = ±0.016 in
RSS: √(0.004² + 0.002² + 0.005² + 0.003² + 0.002²) = ±0.0077 in
Modified RSS (C⊂f = 1.5): 1.5 × 0.0077 = ±0.0115 in
If the design requires ±0.012 in to function and the customer is willing to live with a few PPM defects on uncentered processes, modified RSS clears it. Pure worst-case would force tighter tolerances on at least two features — and those tolerance bumps don’t come for free at the machine shop.
Three Mistakes That Cost the Most Money
1. Treating GD&T positional tolerances like linear tolerances. A ±0.005 position is a diametral zone, not a bilateral linear value. Project it onto your stack direction first, or you’ll over-state the contribution.
2. Ignoring form deviation. A bearing seat called out as ±0.001 in diameter can still have 0.0005 in of out-of-roundness, and that propagates to fit.
3. Using RSS on three-piece stacks. The Central Limit Theorem starts paying off around 5+ contributors. With three normally-distributed inputs you save almost nothing — and you take all the statistical risk.
When to Bring in an Outside Engineer
If you’re fielding rejections on parts that “should fit per the math,” the math is usually fine — the assumptions baked into it aren’t. We do tolerance analysis as part of our engineering services and our prototyping pipeline, often paired with first-article inspection data so the stack reflects the actual process capability, not the catalog tolerance.
For repeat issues, a one-hour review of a five-feature stack-up is typically enough to identify whether the problem is the design, the documentation, or the supplier. Get in touch if you’ve got a recurring assembly issue that the math says shouldn’t exist.
Quick Decision Table
| Situation | Method |
|---|---|
| ≤ 4 contributors, safety-critical | Worst case |
| 5+ contributors, capable processes | RSS |
| 5+ contributors, unknown capability | Modified RSS (C⊂f = 1.5) |
| Single high-PPM mating feature | Worst case, redesign for robustness |
Pick the method that matches your risk tolerance and the data you actually have. Then document the choice on the analysis itself — not in someone’s head — so when the next engineer touches this assembly in five years, they don’t have to re-litigate it from scratch.