Skip to content
← The Lab

Worst-case vs RSS: when each stack-up method lies to you

A practical guide to 1D tolerance stack-ups — how worst-case and RSS work, where each one misleads you, and how to pick the right one for your part.

Melek HakimGD&TWorkflow

Every assembly has a gap that matters: the clearance that keeps a gear from rubbing, the preload on a bearing, the flush line on a cover. A tolerance stack-up answers one question: across all the parts we will actually make, what range will that gap fall in?

There are two standard ways to answer it, worst-case and RSS. Both are “correct”; they answer slightly different questions. This article sets out the method, works through a complete example, and then shows when to trust each one.

The method

A 1D stack-up follows one closed loop of dimensions, from one face of the gap to the other. Each dimension in the loop either opens the gap when it grows or closes it. Add them with the right signs and you get the gap; combine their tolerances and you get the gap’s range.

Notation

Symbol Meaning Unit
did_i Nominal size of dimension ii mm
sis_i Direction: +1+1 if the dimension opens the gap when it grows, −1-1 if it closes it —
Ti+T_i^{+}, Ti−T_i^{-} Upper and lower tolerance, both taken as positive numbers mm
dˉi\bar{d}_i, tit_i Mean size and symmetric half-tolerance of dimension ii mm
GG The gap mm
Gmin⁡G_{\min} Required minimum gap mm

Direction is the part people get wrong. Walk the loop from one face of the gap to the other and the signs fall out naturally.

1. Re-centre every dimension

Tolerances are often one-sided (+0 / −0.12). Before stacking, convert each dimension to its mean size and a symmetric half-tolerance:

dˉi=di+Ti+−Ti−2(1)\bar{d}_i = d_i + \frac{T_i^{+} - T_i^{-}}{2} \tag{1} ti=Ti++Ti−2(2)t_i = \frac{T_i^{+} + T_i^{-}}{2} \tag{2}

2. Nominal and mean gap

The gap is the signed sum of the dimensions. Using nominal sizes gives the nominal gap; using mean sizes gives the gap the process will actually centre on:

Gnom=∑isi di(3)G_{\text{nom}} = \sum_i s_i, d_i \tag{3} Gˉ=∑isi dˉi(4)\bar{G} = \sum_i s_i, \bar{d}_i \tag{4}

Always stack from Gˉ\bar{G}. When a tolerance is one-sided, GnomG_{\text{nom}} is not the centre of the range.

3. Worst-case method

Assume every part lands at the limit that hurts you most, all at the same time. The half-tolerances simply add up:

tWC=∑iti(5)t_{\text{WC}} = \sum_i t_i \tag{5} Gˉ−tWC  ≤  G  ≤  Gˉ+tWC(6)\bar{G} - t_{\text{WC}} ;\le; G ;\le; \bar{G} + t_{\text{WC}} \tag{6}

4. RSS (root sum of squares) method

Treat each dimension as an independent, normally distributed variable whose tolerance equals ±3σi\pm 3\sigma_i, so σi=ti/3\sigma_i = t_i / 3. Variances add, so the half-tolerances combine as the square root of the sum of their squares:

tRSS=∑iti2(7)t_{\text{RSS}} = \sqrt{\sum_i t_i^{2}} \tag{7} Gˉ−tRSS  ≤  G  ≤  Gˉ+tRSS(8)\bar{G} - t_{\text{RSS}} ;\le; G ;\le; \bar{G} + t_{\text{RSS}} \tag{8}

Equation (8) covers about 99.73 % of assemblies. To estimate the share that misses a requirement Gmin⁡G_{\min}, use the gap’s standard deviation and the standard normal distribution Φ\Phi:

σG=tRSS3(9)\sigma_G = \frac{t_{\text{RSS}}}{3} \tag{9} P(G<Gmin⁡)=Φ ⁣(Gmin⁡−GˉσG)(10)P(G < G_{\min}) = \Phi!\left(\frac{G_{\min} - \bar{G}}{\sigma_G}\right) \tag{10}

5. Which dimension drives the variation?

Each dimension’s share of the gap’s variance tells you where tightening a tolerance pays off most:

ci=ti2∑jtj2(11)c_i = \frac{t_i^{2}}{\sum_j t_j^{2}} \tag{11}

Assumptions and limits

  • One direction only. This is a 1D stack: all dimensions are parallel to the gap. Angles, form errors and clearances in other directions need a 2D/3D analysis.
  • Worst-case assumes nothing beyond the drawing limits. Its result is guaranteed, but often far wider than reality.
  • RSS assumes each dimension is independent, normally distributed and centred on its mean, with its tolerance at ±3σ. If processes are off-centre or not capable, RSS is optimistic.
  • With a single dimension, worst-case and RSS give the same result. The difference grows with the number of contributors.

Worked example

A gear shaft assembly: a housing bore holds a bearing, a spacer and a gear hub. We need enough axial clearance at the end of the stack.

Given

Dimension Direction sis_i Nominal did_i (mm) Tolerance (mm)
Housing bore depth +1+1 50.00 ±0.10
Bearing width −1-1 16.00 +0 / −0.12
Spacer −1-1 16.00 ±0.05
Gear hub −1-1 16.00 ±0.05

Requirement: Gmin⁡=1.85G_{\min} = 1.85 mm.

Step 1: Re-centre the dimensions (Eq. 1 and 2)

Only the bearing has a one-sided tolerance:

dˉbearing=16+0−0.122=15.94 mm\bar{d}_{\text{bearing}} = 16 + \frac{0 - 0.12}{2} = 15.94\ \text{mm} tbearing=0+0.122=0.06 mmt_{\text{bearing}} = \frac{0 + 0.12}{2} = 0.06\ \text{mm}
Dimension dˉi\bar{d}_i (mm) tit_i (mm)
Housing bore depth 50.00 0.10
Bearing width 15.94 0.06
Spacer 16.00 0.05
Gear hub 16.00 0.05

Step 2: Nominal and mean gap (Eq. 3 and 4)

Gnom=50−16−16−16=2.00 mmG_{\text{nom}} = 50 - 16 - 16 - 16 = 2.00\ \text{mm} Gˉ=50−15.94−16−16=2.06 mm\bar{G} = 50 - 15.94 - 16 - 16 = 2.06\ \text{mm}

Step 3: Worst-case (Eq. 5 and 6)

tWC=0.10+0.06+0.05+0.05=0.26 mmt_{\text{WC}} = 0.10 + 0.06 + 0.05 + 0.05 = 0.26\ \text{mm} 1.80≤G≤2.32 mm1.80 \le G \le 2.32\ \text{mm}

The minimum, 1.80 mm, is below the 1.85 mm requirement: by worst-case, this design fails.

Step 4: RSS (Eq. 7 and 8)

tRSS=0.102+0.062+0.052+0.052=0.0186≈0.136 mm\begin{aligned} t_{\text{RSS}} &= \sqrt{0.10^2 + 0.06^2 + 0.05^2 + 0.05^2} \ &= \sqrt{0.0186} \approx 0.136\ \text{mm} \end{aligned} 1.92≤G≤2.20 mm1.92 \le G \le 2.20\ \text{mm}

Step 5: Share of assemblies below the requirement (Eq. 9 and 10)

σG=0.1363≈0.0455 mm\sigma_G = \frac{0.136}{3} \approx 0.0455\ \text{mm} P(G<1.85)=Φ ⁣(1.85−2.060.0455)=Φ(−4.6)≈1.9×10−6P(G < 1.85) = \Phi!\left(\frac{1.85 - 2.06}{0.0455}\right) = \Phi(-4.6) \approx 1.9 \times 10^{-6}

About 2 assemblies per million: by RSS, this design is fine.

Step 6: Contributions (Eq. 11)

The sum of squares from Step 4 is 0.0186 mm20.0186\ \text{mm}^2:

Dimension ti2t_i^2 (mm²) Contribution cic_i
Housing bore depth 0.0100 53.8 %
Bearing width 0.0036 19.4 %
Spacer 0.0025 13.4 %
Gear hub 0.0025 13.4 %

Result

Method Gap range Meets Gmin⁡=1.85G_{\min} = 1.85 mm?
Worst-case 1.80 … 2.32 mm No, short by 0.05 mm
RSS 1.92 … 2.20 mm Yes, about 2 ppm below the limit

The two methods disagree, which is exactly the situation where engineering judgement is needed (next section). If worst-case must pass, start with the housing bore: it drives over half of the variation. Tightening it from ±0.10 to ±0.07 brings tWCt_{\text{WC}} to 0.23 mm and the worst-case minimum to 1.83 mm. Opening up the requirement or reworking the bearing-to-spacer interface closes the rest. One tight tolerance is cheaper than four moderately tight ones.

Which method should you use?

Situation Use
Prototype, low volume, safety-critical Worst-case
Production, 4+ contributors, capable suppliers RSS
Production, but process capability unknown RSS with an inflation factor (rule of thumb: ×1.5), or worst-case
Worst-case fails, RSS passes Find the biggest contributor (Eq. 11) and fix that tolerance

Where worst-case lies: it assumes a combination of extremes that, on a well-controlled production line, essentially never happens. Designing to it on a high-volume part means paying for tolerances you don’t need.

Where RSS lies: it quietly assumes every process is centred and normal. A supplier who runs every spacer at the top of its tolerance to “stay safe” breaks that assumption, and your 2 ppm becomes a few percent. Short runs, a single lot of material, or a process that drifts all make RSS optimistic.

Checklist

  • Did you walk a single closed loop from one face of the gap to the other?
  • Are asymmetric tolerances re-centred to their mean (Eq. 1 and 2)?
  • Did you include the parts nobody thinks about: washers, gaskets, plating thickness, retaining-ring grooves?
  • If you used RSS, do you actually know your suppliers’ processes are centred?
  • Have you looked at contributions, not just the result?

References

  • Fischer, B. R., Mechanical Tolerance Stackup and Analysis, 2nd ed., CRC Press, 2011.
  • ASME Y14.5-2018, Dimensioning and Tolerancing.

Run this exact example, and your own stacks, in the tolerance stack-up calculator. It loads this worked example by default and shows worst-case and RSS side by side, the share of assemblies out of spec, and which dimension drives the variation.