How to use the tolerance stack-up calculator
- Sketch the loop. Start at one face of the gap you care about and walk through the parts to the other face. Every dimension you cross is one row.
- Enter each dimension: nominal, plus and minus tolerance, and its direction — “+” if making the dimension larger opens the gap, “−” if it closes it. Paste the whole table from a spreadsheet or upload a file if you prefer.
- Optional: the Cp you expect from each process (1.00 = tolerance is ±3σ), a uniform distribution for dimensions with no central tendency, and the required minimum and maximum gap.
- Read the worst-case and RSS limits, the contribution chart and the predicted reject rate. Print or save the report.
Worst case, RSS and modified RSS
| Method | Tolerance of the gap | Assumes | Use for |
|---|---|---|---|
| Worst case | TWC = Σ tᵢ | Every dimension at its limit at once | Bounds within the stated linear model; independent engineering review still required |
| RSS (statistical) | TRSS = 3·√(Σ σᵢ²) = √(Σ tᵢ²) when tᵢ = 3σᵢ | Independent, centred, normal processes | Volume production, four or more contributors |
| Modified RSS | TMRSS = Cf · TRSS | RSS inflated by a correction factor (Bender’s 1.5 is common) | Processes that drift or are not well centred |
The gap itself is the signed sum of the dimensions, gap = Σ dᵢ·xᵢ with dᵢ = +1 or −1. The worst-case method adds the tolerances arithmetically, which bounds the stated linear model when every input lies within its tolerance; it does not certify the physical assembly. The root-sum-square method adds variances instead: σ²gap = Σ σᵢ², and reports mean ± 3σgap, the range that contains 99.73% of assemblies under the normal model.
From tolerance to standard deviation
- Normal: σᵢ = tᵢ ⁄ (3·Cpᵢ). With the default Cp = 1 the tolerance is ±3σ and the result is the textbook √Σt². A process with Cp = 1.33 contributes less variation; one with Cp = 0.8 contributes more.
- Uniform: σᵢ = tᵢ ⁄ √3 — every value in the tolerance zone equally likely, a conservative choice for purchased parts or tool-wear-driven dimensions.
Asymmetric tolerances
A dimension such as 15.00 +0 ⁄ −0.12 is converted to the equivalent symmetric form 14.94 ± 0.06: mean = nominal + (plus − minus) ⁄ 2 and t = (plus + minus) ⁄ 2. The tool shows every conversion. This is why the “mean gap” can differ from the gap computed from drawing nominals.
Contribution and reject rate
The contribution of a dimension is its share of the gap variance, σᵢ² ⁄ Σσ². Because variances add as squares, the largest tolerance dominates: a dimension with twice the tolerance of another contributes four times as much. Tighten the top contributor first and relax the small ones.
With a required minimum and maximum gap, the normal model gives the predicted fraction out of limits, P = Φ((min − μ) ⁄ σ) + 1 − Φ((max − μ) ⁄ σ), and a predicted Cpk for the assembly — the same index the process capability calculator reports for measured data.
Monte Carlo simulation
The simulation builds N virtual assemblies by drawing every dimension at random from its distribution and summing them. It needs no normality assumption for the gap, so it is the reference when uniform dimensions dominate. The generator is seeded: the same seed and inputs always reproduce exactly the same result, so a report can be re-created. A 95% Wilson score interval describes sampling uncertainty in the simulated reject fraction, conditional on the selected input model. Zero simulated rejects does not establish zero risk; this interval does not include model error. See the NIST proportion interval formulas.
Worked example
Stack: housing depth 50.00 ± 0.10 (+), bearing 15.00 +0 ⁄ −0.12 (−), spacer 20.00 ± 0.05 (−), retaining ring 1.60 +0 ⁄ −0.06 (−). Four dimensions, all assumed normal with Cp = 1.
Centre the asymmetric ones: bearing 14.94 ± 0.06, ring 1.57 ± 0.03.
Mean gap = 50.00 − 14.94 − 20.00 − 1.57 = 13.49, against 13.40 from the drawing nominals. (The built-in sample adds a fifth dimension, 12.90 ± 0.08, which closes the gap to 0.59.)
Worst case T = 0.10 + 0.06 + 0.05 + 0.03 = 0.24 → 13.25 to 13.73.
RSS T = √(0.10² + 0.06² + 0.05² + 0.03²) = √0.0170 = 0.130 → 13.36 to 13.62 — about 54% of the worst-case value, with the housing alone contributing 0.01 ⁄ 0.017 = 59% of the variance.
Assumptions and limitations
- One-dimensional, linear stacks: every sensitivity is +1 or −1. Angles, levers, geometric tolerances with bonus, and 2-D/3-D effects need a vector loop or CAT software.
- RSS assumes independent dimensions and processes centred in the tolerance zone. Mean shifts are the usual reason real assemblies are worse than RSS predicts — hence the modified-RSS factor.
- Normal dimensions are simulated without truncation, so a few simulated parts fall outside their own tolerance (0.27% at Cp = 1), under the assumed centred normal model, not as a guaranteed description of a real process.
- The assumed Cp is only as good as your process knowledge; verify it with measured data, and make sure the gage is adequate with a gage R&R study.
Methods follow the standard treatments in Creveling, Tolerance Design, and Drake, Dimensioning and Tolerancing Handbook.
Frequently asked questions
What is the difference between worst-case and RSS tolerance analysis?
Worst-case analysis adds all tolerances arithmetically and bounds the specified linear model if every part is in tolerance; it does not certify assembly performance. RSS (root sum square) analysis adds them statistically as the square root of the sum of squares, assuming independent, centred, normally distributed dimensions, and predicts the range containing about 99.73% of assemblies. Its usefulness depends on the stated distributions, independence and centring, not merely the number of parts.
How do you calculate RSS tolerance stack-up?
Convert each tolerance to equal bilateral form, square each tolerance, add the squares and take the square root: T = √(t1² + t2² + … + tn²). The gap then ranges from the mean gap minus T to the mean gap plus T, assuming each tolerance represents ±3 standard deviations.
What is modified RSS?
Modified RSS multiplies the RSS result by a correction factor, commonly 1.5 (Bender’s factor), to allow for processes that are not perfectly centred or normal. It is a user-selected inflation, not a guaranteed probability bound, and it can exceed the worst-case range.
How do I handle asymmetric or unilateral tolerances?
Convert them to an equivalent symmetric tolerance about the middle of the tolerance zone. For 15.00 +0/−0.12 the mean is 14.94 and the tolerance is ±0.06. The calculator does this automatically and lists every conversion.
When should I use Monte Carlo simulation for tolerance analysis?
Use it when the input distributions are not normal (for example uniform), when there are only two or three contributors so the central limit theorem does not apply well, or when you want to estimate the reject rate directly rather than rely on the normal model.
What does the contribution percentage mean?
It is the share of the total gap variance caused by one dimension: its variance divided by the sum of all variances. It shows which tolerance to tighten for the greatest reduction in assembly variation.