What is RSS tolerance analysis?
Short answer
RSS (root sum square) tolerance analysis is a statistical way to stack up part tolerances. Instead of adding every tolerance, as worst-case analysis does, it takes the square root of the sum of their squares. The result is smaller because it assumes variations are independent and rarely all sit at their limits together, so it predicts the likely spread of an assembly.
RSS stands for root sum square. It is one of the two standard ways to combine the tolerances in a tolerance stack-up, the chain of dimensions that decides whether parts fit together. The other is worst-case analysis, which assumes every part is at its limit in the same direction at the same time.
What are the worst-case and RSS tolerance formulas?
For a linear stack of n dimensions, each with a bilateral tolerance t (the plus-or-minus value):
- Worst case:
T_wc = |t1| + |t2| + ... + |tn| - RSS:
T_rss = SQRT(t1^2 + t2^2 + ... + tn^2)
Whether a dimension adds to or subtracts from the gap changes the nominal gap, but not the tolerance total, because absolute values and squares remove the sign.
How do I calculate the gap for three parts stacked in a housing?
A housing has an internal length of 50.00 mm. Three parts are stacked inside it, and you want to know the remaining gap.
| Dimension | Nominal (mm) | Tolerance (± mm) | Direction |
|---|---|---|---|
| Housing length | 50.00 | 0.15 | gap grows |
| Part A | 20.00 | 0.10 | gap shrinks |
| Part B | 18.00 | 0.08 | gap shrinks |
| Part C | 8.00 | 0.05 | gap shrinks |
The nominal gap is 50.00 - (20.00 + 18.00 + 8.00) = 4.00 mm.
- Worst case: 0.15 + 0.10 + 0.08 + 0.05 = ±0.38 mm, so the gap runs from 3.62 to 4.38 mm.
- RSS: the squares are 0.0225, 0.0100, 0.0064 and 0.0025, which sum to 0.0414. The square root is ±0.2035 mm, so the gap runs from about 3.80 to 4.20 mm.
The RSS tolerance is about 46 percent smaller than the worst-case tolerance. In a spreadsheet, with the tolerances in C2:C5, the two results are =SUM(C2:C5) and =SQRT(SUMSQ(C2:C5)).
The gap between the two methods widens as the stack grows. Ten parts at ±0.10 mm each give ±1.00 mm worst case but only ±0.316 mm by RSS, because worst-case grows with n while RSS grows with the square root of n.
What assumptions must hold for RSS tolerance analysis?
RSS is only valid when its assumptions hold:
- Each dimension varies independently of the others.
- Each distribution is centered on its nominal value, or any shift is accounted for separately.
- Each stated tolerance represents the same number of standard deviations, commonly ±3 sigma. If the parts are normally distributed, about 99.73 percent of parts then fall inside ±3 sigma, and the RSS result is the matching ±3 sigma limit for the assembly.
RSS is a prediction, not a guarantee. A small share of assemblies can still fall outside the RSS range. Some teams widen the result with a multiplier to allow for mean shift (a factor of 1.5 is often quoted, associated with the engineer Bender), and angled or non-linear dimensions need a sensitivity coefficient, so the formula becomes SQRT(SUM((a_i * t_i)^2)).
When should I use RSS instead of worst case?
Use worst case when every assembly must fit, for example in safety-critical parts, low volumes or anything that cannot be adjusted or reworked. Use RSS when production volumes are high, the processes are capable and stable, and a small, quantified rate of out-of-range assemblies is acceptable in exchange for looser, cheaper tolerances on the parts.
Reporting both numbers side by side is common, since the difference shows how much tolerance budget the statistical assumption buys.
To build the stack in a spreadsheet, see the tolerance stack-up analysis template, which covers worst case and RSS. The blog post on worst-case versus RSS stack-ups goes further into the comparison.