Worst-Case vs RSS Tolerance Analysis: Formulas, Worked Example, and When to Use Each
Choosing between Worst-Case and Root-Sum-Square (RSS) is one of the highest-leverage decisions in mechanical tolerance analysis. Worst-Case guarantees 100% interchangeability at the cost of tight, expensive machining tolerances. RSS models real statistical variation to lower manufacturing cost—provided your production processes actually stay in control.
How Worst-Case (Arithmetic) Stack-Up Works
Worst-Case analysis assumes every part in an assembly lands at its simultaneous extreme limit. For a 1D chain of independent dimensions with signed upper deviations u_i and lower deviations l_i, the stack nominal is the algebraic sum of part nominals, while the extreme bounds sum every upper and lower deviation linearly: Max = Σ(N_i + u_i) and Min = Σ(N_i + l_i).
Because no real part can exceed its drawing limit if inspection is 100% effective, a passing Worst-Case stack guarantees zero assembly failures. The penalty scales linearly with part count: once a loop exceeds four or five components, holding Worst-Case clearance forces individual tolerances down to levels that require grinding, EDM, or selective assembly.
How Root-Sum-Square (RSS) Models Statistical Variation
RSS assumes each dimension follows an independent normal distribution centered on the midpoint of its tolerance zone, with the full tolerance band representing ±3σ (99.73% of produced parts). Instead of adding half-bands t_i = (u_i − l_i) / 2 linearly, RSS combines their variances in quadrature: T_RSS = 2 × √(Σ t_i²).
When tolerances are asymmetric (such as unilateral ISO 286 limits), the statistical center shifts away from the nominal to the sum of tolerance midpoints: μ_stack = Σ(N_i + (u_i + l_i) / 2). The resulting statistical bounds are μ_stack ± √(Σ t_i²). A crucial insight of RSS is variance dominance: because each contributor is squared, a single loose tolerance often drives more than half of total assembly variation.
Worked Example: 3-Part Housing, Spacer, and Bearing Stack
Consider a classic 1D linear chain of three components (the default example loaded in the Precise Fit Pro calculator): Housing = 50.00 ± 0.10 mm, Spacer = 20.00 ± 0.05 mm, and Bearing = 30.00 ± 0.08 mm.
Nominal stack = 50 + 20 + 30 = 100.0000 mm. Under Worst-Case, the half-band is 0.10 + 0.05 + 0.08 = 0.2300 mm (total tolerance 0.4600 mm), giving limits of 99.7700 mm to 100.2300 mm.
Under RSS, the half-band is √(0.10² + 0.05² + 0.08²) = √(0.0189) = 0.1375 mm (total tolerance 0.2750 mm), tightening the predicted range to 99.8625 mm – 100.1375 mm. Looking at variance share (t_i² / Σ t_j²), the Housing accounts for 52.91% of total variation, the Bearing for 33.86%, and the Spacer for only 13.23%. Tightening the Spacer would waste budget; controlling the Housing cuts spread the fastest.
Practical Selection Rules for Design Reviews
- Use Worst-Case for safety-critical mechanisms, low-volume prototypes (<50 units), or single-build tooling where a single non-fitting assembly is unacceptable.
- Use RSS for medium-to-high production volumes when suppliers maintain stable, centered processes (Cp ≥ 1.0–1.33).
- Check variance contribution first before tightening tolerances—reducing the top contributor yields the biggest drop in stack spread per euro spent.
- Pair Worst-Case and RSS with Monte Carlo simulation to inspect tail behavior and estimate actual PPM defect rates.