Numlytix — Tolerance Stackup Methods

Tolerance Analysis

Tolerance Stackup Methods

Predict whether an assembly will meet functional requirements given the dimensional variation of its individual parts. Three core methods: Worst Case, RSS, and Monte Carlo — each with different risk assumptions.

🔩 What is a Tolerance Stackup?

When parts are assembled, their dimensional variations accumulate. A tolerance stackup analysis answers: will the gap, clearance, or interference in my assembly always be acceptable — for every part within tolerance?

Assembly Gap Example

Gap = Housing − (A + B + C)
Each part has its own tolerance ±t. The question: is Gap ≥ 0 always?

Three Analysis Methods

1. Worst Case (WC)

All parts simultaneously at worst extreme. Conservative — guarantees 100% yield. Often over-constrains tolerances.

2. RSS (Statistical)

Root Sum of Squares. Assumes independent normal distributions. Predicts realistic assembly yield (~99.73% at 3σ).

3. Monte Carlo

Random sampling simulation. Handles any distribution shape, correlations, and non-linear stackups. Most realistic.

🧭 Which Method to Use?

Decision Guide

→Safety-critical, zero-defect required → Worst Case (100% compliance guaranteed)

→High-volume production, normal distributions → RSS (realistic yield prediction)

→Non-normal processes, skewed distributions, or correlated parts → Monte Carlo

→First pass / quick check → Run WC first, then RSS to see if tolerances can be relaxed

Tolerance Analysis

Method Comparison

Side-by-side summary of Worst Case, RSS, and Monte Carlo — assumptions, when to use each, and the typical difference in predicted tolerance.

📊 Comparison Table

Property	Worst Case	RSS	Monte Carlo
Predicted gap tolerance	Smallest (tightest)	Moderate (~40–60% looser)	Most realistic
Yield guarantee	100% (by definition)	~99.73% at 3σ	User-defined PPM
Assumes normal dist.	No — any dist.	Yes (required)	No — any dist.
Handles non-linear stackup	Linear only	Linear only	Yes
Part count sensitivity	Worsens rapidly with n	Scales as √n	Independent of n
Computation	Instant	Instant	Requires simulation
Typical use	Safety-critical, low volume	High volume production	Complex assemblies, R&D
Result	Tolerance range	Statistical tolerance	Distribution + PPM

📐 Gap Tolerance Formula Comparison

Worst Case: T_gap = Σ |tᵢ| (sum of ALL individual tolerances)
RSS: T_gap = √(Σ tᵢ²) = √(t₁² + t₂² + ... + tₙ²)
Monte Carlo: Gap = Σ sᵢ · Nᵢ(μᵢ, σᵢ) [simulated, 10k+ runs]
sᵢ = direction coefficient (+1 or −1) | tᵢ = bilateral tolerance (half-range) | σᵢ = tᵢ/3 for 3σ process

⚠ Key Insight

For n equal tolerances t: WC gives T=n·t but RSS gives T=√n·t. For n=9 parts with t=0.1mm: WC→0.9mm, RSS→0.3mm. RSS allows 3× looser tolerances per part for the same assembly performance — but only if the normal distribution assumption holds.

Worst Case

Worst Case Stackup Calculator

Add parts to the loop. Each part contributes its full tolerance in the stackup direction. The gap must remain positive at both extremes to guarantee 100% assembly yield.

🧭 Direction Convention — What is +1 vs −1?

In a tolerance loop, you trace a path through the assembly from one side of the gap to the other. Each dimension either opens (increases) or closes (decreases) the gap.

+1 Opens the gap ↔

Traversing in the same direction as the gap arrow.
Typically the housing / outer envelope.
If this part grows, the gap gets larger.

−1 Closes the gap ←→

Traversing back through a part (opposing the gap arrow).
Typically internal components stacked inside.
If this part grows, the gap gets smaller.

Rule of thumb: Start at one wall of the gap. Every part you cross going right is +1 (adds space). Every part you cross going left (backtracking) is −1 (takes space). Gap = Σ (+1 parts) − Σ (−1 parts).

⚙️ Assembly Chain Builder

Add parts below. Use +1 for the housing/outer reference and −1 for parts stacked inside it. The chain diagram updates live.

Direction: +1 Opens gap (housing, outer ref) −1 Closes gap (inner parts) Positive sign = dimension contributes positively to gap

Part Name Nominal (mm) ±Tol (mm) Direction Sens.

→ Each row contributes: +1 × nominal or −1 × nominal to the gap equation.

Live Assembly Chain — Signed Dimensions

Worst Case

Worst Case — Theory & Formula

The simplest and most conservative stackup method. Every dimension simultaneously at its worst extreme. Guaranteed to work — but often unnecessarily tight.

📐 Formula

Nominal Gap G₀ = Σ sᵢ · Nᵢ (signed sum of nominal dimensions)
Worst Case Tol T_wc = Σ |tᵢ| (sum of all absolute tolerances)
Max Gap = G₀ + T_wc | Min Gap = G₀ − T_wc
sᵢ ∈ {+1, −1} = direction in loop | Nᵢ = nominal dim | tᵢ = bilateral tolerance

🔩 Worked Example — Shaft in Housing

🏭 Bearing Assembly

Housing bore: 50.000 ±0.025 mm (direction +1, opens gap)
Bearing OD: 49.900 ±0.010 mm (direction −1, closes gap)
Required clearance: 0.050 to 0.180 mm

G₀ = (+1)(50.000) + (−1)(49.900) = 0.100 mm
T_wc = 0.025 + 0.010 = 0.035 mm
Max gap = 0.100 + 0.035 = 0.135 mm ✓ ≤ 0.180
Min gap = 0.100 − 0.035 = 0.065 mm ✓ ≥ 0.050
Result: PASSES worst case — 100% yield guaranteed

✅ Assumptions & Limitations

All parts are within their stated tolerance (by inspection or process control)
Dimensions are independent of each other
Conservative: probability of ALL parts simultaneously at worst extreme is extremely low for large n
For n=10 identical ±3σ parts: probability of WC scenario ≈ (0.0027)¹⁰ ≈ 2×10⁻²⁸
Use when yield MUST be 100% regardless of cost (e.g., aircraft, medical devices)

RSS Statistical

RSS Stackup Calculator

Root Sum of Squares. Assumes each part's dimension is normally distributed. The combined assembly gap is also normal — predict the sigma level and yield directly.

🧭 Direction Quick Reference

+1 Opens gap — housing, bore, outer envelope. Growing this dimension increases clearance.

−1 Closes gap — shafts, spacers, inner parts stacked inside. Growing these shrinks clearance.

Gap nominal G₀ = Σ (+1 parts × nominal) − Σ (−1 parts × nominal) | Direction does not affect variance (σ²), only the nominal.

⚙️ Assembly Chain Builder

For RSS, tolerance = kσ of the process. Adjust the Sigma Level (k) per part and the assembly target below.

Direction: +1 Opens gap −1 Closes gap

Part Name Nominal (mm) ±Tol (mm) Direction k (sigma)

Live Assembly Chain

Assembly sigma target 3σ

RSS Statistical

RSS — Theory & Formula

Based on the statistical fact that when independent random variables are summed, their variances add. The combined standard deviation grows as √n, not n.

📐 Formula Derivation

Each part: xᵢ ~ N(Nᵢ, σᵢ²) where σᵢ = tᵢ / kᵢ (kᵢ = process sigma level)
Gap G = Σ sᵢ · xᵢ → G ~ N(G₀, σ_gap²)
σ_gap² = Σ (sᵢ·σᵢ)² = Σ σᵢ² (direction has no effect on variance)
σ_gap = √(Σ σᵢ²) = √(σ₁² + σ₂² + ... + σₙ²)
T_rss (at k·σ) = k · σ_gap = k · √(Σ (tᵢ/kᵢ)²)
For equal kᵢ = k: T_rss = k · √(Σ(tᵢ/k)²) = √(Σ tᵢ²)

📈 WC vs RSS Interactive Comparison

Adjust n (number of equal parts) and individual tolerance to see how WC and RSS totals diverge.

Number of parts (n)

5

Each tolerance ±t (mm)

0.10

Sigma level (k)

3

WC vs RSS tolerance — vs number of parts

✅ RSS Assumptions

All dimensions independently normally distributed
Process is centered (mean = nominal) — zero mean shift
Stackup is a linear function of the component dimensions
RSS under-predicts risk if processes are skewed, bimodal, or correlated
Long-term mean shifts (1.5σ drift) need correction — see Six Sigma Shift page

Monte Carlo

Monte Carlo Simulation

Simulate thousands of assemblies by randomly sampling each part's dimension from its distribution. The result is the actual gap distribution — directly giving PPM out-of-spec.

⚙️ Assembly Setup

Direction: +1 Opens gap (housing/outer) −1 Closes gap (inner parts)

Part Name Nominal (mm) ±Tol (mm) Direction Distribution

Live Assembly Chain

Simulations

Min gap spec

Max gap spec

Monte Carlo

Monte Carlo — Theory & Setup

How random simulation works for tolerance analysis. Distribution types, convergence, and when to use each distribution.

📐 Algorithm

For each part i, define its distribution: Normal(Nᵢ, σᵢ), Uniform(lo, hi), or Triangular(lo, mode, hi)
For each simulation run j = 1…N: sample one value xᵢⱼ from each part's distribution
Compute the gap: Gⱼ = Σ sᵢ · xᵢⱼ (the assembly equation)
After N runs, build the histogram of all Gⱼ values
Count failures: Gⱼ < spec_lo or Gⱼ > spec_hi → PPM = (failures / N) × 10⁶
Fit a normal distribution to Gⱼ to estimate effective sigma level

📋 Distribution Types

Normal N(μ, σ)

Best for: machined dimensions, CNC, grinding
σ = tolerance / k (k=3 for 3σ process)
Most common in precision manufacturing
Default choice

Uniform U(lo, hi)

Best for: inspection-sorted parts, purchased parts with unknown process
Worst-case for RSS comparison
σ = (hi−lo)/(2√3)
Conservative

Triangular T(lo, mode, hi)

Best for: processes with a preferred mode but bounded range
Can model asymmetric processes (e.g., wear, thermal)
σ = √((lo²+mode²+hi²−lo·hi−lo·mode−hi·mode)/18)

Convergence

5,000 runs: good for mean & σ
10,000: reliable to ~100 PPM
100,000: reliable to ~10 PPM
1,000,000: reliable to ~1 PPM
Rule: 10× more runs per decade of PPM

Advanced

Six Sigma Mean Shift (1.5σ)

In long-term production, process means drift by up to ±1.5σ. The modified RSS method accounts for this shift in every component. Critical for automotive and high-volume manufacturing.

📐 Modified RSS with Mean Shift

Short-term (no shift): T_rss = k · √(Σ σᵢ²) → yields matches 3σ, 4σ, etc.
Long-term (1.5σ shift per part): T_rss_lt = k · √(Σ σᵢ²) + 1.5 · Σ σᵢ
The mean shift term 1.5·Σσᵢ is the worst-case of ALL means shifting in the same direction
Motorola convention: 6σ short-term process = 4.5σ long-term = 3.4 PPM defect rate

⭐ Interactive Shift Effect

Adjust mean shift and see how it widens the assembly gap distribution.

Number of parts (n)

5

Each ±tol (mm)

0.10

Mean shift (× σ)

1.5

Sigma level (k)

3.0

Gap distribution: no shift vs with shift

Advanced

Sensitivity Analysis

Which part's tolerance contributes most to assembly variation? Sensitivity analysis ranks parts by their variance contribution — showing where tightening tolerance has the biggest impact.

📐 Sensitivity Coefficients

σ_gap² = Σ (∂G/∂xᵢ)² · σᵢ² = Σ sᵢ² · σᵢ² = Σ σᵢ² (linear stackup)
Contribution of part i: Cᵢ = σᵢ² / σ_gap² × 100% = (tᵢ/kᵢ)² / Σ(tⱼ/kⱼ)²
Sensitivity rank: sort parts by Cᵢ descending
For nonlinear stackup G(x₁…xₙ): sensitivity = (∂G/∂xᵢ)² · σᵢ² at nominal values

⭐ Interactive Sensitivity Explorer

Build an assembly and see which parts dominate variance. Tighten the top contributor first — it gives the most reduction per dollar spent.

Part Name ±Tol (mm) k (sigma)

GD&T

GD&T & Tolerance Loops

How GD&T symbols (position, flatness, runout) translate into stackup loop contributions. Setting up the closed-loop equation from a drawing.

🔁 The Closed-Loop Method

Every tolerance stackup starts from a closed loop: starting at one end of the gap, traverse through each part in the assembly, and return to the other end. Assign +1 to dimensions that open the gap, −1 to those that close it.

Loop equation: G = D_housing − D_part_A − D_part_B − D_spacer + D_shim
At nominal: G₀ = Σ sᵢ · Nᵢ = 0 (balanced loop — all gaps accounted for)
GD&T contributions: add geometric tolerances as bilateral tolerances to the loop

📋 GD&T Symbols → Stackup Contribution

GD&T Symbol	Stackup Contribution	How to Include
⊕ True Position (⌀ t)	±t/2 in each axis	Add ±t/2 as bilateral tol in relevant axis
— Flatness (t)	±t/2 tilt/bow	Add ±t/2 to dimension perpendicular to surface
◎ Runout / Concentricity (t)	Full value t (unilateral)	Add as ±t/2 (bilateral equivalent)
∥ Parallelism (t)	±t/2 over length L	Add ±t·(d/L)/2 where d = distance from datum
⊿ Angularity (t)	±t/2 at nominal angle	Project onto stackup direction: ±(t/2)·cos θ
⊙ Cylindricity (t)	±t/2 radial	Add ±t/2 as radial tolerance
Dimensional tol ±t	Direct ±t	Directly add ±t to the loop

💡 Rules of Thumb

Every loop must close — start and end at the same datum
Assign directions consistently: traversing left-to-right is +1 for openings
Floating fastener: position tol = (hole tol + clearance hole tol) / 2 per axis
Fixed fastener: position tol = clearance / 2 per axis
Do not mix GD&T and ± linear tolerances without converting GD&T to bilateral linear equivalents
Bonus tolerance (MMC/LMC modifiers) changes tolerance value with actual mating size