Stack-Up Analysis Mastery

Master 1-D tolerance stack-up analysis to predict assembly gaps, quantify interference risk, and drive data-driven tolerance decisions. This comprehensive tutorial covers signed dimensional chains, worst-case bounds, Root Sum of Squares (RSS) statistical bounds, Monte Carlo simulation, and practical application to CAPA and PPAP.

Learning Objectives

By the end of this tutorial, you'll understand the complete stack-up analysis workflow from dimensional chain definition to risk-based tolerance decisions.

01Stack-Up Terminology

02Worst-Case Analysis Method

03Root Sum of Squares (RSS) Method

How RSS Works

Root Sum of Squares (RSS) is the classical statistical tolerance method. Instead of assuming every contributor reaches its extreme limit simultaneously (Worst-Case) or sampling from full distributions (Monte Carlo), RSS combines tolerance bands in quadrature — the same way independent random errors add in statistics. This produces bounds that are tighter than Worst-Case and typically aligned with a +/-3 sigma assumption on each contributor.

RSStol = sqrt( Σ ( (lowerToli + upperToli) / 2 )2 )

RSSmin = Gnominal − RSStol

RSSmax = Gnominal + RSStol

Reference: NIST/SEMATECH e-Handbook of Statistical Methods, Section 2.6; ASME Y14.5 statistical tolerancing guidance.

When to Use RSS

• • High-volume production where contributors are independent and approximately normal
• • Cpk >= 1.0 processes on every contributor (the +/-3 sigma assumption holds)
• • Cost-constrained programs where Worst-Case is unnecessarily tight
• • PPAP / APQP submissions that accept statistical tolerancing per AIAG
• • Quick statistical baseline before committing to a full Monte Carlo distribution study

Assumptions and Limitations

• • Independence: contributors must vary independently. Common-cause variation (same operator, same machine, same lot) breaks RSS.
• • Approximate normality: heavily skewed or bimodal contributors invalidate the quadrature combination.
• • +/-3 sigma coverage: RSS bounds approximate a 99.73% interval. Programs requiring 4-6 sigma coverage need Monte Carlo with explicit sigma rules.
• • Optimism risk: when assumptions break, RSS can under-predict risk. Validate with process data or fall back to Worst-Case for safety-critical assemblies.

Method Comparison: Worst-Case vs RSS vs Monte Carlo

Coverage

100% deterministic

~99.73% (+/-3 sigma)

User-defined (3-6 sigma)

Conservatism

Most conservative

Middle ground

Most realistic

Required Inputs

Nominal + tolerances

Nominal + tolerances + distributions + sigma rule

Output

Min/Max bounds

Statistical Min/Max bounds

Full distribution + risk probabilities

Typical Use

Safety-critical, low volume

High volume, independent processes, PPAP

Risk quantification, mixed distributions, correlated factors

Reconciliation principle: The Step 5 Overview KPI grid shows all three methods side-by-side. A robust design typically passes all three. A design that passes RSS but fails Worst-Case is acceptable for high-volume statistical programs but not for safety-critical assemblies.

04Monte Carlo Simulation

05Reading Stack-Up Visualizations

06Practical Planning Tips

Before You Start Your Stack-Up

Define the Functional Gap

• • What clearance or interference are you solving for?
• • What are the LSL and USL for acceptable assembly performance?
• • Is any gap below zero (interference) acceptable?

Gather Your Dimensional Data

• • Collect nominal values and tolerances from drawings or GD&T
• • Identify whether tolerances are symmetric (±) or asymmetric (+upper/−lower)
• • Determine the sign convention: which dimensions add vs. reduce the gap

Characterize Distributions (for Monte Carlo)

• • Do you have process data? Use it to validate distribution assumptions
• • No data? Start with Uniform (conservative) or Triangular (moderate)
• • Use Six Sigma Center's Distribution Characterization module to test for normality

Common Pitfalls to Avoid

Don't Do This

• • Assume all parts are normally distributed
• • Use worst-case for high-volume cost optimization
• • Ignore sign convention errors
• • Skip distribution validation
• • Tighten all tolerances equally

Do This Instead

• • Validate distributions with process data
• • Use Monte Carlo for realistic risk assessment
• • Walk the chain carefully to assign signs
• • Use Distribution Characterization module first
• • Target the top Pareto contributor for CAPA

07Charts You Will See In The Results

08Metrics You Will Read And What They Mean

Every number Gap Stack-Up reports, what it actually measures, how it is computed from your data, the range that counts as healthy, and the way it most often gets misread in a review.

Nominal Gap

What It Means

The gap you get when every part is exactly at its nominal dimension.

How It Is Computed

Each dimension is multiplied by its direction coefficient (+1 or -1) and summed along the loop.

Healthy range: Must sit comfortably inside the requirement window with room on both sides.

Common misread: A healthy nominal says nothing about variation — a centered design can still scrap parts.

Worst-Case Tolerance

The absolute widest the gap can ever be if every tolerance goes the wrong way at once.

Absolute tolerance values are summed arithmetically with no statistical credit.

Healthy range: Fits inside the requirement for 100% interchangeability; often unaffordable on long chains.

Common misread: Worst case is a guarantee, not a prediction — using it as an expected result over-designs the part.

Root Sum Square Tolerance

The statistically expected spread of the gap when contributors vary independently.

Each scaled tolerance is squared, summed, and square-rooted.

Healthy range: Typically 30 to 60 percent of the worst-case value on a chain of four or more contributors.

Common misread: RSS assumes independent, centered, roughly normal contributors — a shifted supplier breaks it.

Contribution Percent

The share of total statistical variation owned by one dimension.

Each squared scaled tolerance is divided by the sum of all squared scaled tolerances.

Healthy range: No single contributor above roughly 50 percent on a mature design.

Common misread: Because it is squared, halving a tolerance cuts contribution far more than the linear intuition suggests.

Simulated Mean and Sigma

The average simulated gap and how much simulated assemblies scatter around it.

Computed across every Monte Carlo trial gap produced by the run.

Healthy range: Mean should track the nominal gap; sigma should be close to one third of the RSS half-tolerance.

Common misread: A mean far from nominal means an input distribution is skewed, not that the design shifted.

Probability Out Of Spec

The share of simulated assemblies that violate the gap requirement.

Trials outside the requirement window are counted and divided by total trials.

Healthy range: Below 0.1 percent for non-critical gaps; parts per million targets for safety features.

Common misread: The estimate cannot resolve rates finer than one over the trial count — 10,000 trials cannot see a 10 ppm risk.

Predicted Gap Ppk

How capable the assembly gap is predicted to be before you cut any metal.

Applies the standard capability ratio to the simulated gap mean and sigma.

Healthy range: At least 1.33; 1.67 or better for critical or safety characteristics at PPAP.

Common misread: This is a prediction from your assumed input distributions, not measured process performance.

Ready to Analyze Your Assembly?

Now that you understand the fundamentals, you're ready to build and analyze your own tolerance stack-up.

Start AnalysisPro

Upgrade to Pro to unlock

Interactive Tutorial

Hands-on practice

Case Study

See stack-up in action