Engineering guide to tolerance stackup analysis for assemblies

How engineers analyze tolerance accumulation

Every manufactured dimension has variation. When parts assemble into a chain of dimensions, individual tolerances accumulate — potentially preventing assembly or degrading function. Tolerance stackup analysis predicts the total variation at a critical assembly dimension using three methods: worst-case (all at extremes), RSS (statistical independence assumed), and Monte Carlo (simulated distribution). The results guide tolerance tightening or loosening decisions that balance cost and function.

Analysis methods

Engineering workflow

1. Identify the critical assembly dimension (gap, clearance, alignment).
2. Map the dimension chain: list every contributing part dimension.
3. Assign tolerance $\pm t_i$ to each dimension from drawing or process capability.
4. Compute worst-case stackup: sum of absolute tolerances.
5. Compute RSS stackup: root of sum of squared tolerances.
6. Optionally run Monte Carlo with specified sample count and distribution type.
7. Compare assembly variation to functional requirements.
8. Tighten tolerances on sensitive dimensions; relax on non-critical dimensions.

Key quantities and formulas

Worst-case total tolerance:

Root-sum-square tolerance:

Monte Carlo statistics:

Worked example

A five-part assembly chain with tolerances: 0.05, 0.10, 0.08, 0.03, 0.12 mm.

• Worst-case: $T_{\text{WC}}=0.05+0.10+0.08+0.03+0.12=0.38$ mm.
• RSS: $T_{\text{RSS}}=\sqrt{0.05^2+0.10^2+0.08^2+0.03^2+0.12^2}=\sqrt{0.0402}=0.191$ mm.
• The RSS result is roughly half the worst-case — this is the statistical advantage when tolerances are independent.
• If the functional requirement is 0.30 mm max gap variation, WC fails but RSS passes.

Common mistakes and checks

• Using RSS when tolerances are not independent: correlated tolerances (same machine, same setup) do not obey RSS assumptions.
• Ignoring thermal expansion: temperature differences between measurement and service add systematic bias, not random variation.
• Forgetting assembly shift: positional tolerance zones shift the mean, not just the spread.
• Not identifying the critical dimension: analyzing the wrong stackup chain wastes effort and misses the real risk.
• Over-tightening all tolerances: tightening non-critical dimensions adds manufacturing cost without improving function.

FAQ

When should I use worst-case vs RSS?

Use worst-case for safety-critical applications where 100% conformance is required. Use RSS when statistical rejection rates are acceptable and tolerances are truly independent.

How many Monte Carlo samples are needed?

At least 10,000 for reliable 3-sigma estimates; 100,000+ for tail probabilities (ppm reject rates).

Can I mix tolerance distributions in Monte Carlo?

Yes — the Monte Carlo method supports uniform, normal, or other distributions per dimension. This is its main advantage over RSS.

What is the Benderization factor?

A semi-empirical correction (typically 1.5) applied to RSS to account for non-normal distributions and mild correlations: $T_{\text{Bender}}=1.5\times T_{\text{RSS}}$.

How does GD&T relate to tolerance stackup?

GD&T defines tolerance zones geometrically. PhyCalcPro’s GD&T stack mode converts size dimensions and feature control frames (position, orientation, profile, runout, etc.) into stack contributors, including MMC/LMC bonus and datum-feature shift, then runs WC / RSS / Monte Carlo on the effective half-zones.

Drawing package (BOM + PDF / ZIP)

Mission: best-in-class drawing-package variation analysis — hierarchical stacks from component drawings up through sub-assembly and assembly — while keeping calculation deterministic and auditable.

1. Upload a ZIP with required BOM.xlsx / CSV (see /templates/PhyCalcPro-BOM-template.csv) plus assembly/part PDFs (or a single PDF for quick trials).
2. BOM defines Level / Parent / Part Number / Revision / Drawing File — the assembly tree and package validation (missing PDF, orphans, duplicate PNs).
3. Extract all (components first): read every component drawing into the annotation library, then SA/assembly sheets. Vision extract is assistive — review quality scores before use.
4. Stack program: create named stacks at sub-assembly and assembly/top levels. Contributors come from component (and SA) annotations under the BOM context node.
5. Confirm each chain; solve WC/RSS/Monte Carlo (P95, yield vs requirement, sensitivity). Optional SA rollup into assembly stacks.
6. Assist tab: propose stacks from notes (suggestions only), explain FCFs/drivers, what-if allocation packages, export DR packet (markdown/JSON).
7. AI does not invent clearances or official pass/fail — you own topology; solvers own numbers.

Requires OPENAI_API_KEY for vision extract (optional OPENAI_VISION_MODEL). Simple bilateral mode remains available without drawings.

Save / retrieve studies

Use Save study / Update study with a study name. When signed in, studies persist in account-backed browser storage (and sync to workspaces when configured). Guests keep studies for the session only. Saved payload includes BOM structure, extracts, multi-stack program, confirmation flags, and results — not original PDF files (re-upload the ZIP only if you need to re-extract).

Use the PhyCalcPro calculator

Open the Tolerance Stackup calculator for drawing package (recommended), Simple bilateral arrays, or GD&T single-drawing mode. Package mode returns a multi-stack program dashboard, WC/RSS/MC (P95/yield), contributor sensitivity, and DR packet export.

Purpose

Analyze dimensional variation accumulation in assemblies using worst-case and statistical (RSS) methods, with optional Monte Carlo simulation and GD&T (MMC/LMC) stack analysis.

Physics & theory

Each dimension in a chain contributes uncertainty $\pm t_i$. Worst-case assumes all tolerances at simultaneous extremes: $T_{\text{WC}}=\sum |t_i|$. RSS assumes independent normal distributions: $T_{\text{RSS}}=\sqrt{\sum t_i^2}$. Monte Carlo draws random deviations per dimension and sums to build the assembly distribution. For GD&T, effective zone $T_{\text{eff}}=T_{\text{spec}}+T_{\text{bonus}}$ at MMC/LMC, with optional datum shift.

Governing equations

Numerical method

Closed-form WC and RSS. Optional Monte Carlo with uniform sampling over monteCarloSamples iterations. Separate X/Y/Z stacks when multi-axis contributors are provided. GD&T path uses solveGdtStackEngine.

Inputs

Outputs

• Worst-case total, RSS total, Monte Carlo mean and standard deviation (if run), per-direction stacks, GD&T contributor/bonus table.

Design codes & checks

• Indicative: Worst-case and RSS stack
• US: ASME Y14.5 dimensioning and tolerancing
• ISO: ISO 286 / ISO GPS principles (related)

Assumptions & limitations

• Linear stack chains projected from geometric zones (half-zone on stack axis).
• Supported characteristics: position, perpendicularity, parallelism, profile, concentricity/coaxiality, circular/total runout, plus size.
• MMC/LMC bonus and datum-feature shift included; composite/simultaneous/pattern fields are captured on extracts for audit (stack uses zone values as modeled).
• Vision extract must be engineer-verified before trusting results.
• RSS assumes normal, independent variations — not valid for skewed processes.
• Monte Carlo quality depends on sample count and distribution assumptions.
• No thermal expansion unless added as dimensions.
• Multi-drawing assemblies: upload the controlling stack sheet or one PDF (≤5 pages).

References

1. ASME Y14.5-2018. Dimensioning and Tolerancing.
2. Wick, C. H., et al. Tolerance Stack Up Analysis, 2nd ed. ASME Press.
3. ISO 286-1:2010. Limits and fits.
4. Srinivasan, V. Statistical Tolerance Analysis. ASME Handbook.