Assemblies that work perfectly on individual part inspection often fail to assemble or function correctly in production — tolerance stack-up analysis prevents this by predicting how individual part tolerances combine to produce variation in critical assembly dimensions.
Tolerance stack-up analysis (also called tolerance chain analysis or dimensional analysis) is the process of calculating the cumulative effect of multiple individual part tolerances on a critical assembly dimension. It is essential for designing assemblies that will consistently achieve their functional requirements across all combinations of manufactured parts. This guide covers the loop diagram method, worst-case analysis, RSS (root sum square) statistical analysis, and a worked assembly example comparing both approaches.
Why Tolerance Stack-Up Matters
Consider an assembly with 10 parts in a dimensional chain. Each part has a ±0.1 mm tolerance. Worst-case, all tolerances can be at their extreme simultaneously, giving a total variation of ±1.0 mm in the assembly gap. If the required gap is 0.3 mm ± 0.2 mm, this assembly will have unacceptable gap variation. Three options exist: tighten individual part tolerances (expensive), change the design to reduce the number of parts in the chain, or accept that a small percentage of assemblies will be out of specification and use selective assembly or shimming.
The statistical approach (RSS) recognizes that the probability of all tolerances being simultaneously at their worst case is extremely small, allowing wider individual part tolerances while still achieving the required assembly yield — a key trade-off between manufacturing cost and assembly quality.
The Loop Diagram Method
The loop diagram is the foundation of tolerance stack-up analysis. The procedure is:
- Identify the assembly gap or resultant dimension G that must be controlled
- Draw a closed loop from one end of G to the other, passing through all parts that contribute dimensions to the chain
- Assign a direction to each dimension in the loop: positive (+) if the dimension goes in the same direction as the traverse, negative (−) if it goes in the opposite direction
- The nominal value of G = sum of (+) dimensions − sum of (−) dimensions
- The tolerance on G depends on how individual tolerances combine (worst-case or RSS)
Every dimension that contributes to the assembly gap must be included in the loop — missing a dimension produces an incorrect analysis. Common sources of dimensions in a loop: part lengths, boss heights, recess depths, spacer thicknesses, bearing inner ring widths, snap ring thicknesses, and clearance gaps themselves.
Worst-Case (WC) Analysis
In worst-case analysis, every individual tolerance is assumed to be at its worst-case extreme simultaneously. The total assembly tolerance TG is the arithmetic sum of all individual tolerances:
TG,WC = Σ |ti|
Where ti is the total tolerance (not half-tolerance) on each dimension in the loop. The assembly gap G will always be within Gnominal ± TG,WC/2 for 100% of manufactured parts.
Worst-case analysis guarantees 100% assembly yield with full interchangeability — every part from any source at any time will assemble and function correctly. The disadvantage is that it often requires very tight individual part tolerances, which increases manufacturing cost. WC analysis is appropriate for: safety-critical assemblies where 100% function is mandatory, small series production where individual part cost is less important than assembly reliability, and assemblies where rework is extremely expensive or impossible.
RSS (Root Sum Square) Statistical Analysis
The RSS method is based on the statistical principle that when multiple independent random variables combine, the standard deviation of the sum is the square root of the sum of the individual variances. For parts where each tolerance represents ±3σ (three-sigma bilateral tolerance), the assembly tolerance at the same confidence level is:
TG,RSS = √(Σ ti²)
This is much smaller than the worst-case sum. For example, with 10 dimensions each having ±0.1 mm tolerance: WC gives ±1.0 mm; RSS gives ±√(10 × 0.01) = ±0.316 mm.
The RSS result means that 99.73% of assemblies (3-sigma) will have the gap within the ±0.316 mm range. The remaining 0.27% will be outside this range — these are the assemblies that need rework or rejection. Whether this yield is acceptable depends on production volume and rework cost.
RSS analysis is appropriate for: high-volume production where a small rejection rate is economically preferable to tighter tolerances, assemblies where the statistical independence assumption is valid (parts come from many different batches), and design exploration to evaluate whether existing tolerances can meet functional requirements.
3-Sigma Approach and Cpk in Context
The RSS method assumes that parts are manufactured to a normal distribution with tolerance = ±3σ. In reality, process capability (Cpk) determines the actual distribution. A process with Cpk = 1.0 produces parts within ±3σ with 99.73% of parts within tolerance. A process with Cpk = 1.33 produces parts within ±4σ of the mean with 99.994% within tolerance. If the manufacturing processes have Cpk < 1.0 (common in legacy manufacturing), the RSS prediction will be optimistic and more assembly rejects will occur than predicted.
In practice, conservative engineers use a modified RSS formula that accounts for process capability: TG = zassembly × √(Σ (ti/zi)²), where zi = sigma level of each process (typically 3 to 6) and zassembly is the target sigma level for the assembly. This allows mixing processes with different capability levels in the same analysis.
Comparison: Worst-Case vs RSS
| Aspect | Worst-Case | RSS Statistical |
|---|---|---|
| Assembly yield guarantee | 100% | Typically 99.73% (3σ) |
| Individual tolerance required | Tightest | Wider (more economical) |
| Best for | Safety-critical, low volume | High-volume, cost-sensitive |
| Key assumption | None | Normal distribution, independence |
| Tolerance ratio (WC/RSS) | 1.0 (baseline) | 1/√n smaller (n = part count) |
| Risk | Tight tolerances increase cost | Small rejection rate possible |
Worked Example: Shaft Assembly Stack-Up
An assembly consists of a shaft end-supported in a housing. The required axial gap (end float) G must be between 0.1 mm and 0.5 mm (nominal 0.3 mm ± 0.2 mm). The dimensional chain along the shaft axis includes four dimensions:
| Dimension | Description | Nominal (mm) | Tolerance ±(mm) | Direction |
|---|---|---|---|---|
| A | Housing bore depth | 80.0 | ±0.10 | + |
| B | Bearing outer ring width | 15.0 | ±0.05 | − |
| C | Shaft shoulder to shoulder | 50.0 | ±0.15 | − |
| D | Second bearing outer ring width | 15.0 | ±0.05 | − |
Nominal gap: G = A − B − C − D = 80.0 − 15.0 − 50.0 − 15.0 = 0.0 mm. Wait — this is zero, which means we need to re-examine: the nominal gap should be 0.3 mm, so adjust A to 80.3 mm. Gnom = 80.3 − 15.0 − 50.0 − 15.0 = 0.3 mm. Correct.
Worst-case: TG,WC = 0.10 + 0.05 + 0.15 + 0.05 = 0.35 mm. G ranges from 0.3 − 0.35 = −0.05 mm to 0.3 + 0.35 = 0.65 mm. The minimum gap is −0.05 mm (interference!) — worst case, bearings jam. Not acceptable with these tolerances.
RSS: TG,RSS = √(0.10² + 0.05² + 0.15² + 0.05²) = √(0.010 + 0.0025 + 0.0225 + 0.0025) = √0.0375 = 0.194 mm. G ranges from 0.3 − 0.194 = 0.106 mm to 0.3 + 0.194 = 0.494 mm. Both within the required 0.1–0.5 mm. RSS predicts 99.73% assembly yield with these tolerances.
Conclusion from example: WC analysis fails; RSS analysis passes. For this high-volume gearbox assembly, the statistical approach is appropriate and saves significant machining cost on the shaft shoulder (±0.15 mm is achievable by turning; ±0.05 mm would require grinding).
When to Use Each Method
Use worst-case analysis for: assemblies with fewer than 4 parts in the chain (statistical benefit is minimal), safety-critical functions (brakes, primary structural joints, nuclear), medical devices, and aerospace primary structure. Use RSS analysis for: consumer products and industrial equipment with high production volumes (10,000+ units/year), assemblies with 5 or more parts in the chain, and when manufacturing process capability is well-characterized and controlled. In any case, the final design tolerance specification should be validated on first-article builds before full production launch.
Conclusion
Tolerance stack-up analysis is the bridge between individual part tolerances and assembly function. The loop diagram method provides a systematic way to identify all contributing dimensions and their directions. Worst-case analysis guarantees 100% assembly yield at the cost of tighter (more expensive) individual tolerances. RSS statistical analysis allows wider individual tolerances with a predictable small rejection rate, appropriate for high-volume production. Always verify nominal gap calculations first, then apply the tolerance method appropriate to the application’s criticality and production volume.
コメント