For a one-dimensional PCB busbar gap, define the dimensional chain and directions first. Worst-case analysis uses dimensional limits; RSS uses standard deviations and correlation assumptions. A drawing tolerance is not automatically a standard deviation, and a statistical range is not an all-combinations fit guarantee.
Draw the chain before calculating
Sigmetrix distinguishes extreme-limit analysis from distribution-based RSS analysis. Identify datums, locating faces and the gap along one direction. Hole-pattern and angular relationships may require a more complete model.
Assume dimensions from a common datum: A=10.00±0.10 mm, B=9.50±0.10 mm and g=A−B. The nominal gap is 0.50 mm. The extreme limits give 9.90−9.60=0.30 mm and 10.10−9.40=0.70 mm. These results belong only to the defined one-dimensional model.
| Approach | Inputs | Question answered |
|---|---|---|
| Worst case | Limits, directions and constraints | Extreme gap combinations |
| RSS | Standard deviations, sensitivities and correlation | Variation under a statistical model |
| Assembly verification | Actual locating, tilt and contact | Physical fit |
Obtain standard deviations separately
NIST's propagation discussion includes covariance, which is zero for independent inputs. Suppose process data separately estimate σA=0.03 mm and σB=0.04 mm with independent inputs. Then σg=√(0.03²+0.04²)=0.05 mm. The same linear propagation mathematics is applied here to dimensional variation, not to redefine manufacturing tolerances as measurement uncertainty.
Three times σg is 0.15 mm, distinct from the worst-case half-width of 0.20 mm. Parts can fall outside a statistical range. Check distributions, means and stability using actual data before making yield claims. Common tooling, datums or material batches require a correlation review.
Check omitted geometry
Support movement, tilt and locating changes can affect the gap. State which effects are already inside each dimensional envelope to avoid omissions or double counting. Use an appropriate geometric model for position tolerances and complex datum systems.
For custom PCB busbars, supply mating-part geometry and assembly datums alongside the drawing. See first-article dimensions and revision control. This mechanical example establishes no electrical-clearance or insulation compliance.
Questions
Can a smaller RSS result justify a smaller gap? Confirm the approved statistical-risk basis and supporting data first.
Can a common dimension be counted twice? Trace the chain; common terms can cancel.
References
Sigmetrix statistical tolerance analysis; NIST propagation and covariance.
The cover is an AI dimensional-chain concept. Assumed values establish no Hongchuan tolerance capability, inspection result or assembly guarantee.