Monte Carlo Tolerance Simulation: Turning Stack-Ups into Yield Forecasts
Worst-Case tells you the absolute mathematical boundary, and RSS gives you an analytical standard deviation. Monte Carlo simulation bridges the gap to manufacturing reality by digitally assembling thousands of virtual parts and showing the exact distribution of clearances or gaps your factory will produce.
How 10,000-Build Monte Carlo Simulation Works
In each iteration, the simulator draws a random value for every dimension in the loop from its probability distribution and computes the resulting assembly dimension. In Precise Fit Pro, each dimension is modeled as a normal distribution centered at μ_i = N_i + (u_i + l_i) / 2 with standard deviation σ_i = (u_i − l_i) / 6 (treating the drawing tolerance zone as ±3σ), sampled via the Box-Muller transform.
Repeating this process 10,000 times generates an empirical histogram of virtual assemblies. From that dataset, you read the simulated mean (μ), combined standard deviation (σ), observed extremes (Sim Min / Sim Max), and process capability indices directly against your specification limits.
Reading Cp and Cpk Correctly
Process capability translates a histogram into expected production yield. Potential capability Cp = (USL − LSL) / (6σ) measures how comfortably the distribution width fits inside your specification window assuming perfect centering. Actual capability Cpk = min((μ − LSL) / 3σ, (USL − μ) / 3σ) penalizes any off-center shift toward either limit.
For a centered normal distribution, standard capability benchmarks map directly to two-sided defect rates: Cpk = 1.00 corresponds to ±3σ (99.73% yield, ~2,700 PPM out of spec); Cpk = 1.33 corresponds to ±4σ (99.994% yield, ~63 PPM — the typical baseline for automotive and industrial production); and Cpk = 1.67 corresponds to ±5σ (99.99994% yield, <1 PPM — common for aerospace and medical devices).
Worked Fit Example: ISO 286 Ø25 H7/g6 Clearance Fit
Consider a Ø25 mm precision running fit using ISO 286 H7/g6 limits. For diameters over 18 mm up to 30 mm, the H7 hole spans 25.000 to 25.021 mm (μ_H = 25.0105 mm, σ_H = 0.0035 mm), while the g6 shaft spans 24.980 to 24.993 mm (μ_S = 24.9865 mm, σ_S = 0.0022 mm).
Worst-Case arithmetic gives a minimum clearance of +0.0070 mm and a maximum of +0.0410 mm—confirming a guaranteed clearance fit. Statistically, however, the clearance centers at μ = 25.0105 − 24.9865 = +0.0240 mm with combined σ = √(0.0035² + 0.0022²) = 0.0041 mm. Across 10,000 simulated builds, 99.73% of real assemblies fall between +0.0117 mm and +0.0363 mm, rarely touching the extreme 0.007 mm or 0.041 mm corners.
Best Practices When Running Simulations
- Use at least 10,000 iterations so standard deviation and tail estimates stabilize between runs.
- Remember that normal ±3σ sampling assumes suppliers control process drift; if a supplier runs near one tool-wear limit, verify sensitivity by shifting the nominal.
- Report both Worst-Case bounds and Monte Carlo Cp/Cpk in design-review PDFs so QA and operations see both the hard boundary and realistic yield.
- Re-run the simulation whenever you swap a manufacturing process (e.g., CNC turning to injection molding) or change a tolerance class.