Tolerance stack-up analysis (TSA) is a systematic method to calculate the total variation in a critical assembly dimension—called a gap, fit, or functional requirement (FR)—caused by the cumulative variation of all individual parts and features that contribute to that dimension.
To put it in plain language: Imagine you are assembling a gearbox where the input shaft must align with the output shaft within 0.1 mm to avoid premature bearing wear. The alignment is not controlled by a single part; it depends on the position of the gearbox housing’s bore, the shaft’s diameter, the bearing’s inner and outer ring dimensions, and even the flatness of the housing’s mounting face. TSA quantifies how all these individual variations add up (or cancel out) to affect the final alignment.
Key terms defined at first use:
TSA directly impacts three core outcomes of manufacturing: part function, assembly yield, and production cost.
The most critical risk of poor TSA is assembly failure. For example, consider a precision medical device where a piston must fit inside a cylinder with a maximum clearance of 0.05 mm to maintain pressure. If the cumulative variation of the piston diameter, cylinder bore, and their mounting features exceeds 0.05 mm, the device will leak and fail. Even seemingly minor misalignments can cause catastrophic failure: a 0.1 mm misalignment in a high-speed spindle assembly can increase bearing fatigue by 300% or more, reducing service life from years to months.
Yield is the percentage of assemblies that meet all functional requirements. Without TSA, manufacturers often set overly tight individual part tolerances to compensate for unknown stack-up, leading to high scrap rates. For example, a consumer electronics manufacturer that initially set a ±0.01 mm tolerance for a plastic housing feature (to avoid fit issues) found that 15% of parts were scrapped due to minor molding variations. After conducting a statistical TSA, they relaxed the tolerance to ±0.02 mm and adjusted only one other part’s tolerance, reducing scrap to 0.5% while maintaining 99.9% assembly yield.
Tight part tolerances directly increase manufacturing costs. Machining a part to ±0.003 mm requires a high-precision machine, longer cycle times, and more expensive inspection, compared to machining the same part to ±0.01 mm. For example, the Gree CNC GA-DV750 vertical machining center has a positioning accuracy of 0.003 mm, making it suitable for tight-tolerance parts, but it has a higher hourly operating cost than the GA-MV856, which has a positioning accuracy of 0.010 mm and is better suited for looser tolerances. TSA helps manufacturers optimize part tolerances to the loosest possible values that still meet functional requirements, reducing both machining and inspection costs.
TSA is performed in two primary ways: worst case analysis and statistical analysis. Both start with the same first step: defining the assembly’s dimensional chain.
The dimensional chain is a sequence of parts and features that contribute to the FR. For example, the dimensional chain for the gearbox shaft alignment FR would be:
The FR is the sum of all these variations, with signs indicating whether the variation increases or decreases the FR.
WC analysis is the simplest TSA method. It assumes that every individual tolerance deviates in the direction that maximizes the total variation of the FR. The formula for WC total tolerance is: [ T_{WC} = \sum |T_i| ] Where ( T_i ) is the tolerance of each individual feature in the dimensional chain.
For example, if the gearbox dimensional chain has seven features each with a tolerance of ±0.01 mm (total tolerance of 0.02 mm per feature), the WC total tolerance is 7 × 0.02 mm = 0.14 mm. If the FR allowable range is 0.1 mm, the WC analysis predicts that the assembly will fail.
WC analysis is often used for small production runs or critical safety components where failure is not acceptable. To perform WC analysis on the shop floor, engineers:
WC analysis is highly conservative. In reality, it is extremely rare for all tolerances to deviate to their maximum limits in the same direction. For example, if a part is machined on the Gree CNC GA-UHD500 (a high-precision 5-axis machining center with a repeatability of 0.004 mm), the chance of producing a part at the extreme upper limit of its tolerance is less than 0.1% for a normally distributed process. For a dimensional chain of 10 parts, the chance of all 10 parts being at their extreme limits is effectively zero. This means WC analysis often overestimates the total variation, leading to unnecessarily tight part tolerances and higher costs.
Statistical analysis uses probability theory to calculate the likelihood of the FR falling outside its allowable range. It assumes that part variations follow a normal distribution (the bell curve), which is typical for well-controlled manufacturing processes. The formula for the statistical total tolerance (at 99.73% confidence, or 3 standard deviations) is: [ T_{Stat} = 3 \times \sqrt{\sum (\sigma_i)^2} ] Where ( \sigma_i ) is the standard deviation of each individual feature’s variation.
For a normally distributed process, 99.73% of parts fall within ±3σ of the nominal value. For a part with a tolerance of ±0.01 mm, the standard deviation σ is typically 0.0033 mm (assuming the process is centered and capable). Using the gearbox example with seven features each with a tolerance of ±0.01 mm, the statistical total tolerance is: [ T_{Stat} = 3 \times \sqrt{7 \times (0.0033)^2} = 3 \times 0.0087 = 0.026 \text{ mm} ] This is far smaller than the WC total tolerance of 0.14 mm, and well within the FR allowable range of 0.1 mm.
Statistical analysis requires data on the actual variation of each part feature. To perform it on the shop floor, engineers:
Statistical analysis is only as accurate as the data used. If the process is not well-controlled (e.g., a worn machine tool producing parts with non-normal variation), the results will be incorrect. For example, if a part is machined on a Gree CNC GA-GF2517 gantry machining center (with a positioning accuracy of 0.015 mm) that is out of calibration, the variation may be skewed or have multiple peaks, violating the normal distribution assumption.
TSA results are only as reliable as the input data. The most common sources of error include:
Incomplete Dimensional Chains: Omitting critical features that contribute to the FR. For example, an engineer may forget to include the flatness of a mounting face in a shaft alignment dimensional chain. Flatness variation can be as large as ±0.02 mm for parts machined on a medium-precision machine, leading to significant undercalculation of total variation.
Incorrect Tolerance Values: Using nominal tolerance values instead of actual process variation. For example, a drawing may specify a tolerance of ±0.01 mm, but the actual process may have a variation of ±0.02 mm due to poor process control.
Non-Normal Distributions: Assuming part variations follow a normal distribution when they do not. Common causes of non-normal variation include tool wear, material inconsistencies, and machine tool drift. For example, tool wear on a high-speed spindle (like the GA-DV750’s 30,000 rpm spindle) can cause part dimensions to increase gradually over a production run, leading to a skewed distribution.
Ignoring Assembly Variation: Focusing only on part tolerances and ignoring variation introduced during assembly. For example, the force used to press a bearing onto a shaft can cause the shaft to bend, leading to misalignment that is not accounted for in the dimensional chain.
Environmental Variation: Ignoring changes in temperature, humidity, or vibration that affect part dimensions. For example, a steel part machined at 20°C will expand by 0.012 mm per meter of length for every 10°C increase in temperature. If an assembly is used in an environment with a 20°C temperature range, this expansion can add significant variation to the FR.
TSA is a powerful tool, but it has important limitations that engineers often overlook:
It Does Not Account for Dynamic Loads: TSA calculates static variation, but assemblies often experience dynamic loads during operation. For example, a gearbox shaft may be aligned within 0.05 mm at rest, but under load, the shaft may bend by 0.1 mm, leading to failure. TSA cannot predict dynamic deformation; this requires finite element analysis (FEA).
It Cannot Predict Random Failures: Statistical analysis calculates the probability of failure based on historical data, but it cannot predict rare, random failures caused by unforeseen events (e.g., a sudden material defect, a machine tool crash).
It Relies on Stable Processes: Statistical analysis is only valid if the manufacturing process is stable and in control. If the process changes (e.g., a new tool is introduced, a machine is repaired), the variation data becomes outdated, and the TSA results are no longer accurate.
It Does Not Account for Wear Over Time: TSA calculates initial variation at the time of assembly, but parts wear over time. For example, a bearing’s inner ring may wear by 0.05 mm after 10,000 hours of operation, leading to misalignment that was not accounted for in the initial TSA.
Over-Confidence in Statistical Results: Engineers often assume that a 99.73% confidence level means the assembly will never fail, but this is not true. For a production run of 1 million assemblies, a 99.73% confidence level means 2,700 assemblies will fail. For critical applications, this may be unacceptable.
| Method | Accuracy | Effort | Typical Use Case |
|---|---|---|---|
| Worst Case | Low (overly conservative) | Low (no process data required) | Small production runs, safety-critical components, initial design phases |
| Statistical | High (when process data is accurate) | Medium (requires process capability data) | Large production runs, non-safety-critical components, mature designs |
Use this checklist to ensure your TSA is accurate and reliable:
Q: When should I use worst case analysis vs statistical analysis? A: Use worst case analysis for small production runs (fewer than 100 units) or safety-critical components where even a 0.1% failure rate is unacceptable. Use statistical analysis for large production runs (more than 1,000 units) where the cost of overly tight part tolerances is significant.
Q: What is a good Cp value for statistical analysis? A: A Cp of 1.33 is generally considered acceptable for most applications, as it means 99.99% of parts fall within the tolerance range. For high-precision applications, a Cp of 1.67 or higher is recommended.
Q: How do I account for temperature variation in TSA? A: Add the thermal expansion of each material to the dimensional chain. For example, if a steel part has a length of 100 mm and the operating temperature range is 20°C, the thermal expansion is 100 mm × 12 ppm/°C × 20°C = 0.024 mm. Include this as an additional tolerance in the chain.
Q: Can I use statistical analysis for custom or low-volume parts? A: Statistical analysis requires process data, which is often not available for custom or low-volume parts. In this case, use worst case analysis or Monte Carlo simulation with estimated variation values based on the machine tool’s specifications (e.g., use the GA-MV856’s positioning accuracy of 0.010 mm as an estimate of part variation).
Q: What is Monte Carlo simulation, and when should I use it? A: Monte Carlo simulation is a statistical method that generates thousands of random combinations of part variations to calculate the probability of the FR falling outside its allowable range. It is more accurate than the 3σ statistical formula for non-normal distributions or complex dimensional chains. Use Monte Carlo simulation for high-precision applications or when part variations are not normally distributed.
Send us your part, batch size and target port. We reply with a configuration, a binding lead window and a total-cost comparison.
Request a quote →