Calculate Fraction Of Rejected Parts For Uniform Distribution

Calculate the fraction of rejected parts when a quality characteristic is modeled as uniformly distributed between a minimum and maximum process value.

How to Calculate the Fraction of Rejected Parts for a Uniform Distribution

In quality engineering, one of the fastest ways to estimate scrap or rejection risk is to model a process output with a probability distribution and compare that distribution to your specification limits. If your measured characteristic is approximately flat across a bounded range, the uniform distribution is often a practical model. This calculator focuses exactly on that case: estimating what fraction of parts fall outside acceptable limits and are therefore rejected.

The idea is simple: if every value between a minimum and maximum process level is equally likely, then probability is just interval length divided by total spread. That gives you a direct geometric way to compute accepted and rejected fractions without advanced statistics software.

Core setup and notation

• Process distribution: \(X \sim U(a, b)\), where a is the minimum possible value and b is the maximum possible value.
• Total process spread: \(b – a\), which must be positive.
• Specification limits: lower spec limit (LSL) and upper spec limit (USL), depending on whether tolerance is one-sided or two-sided.
• Accepted fraction: proportion of the process interval that lies inside the specification interval.
• Rejected fraction: \(1 -\) accepted fraction.

Exact formulas for rejection under uniform assumptions

For a uniform distribution, these equations are exact given your interval assumptions:

1. Two-sided specification (\(LSL \le X \le USL\)): Accepted interval is overlap of \([a,b]\) and \([LSL,USL]\): \[ L_{accept} = \max(0, \min(b,USL)-\max(a,LSL)) \] \[ P(accept) = \frac{L_{accept}}{b-a}, \quad P(reject)=1-P(accept) \]
2. Lower-only specification (\(X \ge LSL\)): \[ L_{accept}=\max(0, b-\max(a,LSL)) \] \[ P(accept)=\frac{L_{accept}}{b-a} \]
3. Upper-only specification (\(X \le USL\)): \[ L_{accept}=\max(0, \min(b,USL)-a) \] \[ P(accept)=\frac{L_{accept}}{b-a} \]

Important interpretation: in a uniform model, probability is proportional to distance, not density shape peaks. Unlike normal distributions, there is no center concentration.

Worked example

Suppose your process output ranges uniformly from 8 to 12 units, and your specification is 9 to 11. Process width = 12 – 8 = 4. Acceptable overlap width = 11 – 9 = 2. Accepted fraction = 2 / 4 = 0.50 (50%). Rejected fraction = 1 – 0.50 = 0.50 (50%).

If you produce 10,000 units in a lot, expected rejects are: \[ 10,000 \times 0.50 = 5,000 \] This is an expectation under the model. Actual lot rejection can vary around that due to finite sampling effects, measurement uncertainty, and process drift.

Comparison table: tolerance coverage versus rejection (uniform case)

Acceptance window as % of process spread	Accepted fraction	Rejected fraction	Expected rejects per 10,000 parts
90%	0.90	0.10	1,000
80%	0.80	0.20	2,000
70%	0.70	0.30	3,000
60%	0.60	0.40	4,000
50%	0.50	0.50	5,000

How this compares to common capability language

Manufacturing teams frequently discuss defects in terms like PPM (parts per million) or DPMO (defects per million opportunities), often in Six Sigma frameworks that assume normality and long-term process shift conventions. Uniform-distribution rejection estimates are conceptually different, but comparison is useful for planning and communication.

Quality level (commonly cited long-term Sigma benchmark)	Approximate DPMO	Equivalent reject fraction
3 Sigma	66,807	6.6807%
4 Sigma	6,210	0.6210%
5 Sigma	233	0.0233%
6 Sigma	3.4	0.00034%

The table above uses standard industry benchmark values frequently cited in quality management training. It is not a uniform-distribution table. Instead, it highlights that if your uniform overlap is narrow, rejection rates can be dramatically larger than high-capability normal-process targets.

When a uniform model is appropriate

• Early process design phase with limited data and only known bounds.
• Mixed-source processes that produce near-flat histograms over a bounded range.
• Tolerance stack scenarios where an intermediate variable is approximately uniform.
• Stress testing and conservative risk planning when center clustering is not guaranteed.

When uniform is not a good fit

• Processes that are naturally bell-shaped around a setpoint.
• Strong skewness, multimodal behavior, drift, or autocorrelation.
• Measurement systems with censoring, clipping, or saturation artifacts.
• Situations where process control data clearly support normal, lognormal, Weibull, or empirical distribution modeling.

Practical implementation workflow in production

1. Define the quality characteristic and verify gauge capability.
2. Estimate realistic process bounds \(a\) and \(b\) from validated data, not assumptions only.
3. Set or confirm specification logic: two-sided, lower-only, or upper-only.
4. Compute overlap length and rejected fraction.
5. Convert rejection fraction into expected scrap count and cost impact per batch.
6. Run sensitivity checks by shifting limits or reducing spread.
7. Re-estimate periodically as process changes and control actions are implemented.

Common mistakes that inflate planning error

• Using wrong bounds: confusing observed sample min/max with true process limits.
• Ignoring one-sided specs: applying two-sided math to single-threshold requirements.
• Not validating units: mixing mm and microns, psi and kPa, etc.
• Assuming static behavior: not updating model after tooling wear, operator changes, or raw material shifts.
• Skipping uncertainty: treating one estimate as exact instead of showing confidence bands and scenarios.

Sensitivity insight: why reducing spread matters immediately

In a uniform model, if the accepted interval stays fixed and process spread \(b-a\) shrinks, accepted fraction rises linearly. This creates a very clear engineering target: reducing range directly improves yield. For teams deciding whether to center the process or narrow variation first, this model often shows that range reduction can produce predictable reject reduction even before advanced control strategies are deployed.

Regulatory and technical references

For deeper statistical and quality framework guidance, consult these authoritative sources:

• NIST/SEMATECH e-Handbook of Statistical Methods (nist.gov)
• FDA Process Validation Guidance for Industry (fda.gov)
• Penn State STAT 414 Probability Theory (psu.edu)

Final takeaway

To calculate the fraction of rejected parts for a uniform distribution, you only need interval geometry: identify your process bounds, measure overlap with specs, divide by total spread, and subtract from one. It is fast, transparent, and excellent for first-pass manufacturing decisions. Use this calculator to quantify current rejection, estimate batch impact, and test what-if improvements. As your data maturity increases, you can extend from uniform assumptions to richer distribution models, but this framework remains a strong baseline for clear operational decision-making.