Reference ID: MET-1F8D | Process Engineering Reference Sheets Calculation Guide
Introduction & Context
Theoretical Random Variance quantifies the minimum statistical scatter expected when a perfectly‑mixed binary powder is sampled. In Process Engineering this value is used as a benchmark: any experimental variance larger than the theoretical floor indicates imperfect mixing, biased sampling, or measurement error. For a deeper dive into how this variance relates to overall blend quality, see our mixing index calculation methodology. Typical applications include validating powder blenders, setting quality‑control limits for pharmaceutical granules, and troubleshooting segregation in pneumatic transfer lines.
Methodology & Formulas
The derivation follows a binomial (Bernoulli) model in which every inspected particle is either component A or not-A. For homogeneous mixing the expected variance of the sample mass fraction is:
\[ \sigma_r^{2} = \frac{p\,(1-p)}{n} \]
and its square root
\[ \sigma_r = \sqrt{\frac{p\,(1-p)}{n}} \]
where
\( p \) (-) = mass fraction of one component (dimensionless range 0–1)
\( n \) (-) = number of particles actually inspected (dimensionless)
\( \sigma_r \) (-) = standard deviation in mass-fraction units (dimensionless)
In code form (mirrored exactly):
sigma_r_squared = p * (1 - p) / n
sigma_r = math.sqrt(sigma_r_squared)
Conditions for normal approximation to binomial distribution
Parameter
Mathematical Condition
np
np ≥ 5
n(1-p)
n(1−p) ≥ 5
Note: The binomial model itself is valid for any \( n \) and \( p \), but these conditions ensure the normal approximation is adequate for statistical tests and confidence intervals.
The relative standard deviation (optional output) is computed only when \( p > 0 \):
\[ \mathrm{RSD} = \frac{\sigma_r}{p} \]
rsd = sigma_r / max(p, 1e-9)
All variables remain dimensionless; no unit conversions are required.
It quantifies the expected dispersion of a measurement that is only influenced by random error. Engineers compare this baseline to observed variance to decide whether additional assignable causes (drifts, fouling, calibration shifts, feed-quality swings) have entered the process.
For continuous flow data, theoretical random variance is calculated using different models, as the binomial model applies to discrete particle counts. Common approaches include:
Poisson model: Variance = Mean when counts are collected at fixed time intervals.
Additive white noise: Variance = sensor noise specification squared.
First-order plus dead-time: Variance ≈ (steady-state gain)² × white noise variance ÷ (2 × time constant); valid when sampling faster than one-tenth of the time constant.
Re-estimate when:
Maintenance replaces or recalibrates a sensor.
Upstream unit redesign alters residence time or mixing.
Variance shifts detectable by a control chart >1 σ beyond the previous baseline occur for 8 consecutive points.
Feed-stream sampling hardware changes.
Otherwise, recalculate annually during scheduled turnarounds.
Verify sensor damping is set to supplier optimum; over-damping hides but does not remove variance.
Reduce sampling interval to capture high-frequency disturbances for root-cause isolation.
Inspect heat-exchanger fouling and clean; temperature oscillations often amplify variance in flow-based loops.
Apply dead-band control on fast proportional-only loops when process gain is low (<0.2) and product quality is insensitive.
Worked Example: Theoretical Random Variance in a Perfectly-Mixed Binary Powder
In a confectionery manufacturing process, a binary powder mixture of white sugar and cocoa is produced with a target composition of 50% by weight for each component. To assess the inherent randomness in sample composition due to ideal mixing, a quality control technician inspects a sample of 1000 particles, assuming the system is perfectly mixed.
Mass fraction of white sugar, \( p = 0.5 \) (dimensionless)
Number of particles inspected in the sample, \( n = 1000 \) (dimensionless)
The theoretical random variance for a perfectly-mixed binary system is given by the binomial formula: \( \sigma_r^2 = \frac{p(1-p)}{n} \).
Substitute the known values into the formula: \( \sigma_r^2 = \frac{0.5 \times (1 - 0.5)}{1000} = \frac{0.5 \times 0.5}{1000} \).
From the provided numerical results, the variance is \( \sigma_r^2 = 2.500 \times 10^{-4} \).
The standard deviation is the square root of the variance: \( \sigma_r = \sqrt{\sigma_r^2} \). From the results, \( \sigma_r = 0.016 \).
The relative standard deviation (RSD) indicates the scatter relative to the mean composition: \( RSD = \frac{\sigma_r}{p} \). From the results, \( RSD = 0.032 \), equivalent to 3.2%.
Final Answer: For the perfectly-mixed binary powder, the theoretical random variance is \( \sigma_r^2 = 2.500 \times 10^{-4} \), with a standard deviation of \( \sigma_r = 0.016 \) and a relative standard deviation of 3.2%. All values are dimensionless as per the model assumptions.
"Un projet n'est jamais trop grand s'il est bien conçu."— André Citroën
"La difficulté attire l'homme de caractère, car c'est en l'étreignant qu'il se réalise."— Charles de Gaulle
