2.1 Probability Distributions, Central Tendency, and Regression

Process Variation and Central Tendency

In chemical engineering, process variables such as temperature, pressure, flow rate, and concentration are subject to random fluctuations due to environmental changes, measurement limitations, and inherent system dynamics. To analyze and optimize processes, chemical engineers rely on probability distributions to model these variations. For example, in a polymerization reactor, monitoring the variance of temperature is as critical as monitoring its mean, as temperature spikes can trigger runaway reactions.

Engineers summarize raw data sets using measures of central tendency (where the data centers) and dispersion (how spread out the data is). The NCEES FE Reference Handbook provides the following standard definitions:

• Arithmetic Mean: The average value, representing the center of gravity of the data. For a sample of size $n$, the mean is $\bar{X} = \frac{1}{n} \sum_{i=1}^{n} X_i$. For a population of size $N$, the mean is $\mu = \frac{1}{N} \sum_{i=1}^{N} X_i$.
• Median: The middle value when the data set is arranged in ascending order. For an odd number of observations, it is the middle data point; for an even number, it is the average of the two middle points. The median is robust against extreme outliers.
• Mode: The value that occurs with the highest frequency in a data set. A distribution can be unimodal, bimodal, or multimodal.

Dispersion and Discrete vs. Continuous Distributions

To quantify the variability or dispersion about the mean, we calculate variance and standard deviation:

• Sample Variance ($s^2$): Quantifies the spread of a sample around the mean. The divisor is $n-1$ (degrees of freedom) to ensure an unbiased estimator of the population variance: $s^2 = \frac{1}{n-1} \sum_{i=1}^{n} (X_i - \bar{X})^2 = \frac{1}{n-1} \left( \sum_{i=1}^{n} X_i^2 - n\bar{X}^2 \right)$
• Sample Standard Deviation ($s$): The square root of the sample variance: $s = \sqrt{s^2}$.
• Population Standard Deviation ($\sigma$): Used when the entire population of size $N$ is known: $\sigma = \sqrt{\frac{1}{N} \sum_{i=1}^{N} (X_i - \mu)^2}$

A probability distribution describes how probabilities are distributed over the values of a random variable. Random variables are categorized as either discrete or continuous:

• Discrete Random Variables: Take on distinct, countable values. Examples include the number of failed valves in a process plant or the count of batch reactions that meet specifications. The probability distribution of a discrete random variable is defined by a Probability Mass Function (PMF), denoted as $P(X = x)$ or $p(x)$, which satisfies: $\sum p(x) = 1 \quad \text{and} \quad 0 \le p(x) \le 1$
• Continuous Random Variables: Take on any value within a continuous interval. Examples include reactor temperature or product density. The probability that $X$ lies within an interval $[a, b]$ is the area under the Probability Density Function (PDF), denoted as $f(x)$: $P(a \le X \le b) = \int_{a}^{b} f(x) \, dx$ The PDF must satisfy $\int_{-\infty}^{\infty} f(x) , dx = 1$ and $f(x) \ge 0$ for all $x$.
• Cumulative Distribution Function (CDF): For both types of variables, the CDF, $F(x)$, represents the probability that the random variable is less than or equal to $x$: $F(x) = P(X \le x)$ For continuous variables, $F(x) = \int_{-\infty}^{x} f(t) , dt$, and by the Fundamental Theorem of Calculus, the PDF is the derivative of the CDF: $f(x) = \frac{dF(x)}{dx}$.

The Binomial Distribution in Process Safety

The binomial distribution is a discrete distribution modeling the number of successes, $x$, in a fixed number of independent trials, $n$, where each trial has only two mutually exclusive outcomes: success (with probability $p$) or failure (with probability $1-p$). In chemical plant safety, this is frequently used to model redundant systems (e.g., the probability of multiple relief valves or interlocks failing simultaneously).

The probability of obtaining exactly $x$ successes in $n$ independent trials is: $P(X = x) = \binom{n}{x} p^x (1-p)^{n-x} = \frac{n!}{x!(n-x)!} p^x (1-p)^{n-x}$ The mean and variance of a binomially distributed random variable are: $\mu = np \quad \text{and} \quad \sigma^2 = np(1-p)$

Consider an overpressure safety system with 4 independent redundant pressure relief valves. If the probability of any single valve failing to open when pressure exceeds the limit is $0.05$, the system fails only if at least 2 valves fail to open. We can calculate this probability using the binomial distribution: $P(X \ge 2) = 1 - P(X = 0) - P(X = 1)$. Evaluating each term, we find $P(X = 0) = \binom{4}{0}(0.05)^0(0.95)^4 \approx 0.8145$ and $P(X = 1) = \binom{4}{1}(0.05)^1(0.95)^3 \approx 0.1715$. The probability that at least 2 valves fail is $1 - 0.8145 - 0.1715 = 0.0140$ or $1.40%$, which represents a significant safety improvement over a single valve system.

The Normal Distribution and Process Analysis

The normal distribution is the most important continuous distribution in engineering because of the Central Limit Theorem, which states that the sum or average of a large number of independent random variables will tend toward a normal distribution, regardless of their underlying distribution. This explains why process noise, temperature variations, and measurement errors often follow a Gaussian profile.

The PDF of a normal distribution is symmetric about its mean $\mu$ and is given by: $f(x) = \frac{1}{\sigma \sqrt{2\pi}} e^{-\frac{1}{2} \left( \frac{x-\mu}{\sigma} \right)^2}$ To find probabilities, engineers convert the random variable $X$ to the standard normal variable $Z$ (which has a mean of 0 and a standard deviation of 1) using the z-score formula: $Z = \frac{X - \mu}{\sigma}$ Using the Standard Normal Cumulative Distribution Table in the FE Reference Handbook, the probability that $X$ is less than a value $x$ is $P(X \le x) = P\left(Z \le \frac{x - \mu}{\sigma}\right) = F(z)$. Due to symmetry, $P(Z > z) = 1 - F(z) = F(-z)$.

For example, if a distillation column overhead temperature is normally distributed with a mean of $78.4^\circ\text{C}$ and a standard deviation of $0.6^\circ\text{C}$, the probability of the temperature exceeding $79.6^\circ\text{C}$ is found by computing $z = (79.6 - 78.4) / 0.6 = 2.0$. The standard normal table gives $F(2.0) = 0.9772$. Thus, the probability of exceeding $79.6^\circ\text{C}$ is $1 - 0.9772 = 0.0228$ or $2.28%$.

Confidence Intervals for Process Parameters

A point estimate (such as the sample mean $\bar{X}$) does not convey the uncertainty of the estimate. A confidence interval provides an range within which the true population parameter is expected to lie with a specified probability ($1-\alpha$, where $\alpha$ is the significance level).

• Case 1: Population Standard Deviation $\sigma$ is Known (or $n \ge 30$): We use the standard normal distribution ($Z$): $\bar{X} - z_{\alpha/2} \frac{\sigma}{\sqrt{n}} \le \mu \le \bar{X} + z_{\alpha/2} \frac{\sigma}{\sqrt{n}}$ For a 95% confidence level, $\alpha = 0.05$, $\alpha/2 = 0.025$, and $z_{0.025} = 1.96$.
• Case 2: Population Standard Deviation $\sigma$ is Unknown and $n < 30$: We must use the Student's t-distribution, which accounts for the additional uncertainty of estimating the standard deviation from a small sample: $\bar{X} - t_{\alpha/2, \nu} \frac{s}{\sqrt{n}} \le \mu \le \bar{X} + t_{\alpha/2, \nu} \frac{s}{\sqrt{n}}$ where $\nu = n-1$ represents the degrees of freedom. The critical value $t_{\alpha/2, \nu}$ is obtained from the t-distribution table in the FE Reference Handbook.

For instance, if a pilot plant runs 16 batch reactions and measures a mean concentration of $24.5\text{ g/L}$ with a sample standard deviation of $1.2\text{ g/L}$, the degrees of freedom are $\nu = 15$. For a 95% confidence level, $\alpha/2 = 0.025$, and $t_{0.025, 15} = 2.131$. The margin of error is $2.131 \times (1.2 / \sqrt{16}) = 0.6393\text{ g/L}$, giving a confidence interval of $23.86$ to $25.14\text{ g/L}$.

Linear Regression and Calibration Curves

Engineers frequently perform curve fitting to relate process variables, such as correlating reaction rate constants with temperature (Arrhenius equation) or calibrating instrumentation. The method of least squares minimizes the sum of the squared residuals between the observed values and the fitted line $y = a + bx$, where $b$ (or $\hat{\beta}_1$) is the slope and $a$ (or $\hat{\beta}_0$) is the y-intercept. The FE Reference Handbook defines these parameters as: $b = \frac{S_{xy}}{S_{xx}} \quad \text{and} \quad a = \bar{y} - b\bar{x}$ where: $S_{xx} = \sum X_i^2 - \frac{(\sum X_i)^2}{n} \quad \text{and} \quad S_{xy} = \sum X_i Y_i - \frac{\sum X_i \sum Y_i}{n}$ The strength of the linear relationship is quantified by the sample correlation coefficient $r$: $r = \frac{S_{xy}}{\sqrt{S_{xx} S_{yy}}} \quad \text{where} \quad S_{yy} = \sum Y_i^2 - \frac{(\sum Y_i)^2}{n}$ The value of $r$ ranges from $-1$ to $+1$. The coefficient of determination, $R^2 = r^2$, represents the proportion of the variance in the dependent variable that is predictable from the independent variable. In chromatography, an $R^2 > 0.99$ is typically required to ensure that concentration is linearly related to the detector peak area.