6.3 Continuous Probability Distributions

Continuous random variables measure quantities along an unbroken continuum—such as component dimensions, process cycle times, mechanical stresses, and equipment lifetimes. Unlike discrete variables, probability for a continuous variable is defined not at a single point, but as an area under a curve. On the FE Industrial examination, continuous distributions appear across multiple topics: statistical process control (Normal distribution), equipment reliability and maintainability (Exponential and Weibull distributions), and statistical hypothesis testing (Student's $t$, Chi-square, and $F$ distributions).

---

1. Continuous Random Variables: PDF and CDF Properties

Probability Density Function (PDF)

A continuous random variable $X$ is governed by a probability density function $f(x)$ that satisfies two fundamental mathematical axioms:

1. $f(x) \ge 0$ for all real numbers $-\infty < x < \infty$
2. The total area underneath the density curve equals unity: $\int_{-\infty}^{\infty} f(x) \, dx = 1$

Important Contrast with Discrete Variables

For any continuous random variable $X$:

• The probability of taking on any single exact mathematical value $c$ is strictly zero: $P(X = c) = \int_c^c f(x) \, dx = 0$
• Consequently, including or excluding boundary endpoints does not change the probability: $P(a \le X \le b) = P(a < X \le b) = P(a \le X < b) = P(a < X < b) = \int_a^b f(x) \, dx$

Cumulative Distribution Function (CDF)

The CDF $F(x)$ represents the accumulated probability from $-\infty$ up to threshold $x$:

$F(x) = P(X \le x) = \int_{-\infty}^x f(u) \, du$

By the Fundamental Theorem of Calculus, the PDF is the first derivative of the CDF wherever the derivative exists:

$f(x) = \frac{d}{dx} F(x)$

Continuous Expectation and Variance

• Expected Value (Mean $\mu$): $E(X) = \mu = \int_{-\infty}^{\infty} x \cdot f(x) \, dx$
• Expected Value of a Function $g(X)$: $E[g(X)] = \int_{-\infty}^{\infty} g(x) \cdot f(x) \, dx$
• Variance ($\sigma^2$): $\text{Var}(X) = \sigma^2 = \int_{-\infty}^{\infty} (x - \mu)^2 f(x) \, dx = E(X^2) - [E(X)]^2 = \left( \int_{-\infty}^{\infty} x^2 f(x) \, dx \right) - \mu^2$

---

2. The Normal (Gaussian) Distribution and Linear Combinations

Density Function and the Standard Normal ($Z$)

The Normal distribution $X \sim N(\mu, \sigma^2)$ is defined by the bell-shaped probability density function:

$f(x) = \frac{1}{\sigma \sqrt{2\pi}} \exp\left( -\frac{(x - \mu)^2}{2\sigma^2} \right), \quad -\infty < x < \infty$

The curve is completely symmetric about the mean $\mu$, with inflection points occurring at $\mu - \sigma$ and $\mu + \sigma$.

To compute probabilities using standard reference tables, any normal random variable $X$ is converted to the standard normal distribution $Z \sim N(0, 1)$ via the standard score transformation:

$Z = \frac{X - \mu}{\sigma}$

The standard normal cumulative distribution function is denoted $\Phi(z) = P(Z \le z)$. Due to symmetry about zero:

• $\Phi(-z) = 1 - \Phi(z)$
• $P(-z \le Z \le z) = 2\Phi(z) - 1$

The Empirical Rule (68–95–99.7 Rule)

For any normally distributed variable:

• $\mu \pm 1\sigma$ encompasses approximately $68.27%$ of the distribution
• $\mu \pm 2\sigma$ encompasses approximately $95.45%$ of the distribution
• $\mu \pm 3\sigma$ encompasses approximately $99.73%$ of the distribution
• In Six Sigma quality engineering, specification limits placed at $\mu \pm 6\sigma$ leave only $3.4$ parts per million defective when accounting for a standard $1.5\sigma$ process mean drift.

Linear Combinations of Independent Normal Variables

In manufacturing tolerance analysis, assemblies consist of multiple mating parts. If $X_1, X_2, \dots, X_n$ are mutually independent normally distributed random variables where $X_i \sim N(\mu_i, \sigma_i^2)$, and $a_1, a_2, \dots, a_n$ and $c$ are real constants, then the linear combination:

$W = c + \sum_{i=1}^n a_i X_i = c + a_1 X_1 + a_2 X_2 + \dots + a_n X_n$

is also normally distributed ($W \sim N(\mu_W, \sigma_W^2)$) with parameters:

$\mu_W = c + \sum_{i=1}^n a_i \mu_i$ $\sigma_W^2 = \sum_{i=1}^n a_i^2 \sigma_i^2 \quad \implies \quad \sigma_W = \sqrt{\sum_{i=1}^n a_i^2 \sigma_i^2}$

Exam Watchout: Assembly Clearances and Variance Addition When finding the clearance gap between a hole and a shaft ($C = X_{\text{hole}} - X_{\text{shaft}}$), the mean is the difference: $\mu_C = \mu_{\text{hole}} - \mu_{\text{shaft}}$. However, because $a_1 = 1$ and $a_2 = -1$, the variance is: $\sigma_C^2 = (1)^2 \sigma_{\text{hole}}^2 + (-1)^2 \sigma_{\text{shaft}}^2 = \sigma_{\text{hole}}^2 + \sigma_{\text{shaft}}^2$ Variances always add. Never subtract variances, and never add standard deviations directly ($\sigma_C \neq \sigma_{\text{hole}} + \sigma_{\text{shaft}}$).

---

3. The Exponential Distribution and Memoryless Reliability Modeling

Probability Density Function, CDF, and Reliability

The exponential distribution models the time or distance between successive Poisson events, such as the operational lifespan of electronic components subjected to random operational shocks. It is parameterized by the constant failure rate $\lambda > 0$:

• PDF: $f(t) = \lambda e^{-\lambda t}$ for $t \ge 0$
• CDF: $F(t) = P(T \le t) = 1 - e^{-\lambda t}$ for $t \ge 0$
• Reliability (Survival) Function: The probability that a component functions without failure beyond time $t$: $R(t) = P(T > t) = 1 - F(t) = e^{-\lambda t}$
• Mean Time to Failure (MTTF / Mean): $\text{MTTF} = E(T) = \mu = \frac{1}{\lambda}$
• Variance: $\text{Var}(T) = \sigma^2 = \frac{1}{\lambda^2} \implies \sigma = \frac{1}{\lambda}$

The Memoryless Property

The exponential distribution is the only continuous probability distribution that possesses the memoryless property:

$P(T > s + t \mid T > s) = \frac{P(T > s + t)}{P(T > s)} = \frac{e^{-\lambda(s + t)}}{e^{-\lambda s}} = e^{-\lambda t} = P(T > t)$

Engineering Consequence: If an electronic circuit has already operated without failure for $s = 1{,}000\text{ hours}$, the conditional probability that it survives an additional $t = 500\text{ hours}$ is identical to the probability of a brand-new component surviving $500\text{ hours}$. The component does not experience wear, aging, or fatigue; failure is driven purely by memoryless, random external events.

The Hazard Rate Function (Failure Rate)

The hazard function $h(t)$ represents the instantaneous failure rate at time $t$ given survival up to time $t$:

$h(t) = \frac{f(t)}{R(t)} = \frac{f(t)}{1 - F(t)} = \frac{\lambda e^{-\lambda t}}{e^{-\lambda t}} = \lambda = \text{constant}$

A constant hazard rate directly characterizes the useful life phase of physical systems.

---

4. The Weibull Distribution and the Engineering Bathtub Curve

Two-Parameter Weibull Distribution

Mechanical, structural, and chemical components typically age, degrade, and fatigue over time, violating the constant hazard rate assumption of the exponential distribution. The Weibull distribution provides the flexibility to model non-constant failure rates via two parameters:

1. $\beta > 0$: Shape parameter (Weibull slope)
2. $\theta > 0$ (or $\alpha$ in some texts): Scale parameter (Characteristic life)

• CDF: $F(t) = 1 - \exp\left[ -\left(\frac{t}{\theta}\right)^\beta \right]$ for $t \ge 0$

• Reliability Function: $R(t) = \exp\left[ -\left(\frac{t}{\theta}\right)^\beta \right]$

• Characteristic Life Property: When operating time equals the scale parameter ($t = \theta$): $R(\theta) = e^{-1} \approx 0.3679 \quad \implies \quad F(\theta) = 1 - e^{-1} \approx 0.6321$ Exactly $63.2%$ of all units will fail by time $t = \theta$, regardless of the shape parameter $\beta$.

• Hazard Rate Function: $h(t) = \frac{f(t)}{R(t)} = \frac{\beta}{\theta} \left( \frac{t}{\theta} \right)^{\beta - 1}$

The Bathtub Curve and the Role of $\beta$

The classical engineering bathtub curve divides the lifecycle of an equipment population into three distinct failure regimes, each directly mapped to the Weibull shape parameter $\beta$:

Failure Rate h(t)
  │  \                                              / Wearout (β > 1)
  │   \                 Constant Rate              /
  │    \──────────────────────────────────────────/
  │   Infant Mortality (β < 1)     Useful Life (β = 1)
  └─────────────────────────────────────────────────────────> Operating Time t

---

5. Fundamental Sampling Distributions for Statistical Inference

Sampling distributions describe the probabilistic behavior of sample statistics (e.g., sample mean $\bar{X}$, sample variance $s^2$) computed from random samples of size $n$.

Student's $t$-Distribution

• Physical Genesis: Derived by William Sealy Gosset when estimating the population mean $\mu$ of a normal distribution when the true population variance $\sigma^2$ is unknown and replaced by the sample variance $s^2$: $t = \frac{\bar{X} - \mu}{s / \sqrt{n}} \sim t_{\nu}, \quad \text{where } \nu = n - 1 \text{ degrees of freedom}$
• Mathematical Shape: Symmetrical and bell-shaped around zero, but possessing heavier tails (higher kurtosis) than the standard normal distribution to reflect the added sampling variability introduced by estimating $\sigma$ with $s$.
• Limiting Behavior: As sample size $n \to \infty$ (degrees of freedom $\nu \to \infty$), the sample variance $s$ converges to $\sigma$, and the $t$-distribution converges identically to the standard normal distribution $Z$.

Chi-Square ($\chi^2$) Distribution

• Physical Genesis: The sum of squares of $k$ independent standard normal random variables follows a Chi-square distribution with $k$ degrees of freedom: $\chi^2 = \sum_{i=1}^k Z_i^2 \sim \chi^2_k$
• Distribution Properties:
  • Support is strictly non-negative: $0 \le \chi^2 < \infty$.
  • Positively skewed (skewed to the right); approaches normality as degrees of freedom become large.
  • Mean: $E(\chi^2) = k$
  • Variance: $\text{Var}(\chi^2) = 2k$
• Primary Applications in Industrial Engineering:
  • Inference on a single population variance: $\frac{(n - 1)s^2}{\sigma^2} \sim \chi^2_{n - 1}$.
  • Goodness-of-Fit testing (Pearson's $\chi^2$ test for comparing observed versus theoretical frequencies).
  • Contingency table analysis (tests of independence and homogeneity across categorical attributes).

Fisher-Snedecor $F$-Distribution

• Physical Genesis: The ratio of two independent chi-square random variables, each divided by its respective degrees of freedom: $F = \frac{\chi_1^2 / d_1}{\chi_2^2 / d_2} \sim F_{d_1, d_2}$ where $d_1$ is numerator degrees of freedom and $d_2$ is denominator degrees of freedom.
• Support and Shape: Strictly non-negative ($F \ge 0$) and positively skewed.
• Primary Applications in Industrial Engineering:
  • Comparing the variances of two independent normal populations: $F = \frac{s_1^2 / \sigma_1^2}{s_2^2 / \sigma_2^2}$. Under the null hypothesis $H_0: \sigma_1^2 = \sigma_2^2$, $F = s_1^2 / s_2^2$.
  • Analysis of Variance (ANOVA): Partitioning total process variation into Between-Treatment Mean Square ($MS_{\text{Treatment}}$) and Within-Error Mean Square ($MS_{\text{Error}}$): $F = MS_{\text{Treatment}} / MS_{\text{Error}}$.
  • Testing overall model significance in linear regression analysis.

---

6. Comparative Summary of Continuous and Sampling Distributions

---

7. Step-by-Step Worked Engineering Examples

Worked Example 6.3.1: Assembly Clearance Gap Tolerance Stack-Up Analysis

Problem: A critical mechanical subassembly pairs a cylindrical steel pin into a brass bushing.

• The internal diameter of the bushing is normally distributed with mean $\mu_B = 25.15\text{ mm}$ and standard deviation $\sigma_B = 0.04\text{ mm}$.
• The external diameter of the pin is normally distributed with mean $\mu_P = 25.05\text{ mm}$ and standard deviation $\sigma_P = 0.03\text{ mm}$.
• Bushing and pin dimensions are statistically independent.

Assembly clearance is defined as $C = X_B - X_P$.

1. Determine the probability distribution, mean, and standard deviation of clearance $C$.
2. Compute the probability of an interference fit (where the pin is larger than the bushing, meaning $C < 0$).

Solution:

1. Identify the Distribution of Clearance $C$:
   • Since $C$ is a linear combination of independent normal random variables ($C = 1 \cdot X_B - 1 \cdot X_P$), $C$ is itself normally distributed.
   • Mean clearance: $\mu_C = \mu_B - \mu_P = 25.15\text{ mm} - 25.05\text{ mm} = 0.10\text{ mm}$
   • Variance of clearance (coefficients are $a_1 = 1$ and $a_2 = -1$): $\sigma_C^2 = (1)^2 \sigma_B^2 + (-1)^2 \sigma_P^2 = (0.04)^2 + (0.03)^2 = 0.0016 + 0.0009 = 0.0025\text{ mm}^2$
   • Standard deviation of clearance: $\sigma_C = \sqrt{0.0025} = 0.05\text{ mm}$
   • Therefore, $C \sim N(0.10, 0.05^2)$.
2. Compute Probability of Interference ($C < 0$):
   • Standardize to find the $Z$-score for $C = 0$: $Z = \frac{0 - \mu_C}{\sigma_C} = \frac{0 - 0.10}{0.05} = -2.00$
   • Using the standard normal cumulative table: $P(C < 0) = P(Z < -2.00) = \Phi(-2.00) = 1 - \Phi(2.00)$ From the standard normal table, $\Phi(2.00) = 0.9772$: $P(C < 0) = 1 - 0.9772 = 0.0228 \approx 2.28\%$
3. Engineering Conclusion: Exactly $2.28%$ of the assembled pairs will experience an interference fit requiring rework or mechanical press-fitting.

Worked Example 6.3.2: Hydraulic Pump Reliability and Mission Success Probability

Problem: An industrial hydraulic pump operating on an automated stamping press has a time-to-failure governed by an exponential distribution with a mean time to failure (MTTF) of $\mu = 4{,}000\text{ operating hours}$.

1. What is the failure rate $\lambda$ per hour?
2. What is the probability that the pump survives a $1{,}000\text{-hour}$ production run without failure?
3. If the pump has already operated successfully for $3{,}000\text{ hours}$ without incident, what is the conditional probability that it survives an additional $1{,}000\text{ hours}$?

Solution:

1. Calculate the Failure Rate $\lambda$: $\lambda = \frac{1}{\text{MTTF}} = \frac{1}{4{,}000} = 0.00025\text{ failures/hour}$
2. Surviving a $1{,}000\text{-Hour}$ Mission ($R(1{,}000)$): $R(1{,}000) = P(T > 1{,}000) = e^{-\lambda t} = e^{-(0.00025)(1{,}000)} = e^{-0.25}$ $e^{-0.25} \approx 0.77880 \approx 77.88\%$
3. Conditional Survival Given Past Operation ($3{,}000\text{ hours}$):
   • By the memoryless property of the exponential distribution: $P(T > 3{,}000 + 1{,}000 \mid T > 3{,}000) = P(T > 1{,}000) = e^{-0.25} \approx 0.77880$
   • Verification via Conditional Probability Definition: $P(T > 4{,}000 \mid T > 3{,}000) = \frac{P(T > 4{,}000)}{P(T > 3{,}000)} = \frac{e^{-0.00025 \times 4{,}000}}{e^{-0.00025 \times 3{,}000}} = \frac{e^{-1.0}}{e^{-0.75}} = e^{-0.25} \approx 77.88\%$
4. Final Result: The probability of surviving the next $1{,}000$ hours remains exactly $77.88%$, entirely unaffected by the previous $3{,}000$ hours of service.

---

8. NCEES Reference Handbook Tips & Realistic Exam Traps

• Standard Deviation Addition Error: When computing the variance of a linear combination $W = a_1 X_1 + a_2 X_2$, candidates routinely make the fatal error of writing $\sigma_W = a_1 \sigma_1 + a_2 \sigma_2$. Standard deviations are geometric vectors; only variances add algebraically ($\sigma_W = \sqrt{a_1^2 \sigma_1^2 + a_2^2 \sigma_2^2}$). Always compute variance first, then take the square root.
• Table Lookup Conventions in the NCEES Handbook: Check the diagram at the head of the Normal distribution table in the Reference Handbook. The NCEES table displays the cumulative probability $\Phi(z) = P(Z \le z)$ from $-\infty$ to $z$ (or in some editions, the upper-tail area $P(Z > z)$). If you confuse cumulative area with upper-tail area, your probability will equal $1 - p$, which is guaranteed to be one of the three distractor choices.
• Exponential Parameter Inversion: Carefully distinguish between rate parameter $\lambda$ (events per unit time) and mean parameter $\theta = 1/\lambda$ (time per event). If a problem specifies "mean life is 500 hours", then $\mu = 500$ and $\lambda = 1/500 = 0.002$. Substituting $500$ in place of $\lambda$ in $e^{-\lambda t}$ yields $e^{-500 t} = 0$, ruining the calculation.
• Weibull Shape Identification Shortcut: For fast points on conceptual questions:
  • $\beta < 1 \implies$ Infant mortality / decreasing failure rate (burn-in period).
  • $\beta = 1 \implies$ Exponential distribution / constant failure rate (useful life).
  • $\beta > 1 \implies$ Wearout / increasing failure rate (aging and fatigue).