13.1 Reliability Modeling, Failure Rates (lambda) & Life Distributions (Exponential, Weibull)

13.1 Reliability Modeling, Failure Rates (λ) & Life Distributions

Reliability engineering provides the quantitative framework for predicting, assessing, and optimizing the operational life cycle of components, production equipment, and complex industrial systems. In modern manufacturing and systems engineering, understanding failure probability over time is essential for establishing safety envelopes, warranty periods, spare parts provisioning, and condition-based maintenance policies.

---

1. Fundamentals of Reliability Mathematics

Let the operating time to failure of an engineering component or system be modeled as a continuous, non-negative random variable $T \ge 0$, characterized by a probability density function (PDF) denoted by $f(t)$.

Reliability Function $R(t)$

The Reliability function (also termed the Survival function $S(t)$) defines the probability that the asset functions without failure beyond operating time $t$ under specified operating and environmental conditions:

$R(t) = P(T > t) = \int_t^\infty f(u)\,du$

Boundary conditions require:

• At time zero: $R(0) = 1$ (the asset is assumed operational at start of mission).
• As time approaches infinity: $\lim_{t \to \infty} R(t) = 0$ (all physical systems eventually fail).
• Monotonicity: $\frac{dR(t)}{dt} \le 0$ for all $t \ge 0$.

Cumulative Failure Probability $F(t)$

The Cumulative Failure Distribution function (or Unreliability function) represents the complementary cumulative probability that the system fails at or before operating time $t$:

$F(t) = P(T \le t) = 1 - R(t) = \int_0^t f(u)\,du$

Probability Density Function $f(t)$

The failure probability density function $f(t)$ represents the instantaneous failure probability per unit time in the interval $[t, t + dt]$:

$f(t) = \frac{dF(t)}{dt} = -\frac{dR(t)}{dt}$

Mean Time To Failure (MTTF)

For non-repairable assets, the expected operational life is the Mean Time To Failure (MTTF). Applying integration by parts over the domain $[0, \infty)$:

$\text{MTTF} = E[T] = \int_0^\infty t f(t)\,dt = \left[ -t R(t) \right]_0^\infty + \int_0^\infty R(t)\,dt$

Assuming $\lim_{t \to \infty} t R(t) = 0$, the standard identity simplifies to the integral of the reliability function:

$\text{MTTF} = \int_0^\infty R(t)\,dt$

For repairable systems undergoing renewal or replacement upon breakdown, this metric is expressed as the Mean Time Between Failures (MTBF).

---

2. Hazard Rate and Instantaneous Failure Rate Function

A critical distinction in reliability theory exists between the absolute failure density $f(t)$ and the conditional failure intensity given survival to time $t$. The hazard rate function $h(t)$ (also designated instantaneous failure rate $z(t)$ or $\lambda(t)$) represents the conditional probability per unit time that an item fails in the infinitesimal interval $(t, t + \Delta t]$, given that it survived up to time $t$:

$h(t) = \lim_{\Delta t \to 0} \frac{P(t < T \le t + \Delta t \mid T > t)}{\Delta t}$

Applying conditional probability $P(A \mid B) = \frac{P(A \cap B)}{P(B)}$:

$P(t < T \le t + \Delta t \mid T > t) = \frac{F(t + \Delta t) - F(t)}{R(t)}$

Taking the limit as $\Delta t \to 0$:

$h(t) = \frac{1}{R(t)} \lim_{\Delta t \to 0} \frac{F(t + \Delta t) - F(t)}{\Delta t} = \frac{f(t)}{R(t)} = -\frac{1}{R(t)}\frac{dR(t)}{dt} = -\frac{d}{dt} \ln R(t)$

Derivation of the General Reliability Relationship

Integrating both sides of $h(u) = -\frac{d}{du} \ln R(u)$ from $0$ to $t$:

$\int_0^t h(u)\,du = -\int_0^t d(\ln R(u)) = -\left[\ln R(t) - \ln R(0)\right]$

Because $R(0) = 1$, $\ln R(0) = 0$. Defining the Cumulative Hazard Function as $H(t) = \int_0^t h(u)\,du$:

$\ln R(t) = -\int_0^t h(u)\,du = -H(t)$

Exponentiating both sides yields the foundational equation connecting reliability directly to the hazard rate:

$R(t) = \exp\left( -\int_0^t h(u)\,du \right) = e^{-H(t)}$

Consequently, the failure density function can be expressed purely in terms of the hazard rate:

$f(t) = h(t) \exp\left( -\int_0^t h(u)\,du \right) = h(t) e^{-H(t)}$

---

3. The Classic Bathtub Curve

In physical and mechanical systems, the instantaneous hazard rate $h(t)$ across an asset's full operating life typically exhibits three distinct chronological regimes, forming the classic Bathtub Curve.

  Hazard Rate h(t)
    ^
    |
    | \                                          /
    |  \                                        /
    |   \______________________________________/
    |    <-- Infant --> <--- Useful Life ---> <-- Wear-out -->
    |    Mortality (DFR)    Constant (CFR)          (IFR)
    +----------------------------------------------------> Time (t)

Strategic Engineering Implications

1. During Phase I, scheduled preventive maintenance is counterproductive. Disassembling assets experiencing early infant mortality introduces human maintenance-induced errors, resetting components into infant mortality failure risk.
2. During Phase II, the asset does not experience age-related degradation. The failure probability in any subsequent hour is entirely independent of prior operational service time.
3. During Phase III, failure intensity escalates rapidly. Timely scheduled overhaul or component retirement based on cumulative operating cycles or run hours prevents catastrophic in-service breakdowns.

---

4. The Exponential Distribution (Constant Hazard Rate)

The Exponential Distribution represents the foundational life distribution in reliability modeling, characterizing assets operating in Phase II of the bathtub curve where failure is governed strictly by random, Poisson-distributed shocks.

Mathematical Formulation

Setting the hazard rate to a time-invariant constant $h(t) = \lambda > 0$:

$H(t) = \int_0^t \lambda\,du = \lambda t$

$R(t) = e^{-\lambda t}$

$F(t) = 1 - e^{-\lambda t}$

$f(t) = \lambda e^{-\lambda t}$

Expected Life and Variance

$\text{MTTF} = \int_0^\infty e^{-\lambda t}\,dt = \left[ -\frac{1}{\lambda} e^{-\lambda t} \right]_0^\infty = \frac{1}{\lambda}$

$\text{Var}(T) = \frac{1}{\lambda^2}, \quad \sigma = \frac{1}{\lambda}$

The Memoryless Property

A continuous random variable $T$ is memoryless if and only if for all $t > 0$ and $s > 0$:

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

Physical Meaning: A component operating under a constant failure rate that has run without failure for $s = 5{,}000$ hours has the exact same probability of surviving the next $t = 1{,}000$ hours as a brand new component fresh off the production line. Wear, usage, and age do not accumulate. Consequently, replacing unfailed exponential components based on calendar time or operational hours wastes capital and provides zero reliability benefit.

Reliability at $t = \text{MTTF}$

Evaluating the reliability function at an operating time equal to the mean life $t = 1/\lambda$:

$R(\text{MTTF}) = e^{-\lambda(1/\lambda)} = e^{-1} = \frac{1}{2.71828} \approx 0.367879 \quad (36.79\%)$

Key Takeaway: An asset governed by an exponential life distribution has only a 36.8% chance of surviving its full MTTF. Conversely, 63.2% of such components fail before reaching MTTF.

---

5. The Weibull Distribution (Time-Dependent Hazard)

Developed by Waloddi Weibull in 1951, the Weibull distribution is the most versatile parametric distribution in reliability engineering because its hazard function can model decreasing, constant, or increasing failure rates depending on the value of its shape parameter.

Two-Parameter Weibull Equations

The two-parameter Weibull reliability function is parameterized by shape parameter $\beta > 0$ (dimensionless) and scale parameter $\theta > 0$ (characteristic life, having units of time):

$R(t) = \exp\left( -\left(\frac{t}{\theta}\right)^\beta \right)$

$F(t) = 1 - \exp\left( -\left(\frac{t}{\theta}\right)^\beta \right)$

Taking the derivative gives the probability density function $f(t)$:

$f(t) = -\frac{dR(t)}{dt} = \frac{\beta}{\theta} \left( \frac{t}{\theta} \right)^{\beta - 1} \exp\left( -\left(\frac{t}{\theta}\right)^\beta \right)$

The instantaneous hazard rate $h(t) = f(t)/R(t)$ simplifies directly to:

$h(t) = \frac{\beta}{\theta} \left( \frac{t}{\theta} \right)^{\beta - 1}$

Physical Interpretation of the Shape Parameter $\beta$

The exponent $(\beta - 1)$ in $h(t)$ directly governs the slope of the hazard function:

1. $\beta < 1$ (Decreasing Failure Rate, DFR): Infant mortality regime. Hazard rate declines over time ($h'(t) < 0$). Assets weed out early defects. Burn-in screening is advantageous.
2. $\beta = 1$ (Constant Failure Rate, CFR): Hazard rate reduces to $h(t) = \frac{1}{\theta} = \lambda$. The Weibull reduces identically to the Exponential distribution with $\text{MTTF} = \theta$.
3. $\beta > 1$ (Increasing Failure Rate, IFR): Progressive wear-out regime ($h'(t) > 0$). Assets degrade with operating cycles, friction, fatigue, or aging. Preventive replacement strategies are highly effective.
   • $\beta = 2$: Rayleigh distribution, where the hazard rate increases linearly with time: $h(t) = \frac{2t}{\theta^2}$.
   • $\beta \approx 3.44$: The Weibull distribution exhibits zero skewness, closely approximating a symmetrical Gaussian (Normal) distribution with mean $\mu \approx 0.90\theta$ and standard deviation $\sigma \approx 0.29\theta$.
   • $\beta \ge 4$: High wear-out steepness, characteristic of severe mechanical fatigue, rapid bearing degradation, or thermal embrittlement.

Characteristic Life $\theta$

Evaluating the Weibull reliability function at $t = \theta$:

$R(\theta) = \exp\left( -\left(\frac{\theta}{\theta}\right)^\beta \right) = e^{-1^\beta} = e^{-1} \approx 0.367879 \quad (36.79\%)$

Regardless of the value of shape parameter $\beta$, the scale parameter $\theta$ is always the operational age at which 63.2% of the asset population has failed and 36.8% remains operational. For this reason, $\theta$ is formally defined as the Characteristic Life.

Mean Time To Failure for the Weibull Model

The expected life involves the Euler Gamma function $\Gamma(x) = \int_0^\infty u^{x-1} e^{-u}\,du$:

$\text{MTTF} = \theta\,\Gamma\left( 1 + \frac{1}{\beta} \right)$

$\text{Var}(T) = \theta^2 \left[ \Gamma\left( 1 + \frac{2}{\beta} \right) - \left( \Gamma\left( 1 + \frac{1}{\beta} \right) \right)^2 \right]$

Three-Parameter Weibull (Guaranteed Operating Life)

When an asset exhibits zero probability of failure prior to a threshold operating duration $\gamma > 0$ (the location parameter or failure-free life), the distribution shifts rightward for $t \ge \gamma$:

$R(t) = \exp\left( -\left(\frac{t - \gamma}{\theta}\right)^\beta \right), \quad t \ge \gamma$

$h(t) = \frac{\beta}{\theta} \left( \frac{t - \gamma}{\theta} \right)^{\beta - 1}, \quad t \ge \gamma$

---

6. Normal and Lognormal Life Distributions

While the Exponential and Weibull distributions dominate standard reliability modeling, structural fatigue and electronic degradation frequently adhere to Normal and Lognormal behaviors.

Normal (Gaussian) Life Distribution

The Normal distribution models degradation where total wear results from the additive accumulation of many small, independent physical increments (Central Limit Theorem):

$f(t) = \frac{1}{\sigma \sqrt{2\pi}} \exp\left( -\frac{(t - \mu)^2}{2\sigma^2} \right)$

$R(t) = 1 - \Phi\left( \frac{t - \mu}{\sigma} \right)$

Where $\Phi(z)$ is the standard normal cumulative distribution function. The hazard rate $h(t) = f(t)/R(t)$ for a Normal distribution increases monotonically with operating time ($h'(t) > 0$), making it suitable for mechanical structural wear-out with fixed service lives.

Lognormal Life Distribution

The Lognormal distribution governs failure processes where degradation is multiplicative rather than additive. If the random variable $T$ is lognormally distributed, then $Y = \ln(T)$ is normally distributed: $Y \sim N(\mu_{\ln}, \sigma_{\ln}^2)$.

$f(t) = \frac{1}{t \sigma_{\ln} \sqrt{2\pi}} \exp\left( -\frac{(\ln t - \mu_{\ln})^2}{2\sigma_{\ln}^2} \right), \quad t > 0$

$R(t) = 1 - \Phi\left( \frac{\ln t - \mu_{\ln}}{\sigma_{\ln}} \right)$

Engineering Applications of the Lognormal Distribution

• Fatigue crack growth: Linear elastic fracture mechanics (Paris' Law) exhibits exponential crack propagation rates $da/dN = C(\Delta K)^m$, leading to multiplicative crack enlargement.
• Semiconductor physics: Electromigration in metallic interconnects (governed by Black's Equation), dielectric breakdown, and corrosion kinetics.
• Hazard rate profile: Unlike the Weibull IFR, the Lognormal hazard rate starts at zero, climbs to a peak, and then decreases asymptotically toward zero at very large $t$. The physical rationale is that surviving units past the peak represent an exceptionally robust subpopulation.

7. Worked Numerical Example: Comparative Life Analytics

Problem Formulation

A high-reliability variable-speed feedwater pump installed in an industrial thermal plant has a characteristic life of $\theta = 8{,}000$ operating hours. A reliability engineering team must evaluate the pump's survival metrics over an operating mission of $t_1 = 2{,}000$ hours and an extended operating cycle of $t_2 = 10{,}000$ hours under two competing lifecycle models:

1. Model A (Exponential Model): Assumes constant failure rate throughout life with $\text{MTTF} = 8{,}000$ hours, so $\lambda = 1/8{,}000 = 1.25 \times 10^{-4}\text{ failures/hour}$.
2. Model B (Weibull Wear-Out Model): Assumes mechanical bearing wear-out with shape parameter $\beta = 2.5$ and scale parameter $\theta = 8{,}000$ hours.

Determine for both models:

• The reliability $R(t)$ at $t = 2{,}000$ hr and $t = 10{,}000$ hr.
• The cumulative failure probability $F(t)$ at both horizons.
• The instantaneous hazard rate $h(t)$ at both horizons.

---

Step-by-Step Solution

Part 1: Model A (Exponential Distribution, $\lambda = 1.25 \times 10^{-4}\text{ hr}^{-1}$)

At $t_1 = 2{,}000$ hours:

• Operating exponent: $\lambda t_1 = (1.25 \times 10^{-4}\text{ hr}^{-1})(2{,}000\text{ hr}) = 0.250$
• Reliability: $R_A(2{,}000) = e^{-0.250} = 0.778801 \implies 77.88\%$
• Cumulative failure probability: $F_A(2{,}000) = 1 - 0.778801 = 0.221199 \implies 22.12\%$
• Instantaneous hazard rate (constant for all $t$): $h_A(2{,}000) = \lambda = 1.250 \times 10^{-4}\text{ failures/hour}$

At $t_2 = 10{,}000$ hours:

• Operating exponent: $\lambda t_2 = (1.25 \times 10^{-4}\text{ hr}^{-1})(10{,}000\text{ hr}) = 1.250$
• Reliability: $R_A(10{,}000) = e^{-1.250} = 0.286505 \implies 28.65\%$
• Cumulative failure probability: $F_A(10{,}000) = 1 - 0.286505 = 0.713495 \implies 71.35\%$
• Instantaneous hazard rate: $h_A(10{,}000) = \lambda = 1.250 \times 10^{-4}\text{ failures/hour}$

---

Part 2: Model B (Weibull Distribution, $\beta = 2.5, \theta = 8{,}000\text{ hr}$)

At $t_1 = 2{,}000$ hours:

• Normalized time ratio: $\frac{t_1}{\theta} = \frac{2{,}000}{8{,}000} = 0.250$
• Power term: $\left(\frac{t_1}{\theta}\right)^\beta = (0.250)^{2.5} = (0.250)^2 \times \sqrt{0.250} = 0.0625 \times 0.500 = 0.03125$
• Reliability: $R_B(2{,}000) = \exp(-0.03125) = e^{-0.03125} = 0.969233 \implies 96.92\%$
• Cumulative failure probability: $F_B(2{,}000) = 1 - 0.969233 = 0.030767 \implies 3.08\%$
• Instantaneous hazard rate: $h_B(t) = \frac{\beta}{\theta} \left( \frac{t}{\theta} \right)^{\beta - 1}$ $\frac{\beta}{\theta} = \frac{2.5}{8{,}000} = 3.125 \times 10^{-4}\text{ hr}^{-1}$ $\left(\frac{t_1}{\theta}\right)^{\beta - 1} = (0.250)^{1.5} = 0.250 \times \sqrt{0.250} = 0.125$ $h_B(2{,}000) = (3.125 \times 10^{-4})(0.125) = 3.90625 \times 10^{-5}\text{ failures/hour}$

At $t_2 = 10{,}000$ hours:

• Normalized time ratio: $\frac{t_2}{\theta} = \frac{10{,}000}{8{,}000} = 1.250$
• Power term: $\left(\frac{t_2}{\theta}\right)^\beta = (1.250)^{2.5} = (1.250)^2 \times \sqrt{1.250} = 1.5625 \times 1.118034 = 1.746928$
• Reliability: $R_B(10{,}000) = \exp(-1.746928) = e^{-1.746928} = 0.174308 \implies 17.43\%$
• Cumulative failure probability: $F_B(10{,}000) = 1 - 0.174308 = 0.825692 \implies 82.57\%$
• Instantaneous hazard rate: $\left(\frac{t_2}{\theta}\right)^{\beta - 1} = (1.250)^{1.5} = 1.250 \times 1.118034 = 1.397542$ $h_B(10{,}000) = (3.125 \times 10^{-4})(1.397542) = 4.36732 \times 10^{-4}\text{ failures/hour}$

---

Comparative Engineering Synthesis Table

Critical Engineering Takeaway

At $t = 2{,}000$ hours, assuming an exponential distribution drastically underestimates the true reliability of a mechanical component ($77.88\%$ vs $96.92\%$) because it presumes the component has been subject to full random failure intensity from time zero. Conversely, at $t = 10{,}000$ hours, the exponential model dangerously overestimates reliability ($28.65\%$ vs $17.43\%$) and underestimates the failure hazard rate by a factor of 3.5 ($1.25 \times 10^{-4}$ vs $4.37 \times 10^{-4}\text{ hr}^{-1}$) because it ignores progressive wear-out.