Weibull Analysis Notes (Two-Parameter Model)

Breaking strength as a distribution

• Breaking strength depends on material history and condition; it is not a single fixed value. We can describe it by a distribution of fracture strengths (the grand average idea) rather than one point value.
• The appropriate distribution for brittle materials under a weakest-link scenario is the Weibull distribution (two-parameter form used here).
• The distribution arises from a weakest-link analysis: fracture strength is governed by the largest flaw in the material; the overall strength depends on the weakest link in the chain of flaws.
• This leads to the concept that the breaking strength is a random variable with a distribution characterized by two parameters.

The Weibull model and its two parameters

• The fracture (breaking) strength distribution is described by the two-parameter Weibull distribution with the cumulative distribution function (CDF): $F(s) = P( ext{fracture at or below } s) = 1 - \,\exp\left(-\left(\frac{s}{\sigma_0}\right)^{m}\right)$ where:
  • $m$ is the Weibull modulus (shape parameter)
  • $\sigma<em>0$ (often written as $\sigma</em>naught$) is the scaling (characteristic) strength
• The probability of failure at a given applied stress $s$ is $p(s) = F(s)$.
• The survival function (not failed at stress $s$) is
  $S(s) = 1 - F(s) = \exp\left(-\left(\frac{s}{\sigma_0}\right)^{m}\right).$
• At the stress level $s = \sigma<em>0$, the failure probability is $F(\sigma</em>0) = 1 - e^{-1} \approx 0.632.$
• The shape of the distribution becomes steeper (less scatter) as $m$ increases; larger $m$ means less variability in strength for similar samples.
• The two free parameters completely characterize the distribution; determining $m$ and $\sigma_0$ from data allows us to know the entire probability of failure function.

From data to the probability of failure function p(s)

• Experimental data consist of fracture stresses observed for a number of specimens under various conditions.
• For each stress value $s_i$, record how many specimens fractured at or below that stress.
• If there are $n$ specimens in a condition, a common practice (order-statistics approach) is to estimate the probability of failure at $s<em>i$ as: pi = \frac{\text{# of specimens that failed at or below } s_i}{n+1}.
• This $p<em>i$ serves as the empirical estimate of the failure probability at stress $s</em>i$.
• Example from the transcript (two conditions, 12 specimens per condition):
  • There are 12 observed fracture stresses. The denominator used is $n+1 = 13$.
  • At $s = 15$ MPa, 1 specimen failed → $p = \frac{1}{13}$.
  • At $s = 21$ MPa, 2 specimens failed or below → $p = \frac{2}{13}$.
  • At $s = 23$ MPa, 3 specimens failed or below → $p = \frac{3}{13}$.
  • At the maximum observed stress (34 MPa) 12 specimens failed or below → $p = \frac{12}{13}$.
• The result is a discrete set of points $(s<em>i, p</em>i)$ that approximates the underlying $p(s)$ curve.
• The plotted points typically show a low probability of failure at low $s<em>i$ and a high probability at high $s</em>i$, with a smooth transition in between.

Linearization strategy to estimate $m$ and $\sigma_0$

• Directly fitting the nonlinear form is difficult, so we linearize the Weibull relation using logarithmic transformations.
• Start from the CDF form:
  $F(s) = p = 1 - \exp\left(-\left(\frac{s}{\sigma_0}\right)^{m}\right).$
• Then the survival is
  $S(s) = 1 - p = \exp\left(-\left(\frac{s}{\sigma_0}\right)^{m}\right).$
• Take natural logs twice:
  $\ln S(s) = -\left(\frac{s}{\sigma<em>0}\right)^{m},$$\ln(-\ln S(s)) = m \ln s - m \ln \sigma</em>0.$
• Define the transformed variables for linear regression:
  • $x = \ln s$
  • $y = \ln\left(-\ln(1 - p)\right) = \ln\left(-\ln S(s)\right)$
• Then the linear relationship is: $y = m x - m \ln \sigma_0.$ That is, a linear model with:
  • slope: $m$ (Weibull modulus)
  • intercept: $b = -m \ln \sigma_0$
• In practice, compute for each data point:
  • $x<em>i = \ln s</em>i$
  • $y<em>i = \ln\left(-\ln(1 - p</em>i)\right)$
• Fit the line $y = m x + b$ by linear least squares to obtain the slope $m$ and intercept $b$.
• From the intercept, recover the scale parameter:
  $\sigma_0 = \exp\left(-\frac{b}{m}\right).$
• Once you have $m$ and $\sigma<em>0$, you can reconstruct the probability of failure function for any $s$ via $p(s) = 1 - \exp\left(-\left(\frac{s}{\sigma</em>0}\right)^{m}\right).$

Worked example outline (based on the transcript data)

• Data: 12 observed fracture stresses with corresponding failure counts, leading to a table of $(s<em>i, p</em>i)$ with $p<em>i = \frac{\text{fails at or below } s</em>i}{13}$.
• Example points (illustrative):
  • At $s = 15$ MPa: $p_1 = \frac{1}{13}$
  • At $s = 21$ MPa: $p_2 = \frac{2}{13}$
  • At $s = 23$ MPa: $p_3 = \frac{3}{13}$
  • …
  • At $s = 34$ MPa: $p_{12} = \frac{12}{13}$
• Compute transformed coordinates:
  • $x<em>i = \ln s</em>i$
  • $y<em>i = \ln\left(-\ln(1 - p</em>i)\right)$
• Perform a linear least-squares fit to obtain slope $m$ and intercept $b$.
• Calculate $\sigma_0 = \exp(-b/m)$.
• Report the Weibull parameters $(m, \sigma_0)$ for each condition and compare between conditions.
• Use the fitted model to evaluate $p(s)$ at any stress of interest.

Practical interpretation and implications

• What $m$ tells you:
  • Higher $m$ => steeper rise in failure probability with stress => less scatter in strength across specimens.
• What $\sigma_0$ tells you:
  • It is the stress level at which the failure probability is $1 - e^{-1} \approx 0.632$; acts as a scaling factor for the distribution.
• The procedure provides a practical way to summarize brittle material strength data with two interpretable parameters, enabling prediction of failure probabilities at different loading levels.
• This analysis is widely used in materials science for brittle solids where flaws control strength via weakest-link behavior.

Notes on the data handling and analysis flow

• Treat each material condition separately: perform the Weibull fit for each condition to obtain $m$ and $\sigma_0$ for that condition.
• Use the lab handout as guidance for the exact data handling steps and the required outputs for the assignment (the two parameters for both conditions).
• The transformation to a linear form greatly simplifies the estimation via standard linear regression techniques.
• Remember the underlying assumptions:
  • The material family and flaw population follow a Weibull distribution due to the weakest-link mechanism.
  • The dataset is representative of the fracture-strength distribution for the material in that condition.

Connections to broader concepts and implications

• This approach ties statistical modeling to physical intuition: fracture strength variability stems from flaw distribution, and the Weibull form captures that physics through $m$ and $\sigma_0$.
• The method uses order-statistics-inspired probabilities ($p_i$ computed with $n+1$ in the denominator) to provide unbiased-ish estimates of the cumulative distribution at observed stresses.
• The linearization technique is a common data-analysis trick in the physical sciences: transform nonlinear models into linear ones to enable straightforward parameter estimation.
• Ethical/practical relevance: understanding strength variability and reliable failure probabilities informs safety, reliability, and design margins in engineering applications.

Summary of key formulas (LaTeX)

• Weibull CDF:
  $F(s) = P( ext{fracture at or below } s) = 1 - \exp\left(-\left(\frac{s}{\sigma_0}\right)^{m}\right)$
• Failure probability:
  $p(s) = F(s)$
• Survival:
  $S(s) = \exp\left(-\left(\frac{s}{\sigma_0}\right)^{m}\right)$
• Linearized form (transforms):
  x = \ln s,\n \ y = \ln\left(-\ln(1 - p)\right)
  $y = m x + b, ext{ with } b = -m \ln \sigma_0$
• Parameter recovery:
  $m = \text{slope of the fitted line},\quad \sigma_0 = \exp\left(-\frac{b}{m}\right)$
• Reconstructed failure probability:
  $p(s) = 1 - \exp\left(-\left(\frac{s}{\sigma_0}\right)^{m}\right)$

Action items for the assignment

• Perform the Weibull (two-parameter) analysis for both material conditions:
  • Compute $p<em>i$ from your data using pi = \frac{\text{# failures at or below } s_i}{n+1}.
  • Form $x<em>i = \ln s</em>i$ and $y<em>i = \ln(-\ln(1 - p</em>i))$.
  • Fit the line $y = m x + b$ to obtain $m$ and $b$.
  • Compute $\sigma_0 = \exp(-b/m)$.
  • Report $(m, \sigma_0)$ for each condition and compare.
• Use the fitted model to plot or compute $p(s)$ across a range of stresses as needed for interpretation.
• Upload the parameter values as required by the lab handout.