../_images/logo.png

Statistical parametrization#

For DFNs it is essential to provide a length distribution model. This is used in model construction to guide the creation of synthetic fractures with lengths that follow the given distribution. The immediate problem that arises is that often fractures are cut by boundaries thus giving partial length information. This type of data is defined as censored data.

Censoring#

Survival (or reliability) analysis are rooted in the medical, biological and engineering sphere, thus the continuous measured variable is time. In our case the measured variable is space (length) but the principles of survival analysis are maintained since it is still a continuous variable censored by an observational limit. We can thus translate the theory and definitions from a function of time to function of space.

In general censoring can be of three main types:

  1. Right censoring: the fracture starts inside the boundary but terminates outside of it

  2. Left censoring: The fractures starts and terminates outside the boundary

  3. Interval censoring: The fracture is cut by internal holes (interpretational voids).

When trying to estimate a parameter from the data, and eventually fit a distribution, censored data must be corrected or underestimation is assured.

In this case the data is right-censored thus we can apply survival analysis techniques to correct for such bias. A in depth explanation is provided in the `reliability page

Length parameter#

By following the schematic below we can introduce the problem

../_images/example_diagram.png