Wöhler likelihood estimator estimation strategy

[johannes-mueller]

Current behavior

Infinite likelihood

In the current likelihood estimation we use only the infinite zone to estimate the infinite likelihood. According to our tests it is better to also take the load levels of the finite zone into account, because those indicate a fracture probability of 1.0 for those load levels. This will lead to a generally increased confidence.

Finite likelihood

Currently we use all fractures to estimate the finite likelihood. This can lead to perturbed results. For example a single late fracture on a high mixed load level can spoil the result completely. As of now, we don't have a means to avoid that other than labeling the concerning datapoint as a runout.

In #80 we changed the behavior of the Elementary analyzer to take only the finite zone fractures into account when estimating the Wöhler slope. Analogously to that we should do the same thing when estimating the finite likelihood of the finite parameters of a Wöhler curve.

Planned behavior

For the infinite likelihood we will take all the load levels into account, whereas we will limit the finite likelihood finite fractures. An expert user can manipulate this selection using set_finite_infinite_transition(). That will make a infinite load level considered to be finite.

Required changes

Two behaviors are to be changed:

• The finite likelihood now considers all fractures, in the future only the finite fractures
• The infinite likelihood now considers only finite fractures, in the future all fractures
• Better documentation of the behavior