Size effect in fatigue based on the extreme value distribution of defects

Fatigue limits need to be extrapolated from test specimens to manufactured products. The relevant industry standards provide a method for this by utilizing the statistics of defects in the material. We show here that the standard method involves an inappropriate definition. Moreover, it relates to the characteristic size of the largest defects, which is not associated with any unique exceedance probability. We outline a more consistent method which, by a quantile of the largest defects, relates the sample size effect to the desired failure probability. This method is applicable also to samples smaller than the test specimen.