An investigation of the location of fatigue initiation – Deterministic and probabilistic aspects

•There is large scatter in the measured fatigue failure location.•A gradient adjusted stress describes the fatigue failure location best.•The fatigue lives are generally longer for the larger notch.•Competing notch tests are not captured well by any of the four models.

Fatigue limit tests where there are two competing fatigue failure locations, two notches of different size, are performed in this paper. The normal distribution is used for describing the experimental outcome, where the largest occurring stress value or the gradient adjusted stress is used as a stress measure. Another model for describing the experimental outcome is the Weakest Link (WL)-integral where the integration is performed over the specimen surface area or the specimen volume. In addition, the positions of the fatigue failure initiation sites are measured and investigated. The predictive capabilities of the different models are evaluated with respect to determination of the longitudinal failure location.A notable result is that the longitudinal location of fatigue failure differs considerably from one experiment to the other, even when they are numerically equal. Nominally the stress at the fatigue failure site is much lower than the maximal stress. The load cases where there should be an equal amount of failures in both notches according to the models used do not agree with the experimental findings. Neither the point stress model nor the gradient adjusted point stress, nor the WL-integral can explain the experimental outcome from the competing notch tests.