A methodology for probabilistic prediction of fatigue crack initiation taking into account the scale effect

An approach for probabilistic prediction of fatigue crack initiation lifetime of structural details and mechanical components is presented. The methodology applied is an extension of the generalized local model (GLM) to the fatigue case using the fatigue Weibull regression model proposed by Castillo-Canteli. First, the primary failure cumulative distribution function (PFCDF) of the generalized failure parameter is derived from experimental results for a given reference size, taking into account the non-uniform distribution of the generalized parameter (GP) the specimens are submitted to. The adequacy of the GP is presumed, ensuring uniqueness of the derived PFCDF as a material property, irrespective of the specimen shape and size, and the test chosen to this end. Next, the GP distribution is obtained by a finite element calculation and the PFCDF is applied to each finite element, considering the scale effect, to derive the probability of failure for the whole component. The suitability of the proposed approach is illustrated twice: first, assessing simulated data in a theoretical example, and second, evaluating experimental fatigue life results for riveted joints from the historical Fão Bridge. The PFCDF for the puddle iron from the bridge is calculated from standard tensile specimens, from which the initiation fatigue lifetime of the riveted connections is predicted and compared with the experimental results. In this way, the transferability from standard tests to real components is demonstrated.