Search

Fatigue strength is known to decrease with increasing dimension of the component. This is due to a technological size effect, related to the production process, and to a geometrical size effect, due to a higher probability of finding a large defect. To investigate the latter, an heavy-walled component made of Ductile Cast Iron (DCI) has been trepanned and a fatigue test plan has been carried out using 4 different specimen geometries. An attempt has been made to relate the resulting fatigue strength using a weakest-link approach based on the effective volumes and surfaces. This approach seems to work well only in cases of different specimen's lengths. Some of the fracture surfaces were analyzed by means of SEM and the initiating defects were identified and measured. An approach in which the defects population can be randomly distributed in the specimen has been tried. Virtual fatigue tests have been carried out by considering pure propagation of the worst defect. The resulting fatigue curves showed that this approach is promising but needs further description of the initiation phase.

The definition of the relationship between probability, fatigue stress and cycles to failure is of great importance, especially in applications which requires very low probability of failure (e.g. Pf=0.1%). In this paper a new formulation is presented, which allows to separately consider the probability of the endurance and the probability of existence of the initiating defect. This approach is then compared to a number of known models. For this purpose, the proprietary results of 8 fatigue test sets, each with at least 24 data points, have been analyzed. A ranking of goodness-of-fit based on the Relative Likelihood can be used to choose the best distribution within each model, but unfortunately cannot be used to compare different models. The conclusions were: 1- the better description of the HCF strength, in terms of performance and robustness, were obtained by the 2-p Weibull distribution; 2- the choice of the model has a great influence in the estimate of the low probability quantiles, but it still subjective and a definitive answer cannot be given after this benchmark; 3- the strength of the proposed model is its flexibility.