International Journal of Engineering and Technology

The present paper gives a survey of bibliographical recent models generating statistical predictions of minimum sample size required for fatigue characterisation. Metallurgical defects, roughness and other geometrical defaults are most often considered as the sources of deviations between fatigue lives for the same stress level. The sample size at one given stress level must therefore be carefully chosen in order to find the good compromise between the accuracy of the fatigue strength estimation and the cost of the experiments. The estimations are dependent on the desired confidence level and on the given survival probability. The application of the models to several series of tests issued from the relevant literature or realised within our laboratories is made. The proposed method is based on Gope method which establishes a relationship between the size of the sample, the probability of survival, confidence level and the parameters of Weibull distribution and log-normal. The comparison between the information taken from results of the predicted sample size and the information given by real tested specimens is then developed. It allows to study and to discuss the validity of these models with respect to the former objectives of these theories.