ABSTRACT
Fatigue reliability is an important aspect of the reliability problem of mechanical component in the engineering field (Johannesson 1999, Rambabu et al. 2010, Tovo 2001). It has been noted that the prediction of P-S-N curve is very important in fatigue reliability estimation of the mechanical component working under variable-amplitude loading (Wang 2010).Therefore, the investigations on the prediction of P-S-N curve of the structure or material have attracted many attentions (Ling & Pan 1997, Shimizu et al. 2010, Tosha et al. 2008, Zheng & Wei 2005). Traditionally, the expression of the P-S-N curve is predicted through constant-amplitude fatigue tests in several stress levels, say 4, with specified sample size in each level, say 5. The conventional method is costly and timeconsuming. Hence, the methods using less specimens and experimental time have been the interesting and important topics in the area of mechanical fatigue. Ling & Pan (1997) has proposed a maximum likelihood method for predicting the P-S-N curves based on fatigue data in different stress levels with 1 specimen except in a medium level with 6 specimens. This approach needs less specimens and can save the experimental time, but it still needs to conduct multiple constant-amplitude fatigue tests in different stress levels. Zheng & Wei (2005) proposed a method to predict the P-S-N curve of 45 steel notched elements based on the tensile properties. Shimizu et al. (2010) proposed methods for predicting the P-S-N curves based on
two-parameter Weibull distribution, three-parameter Weibull distribution and lognormal distribution, and a comparison was carried out showing that the lognormal distribution with identical variation coefficient in different stress levels can provide accurate agreement with the test result.
