Search

In an attempt to develop acceptably accurate alternative to complex, expensive and long-lasting experimental determination of fatigue data, numerous methods for estimation of fatigue parameters from materials’ monotonic properties have been proposed in the literature so far. New approaches and estimation methods are still being developed and proposed, with recent efforts shifting notably from empirical/analytical approaches and models to machine learning-based predictive models. For evaluation and comparison of performance of estimation methods, the error criterion (fraction of data points within a certain scatter band) is most frequently used along with additional criteria proposed in [1]. However, existing evaluations and comparisons of estimated and experimental fatigue lives are regularly performed for complete range of fatigue lives and materials with wide range of monotonic properties, resulting in significant averaging of results of evaluations and possibly erroneous conclusions on individual estimation methods accuracy. This was confirmed in a preliminary study [2] of a few selected estimation methods using very compact and consistent dataset on a single low-alloy steel featuring different heat treatments. Significant differences which were determined in performance and accuracy of the reviewed estimation methods for low- and high- cycle fatigue lives and for two subgroups of materials regarding their ultimate strength were the main motivation for this research. According to the detailed evaluation methodology described above, a comprehensive evaluation of the established fatigue parameters estimation methods, as well as those proposed more recently, was performed. Analysis was based on an extensive number of detailed fatigue datasets on unalloyed, low-alloy and high alloy steels and on aluminum and titanium alloys, collected from relevant literature. Obtained results provide detailed insight on the accuracy of the individual estimation methods and their applicability for the estimation of fatigue parameters of various materials divided to strength-based subgroups and for particular fatigue regimes (low- and high- cycle fatigue). For certain methods, the variability in their accuracy between strength- based subgroups and in different fatigue regimes/lives is particularly pronounced. The determined values of evaluation criteria can be used as a reference for evaluations of newly developed estimation methods as well as of predictive models based on machine learning.