Statistical evaluation of the softening or hardening behaviour of metallic materials

Detailed knowledge of the cyclic behaviour of a material is essential for the design of components which can be plastically deformed by cyclic loads. The design process based on the sole static characteristics could lead, in case of softening behaviour, to a significant underestimation of the plastic strain amplitude and to a consequential overestimation in terms of component life. The experimental determination of cyclic characteristics is onerous compared to static properties, which can be obtained by a simple tensile test. The objective of this study is to derive a model to evaluate, starting from the knowledge of tensile variables alone, whether the material subject to cyclic loads hardens or softens. The proposed approach is statistical, based on a sample of about 240 materials. The predictive model results from a multinomial logistic regression, which allows deriving a relationship between independent input variables (tensile variables of the materials) and a dependent output variable, represented by the material belonging to one of the following behaviour categories: hardening, softening, mixed (hardening or softening depending on the load level), stable (small differences between static and cyclic behaviour). To determine which tensile variables significantly influence the cyclic behaviour, different regressions based on no more than three parameters are compared, to result in a simple and functional calculation tool. Finally, several correlations from the literature are considered, comparing the associated results with those from the derived models. The comparison highlights higher goodness-of- fit for the proposed approach with respect to the state of the art, demonstrating its predictive potential.