engleski

Estimation of materials' parameters of strain-life fatigue behavior using empirical and artificial neural networks based approach

Estimation of fatigue lives and material behavior, along with determination of corresponding material parameters, is needed in early design stages that precede experimental testing. Since the experimental characterization is the most accurate, but also time and resource consuming, a number of methods for estimation of fatigue parameters from easily obtainable monotonic properties exist in the literature. Most commonly used methodologies for estimation of strain life parameters nowadays include widely used empirical estimation methods and machine-learning based methods, mainly artificial neural networks (ANNs). The latter, when correctly developed, facilitate capturing complex relationships among input and target variables. ANNs were developed for estimation of strain-life parameters of unalloyed, low-alloy and high-alloy steels on the basis of monotonic properties which were previously determined as relevant by performing a detailed statistical analysis. Previous statistical analyses indicated that different monotonic properties are statistically significant for estimation of fatigue parameters of unalloyed, low- and high-alloy steel subgroups. Results were evaluated on an independent set of data, both for aforementioned groups of steels and for individual materials. Evaluations showed that when developed correctly, ANNs are a promising method for estimation of materials’ strain-life parameters and behavior of particular material. Should more data be included in developing of ANNs for the given purpose, a fast, robust and efficient solution can be obtained and used for the estimation of strain- life parameters and behavior of metallic materials.

estimation methods ; monotonic properties ; strain-life fatigue behavior ; artificial neural networks

nije evidentirano