Application of machine learning models for estimating the material parameters for multiaxial fatigue strength calculation

This paper deals with a practical task of estimating missing material fatigue strengths required for the evaluation of multiaxial fatigue strength criteria, knowing other static or fatigue material parameters. Instead of searching for various analytical equations describing the dependencies between different material parameters, several machine learning models implemented in the caret R package are used here. The dataset used to train and test these models is based on the FatLim dataset with different material parameters, which has been redesigned for this new purpose. It is demonstrated that substantially more data points, such as were available in this study, are needed to achieve the goal set here. Although the results obtained at the current scale may be improved by the addition of new data points, the best performance of the random forest model rf and the worst performance of the pcr model are evident.

Automated summary

This paper discusses the practical task of estimating missing material fatigue strengths for the evaluation of multiaxial fatigue strength criteria, using machine learning models implemented in R. The dataset used for training and testing the models is based on the FatLim dataset with different material parameters. The results show that more data points are needed to achieve the desired goal, and the random forest model rf performs the best while the pcr model performs the worst.