The interpretable descriptors for fatigue performance of wrought aluminum alloys

Abstract

Aluminum alloys are extensively utilized in the transportation and aerospace industries due to their excellent processability and corrosion resistance. A key aspect affecting the application of aluminum alloys is fatigue failure, with fatigue strength constituting an essential indicator for assessing the failure mode. In this study, 859 data sets were collected to evaluate the effects of aluminum alloy composition and mechanical properties on fatigue strength. To comprehensively evaluate the features that exert a great influence on the fatigue strength of wrought aluminum alloys, atomic physical quantities of aluminum alloys were added to the original data set. To improve the computational efficiency and physical interpretability of the model, Spearman correlation analysis, optimal subset method, and other methods were used to perform feature selection on the atomic feature quantities and mechanical properties in the data set. Finally, a mathematical relationship between material descriptors and fatigue strength was established based on the alloy composition and the three most relevant features in the feature screening, which helps to better understand and predict the fatigue behavior of wrought aluminum alloys.

Keywords

• Data-driven design
• Wrought aluminum alloy
• Fatigue strength
• Machine learning