Composition-based aluminum alloy selection using an artificial neural network

Materials selection for aluminum alloys with desired fatigue properties and other mechanical properties is very difficult. Usually, when fatigue properties are maximized, other mechanical properties should be compromised. In this paper, an artificial neural network, was utilized to build two prediction models that has the purpose of predicting fatigue life from composition and inverse design to predict composition from fatigue properties as a tool for materials selection. A first model was built to predict fatigue life using information on alloy composition, heat treatment, and other mechanical properties. The second model is an inversion of the first model, which predicts the material compositions to get the desired fatigue performance and other mechanical properties. Both models produce good performances based on the R² scoring metric, where the values were found to be 0.92 and 0.96 for the first and second models, respectively. This study proved that the inversion model for predicting composition based on fatigue properties can reach acceptable accuracy and can be used as a materials selection tool. In addition, to investigate how atomic properties can affect fatigue life, the third model was built. It was found that atomic properties, such as electronegativity and the radius of alloying elements, are closely related to fatigue life and can be used to predict fatigue life as well. The significance of our work is that users can design new alloys for specific applications as well as select available alloys based on fatigue property criteria.