Manipulation of mechanical properties of 7xxx aluminum alloy via a hybrid approach of machine learning and key experiments

Abstract

Considering the complex relationship among mechanical properties of 7xxx aluminum alloy, it is very crucial to optimize two or more target properties simultaneously in developing new materials. In this paper, three different machine learning assisted strategies are used to study the relationship between alloy composition, process parameters and mechanical properties of 7xxx aluminum alloy, which is expected to accelerate the development of new materials. Firstly, the mechanical properties prediction model of 7xxx aluminum alloy was established by back propagation (BP) neural network. Then, on this basis, genetic algorithm (GA) is used to optimize the prediction accuracy of back propagation (BP) neural network. In addition, a radial basis function (RBF) neural network is used for modeling analysis, and the prediction results of the three models are compared. The results show that back propagation (BP) neural network optimized by genetic algorithm (BP-GA) has higher prediction accuracy than BP and RBF neural network, the coefficient of correlation (R), average absolute relative error (AARE) and root mean squared error (RMSE) value are 0.948, 4.28%, 0.087, respectively. In addition, scanning electron microscope (SEM), electron backscatter diffraction (EBSD), high resolution transmission electron microscope (HRTEM) and the tensile test were used to carry out the related experimental verification work. A large number of fine and dispersed spherical precipitates can be observed after ageing heat treatment. The experimental value is close to the target value, which further confirms that BP neural network model optimized by genetic algorithm can be used to predict and design 7xxx aluminum alloy.

Keywords

• Aluminum alloy
• Alloy design
• Machine learning
• Multi-objective optimization
• Mechanical properties