High-Throughput Design of Micro-Alloyed Al-Mg-Si Alloys

Abstract

Micro-alloying has been widely used as an effective way to improve the comprehensive performance of metal structural materials. However, there are many kinds of micro-alloying elements and the content is less, so the traditional ‘trial and error’ experiment is slightly weak in material design. How to efficiently and quickly design the composition and process of micro-alloyed Al-Mg-Si alloys remains a huge challenge. In this work, the quantitative relationship of material composition–process–performance has been established by convolutional neural network combined with multi-model framework of six independent machine learning algorithms. A random optimization method for aluminum alloy samples was proposed and a variety of new components of materials were screened by high-throughput screening to narrow the range of micro-alloying elements. Then, the first-principles method was used to calculate the substitution, energy, and many properties of the L1₂-type precipitated phase. The results show that Yb, Sc and Ce elements are excellent micro-alloying elements. After designing the material composition and process, the experimental test results proved the accuracy of machine learning prediction and the strength and elongation of the material after micro-alloying were greatly improved. At the same time, a large number of rare-earth strengthening phases were observed, which was the key factor in improving the properties of the material, and was also consistent with the characteristic calculation results of the first-principles alloy phase. This study opens up a new way for a variety of material genetic engineering methods and efficient design of materials.