Applications of machine learning on magnesium alloys

In materials genetic engineering, data-driven machine learning techniques have garnered significant attention as a powerful new tool in the field of magnesium alloys. Traditional empirical trial-and-error methods and those based on density functional theory have struggled to keep pace with the continuous advancements in material science needs owing to high time costs and low efficiency. By relying on statistical methods instead of solving physical equations, machine learning can quickly predict material properties at a low cost, provided the connection between descriptors and target properties is identified. This capability can streamline the experimental process. Magnesium and its alloys show tremendous potential in aerospace, automotive, and other fields owing to their low density and high specific strength. However, their industrialization has been limited by several challenges, including the varied effects of different alloying elements, preparation and processing defects, deformation difficulties, and the common trade-off between strength and ductility. Machine learning can accelerate the discovery of novel magnesium alloys or processing parameters, and explore the relationships between their physicochemical characteristics and target properties. This paper comprehensively and systematically reviews the research progress of machine learning applications in magnesium alloys. It introduces the basic processes and various methods of machine learning, including data set collection, data preprocessing, model building, and performance evaluation. The classification of machine learning algorithms is summarized briefly. The paper then focuses on the research achievements of machine learning applied in many aspects, such as machining processes, microstructure, mechanical properties, corrosion resistance, hydrogen storage properties, intrinsic properties (reinforcement mechanism, anisotropy, etc.) and inverse design. Factors such as alloy compositions, test temperature and time, second phase, and Schmid factor can be considered as features and input into machine learning models for training. These models not only accelerate the design of novel high-performance magnesium alloys but also enhance the understanding of magnesium alloy mechanisms. Additionally, the paper analyzes some urgent issues in the research and application of machine learning in magnesium alloys. These include insufficient prediction of the chemical and physical properties of magnesium alloys, the nascent stage of predicting the design and service performance of magnesium alloy components, and the lack of high-quality data sets. Finally, the paper proposes future research directions and development trends in the application of machine learning in magnesium alloys.