Application of machine learning for predicting the service performance of metallic materials

In materials genetic engineering, data-driven machine learning (ML) technology is driving materials research into a new paradigm after theory, experiment, and computation, which is the fourth paradigm. Through ML, we can fully use the existing experimental data and exploit hidden connections underlying the data to achieve a more accurate prediction of material service performance despite not knowing the underlying principles. Therefore, ML can greatly reduce the time and cost required for experiments. Further, it shows remarkable vitality in predicting material performance. The service behavior of a material is one of the key factors affecting its performance and applications. The service performance prediction of materials has been initially achieved using the data on materials from previous experiments and established databases. Since different ML algorithms greatly affect the accuracy and generalization of the prediction results, selecting a suitable ML algorithm is crucial. In this paper, we summarize and analyze the following standard models for predicting the service performance of metallic materials: random forest, support vector machine, cluster analysis etc. In addition, the development history, advantages, and disadvantages of these models are briefly described. These models have obvious advantages in predicting the service performance of metallic materials and designing new high-performance metallic materials. Furthermore, we present the practical applications of ML algorithms in predicting several typical service performances of metallic materials. In materials research, the chemical composition of metallic materials, test-environment conditions, and other factors can be considered as features and input into ML training models to save time and cost. The models can achieve accurate and effective predictions of the service performance of metallic materials and provide reliable ideas for designing high-performance metallic materials. In this paper, we introduce the application of ML to the three most typical service properties, namely fatigue, creep, and corrosion, whose testing generally has a long time cycle and high cost. Further, we analyze some specific cases and briefly introduce the application of ML in predicting other service properties, such as hydrogen embrittlement and irradiation damage. Finally, the characteristics of ML for the service performance prediction of metallic materials are summarized, a few unresolved, related current problems are analyzed, and related development prospects are presented.