OBRABOTKAMETALLOV MATERIAL SCIENCE Vol. 26 No. 4 2024 supercomputers and high-performance computing allows modeling of complex systems with millions of atoms and molecules. One of the key methods in multiscale modeling is molecular dynamics, which allows simulating the dynamics of atoms and molecules at the microscopic level [58]. This method is used to study material properties such as strength, elasticity, thermal conductivity, and others. Figure 4 shows how computer-assisted learning is applied in HEAs research. Fig. 4. Schematic diagram illustrating the application of multiscale modeling and machine learning in HEA research [58] Moreover, machine learning methods also play a signifi cant role in analyzing material data. It is used to classify materials, predict its properties, and optimize the production process. For example, machine learning algorithms can be used to determine the optimal material structure or predict its properties based on its composition and structure. Thus, the combination of multiscale modeling and machine learning provides a deeper understanding of material properties and improves the design and production process. By combining machine learning, phenomenological rules, and CALPHAD modeling, new promising compositions of refractory HEAs with specifi ed phase compositions and mechanical properties (such as yield strength) were predicted. It is emphasized that the creation and modifi cation of the properties of fi vecomponent HEAs can be achieved using CALPHAD software, designed to calculate phase diagrams. Studies conducted at the Siberian State Industrial University showed that CALPHAD phase diagram calculations are confi rmed by experimental data, allowing the development of next-generation alloys with specifi ed properties [59]. Below, in Table 2, a comparison of the predicted and experimental result of the yield strength for various alloys is presented. It can be noted that the predicted yield strength values for the alloys generally show a good match with the experimental data, although there are cases where the errors in the estimates are signifi cant. This may be due to various factors such as the complexity of the alloy structures, environmental eff ects, and others. Regarding the prospects for using machine learning in this context, it can be highlighted that machine learning methods can be eff ectively applied to predict material properties based on its composition, structure,
RkJQdWJsaXNoZXIy MTk0ODM1