Shape memory alloys are widely used in various industries due to their superelastic and shape-memory properties, and they have been the subject of extensively researched. The temperature and strain of martensitic transformation are key factors influencing these alloys' properties, and molecular dynamics simulations help understand their effects. Consequently, improving the accuracy of new models for atomic-scale simulations is essential for advancing research in this field. Recently, new data-driven atomic potentials that leverage neural networks have been introduced to enhance simulation results. Despite the potential applications, few studies have explored the effects of microstructural factors on martensitic transformation boundaries. Studying the mechanical and thermal transformation properties of the austenite phase can help validate these models. This research examines the influence of grain boundaries in proximity to the ideal arrangement of austenitic structures, using both capacity eva‎luation and in-depth analysis. The first step involved the creating cubic nanopillars (2 nm wide with varying heights) in a bi-crystal form, which were used for characterizing thermal and mechanical properties. Various samples underwent cooling and heating cycles to compare with previous studies, and the temperatures at which thermal transformations occurred were validated. Subsequently, mechanical loading was applied along the [001] crystalline direction of the bi-crystal samples, inducing martensitic shear transformation and allowing for an examining of their boundary conditions. The atomic arrangement during temperature changes and strain was analyzed using multivariate fitting and atomic density distribution methods. The results showed a correlation between the transformation path and grain size, in agreement with experimental findings, within the grain size range of 6 to 30 nm. As a result, the thermal transformation temperatures from austenite to martensite and vice versa were obtained with high reliability and significantly lower computational cost compared to previous studies. Additionally, the stress and strain at the start and end of the mechanical transformation, were also determined, along with the elastic modulus of austenite. The effect of maximum distance from the boundary on the stress-strain curve was also discussed. Finally, by analyzing defects using the Wigner-Seitz defect analysis method, the relationship between these defects and previous findings on residual strain in strain cycles was examined. The main conclusions are as follows: a) The bi-crystal nanopillar model provides highly accurate results on the effect of grain size on transformation temperature, all while maintaining a significantly lower computational cost than prior simulations. b) The effect of grain size on the transformation mechanism was successfully simulated, with results matching previous experimental findings. c) Molecular dynamics simulations using data-driven interatomic potentials demonstrate both high accuracy and capability, and further development will reduce computational effort while increasing accuracy and generalizability.

مسعود پنجه پور

محمود مرآتيان اصفهاني , ابوذر طاهري زاده