Molecular dynamics can bring completely new insight into the microstructure evolution, however, the quality of a molecular dynamics (MD) simulation strongly depends on the interatomic potential. In recent decades, this interaction has been driven by empirical potentials (such as EAM, MEAM). Such approaches have been able to capture the basic properties of simple systems but struggle with more complex problems. Thanks to advances in computer science, today we are able to use a completely new approach based on machine learning algorithms. The goal of these approaches is to bring density functional theory (DFT) accuracy to MD and capture the complex interaction in the system.

In this contribution, we will introduce the methodology of high dimensional neural network (NN) potential (HDNNP) with the atom-centered symmetry function descriptors [1]. This methodology is applied to the extensively studied smart alloy NiTi which represents a challenging task for both the experimental and theoretical levels. We will focus on the development of the NN potential for the martensitic B19’ phase suitable for simulating the complex microstructure evolution during combined twinning and plastic slip [2]. Specifically, we demonstrate that our interatomic potential preserves the strong plastic slip anisotropy and that the HDNNP methodology is suitable for simulating the evolution of the martensitic microstructure.

Ref.
[1] Behler, J., 2011. Neural network potential-energy surfaces in chemistry: A tool for large-scale simulations. Physical Chemistry Chemical Physics 13, 17930–17955. https://doi.org/10.1039/c1cp21668f


[2] Seiner, H., Sedlák, P., Frost, M., Šittner, P., 2023. Kwinking as the plastic forming mechanism of B19′ NiTi martensite. International Journal of Plasticity 168, 103697. https://doi.org/10.1016/j.ijplas.2023.103697