Speaker
Description
NiTi shape memory alloy has become the most widely used shape memory material in industrial, high-tech, and medical applications due to its unique thermal and mechanical properties, primarily represented by the shape memory effect and superelasticity. Due to these unique characteristics, this alloy has been used in numerous practical applications since its discovery in 1963. Despite having only a single active slip system, which theoretically limits the plastic deformability of individual grains in a polycrystalline material, NiTi martensite exhibits remarkable plasticity. The underlying mechanism behind this behavior remained unclear for a long time.
Recently, a novel deformation mechanism termed "kwinking" was identified [1]. The term originates from a combination of "kink" and "twin," reflecting the unique nature of this planar crystal defect. Kwinking exhibits characteristics of both twinning and kinking: it forms twin-related regions while simultaneously accommodating plastic deformation through geometrically necessary dislocation walls. It has been observed experimentally [2] that kwinking is the dominant mechanism of plastic deformation of the B19' martensitic phase and provides this phase with high ductility.
To understand the martensitic phase at the atomic scale, we will use the most modern methods of atomistic simulations that combine quantum-mechanical calculations using density functional theory (DFT) and molecular dynamics (MD) simulations. The quality of atomistic simulations in MD strongly depends on the accuracy of the available interatomic potentials. In the past decades, MD simulations have been employing semiempirical interatomic potentials but, recently, new techniques based on machine learning (ML) became available for potential generation. In our study, we will explore reliability of our new ML interatomic potentials by comparing their prediction with DFT data. In particular, we will calculate elastic constants of the martensite phase and the energy as a function of the monoclinic angle.
The ML potentials are constructed from the ab initio training set obtained either from the ab initio molecular dynamics simulations or from many ab initio static configurations by fitting free energies and forces acting on individual atoms. All data necessary for the construction of the training set were obtained using the ab initio software package VASP [3], [4]. For the fitting of the ML potential, we employ i) the VASP routine that collects the selected configurations occurring during ab initio MD simulations and includes them in the data set, ii) the neural network approach implemented in the code RuNNer [5] and iii) the atomic cluster expansion as employed in the code Pacemaker [6].
References
[1] H. Seiner, P. Sedlák, M. Frost and P. Šittner, Int. J. Plast. 168, 103697 (2023).
[2] O. Molnárová, M. Klinger, J. Duchoň, H. Seiner and P. Šittner, Acta Mater. 258, 119242 (2023).
[3] G. Kresse and J. Hafner, Phys. Rev. B 48, 13115 (1993).
[4] G. Kresse and J. Furthmüller, Phys. Rev. B 54, 11169 (1996).
[5] J. Behler, J. Chem. Phys. 13, 170901 (2011).
[6] Y. Lysogorskiy, C. v.d. Oord, A. Bochkarev, S. Menon, M. Rinaldi, T. Hammerschmidt, M. Mrovec, A. Thompson, G. Csányi, C. Ortner and R. Drautz, npj Comput. Mater. 7, 1 (2021).
