Molecular dynamics dataset of cyclic mechanical responses for machine-learning-based gradient design of TiNi shape memory alloys

It contains 78,208 ordered sequence samples derived from molecular dynamics simulations of nanograined TiNi shape memory alloys with different grain-size distributions and gradient architectures. The dataset includes structural descriptors, including minimum grain size, maximum grain size, average grain size, gradient coefficient, and grain-boundary volume fraction; loading conditions, including temperature and maximum strain amplitude; state variables, including cycle number, loading–unloading state, and instantaneous strain; and the corresponding instantaneous stress.

The samples are identified by MD Model ID and arranged chronologically to preserve the path-dependent characteristics of cyclic deformation. The data were used for group cross-validation, cyclic stress–strain response prediction, virtual screening, Pareto analysis, functional stability evaluation, and SHAP interpretation. The dataset can be used to reproduce the machine-learning analyses and facilitate further research on the data-driven design of gradient nanograined TiNi shape memory alloys.