This work proposes a novel artificial neural network (ANN)-based framework to explore the mechanical behaviour of shape memory alloy (SMA) Schwartz primitive triply periodic minimal surfaces (TPMSs) architectures where a loop with multiple conditions (LMCs) is introduced into the framework to improve its accuracy. Remarkably, it is found that the introduced ANN-based framework comprehends extremely well the example data and then provides a highly accurate prediction for the topology-driven mechanical behaviour of the SMA TPMS lattice architectures, which do not belong to the example. Indeed, the mechanical responses obtained from ANN-based approach excellently agree with that from numerical homogenization, with the maximum percent difference below 2.7% attained when the TPMS is subjected to tension or shear during the loading and unloading processes. More interestingly, the ANN-based approach can perform well with the increment of relative density or temperature 1000 times smaller than that in the homogenization, only requiring a little simulation time or thousands of times faster than that in the homogenization at the same computer setup. Hence, the mechanical behaviour of SMA TPMSs can be accurately characterized by the proposed framework at almost any value of the relative density or the temperature after the training. Subsequently, the ANN-based framework is used to map two-dimensionally the impact of varying relative density, temperature, and deformation on the effective hardness and superelasticity of the SMA TPMS architectures. The results show that the superelasticity of SMA TPMS scaffolds increases with decreased temperature and relative density and with increased deformation, in a nonlinear pattern. This work lays the foundation for further research on using ANN to explore the mechanical response and advanced applications of the complex SMA-based structures, such as smart composite systems, helical springs, and metamaterials at considerably reduced computational and financial costs.
Read full abstract