Abstract—

A technique has been developed for the use of artificial neural networks to describe the nonlinear relationship between the components of stresses and strains (tensor equation of state) and the onset of plastic flow (homogeneous nucleation of dislocations) in metal single crystals by the example of aluminum. Datasets for training neural networks are generated using molecular dynamics (MD) modeling of uniform deformation of representative volumes of a single crystal. Axisymmetric deformed states are considered when the symmetry axis coincides with the [100] direction of the single crystal. The trained neural networks are used as approximating functions within the dislocation plasticity model generalized to the case of finite deformations. It is used to simulate the propagation of shock waves arising from the collision of plates. In the case of nanoscale plates, a comparison is made with direct MD simulation of the process. In an ideal single crystal, the elastic precursor retains a constant amplitude corresponding to the threshold of homogeneous nucleation of dislocations, while in a deformed single crystal it has a significantly lower amplitude and rapidly decays with distance.

中文翻译：

---

神经网络在铝材冲击波过程建模中的应用

摘要-

已经开发了一种技术，用于使用人工神经网络来描述金属单晶中应力和应变分量（状态张量方程）与塑性流动的开始（位错的均匀成核）之间的非线性关系，例如铝。用于训练神经网络的数据集是使用单晶代表性体积的均匀变形的分子动力学 (MD) 建模生成的。当对称轴与单晶的[100]方向重合时，考虑轴对称变形状态。经过训练的神经网络用作位错塑性模型中的近似函数，该模型推广到有限变形的情况。它用于模拟板块碰撞产生的冲击波的传播。在纳米板的情况下，与该过程的直接 MD 模拟进行了比较。在理想的单晶中，弹性前驱体保持与位错均匀成核阈值相对应的恒定幅度，而在变形单晶中，它具有明显较低的幅度并随距离迅速衰减。