In this paper, a machine learning potential of titanium in the form of moment tensor potential (MTP) is constructed to study the crack propagation behavior of titanium with a pre-crack via molecular dynamics (MD) simulations. The MTP of titanium is obtained by learning the results of the first-principles calculation for various configurations, and is verified to rationally describe the statistic and dynamic property of titanium. Using MTP, MD simulations of a pre-crack on $\left\{10\bar{1}0\right\}$ and $\left\{0001\right\}$ planes upon loading are performed. The results show that generation of twinning occurs in the $\left\{10\bar{1}0\right\}$ plane, while generation of partial dislocation occurs in the $\left\{0001\right\}$ plane. MD simulations of a preset microcrack using traditional potential function are also presented and compared with those with MTP.

在本文中，以矩张量势 (MTP) 形式构建了钛的机器学习势，以通过分子动力学 (MD) 模拟研究具有预裂纹的钛的裂纹扩展行为。钛的MTP是通过学习各种构型的第一性原理计算结果而得到的，并被验证可以合理地描述钛的统计和动态特性。使用 MTP、MD 模拟预裂纹$\left\{10\bar{1}0\right\}$ 和 $\left\{0001\right\}$执行加载时的平面。结果表明孪生的产生发生在$\left\{10\bar{1}0\right\}$ 平面，而局部位错的产生发生在 $\left\{0001\right\}$飞机。还介绍了使用传统势函数对预设微裂纹的 MD 模拟，并与 MTP 进行了比较。