基于机器学习型分子力场的纯锆冲击相变行为模拟研究

Materials dynamics in zirconium, particularly the phase transition behavior under shock compression have been widely studied as it shows the complexity of processes that are typical in nuclear materials. Lack of reliable interatomic potential that is capable of describing the shock induced solid-solid phase transformation in Zr, the understanding on its microscopic picture is based on the atomic simulations on pure titanium with a unique modified embedded atomic method （MEAM） potentials. It is commonly believed that both Zr and Ti possess similar phase diagrams and phase transition processes. However, the very recent studies have shown that the two metals show enormous difference in the aspect of pressure induced phase transition, thus challenging existing theoretical studies. In this proposal, we plan to adopt machine learning (ML) method to construct high-accurate Zr interatomic potential, with which the shock induced phase transition can be directly simulated and understood. .The project is organized as follows. Firstly, with the help of the machine learning model as well as the domain knowledge of phase transition theory, the applicant intends to construct an interatomic potential that can reasonably capture the shock induced phase transition in Zr. Secondly, the phase transformation mechanisms and kinetics will be uncovered by using a combination of nonequilibrium molecular dynamics and the developed ML potential. Based on it, the controversy of the existing shock-recover experimental measurements will be clarified and explained. .The present project not only helps us understand dynamic mechanical responses and phase transition behaviors of metals and alloys at high strain rates, but also provides theoretic guidance in developing new materials served in extreme conditions.

纯锆作为研究核材料冲击相变行为的模型材料而受到广泛的关注。由于缺乏准确描述纯锆原子间作用力的分子力场，目前对其冲击相变微观物理图像的理解主要是基于纯钛半经验势函数的原子模拟。最新文献和申请人前期研究都表明两种金属在冲击相变行为方面存在差异，进而质疑现有模拟研究的理论基础。本申请项目拟通过机器学习方法构造高精度的纯锆分子力场，并据此直接模拟和理解纯锆的冲击相变行为。首先，以冲击相变理论为先导知识，运用机器学习模型构造高精度的分子力场，使其能够有效地描述纯锆的冲击相变；进而采用非平衡分子动力学方法模拟纯锆的冲击相变过程，从相变机制和动力学特性两方面揭示微观物理图像；在此基础上，利用原子模拟与实验研究相结合的方法理解现有冲击“软回收”实验研究的结果。项目的完成将拓展材料动态行为的相关研究领域，同时为极端服役环境下材料的优选与设计提供理论支持。

纯锆是研究核结构材料动态力学响应的模型材料，理论计算是研究冲击相变行为的主要手段，然而缺乏高精度的分子力场成为揭示微观机制及其宏观力学行为的瓶颈问题。本项目融合机器学习势函数和大规模分子动力学方法研究了纯锆的冲击相变行为，取得如下成果：首先利用融合领域知识的机器学习模型开发了准确描述纯锆相变行为的机器学习势函数，复现了温度-压力相图和主要热物性参数；进一步，大规模分子动力学模拟结果表明纯锆的冲击相变机理与纯钛不同，即低冲击强度下的α→ω相变机制与加载方向、预制缺陷密度无关，而强冲击载荷下hcp→ bcc相变路径具有明显的各向异性，并首次发现了亚稳非晶态中间相主导的新相变机制，加速了相变动力学过程；最后，本项目还提出了逆相变和多级孪生变形是主导冲击“软回收”样品奇异塑性的微观力学响应机制。本项目的研究成果不仅从原子尺寸揭示了纯锆的冲击相变机制及其动力学过程，也拓展了金属材料动态响应机理的研究手段。相关成果在材料及物理领域权威期刊上共计发表SCI 论文15 篇，包括Acta Mater，Phys Rev B，JMST 和IJMS 等，培养研究生3 人 。