单晶金属中微孔洞生长过程的深度学习预测方法

苏浩; 赵雷洋; 丛龙跃; 陈聪; 关添元; 刘岩

doi:10.11883/bzycj-2025-0324

单晶金属中微孔洞生长过程的深度学习预测方法

doi: 10.11883/bzycj-2025-0324

苏浩^{1, 2,},
赵雷洋^{1, 2},
丛龙跃^{1, 3},
陈聪⁴,
关添元^{1, 3},
刘岩^{1, 3, ,}

1.
清华大学航天航空学院，北京 100084
2.
北京宇航系统工程研究所，北京 100076
3.
清华大学空间高效能推进技术及应用全国重点实验室，北京 100084
4.
北京理工大学（珠海）智能制造技术研究中心，广东珠海 519088

基金项目: 国家自然科学基金（12572228）

详细信息

作者简介:
苏　浩（1994－　），男，博士，工程师，suhao9406@163.com

通讯作者:
刘　岩（1978－　），男，博士，长聘副教授，yan-liu@tsinghua.edu.cn

中图分类号: O341
计量
- 文章访问数: 110
- HTML全文浏览量: 10
- PDF下载量: 46
- 被引次数: 0
出版历程
- 收稿日期: 2025-09-29
- 修回日期: 2026-03-17
- 网络出版日期: 2026-03-20

A deep learning prediction method for growth of micro voids in single-crystal metal

SU Hao^{1, 2
,},
ZHAO Leiyang^{1, 2},
CONG Longyue^{1, 3},
CHEN Cong⁴,
GUAN Tianyuan^{1, 3},
LIU Yan^{1, 3
, ,}

1.
School of Aerospace Engineering, Tsinghua University, Beijing 100084, China
2.
Beijing Institute of Astronautical Systems Engineering, Beijing 100076, China
3.
State Key Laboratory of Advanced Space Propulsion, Tsinghua University, Beijing 100084, China
4.
Research Center of Intelligent Manufacturing Technology, Beijing Institute of Technology (Zhuhai), Zhuhai 519088, Guangdong, China

摘要

摘要: 针对单晶金属中微孔洞生长过程的预测问题，建立了一种基于U-Net和Transformer的深度神经网络模型：基于包含初始椭球双孔洞的单晶铜原子模型的分子动力学模拟结果构建数据集；提出了一种基于背景网格的数据预处理方法，在数据集中对模拟结果进行局部统计。算例结果表明，上述深度学习方法能够对单晶金属中微孔洞生长过程中的整体物理量和局部细节信息进行准确预测。
- 深度学习 /
- 微孔洞生长 /
- 分子动力学 /
- U-Net /
- Transformer
Abstract: A novel deep neural network is proposed to predict the growth of micro voids in single-crystal metal based on U-Net and Transformer in this paper. The dataset was constructed through molecular dynamics (MD) simulation results of a single-crystal copper atom model with initial double ellipsoidal voids. A data preprocessing scheme based on background mesh was proposed to perform local statistics on the simulation results. The information obtained from simulation results, such as void morphology, dislocation distribution, and von Mises effective stress, was converted into local statistics on the background mesh. These statistics were then converted into pixel matrix format as the input of the deep neural network. Multiple data samples can be generated from the results of one single MD simulation, which significantly reduces the computational resources required for dataset generation. The samples encompass typical stages of the void growth, which enables the network to capture key features and to facilitate data augmentation conveniently. The deep neural network model consists of four parts: U-Net composed of down-sampling and up-sampling networks, a generation model, a Query network model, and a regression prediction network. The model input includes both physical information and positional information. The output is the predicted physical information for the next time step. The loss function is a superposition of loss functions for each predicted variable. Numerical examples demonstrate that the aforementioned deep-learning method can accurately predict the global porosity ratio, dislocation density, and von Mises stress during growth of micro voids in single-crystal metal. The time for the network prediction can reach two orders of magnitude lower than that of MD simulation.
- deep learning /
- micro void growth /
- molecular dynamics /
- U-Net /
- Transformer

HTML全文

图 1 含有椭球双孔洞的单晶铜原子模型示意图

Figure 1. Schematic figure of single-crystal copper model with two ellipsoidal voids

下载: 全尺寸图片幻灯片

图 2 典型模型在不同应变下的孔洞形状和位错分布

Figure 2. The void morphology and dislocation distribution in the typical model under various strains

下载: 全尺寸图片幻灯片

图 3 MD模拟不同参数的孔洞在拉伸应变下的演化

Figure 3. Evolution of voids with different parameters under tensile strain by MD simulation

下载: 全尺寸图片幻灯片

图 4 金属微孔洞生长过程预测的深度学习网络模型

Figure 4. Deep learning network model for predicting the growth of micro voids in metals

下载: 全尺寸图片幻灯片

图 5 深度神经网络模型中U-Net模型示意图

Figure 5. The U-Net model in the deep neural network

下载: 全尺寸图片幻灯片

图 6 回归预测网络模型结构示意图

Figure 6. Schematic figure of the regression prediction network architecture

下载: 全尺寸图片幻灯片

图 7 深度神经网络的训练和验证曲线

Figure 7. Training and validation curves of the deep neural network

下载: 全尺寸图片幻灯片

图 8 深度神经网络在测试集上预测得到的物理量与目标物理量的对比

Figure 8. Comparison between the physical quantities predicted by the deep neural network on the test set and the target physical quantities

下载: 全尺寸图片幻灯片

图 9 孔隙率及孔隙率增量随应变变化结果

Figure 9. Variation of porosity ratio and its increment versus strain

下载: 全尺寸图片幻灯片

图 10 测试集中典型模型在不同应变下的孔洞形状结果

Figure 10. Void shapes under various strains of a typical model in the test set

下载: 全尺寸图片幻灯片

图 11 测试集中典型模型在不同应变下的归一化局部位错密度

Figure 11. Normalized local dislocation densities under various strains of a typical model in the test set

下载: 全尺寸图片幻灯片

图 12 测试集中典型模型在不同应变下的归一化von Mises应力结果

Figure 12. Normalized von Mises stress under various strains of a typical model in the test set

下载: 全尺寸图片幻灯片

参考文献(28)

[1]	朱建士, 胡晓棉, 王裴, 等. 爆炸与冲击动力学若干问题研究进展 [J]. 力学进展, 2010, 40(4): 400–423. DOI: 10.6052/1000-0992-2010-4-J2009-144. ZHU J S, HU X M, WANG P, et al. A review on research progress in explosion mechanics and impact dynamics [J]. Advances in Mechanics, 2010, 40(4): 400–423. DOI: 10.6052/1000-0992-2010-4-J2009-144.
[2]	彭辉, 裴晓阳, 陈实, 等. 延性金属动态拉伸断裂的损伤演化研究 [J]. 中国科学: 物理学力学天文学, 2017, 47(7): 070002. DOI: 10.1360/SSPMA2016-00392. PENG H, PEI X Y, CHEN S, et al. Damage evolution on dynamic tensile fracture of ductile metals [J]. SCIENTIA SINICA Physica, Mechanica & Astronomica, 2017, 47(7): 070002. DOI: 10.1360/SSPMA2016-00392.
[3]	贺红亮. 动态拉伸断裂的物理判据研究 [J]. 高压物理学报, 2013, 27(2): 153–161. DOI: 10.11858/gywlxb.2013.02.001. HE H L. Physical criterion of dynamic tensile fracture [J]. Chinese Journal of High Pressure Physics, 2013, 27(2): 153–161. DOI: 10.11858/gywlxb.2013.02.001.
[4]	NOELL P J, CARROLL J D, BOYCE B L. The mechanisms of ductile rupture [J]. Acta Materialia, 2018, 161: 83–98. DOI: 10.1016/j.actamat.2018.09.006.
[5]	周洪强, 孙锦山, 王元书. 动载荷下延性材料中微孔洞的增长模型 [J]. 爆炸与冲击, 2003, 23(5): 415–419. DOI: 10.11883/1001-1455(2003)05-0415-5. ZHOU H Q, SUN J S, WANG Y S. The growth of microvoids in ductile materials under dynamic loading [J]. Explosion and Shock Waves, 2003, 23(5): 415–419. DOI: 10.11883/1001-1455(2003)05-0415-5.
[6]	WU X Y, RAMESH K T, WRIGHT T W. The effects of thermal softening and heat conduction on the dynamic growth of voids [J]. International Journal of Solids and Structures, 2003, 40(17): 4461–4478. DOI: 10.1016/S0020-7683(03)00214-2.
[7]	董杰, 李永池, 陈学东. 微孔洞唯象损伤力学模型及其应用 [J]. 爆炸与冲击, 2008, 28(5): 443–447. DOI: 10.11883/1001-1455(2008)05-0443-05. DONG J, LI Y C, CHEN X D. A phenomenological damage model of microvoids and its application [J]. Explosion and Shock Waves, 2008, 28(5): 443–447. DOI: 10.11883/1001-1455(2008)05-0443-05.
[8]	ORTIZ M, MOLINARI A. Effect of strain hardening and rate sensitivity on the dynamic growth of a void in a plastic material [J]. Journal of Applied Mechanics, 1992, 59(1): 48–53. DOI: 10.1115/1.2899463.
[9]	GURSON A L. Continuum theory of ductile rupture by void nucleation and growth: part I-Yield criteria and flow rules for porous ductile media [J]. Journal of Engineering Materials and Technology, 1977, 99(1): 2–15. DOI: 10.1115/1.3443401.
[10]	TVERGAARD V. Influence of voids on shear band instabilities under plane strain conditions [J]. International Journal of Fracture, 1981, 17(4): 389–407. DOI: 10.1007/BF00036191.
[11]	JOHNSON J N. Dynamic fracture and spallation in ductile solids [J]. Journal of Applied Physics, 1981, 52(4): 2812–2825. DOI: 10.1063/1.329011.
[12]	黄敏生, 李振环, 王乘, 等. 基体微尺度效应对弹塑性多孔洞材料本构势及孔洞长大的影响 [J]. 固体力学学报, 2003, 24(2): 137–147. DOI: 10.19636/j.cnki.cjsm42-1250/o3.2003.02.002. HUANG M S, LI Z H, WANG C, et al. The influences of microscope size effect of matrix on plastic potential and void growth of porous materials [J]. Acta Mechanica Solida Sinica, 2003, 24(2): 137–147. DOI: 10.19636/j.cnki.cjsm42-1250/o3.2003.02.002.
[13]	WEN J, HUANG Y, HWANG K C, et al. The modified Gurson model accounting for the void size effect [J]. International Journal of Plasticity, 2005, 21(2): 381–395. DOI: 10.1016/j.ijplas.2004.01.004.
[14]	王永刚, 贺红亮, 陈登平, 等. 延性金属层裂模型比较 [J]. 爆炸与冲击, 2005, 25(5): 467–471. DOI: 10.11883/1001-1455(2005)05-0467-05. WANG Y G, HE H L, CHEN D P, et al. Comparison of different spall models for simulating spallation in ductile metals [J]. Explosion and Shock Waves, 2005, 25(5): 467–471. DOI: 10.11883/1001-1455(2005)05-0467-05.
[15]	TRAIVIRATANA S, BRINGA E M, BENSON D J, et al. Void growth in metals: atomistic calculations [J]. Acta Materialia, 2008, 56(15): 3874–3886. DOI: 10.1016/j.actamat.2008.03.047.
[16]	MI C W, BUTTRY D A, SHARMA P, et al. Atomistic insights into dislocation-based mechanisms of void growth and coalescence [J]. Journal of the Mechanics and Physics of Solids, 2011, 59(9): 1858–1871. DOI: 10.1016/j.jmps.2011.05.008.
[17]	ZHANG Y Q, JIANG S Y. Investigation on dislocation-based mechanisms of void growth and coalescence in single crystal and nanotwinned nickels by molecular dynamics simulation [J]. Philosophical Magazine, 2017, 97(30): 2772–2794. DOI: 10.1080/14786435.2017.1352108.
[18]	JIANG D D, SHAO J L, HE A M, et al. Dynamic fracture characteristics of nanocrystalline Al containing He bubbles [J]. Scripta Materialia, 2023, 234: 115546. DOI: 10.1016/j.scriptamat.2023.115546.
[19]	YANG X, TIAN Y, ZHAO H, et al. Coupling of spallation and microjetting in aluminum at the atomic scale [J]. Physical Review B, 2024, 110(2): 024113. DOI: 10.1103/PhysRevB.110.024113.
[20]	ZHAO L Y, LIU Y. Investigation on void growth and coalescence in single crystal copper under high-strain-rate tensile loading by atomistic simulation [J]. Mechanics of Materials, 2020, 151: 103615. DOI: 10.1016/j.mechmat.2020.103615.
[21]	BISHARA D, XIE Y X, LIU W K, et al. A state-of-the-art review on machine learning-based multiscale modeling, simulation, homogenization and design of materials [J]. Archives of Computational Methods in Engineering, 2023, 30(1): 191–222. DOI: 10.1007/s11831-022-09795-8.
[22]	XIAO S P, HU R J, LI Z, et al. A machine-learning-enhanced hierarchical multiscale method for bridging from molecular dynamics to continua [J]. Neural Computing and Applications, 2020, 32(18): 14359–14373. DOI: 10.1007/s00521-019-04480-7.
[23]	TIONG L C O, LEE G, YI G H, et al. Predicting failure progressions of structural materials via deep learning based on void topology [J]. Acta Materialia, 2023, 250: 118862. DOI: 10.1016/j.actamat.2023.118862.
[24]	LIU D P, YANG H, ELKHODARY K I, et al. Mechanistically informed data-driven modeling of cyclic plasticity via artificial neural networks [J]. Computer Methods in Applied Mechanics and Engineering, 2022, 393: 114766. DOI: 10.1016/j.cma.2022.114766.
[25]	SU H, GUAN T Y, LIU Y. A three-dimensional prediction method of stiffness properties of composites based on deep learning [J]. Computational Mechanics, 2023, 71(3): 583–597. DOI: 10.1007/s00466-022-02253-z.
[26]	MISHIN Y, MEHL M J, PAPACONSTANTOPOULOS D A, et al. Structural stability and lattice defects in copper: ab initio, tight-binding, and embedded-atom calculations [J]. Physical Review B, 2001, 63(22): 224106. DOI: 10.1103/PhysRevB.63.224106.
[27]	STUKOWSKI A, ALBE K. Extracting dislocations and non-dislocation crystal defects from atomistic simulation data [J]. Modelling and Simulation in Materials Science and Engineering, 2010, 18(8): 085001. DOI: 10.1088/0965-0393/18/8/085001.
[28]	STUKOWSKI A, BULATOV V V, ARSENLIS A. Automated identification and indexing of dislocations in crystal interfaces [J]. Modelling and Simulation in Materials Science and Engineering, 2012, 20(8): 085007. DOI: 10.1088/0965-0393/20/8/085007.