基于CVAE的超高速碰撞碎片云运动过程的快速预测技术

周浩; 李毅; 张浩; 陈鸿; 任磊生

doi:10.11729/syltlx20200058

基于CVAE的超高速碰撞碎片云运动过程的快速预测技术

doi: 10.11729/syltlx20200058

周浩^1,,
李毅^1, ,,
张浩²,
陈鸿¹,
任磊生¹

1.
中国空气动力研究与发展中心超高速空气动力研究所，四川绵阳　621000
2.
中国工程物理研究院计算机应用研究所，四川绵阳　621000

详细信息

作者简介:
周浩：（1986-），男，湖南常德人，工程师。研究方向：超高速碰撞数值模拟。通信地址：四川省绵阳市二环路南段6号（621000）。E-mail：499907834@qq.com

通讯作者:
E-mail：liyi@cardc.cn

中图分类号: V19
计量
- 文章访问数: 668
- HTML全文浏览量: 214
- PDF下载量: 35
- 被引次数: 0
出版历程
- 收稿日期: 2020-04-17
- 修回日期: 2020-09-13
- 网络出版日期: 2021-11-15
- 刊出日期: 2021-11-05

Evolution prediction of HVI debris based on CVAE model

ZHOU Hao^1
,,
LI Yi^{1
, ,},
ZHANG Hao²,
CHEN Hong¹,
REN Leisheng¹

1.
Hypervelocity Aerodynamics Institute, China Aerodynamics Research and Development Center, Mianyang Sichuan　621000, China
2.
Institute of Computer Application, China Academy of Engineering Physics, Mianyang Sichuan　621000, China

摘要

摘要: 在航天器防护构型设计中，需要快速、精确预测空间碎片超高速撞击防护屏产生碎片云的质量分布及其运动过程。采用深度学习方法，基于条件变分自编码器（CVAE）模型和大量铝球超高速正撞击铝板的光滑粒子流体动力学（SPH）方法的数值模拟结果，初步构建了碎片云空间质量分布与运动特征的快速预测模型。数值模拟中把铝球速度（3.00～8.00 km/s）、铝球半径（2.00～8.00 mm）、铝板厚度（1.000～4.000 mm）以及观测时间（1.0～12.0 μs） 4个变量作为输入控制参数，生成大量格式统一的训练集数据。模型隐藏层采用200个特征数据来描述碎片云质量分布，训练集参数范围内平均误差在0.6%以内，生成一个碎片云质量分布的平均时间小于7 ms。
- 超高速碰撞 /
- 碎片云 /
- 深度学习 /
- 条件变分自编码器 /
- 防护屏
Abstract: Efficiently and accurately predicting the evolution of hypervelocity impact debris is crucial in the design of spacecraft protective structures. To this end a deep learning model is constructed. The model is based on the conditional variation auto encoder (CVAE) and massive smoothed particle hydrodynamics (SPH) simulations. The model includes four controllable labels, namely projectile velocity (3.00-8.00 km/s) radius (2.00-8.00 mm), target plate thickness (1.000-4.000 mm), and time instant (1.0-12.0 μs). The model uses 200 parameters to describe the debris mass distribution, resulting in an average error less than 0.6%. It takes less than 7 micro seconds to predict one debris mass distribution.
- HVI /
- debris /
- deep learning /
- CVAE /
- protective structures

HTML全文

图 1 典型超高速碰撞碎片云工程模型

Figure 1. Typical HVI debris engineering model

下载: 全尺寸图片幻灯片

图 2 锌弹丸超高速撞击锌板

Figure 2. Zinc projectile impact Zinc plate

下载: 全尺寸图片幻灯片

图 3 铜弹丸超高速撞击铝板

Figure 3. Copper projectile impact aluminum plate

下载: 全尺寸图片幻灯片

图 4 典型超高速碰撞碎片云发展过程试验（左）与SPH数值模拟（右）结果对比

Figure 4. Comparison of experiment and SPH simulation for a typical HVI impact

下载: 全尺寸图片幻灯片

图 5 CVAE模型

Figure 5. Structure of the CVAE model

下载: 全尺寸图片幻灯片

图 6 CVAE模型训练过程

Figure 6. Training of the CVAE model

下载: 全尺寸图片幻灯片

图 7 碎片云发展过程的CVAE模型预测与数值模拟结果对比（v = 8.00 km/s，d = 4.000 mm，r = 2.00 mm）

Figure 7. Comparison of the CVAE model prediction and numerical simula-tion of the evolution of debris(v = 8.00 km/s，d = 4.000 mm，r =2.00 mm)

下载: 全尺寸图片幻灯片

图 8 碎片云发展过程的CVAE模型预测与数值模拟结果对比（v = 8.00 km/s，d = 1.000 mm，r = 8.00 mm）

Figure 8. Comparison of the CVAE model prediction and numerical simu-lation of the evolution of debris(v = 8 km/s，d = 1 mm，r = 8 mm)

下载: 全尺寸图片幻灯片

图 9 训练集中2016个数据的平均误差

Figure 9. Average error of 2016 data on the training set

下载: 全尺寸图片幻灯片

图 10 模型在速度上的内插能力（d = 1.000 mm，r = 8.00 mm，t = 12.0 μs）

Figure 10. Interpolation capability of the model at the velocity direction(d = 1.000 mm， r = 8.00 mm， t = 12.0 μs)

下载: 全尺寸图片幻灯片

图 11 测试集中81个数据的平均误差

Figure 11. Average error of 81 data on the testing set

下载: 全尺寸图片幻灯片

图 12 模型在速度上的外插能力

Figure 12. Extrapolation capability of the model at the velocity direction

下载: 全尺寸图片幻灯片

参考文献(20)

[1]	CHRISTIANSEN E L. Meteoroid/debris shielding[R]. TP-2003-210788, 2003.
[2]	WHIPPLE F L. Meteorites and space travel[J]. The Astronomical Journal,1947,52(5):131-137. doi: 10.1086/106009
[3]	SWIFT H F,BAMFORD R,CHEN R. Designing space vehicle shields for meteoroid protection: a new analysis[J]. Advances in Space Research,1982,2(12):219-234. doi: 10.1016/0273-1177(82)90311-8
[4]	SCHÄFER F K. An engineering fragmentation model for the impact of spherical projectiles on thin metallic plates[J]. International Jour-nal of Impact Engineering,2006,33(1):745-762. doi: 10.1016/j.ijimpeng.2006.09.067
[5]	CORVONATO E,DESTEFANIS R,FARAUD M. Integral model for the description of the debris cloud structure and impact[J]. Inter-national Journal of Impact Engineering,2001,26(1):115-128. doi: 10.1016/S0734-743X(01)00074-4
[6]	HUANG J,MA Z X,REN L S,et al. A new engineering model of debris cloud produced by hypervelocity impact[J]. International Jour-nal of Impact Engineering,2013,56:32-39. doi: 10.1016/j.ijimpeng.2012.07.003
[7]	MA Z X, HUANG J, LIANG S C. The technology of modeling debris cloud produced by hypervelocity impact[C]//Proc of the 6th European Conference on Space Debris. 2013.
[8]	RYAN S,THALER S,KANDANAARACHCHI S. Machine learning methods for predicting the outcome of hypervelocity impact events[J]. Expert Systems With Applications,2016,45:23-39. doi: 10.1016/j.eswa.2015.09.038
[9]	HOSSEINI M,ABBAS H. Neural network approach for estimation of hole-diameter in thin plates perforated by spherical projectiles[J]. Thin-Walled Structures,2008,46(6):592-601. doi: 10.1016/j.tws.2008.01.012
[10]	刘源,庞宝君. 基于贝叶斯正则化BP神经网络的铝平板超高速撞击损伤模式识别[J]. 振动与冲击,2016,35(12):22-27. LIU Y,PANG B J. Hypervelocity impact damage pattern recognition on aluminum plates based on Bayesian Regularization BP neural network[J]. Journal of Vibration and Shock,2016,35(12):22-27.
[11]	KINGMA D P, WELLING M. Auto-encoding variational Bayes[EB/OL]. [2013-12-20]. https://arxiv.org/abs/1312.6114.
[12]	KIHYUK S. Learning structured output representation using deep conditional generative models[C]//Proc of the International Conferen-ce on Neural Information Processing Systems. 2015.
[13]	T HENRY DE FRAHAN M,YELLAPANTULA S,KING R,et al. Deep learning for presumed probability density function models[J]. Combustion and Flame,2019,208:436-450. doi: 10.1016/j.combustflame.2019.07.015
[14]	EISMANN S, BARTZSCH S, ERMON S. Shape optimization in la-minar flow with a label-guided variational autoencoder[EB/OL]. [2017-12-10]. https://arxiv.org/abs/1712.03599.
[15]	BHOWMIK D,GAO S,YOUNG M T,et al. Deep clustering of protein folding simulations[J]. BMC Bioinformatics,2018,19(18):47-58. doi: 10.1186/s12859-018-2507-5
[16]	CHEN H M,ENGKVIST O,WANG Y H,et al. The rise of deep learning in drug discovery[J]. Drug Discovery Today,2018,23(6):1241-1250. doi: 10.1016/j.drudis.2018.01.039
[17]	文永,张浩,李毅,等. 基于深度学习的超高速碰撞碎片云生成模型[J]. 固体力学学报,2020,41(5):455-469. doi: 10.19636/j.cnki.cjsm42-1250/o3.2020.029 Wen Y,Zhang H,Li Y,et al. Deep learning based HIV debris generate model[J]. Acta Mechanica Solida Sinica,2020,41(5):455-469. doi: 10.19636/j.cnki.cjsm42-1250/o3.2020.029
[18]	李毅, 柳森, 陈鸿, 等. 并行SPH仿真软件PTS的开发与应用[C]//第二届全国超高速碰撞学术会议论文集. 2016. Li Y, Liu S, Chen H, et al. Develop and application of the parallel SPH software PTS[C]//Proc of the Second National Symposium on HVI. 2016.
[19]	MULLIN S A,LITTLEFIELD D L,ANDERSON C E Jr,et al. An examination of velocity scaling[J]. International Journal of Impact Engineering,1995,17(4):571-581. doi: 10.1016/0734-743x(95)99881-q
[20]	柯发伟,黄洁,李鑫,等. 超高速撞击铝板碎片云测量[J]. 空间碎片研究,2019,19(2):18-24. KE F W,HUANG J,LI X,et al. Measurement on debris cloud produced by impacting Al plate with hypervelocity[J]. Space Debris Research,2019,19(2):18-24.