基于机器学习的混凝土侵彻深度预测模型

doi:10.12382/bgxb.2023.0291

摘要/Abstract

摘要：

针对侵彻试验数据少且离散性较大导致预测侵彻深度的机器学习模型精度不高的问题,在收集大量侵彻试验数据的基础上,通过线性插值和添加高斯噪声等数据增强方法扩展侵彻数据,增大可用数据数量。采用遗传算法和贪心算法优化多层感知器、径向基神经网络、支持向量回归和极限梯度提升树4种常用机器学习模型的超参数,实现基于机器学习的混凝土侵彻深度预测。采用灵敏度分析方法分析侵彻深度对各弹靶参数的敏感程度。研究结果表明:采用线性插值和添加高斯噪声的方法可以有效地缓解数据不足的问题;采用数据增强后,多层感知器、径向基神经网络和极限梯度提升树的精度分别提高了2.49%、0.99%、0.74%和0.72%;弹体直径、着靶速度、弹体质量是对侵彻深度影响最大的参数;最优混凝土侵彻深度预测机器学习模型的平均误差为8.28%,该模型精度优于常用的侵彻深度预测经验公式。

关键词: 混凝土侵彻深度, 机器学习, 神经网络, 数据增强, 灵敏度分析

Abstract:

The test data about concrete penetration are often limited in quantity and unevenly distributed, which leads to the poor accuracy of machine learning-based models for predicting the penetration depth. This study aims to mitigate the unevenness of data distribution and increase the amount of available data to obtain an optimal machine learning model under the limitation of limited penetration test data. Based on collecting a large amount of penetration test data, the penetration data are extended by data augmentation methods such as linear interpolation and adding Gaussian noise. The genetic algorithm and greedy algorithm are used to optimize the hyperparameters of four common machine learning models: multilayer perceptron, radial basis neural network, support vector regression and extreme gradient boosting tree. The prediction of concrete penetration depth based on machine learning is realized. Sensitivity analysis method is used to analyze the influence of input factors on the penetration depth. The results show that the problem of insufficient data can be effectively alleviated by using linear interpolation and adding Gaussian noise. The accuracies of multilayer perceptron, radial basis neural network, support vector regression and extreme gradient boosting tree are improved by 2.49%, 0.99%, 0.74%, and 0.72%, respectively, after using data augmentation. The diameter, impact velocity and mass of projectile have the dominant influence on penetration depth. In addition, the average error of the optimal concrete penetration depth prediction machine learning model is 8.28%, and its global accuracy is better than the commonly used empirical formulas for predicting the penetration depth.

Key words: penetration depth of concrete, machine learning, neural network, data augmentation, sensitivity analysis

中图分类号:

O385

李萌, 武海军, 董恒, 任光, 张鹏, 黄风雷. 基于机器学习的混凝土侵彻深度预测模型[J]. 兵工学报, 2023, 44(12): 3771-3782.

LI Meng, WU Haijun, DONG Heng, REN Guang, ZHANG Peng, HUANG Fenglei. Machine Learning-based Models for Predicting the Penetration Depth of Concrete[J]. Acta Armamentarii, 2023, 44(12): 3771-3782.

图/表 22

参考文献 46

[1]	张树霞, 赵捍东, 韩志高. 基于 PSO-SVM 的侵彻效果预测方法[J]. 中北大学学报(自然科学版), 2015, 36(2):166-175.
	ZHANG S X, ZHAO H D, HAN Z G. Method of penetrate result prediction based on PSO-SVM[J]. Journal of North University of China(Natural Science Edition), 2015, 36(2):166-175. (in Chinese)
[2]	RYAN S, KANDANAARACHCHI S, SMITH-MILES K. Support vector machines for characterizing Whipple shield performance[J]. Procedia Engineering, 2015, 103: 522-529. doi: 10.1016/j.proeng.2015.04.068 URL
[3]	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.
[4]	RYAN S, THALER S. Artificial neural networks for characterizing Whipple shield performance[J]. International Journal of Impact Engineering, 2013, 58: 31-38. doi: 10.1016/j.ijimpeng.2013.03.003 URL
[5]	熊秋鹏. 基于神经网络技术的空间碎片损伤模式识别研究[D]. 哈尔滨: 哈尔滨工业大学, 2012.
	XIONG Q P. Research on space debris damage pattern recognition based on neural network technology[D]. Harbin:Harbin Institute of Technology, 2012. (in Chinese)
[6]	张帅. 弹丸侵彻钢筋混凝土多层靶板的数值模拟分析[D]. 南京: 南京理工大学. 2017.
	ZHANG S. Numerical simulation analysis of projectile penetration into reinforced concrete multilayer target slab[D]. Nanjing: Nanjing University of Science & Technology. 2017. (in Chinese)
[7]	王烁, 石全. 基于人工神经网络的侵彻毁伤模式识别[J]. 四川兵工学报, 2017, 38(10): 60-64.
	WANG S, SHI Q. Penetration pattern recognition based on artificial neural network[J]. Journal of Ordnance Equipment Engineering, 2017, 38(10): 60-64. (in Chinese)
[8]	李建光, 李永池, 王玉兰. 人工神经网络在弹体侵彻混凝土深度中的应用[J]. 中国工程科学, 2007, 9(8): 77-81.
	LI J G, LI Y C, WANG Y L. Penetration depth of projectiles into concrete using artificial neural network[J]. Strategic Study of Chinese Academy of Engineering, 2007, 9(8):77-81. (in Chinese)
[9]	KILIÇ N, EKICI B, HARTOMACIOĞLU S. Determination of penetration depth at high velocity impact using finite element method and artificial neural network tools[J]. Defence Technology, 2015, 11(2): 110-122. doi: 10.1016/j.dt.2014.12.001
[10]	AUTEN S J R, HAMMELL II R J. Comparing the prediction capabilities of an artificial neural network vs a phenomenological model for predicting the terminal ballistics of kinetic energy projectiles[C]//Proceedings of the Conference on Information Systems Applied Research. Wilmington, NC, US: Information Systems & Computing Academic Professionals, 2015.
[11]	GONZALEZ-CARRASCO I, GARCIA-CRESPO A, RUIZ-MEZCUA B, et al. Dealing with limited data in ballistic impact scenarios: an empirical comparison of different neural network approaches[J]. Applied Intelligence, 2011, 35(1): 89-109. doi: 10.1007/s10489-009-0205-8 URL
[12]	GONZALEZ-CARRASCO I, GARCIA-CRESPO A, RUIZ-MEZCUA B, et al. A neural network—based methodology for the recreation of high-speed impacts on metal armours[J]. Neural Computing and Applications, 2012, 21(1): 91-107. doi: 10.1007/s00521-011-0635-1 URL
[13]	GONZALEZ-CARRASCO I, GARCIA-CRESPO A, RUIZ-MEZCUA B, et al. An optimization methodology for machine learning strategies and regression problems in ballistic impact scenarios[J]. Applied Intelligence, 2012, 36(2): 424-441. doi: 10.1007/s10489-010-0269-5 URL
[14]	杨江, 张磊, 王继民, 等. 基于数据融合的混凝土侵彻深度预测[J]. 兵器材料科学与工程, 2020, 43(4): 40-45.
	YANG J, ZHANG L, WANG J M, et al. Prediction of concrete penetration depth based on data fusion[J]. Ordnance Material Science and Engineering, 2020, 43(4): 40-45. (in Chinese)
[15]	张磊, 吴昊, 赵强, 等. 基于数据挖掘技术的地下工程目标毁伤效应计算方法[J]. 爆炸与冲击, 2021, 41(3): 4-13.
	ZHANG L, WU H, ZHAO Q, et al. Calculation method of damage effects of underground engineering objectives based on data mining technology[J]. Explosion and Shock Waves, 2021, 41(3): 4-13. (in Chinese)
[16]	THOMPSON S, TEIXEIRA-DIAS F, PAULINO M, et al. Ballistic response of armour plates using generative adversarial networks[J]. Defence Technology, 2022, 18(9): 1513-1522. doi: 10.1016/j.dt.2021.08.001
[17]	CARGILE J D, O’NEIL E F, NEELEY B D. Very-high-strength concretes for use in blast- and penetration-resistant structures[J]. AMPTIAC Quarterly, 2002, 6(4): 61-66.
[18]	GOEL R A, CHANDRA S, CHANDOLA U C, et al. Behavior of concrete under high velocity impact[R]. New Delhi, India: Defence Research and Development Organisation, 1990.
[19]	WU H, FANG Q, CHEN X W, et al. Projectile penetration of ultra-high performance cement based composites at 510-1320 m/s[J]. Construction & Building Materials, 2015, 74: 188-200.
[20]	WU H, FANG Q, GONG J, et al. Projectile impact resistance of corundum aggregated UHP-SFRC[J]. International Journal of Impact Engineering, 2015, 84: 38-53. doi: 10.1016/j.ijimpeng.2015.05.007 URL
[21]	吴昊, 方秦, 龚自明. HSFRC靶体的弹体侵彻试验与理论分析[J]. 弹道学报, 2012, 24(3): 19-24,53.
	WU H, FANG Q, GONG Z M. Experiments and theoretical analyses on HSFRC target under the impact of rigid projectile[J]. Journal of Ballistics, 2012, 24(3): 19-24,53. (in Chinese)
[22]	KATAOKA S, BEPPU M, ICHINO H, et al. Failure behavior of reinforced concrete slabs subjected to moderate-velocity impact by a steel projectile[J]. International Journal of Protective Structures, 2017, 8(3): 384-406. doi: 10.1177/2041419617721550 URL
[23]	顾晓辉, 王晓鸣, 陈惠武, 等. 动能弹低速垂直侵彻钢筋混凝土的试验研究[J]. 南京理工大学学报(自然科学版), 2006, 30(1): 1-4.
	GU X H, WANG X M, CHEN H W, et al. Experimental studies on kinetic projectile’s direct penetrations with low-speed against reinforced concrete targets[J]. Journal of Nanjing University of Science and Technology, 2006, 30(1): 1-4. (in Chinese)
[24]	ZHANG X, WU H, ZHANG S, et al. Projectile penetration of reinforced concrete considering the effect of steel reinforcement: experimental study and theoretical analysis[J]. International Journal of Impact Engineering, 2020, 144: 103653. doi: 10.1016/j.ijimpeng.2020.103653 URL
[25]	CANFIELD J A, CLATOR I G. Development of a scaling law and techniques to investigate penetration in concrete[R]. Vicksburg, MI, US: United States Naval Weapons Laboratory, 1966.
[26]	FORRESTALT M J, ALTMAN B S, CARGILE J D, et al. An empirical equation for penetration depth of ogive-nose projectiles into concrete targets[J]. International Journal of Impact Engineering, 1994, 15(4): 395-405. doi: 10.1016/0734-743X(94)80024-4 URL
[27]	FORRESTAL M J, FREW D J, HICKERSON J P, et al. Penetration of concrete targets with deceleration-time measurements[J]. International Journal of Impact Engineering, 2003, 28(5): 479-497. doi: 10.1016/S0734-743X(02)00108-2 URL
[28]	FREW D J, HANCHAK S J, GREEN M L, et al. Penetration of concrete targets with ogive-nose steel rods[J]. International Journal of Impact Engineering, 1998, 21(6): 489-497. doi: 10.1016/S0734-743X(98)00008-6 URL
[29]	HANSSON H. Penetration in concrete for projectiles with L/D-9:FOI-R-1659-SE[R]. Tumba, Sweden: Swedish Defence Reasearch Agency (FOI), 2005.
[30]	SJØL H, TELAND J A, KALDHEIM Ø. Penetration into concrete-analysis of small scale experiments with 12 mm projectiles[R]. Kjeller, Norway: Norwegian Defence Research Establishment, 2002.
[31]	张雪岩. 高强混凝土的本构模型与抗侵彻机理研究[D]. 北京: 北京理工大学, 2021.
	ZHANG X Y. Research on constitutivemodel and anti-penetration mechanism of high-strength concrete[D]. Beijing: Beijing Institute of Technology, 2021. (in Chinese)
[32]	徐建波. 长杆射弹对混凝土的侵彻特性研究[D]. 长沙: 国防科学技术大学, 2001.
	XU J B. Investigations on long projectiles penetrating into concrete targets[D]. Changsha: National University to Defense Technology, 2021. (in Chinese)
[33]	徐建波, 林俊德, 唐润棣, 等. 长杆射弹侵彻混凝土实验研究[J]. 爆炸与冲击, 2002(2): 174-178.
	XU J B, LIN J D, TANG R D, et al. Experimental study on penetration of long rod projectile into concrete[J]. Explosion and Shock Waves, 2002(2): 174-178. (in Chinese)
[34]	武海军, 张雪岩, 张爽, 等. 刚性弹体侵彻/贯穿有限厚钢筋混凝土靶的弹道预测[J]. 兵工学报, 2020, 41(增刊2): 12-25.
	WU H J, ZHANG X Y, ZHANG S, et al. Prediction of trajectory of rigid projectile penetrating/perforating a finite-thick reinforced concrete target[J]. Acta Armamentarii, 2020, 41(S2): 12-25. (in Chinese)
[35]	吴成. 细观尺度下混凝土靶的抗侵彻性能研究[D]. 南京: 南京理工大学, 2019.
	WU C. Research on anti-penetration performance of concrete target at mesoscale[D]. Nanjing: Nanjing University of Science & Technology, 2019. (in Chinese)
[36]	沈兆武, 孙宇新, 李永池, 等. 机场跑道混凝土靶板贯穿的试验研究[J]. 实验力学, 2001(3): 264-269.
	SHEN Z W, SUN Y X, LI Y C, et al. Experimental study on concrete target plate penetration of airport runway[J]. Journal of experimental mechanics, 2001(3): 264-269. (in Chinese)
[37]	卢浩, 岳松林, 孙善政, 等. 混凝土靶侵爆条件下破坏深度的模型实验研究[J]. 爆炸与冲击, 2021, 41(7): 80-87.
	LU H, YUE S L, SUN S Z, et al. Model test study on damage depth of concrete target under penetration and explosion[J]. Explosion and Shock Waves, 2021, 41(7): 80-87. (in Chinese)
[38]	宋梅利, 李文彬, 王晓鸣, 等. 弹体高速侵彻效率的实验和量纲分析[J]. 爆炸与冲击, 2016, 36(6): 752-758.
	SONG M L, LI W B, WANG X M, et al. Experimental and dimensional analysis of high speed penetration efficiency of projectile body[J]. Explosion and Shock Waves, 2016, 36(6): 752-758. (in Chinese)
[39]	FENG J, SONG M L, SUN W W, et al. Thick plain concrete targets subjected to high speed penetration of 30CrMnSiNi2A steel projectiles: tests and analyses[J]. International Journal of Impact Engineering, 2018, 122: 305-317. doi: 10.1016/j.ijimpeng.2018.09.005 URL
[40]	GENG H, LU H, SUN S, et al. Experimental study and damage effect analysis of concrete structures under the combined loadings of penetration and explosion[J]. Shock and Vibration, 2020, 2020: 2137945.
[41]	房安琪, 李蓉. 基于数据增强的侵彻引信准确层识别神经网络方法[J]. 探测与控制学报, 2022, 44(1): 1-6.
	FANG A Q, LI R. Penetration fuze accurate layer recognition neural network data enhancement method[J]. Journal of detection & control, 2022, 44(1): 1-6. (in Chinese)
[42]	赵宏远, 武海军, 董恒, 等. 蜂窝钢管混凝土抗侵彻性能实验研究[J]. 爆炸与冲击, 2023, 5(4): 1-13.
	ZHAO H Y, WU H J, DONG H, et al. Experimental study on anti-penetration performance of concrete-filled steel tubular with honeycomb structure[J]. Explosion and Shock Waves, 2023, 5(4): 1-13. (in Chinese) doi: 10.1007/BF00757446 URL
[43]	雷英杰, 张善文. MATLAB遗传算法工具箱及应用[M]. 西安: 西安电子科技出版社, 2014.
	LEI Y J, ZHANG S W. MATLAB genetic algorithm toolbox and its application[M]. Xi’an: Xi’an Electronic Technology Publishing House, 2014. (in Chinese)
[44]	张爽. 弹体侵彻/贯穿钢筋混凝土靶机理与弹道轨迹研究[D]. 北京: 北京理工大学, 2018.
	ZHANG S. Investigation on the mechanism and trajectory of projectile penetrating / perforating reinforced concrete target[D]. Beijing: Beijing Institute of Technology, 2018. (in Chinese)
[45]	魏伟, 张竞文, 常思江, 等. 基于Sobol和EFAST法的枪弹转动角速度全局灵敏度仿真分析[J]. 兵工学报, 2022, 43(9): 2408-2416. doi: 10.12382/bgxb.2022.0546
	WEI W, ZHANG J W, CHANG S J, et al. Global sensitivity analysis of bullet’s angular velocity based on Sobol and EFAST method[J]. Acta Armamentarii, 2022, 43(9): 2408-2416. (in Chinese)
[46]	罗承昆, 陈云翔, 张执国, 等. 基于Sobol’法的歼击机需求影响因素灵敏度分析[J]. 火力与指挥控制, 2016, 41(10): 122-125,129.
	LUO C K, CHEN Y X, ZHANG Z G, et al. Sensitivity analysis on influencing factor s of fighter’s demand based on Sobol’ method[J]. Fire Control & Command Control, 2016, 41(10): 122-125,129. (in Chinese)

数据来源	试验数据数量	弹体质量 m/kg	弹体直径 d/mm	头部曲率半径CRH	速度v/ (m·s^-1)	靶板强度 f_c/MPa	钢纤维体积分数ρ_v/%	侵深 H/mm
Cargile等^[17]	11	0.91	26.9	2	239~780	85,157	2~3	78~687
Goel等^[18]	19	0.47	30.0	1	60~463	34	1.3	15~114
Wu等^[19]	31	0.33~0.40	25.3	3	510~1320	35~142	4	115~748
Wu等^[20]	14	0.34	25.3	3	503~846	103~129	2	21~191
吴昊等^[21]	16	0.33~0.35	25.3	3	491~858	67~125	1.5~3.0	116~346

数据来源	试验数据数量	弹体质量 m/kg	弹体直径 d/mm	头部曲率半径CRH	速度v/ (m·s^-1)	靶板强度 f_c/MPa	钢纤维体积分数ρ_v/%	侵深 H/mm
Cargile等^[17]	11	0.91	26.9	2	239~780	85,157	2~3	78~687
Goel等^[18]	19	0.47	30.0	1	60~463	34	1.3	15~114
Wu等^[19]	31	0.33~0.40	25.3	3	510~1320	35~142	4	115~748
Wu等^[20]	14	0.34	25.3	3	503~846	103~129	2	21~191
吴昊等^[21]	16	0.33~0.35	25.3	3	491~858	67~125	1.5~3.0	116~346

数据来源	试验数据数量	m/kg	d/mm	CRH	v/(m·s^-1)	f_c/MPa	体积配筋率r/%	H/mm
Geol等^[18]	19	0.47	30	1	60~463	34	4.31	15~114
Kataoka等^[22]	11	8.4	80	0.5	64~90	44	1.43	38~53
顾晓辉等^[23]	5	4.43	57	4	313~387	28	2.52	264~412
Zhang等^[24]	4	4.9	64	3	430~439	24	1.28	184~568

数据来源	试验数据数量	m/kg	d/mm	CRH	v/(m·s^-1)	f_c/MPa	体积配筋率r/%	H/mm
Geol等^[18]	19	0.47	30	1	60~463	34	4.31	15~114
Kataoka等^[22]	11	8.4	80	0.5	64~90	44	1.43	38~53
顾晓辉等^[23]	5	4.43	57	4	313~387	28	2.52	264~412
Zhang等^[24]	4	4.9	64	3	430~439	24	1.28	184~568

数据来源	试验数据数量	m/kg	d/mm	CRH	v/(m·s^-1)	f_c/MPa	H/mm
Canfield等^[25]	13	5.9	76.2	1.5	307~828	35	199~747
Forrestal等^[26]	17	0.9~0.91	26.9	2	277~800	32~108	173~958
Forrestal等^[27]	15	12.9~13.2	76.2	3,6	139~456	23,39	240~1250
Frew等^[28]	18	0.48~1.62	20.3,30.5	3	442~1225	58.4	287~2830
Hansson^[29]	22	3.65~4.5	50	3~12	406~464	48~146	250~1380
Sjøl等^[30]	77	0.87~454	11~307	0.7~1.5	160~908	24~200	71~1700
张雪岩^[31]	26	0.1	15	3	615~1325	35~117	117~1035
徐建波^[32]	11	0.46~6.21	10~40	3	206~657	15,17	145~1690
徐建波等^[33]	16	0.48~1.6	13,31	3,4.25	405~1201	51,63	300~2030
武海军等^[34]	16	0.1~0.13	15	3	498~1269	28,35	231~945
吴成^[35]	40	0.12~7.81	15~60	4	262~709	31~71	100~862
沈兆武等^[36]	19	0.24~1.12	27,46	5.3	148~465	24,30	96~260
卢浩等^[37]	10	0.168	20	2.5	479~675	31,40	211~279
宋梅利等^[38]	12	0.29~0.32	27,30	3	885~1872	40.8	170~780
Feng等^[39]	15	0.31~4.73	30~60	3	841~1872	48	310~3200
Geng等^[40]	10	0.168	20	2.5	465~671	35	197~352