Prediction of Modulus of Composite Materials by BP Neural Network Optimized by Genetic Algorithm

doi:10.16183/j.cnki.jsjtu.2021.126

Abstract

Abstract:

In order to reduce the cost of testing and shorten the design cycle, this paper studies the prediction method of the modulus of resin matrix composites based on the machine learning method. Using a new prediction method — the neural network in combination with the genetic algorithm (GA-ANN), the strength, the Poisson’s ratio, and the failure strain of the T800/epoxy composite material are used as three input variables of the back propagation (BP) neural network. Then, the optimal threshold and weight are obtained in the genetic algorithm (GA), which are assigned to the corresponding network parameters, and the BP neural network is updated for higher accuracy to predict the modulus of resin matrix composites. Under the same conditions, the Adam algorithm is used to predict. A comparison of these two methods fully proves the feasibility of the GA-ANN algorithm.

Key words: machine learning, back propagation (BP) neural network, genetic algorithm, composite material modulus, Adam algorithm

CLC Number:

TP181
V 258

WANG Zhuoxin, ZHAO Haitao, XIE Yuehan, REN Hantao, YUAN Mingqing, ZHANG Boming, CHEN Ji’an. Prediction of Modulus of Composite Materials by BP Neural Network Optimized by Genetic Algorithm[J]. Journal of Shanghai Jiao Tong University, 2022, 56(10): 1341-1348.

Figures/Tables 11

Tab.1

Fig.1

Tab.2

Tab.3

Fig.2

Fig.3

Tab.4

Comparison of GA-ANN and Adam-ANN training set prediction results

模型	$e -$ /%		RMS/GPa		SEP/GPa
模型	z₁	z₂	z₁	z₂	z₁	z₂
GA-ANN	1.66	2.47	3.939 9	4.614 8	3.939 9	4.614 3
Adam-ANN	1.64	2.50	3.847 3	4.674 7	3.847 2	4.673 7

Tab.4

Tab.5

Comparison of prediction results of GA-ANN and Adam-ANN test sets

模型	$e -$ /%		RMS/GPa		SEP/GPa
模型	z₁	z₂	z₁	z₂	z₁	z₂
GA-ANN	1.83	3.22	4.084 9	7.053 7	4.079 8	5.764 3
Adam-ANN	1.94	3.71	4.254 0	7.960 5	4.203 9	5.923 5

Tab.5

Fig.4

Fig.5

Fig.6

References 17

[1]	刘陈续, 于桂兰. 基于神经网络的层状周期结构能量传输谱预测[J]. 上海交通大学学报, 2021, 55(1): 88-95.
	LIU Chenxu, YU Guilan. Prediction of energy transmission spectrum of layered periodic structure based on neural network[J]. Journal of Shanghai Jiao Tong University, 2021, 55(1): 88-95.
[2]	GELAYOL G, MINOO N, KHASHAYAR B, et al. A machine learning case study with limited data for prediction of carbon fiber mechanical properties[J]. Computers in Industry, 2019, 105: 123-132. doi: 10.1016/j.compind.2018.11.004 URL
[3]	ZHANG Z S, HONG Y, HOU B, et al. Accelerated discoveries of mechanical properties of graphene using machine learning and high-throughput computation[J]. Carbon, 2019, 148: 115-123. doi: 10.1016/j.carbon.2019.03.046 URL
[4]	CHEN G, WANG H Y, BEZOLD A, et al. Strengths prediction of particulate reinforced metal matrix composites (PRMMCs) using direct method and artificial neural network[J]. Composite Structures, 2019, 223: 89-91.
[5]	QI Z C, ZHANG N X, LIU Y, et al. Prediction of mechanical properties of carbon fiber based on cross-scale FEM and machine learning[J]. Composite Structures, 2019, 212: 199-206. doi: 10.1016/j.compstruct.2019.01.042 URL
[6]	杨红, 程万里, 任丽丽. 高温高压蒸汽改性落叶松木材力学性能预测模型的建立[J]. 东北林业大学学报, 2016, 44(4): 77-80.
	YANG Hong, CHENG Wanli, REN Lili. Establishment of prediction model for mechanical properties of larch wood modified by high temperature and high pressure steam[J]. Journal of Northeast Forestry University, 2016, 44(4): 77-80.
[7]	白晓明. 基于数据挖掘的复合材料宏—细观力学模型研究[D]. 哈尔滨: 哈尔滨工业大学, 2016.
	BAI Xiaoming. Research on macro-micromechanics model of composite materials based on data mining[D]. Harbin:Harbin Institute of Technology, 2016.
[8]	张博. 稀土基化合物的磁熵变及其机器学习研究[D]. 北京: 中国科学院大学(中国科学院物理研究所), 2018.
	ZHANG Bo. Magnetic entropy change of rare earth-based compounds and its machine learning research[D]. Beijing: University of Chinese Academy of Sciences (Institute of Physics, Chinese Academy of Sciences), 2018.
[9]	QI Z C, LIU Y, CHEN W L. An approach to predict the mechanical properties of CFRP based on cross-scale simulation[J]. Composite Structures, 2018, 210: 339-347. doi: 10.1016/j.compstruct.2018.11.056 URL
[10]	CORTES C, VAPNIK V. Support-vector networks[J]. Machine Learning, 1995, 20(3): 273-297.
[11]	FU L, LI P F. The research survey of system identification method[C]//International Conference on Intelligent Human-machine Systems & Cybernetics. Hangzhou, China: IEEE Computer Society, 2013: 397-401.
[12]	DING S F, SU C Y, YU J Z. An optimizing BP neural network algorithm based on genetic algorithm[J]. Artificial Intelligence Review, 2011, 36(2): 153-162. doi: 10.1007/s10462-011-9208-z URL
[13]	曾坤, 姜志侠. 基于多遗传算法的BP神经网络人脸识别[J]. 计算机技术与发展, 2021, 31(1): 77-82.
	ZENG Kun, JIANG Zhixia. Face recognition of BP neural network based on multiple genetic algorithms[J]. Computer Technology and Development, 2021, 31(1): 77-82.
[14]	ZHANG H, LI S X, LIU X Y. Prediction of total order amount based on BP neural network optimized by genetic algorithm[C]//Proceedings of 2019 2nd International Conference on Mechanical, Electronic and Engineering Technology. Shanxi, China: Computer Science and Electronic Technology International Society, 2019: 106-111.
[15]	WANG X P, CAO L M. Genetic algorithm theory, application and software implementation[M]. Xi’an: WestAnn Traffic University Press, 2002.
[16]	KINGMA D, BA J. Adam: A method for stochastic optimization[DB/OL]. (2014-11-02)[2021-04-05]. https://xueshu.baidu.com/usercenter/paper/show?paperid=37a73866f09edd03830b234716447e4f.
[17]	朱攀星, 杨绍昌. X850树脂预浸料材料工艺性研究[J]. 科技展望, 2016, 26(24): 70+72.
	ZHU Panxing, YANG Shaochang. Research on the processability of X850 resin prepreg material[J]. Science and Technology Outlook, 2016, 26(24): 70+72.

试验编号	x	ν	y	z₁	z₂
1	0.463 068	0.637 681	0.403 461	0.652 174	0.60
2	0.414 773	0.543 478	0.234 987	0.608 696	0.60
3	0.696 023	0.471 014	0.388 657	0.608 696	0.20
4	0.510 417	0.710 145	0.434 320	0.608 696	0.55
5	0.754 735	0.731 884	0.645 538	0.521 739	0.70
6	0.775 568	0.789 855	0.604 671	0.565 217	0.70
7	0.778 409	0.347 826	0.576 939	0.608 696	0.70

隐藏层结构	e₁	e₂
2	2.087 7	2.765 7
6	2.094 0	2.739 1
10	2.076 5	2.533 3
14	2.068 1	2.647 5

隐藏层结构	e₁	e₂
6, 6	2.062 4	2.568 7
10, 6	2.071 9	2.759 9
12, 10	2.070 3	2.667 9
14, 12	2.027 8	2.580 9