Research on Machine Learning Based Model for Predicting the Impact Status of Laminated Glass

doi:10.15541/jim20200187

Abstract

Abstract:

Architectural laminated glass exhibits significant vulnerability under hard body impacts such as windborne debris impacts. In this work, a prediction model is proposed for assessing the impact status of laminated glass under hard body impact. Multiple design variables including the glass make-ups, interlayer types, support conditions and size are considered. The impact tests with consecutive impact attempts are first conducted. A comprehensive database encompassing the failure condition of each glass layer is then established. This database has 567 groups of PVB laminated glass data and 210 groups of SGP laminated glass data. A combined WOA-KELM machining learning based model is subsequently developed to predict the impact status of laminated glass. The modelling results are compared with that from SVM and LSSVM based models. The results show that the proposed model has a prediction accuracy of 88.45% in failure status of each glass layer. Such model can well predict the impact status of laminated glass and shows better performance in both accuracy and computation cost than other models.

Key words: laminated glass, impact failure, machine learning, kernel extreme learning machine

CLC Number:

TQ171

MENG Yanran, WANG Xinger, YANG Jian, XU Han, YUE Feng. Research on Machine Learning Based Model for Predicting the Impact Status of Laminated Glass[J]. Journal of Inorganic Materials, 2021, 36(1): 61-68.

Figures/Tables 15

References 30

[1]	KAISER NATHAN D, BEHR RICHARD A, MINOR JOSEPH E , et al. Impact resistance of laminated glass using “sacrificial ply” design concept. Journal of Architectural Engineering, 2000,6(1):24-34.
[2]	SAXE TIMOTHY J, BEHR RICHARD A, MINOR JOSEPH E , et al. Effects of missile size and glass type on impact resistance of “sacrificial ply” laminated glass. Journal of Architectural Engineering, 2002,8(1):24-39.
[3]	WANG XING-ER, YANG JIAN, LIU QIANG , et al. Experimental investigations into SGP laminated glass under low velocity impact. International Journal of Impact Engineering, 2018,122:91-108.
[4]	ZHANG XI-HONG, HAO HONG, MA GUO-WEI . Laboratory test and numerical simulation of laminated glass window vulnerability to debris impact. International Journal of Impact Engineering, 2013,55:49-62.
[5]	CHEN SHUN-HUA, ZANG MENG-YAN, WANG DI , et al. Finite element modelling of impact damage in polyvinyl butyral laminated glass. Composite Structures, 2016,138:1-11.
[6]	WANG XING-ER, YANG JIAN, LIU QING-FENG , et al. A comparative study of numerical modelling techniques for the fracture of brittle materials with specific reference to glass. Engineering Structures, 2017,152:493-505.
[7]	MOHAGHEGHIAN IMAN, WANG Y, JIANG L , et al. Quasi-static bending and low velocity impact performance of monolithic and laminated glass windows employing chemically strengthened glass. European Journal of Mechanics-A/Solids, 2017,63:165-186.
[8]	ALTER CHRISTIAN, KOLLING STEFAN, SCHNEIDER JENS . An enhanced non-local failure criterion for laminated glass under low velocity impact. International Journal of Impact Engineering, 2017,109:342-353.
[9]	ZHANG YANG-MEI, WANG XING-ER, YANG JIAN . Experimental study of multiple layered SGP laminated glass under hard body impact. Journal of Inorganic Materials, 2018,33(10):1110-1118.
[10]	WANG XING-ER, YANG JIAN, WANG FEILIANG , et al. Simulating the impact damage of laminated glass considering mixed mode delamination using FEM/DEM. Composite Structures, 2018,202:1239-1252.
[11]	LIU XIAO-GEN, BAO YI-WANG, SONG YI-LE , et al. The calculation method The calculation method and influence factor to the natural frequency of laminated glass. Bulletin of the Chinese Ceramic Society, 2008, (27) 5:918-923.
[12]	DAS SANTANU, SRIVASTAVA ASHOK N, CHATTOPADHYAY ADITI . Classification of Damage Signatures in Composite Plates using One-class SVMs. 2007 IEEE Aerospace Conference. MT, USA. 2007: 1-19.
[13]	LI HONG-NAN, GAO DONG-WEI, YI TING-HUA . Advances in structural health monitoring systems in civil engineering. Advances in Mechanics, 2008(02):151-166.
[14]	JIANG ZHEN-YU, ZHANG ZHONG, FRIEDRICH KLAUS . Prediction on wear properties of polymer composites with artificial neural networks. Composites Science and Technology, 2007,67(2):168-176.
[15]	CHATTERJEE SANKHADEEP, SARKAR SARBARTHA, HORE SIRSHENDU , et al. Particle swarm optimization trained neural network for structural failure prediction of multistoried RC buildings. Neural Computing and Applications, 2017,28(8):2005-2016.
[16]	XU HAN, YANG JIAN, WANG XING-ER , et al. Application of back propagation neural network on debonding prediction of glass curtain walls with concealed frames. Journal of the Chinese Ceramic Society, 2019,47(8):1073-1079.
[17]	SHENG MING-JIAN, CHEN PU-HUI, QIAN YI-BIN . An estimating method of compressive strength of composite laminates after low-velocity impact. Journa of Shanghai Jiao Tong University, 2019,53(10):1182-1186.
[18]	ZHANG YONG-ZHEN, TONG XIAO-YAN, YAO LEI-JIANG , et al. Acoustic emission pattern recognition on tensile damage process of C/SiC composites using an improved genetic algorithm. Journal of Inorganic Materials, 2020,35(5):593-600.
[19]	HUANG GUANG-BIN, DING XIAO-JIAN, ZHOU HONG-MING . Optimization method based extreme learning machine for classification. Neurocomputing, 2010,74(1/2/3):155-163.
[20]	MIRJALILI SEYEDALI, LEWIS ANDREW . The whale optimization algorithm. Advances in Engineering Software, 2016,95:51-67.
[21]	VEER FA, LOUTER PIETER CHRISTIAAN, BOS FP . The strength of annealed, heat-strengthened and fully tempered float glass. Fatigue & Fracture of Engineering Materials & Structures, 2009,32(1):18-25.
[22]	FAWCETT TOM . An introduction to ROC analysis. Pattern Recognition Letters, 2006,27(8):861-874.
[23]	HUANG J, LU J, LING C X. Comparing Naive Bayes, Decision trees, and SVM with AUC and Accuracy. Third IEEE International Conference on Data Mining. Melbourne, FL, USA, USA. 2003: 553-556.
[24]	BARAKAT N, BRADLEY A P . Rule Extraction from Support Vector Machines: Measuring the Explanation Capability Using the Area under the ROC Curve. 18th International Conference on Pattern Recognition (ICPR'06). 2006: 812-815.
[25]	ZHANG X, JIANG C . Improved SVM for Learning Multi-Class Domains with ROC Evaluation. 2007 International Conference on Machine Learning and Cybernetics. Hong Kong, China. 2007: 2891-2896.
[26]	TESFAMARIAM SOLOMON, LIU ZHENG . Earthquake induced damage classification for reinforced concrete buildings. Structural Safety, 2010,32(2):154-164.
[27]	GUI GUO-QING, PAN HONG, LIN ZHI-BIN, et al. Data-driven support vector machine with optimization techniques for structural health monitoring and damage detection. KSCE Journal of Civil Engineering, 2017,21(2):523-534.
[28]	HUANG GUANG-BIN, WANG DIAN HUI, LAN YUAN . Extreme learning machines: a survey. International. Journal of Machine Learning and Cybernetics, 2011,2(2):107-122.
[29]	HO SL, YANG SHI-YOU, NI GUANG-ZHENG , et al. A particle swarm optimization-based method for multiobjective design optimizations. IEEE Transactions on Magnetics, 2005,41(5):1756-1759.
[30]	YANG XI-YUN, GUAN WEN-YUAN, LIU YU-QI , et al. Prediction intervals forecasts of wind power based on PSO-KELM. Proceedings of the CSEE, 2015,35(S1):146-153.

ID	Material	Make-up (o/m/i)	Dimensional of glass/mm	Support condition	Quantity
P01	FTG/PVB/FTG	8/1.52/8	1000 × 1000	Edge clamped	12
P02	FTG/PVB/FTG	8/1.52/8	1000 × 1000	Bolted connection	6
P03	FTG/PVB/FTG	8/1.52/8	1500 × 1500	Edge clamped	3
P04	FTG/PVB/FTG	8/1.52/8	1500 × 1500	Bolted connection	3
P05	HSG/PVB/HSG	8/1.52/8	1000 × 1000	Bolted connection	3
P06	FTG/PVB/FTG	8/0.76/8	1000 × 1000	Bolted connection	3
P07	FTG/PVB/FTG	8/3.04/8	1000 × 1000	Bolted connection	3
P08	FTG/PVB/FTG	8/1.52/10	1000 × 1000	Bolted connection	3
P09	FTG/PVB/FTG	6/1.52/10	1000 × 1000	Bolted connection	3
P10	FTG/PVB/HSG	8/1.52/8	1000 × 1000	Bolted connection	3
P11	HSG/PVB/FTG	8/1.52/8	1000 × 1000	Bolted connection	3
S01	ANG/SGP/FTG	8/3/8	1000 × 1000	Edge clamped	3
S02	FTG/SGP/ANG	8/3/8	1000 × 1000	Edge clamped	3
S03	FTG/SGP/FTG	8/3/8	1000 × 1000	Bolted connection	6
S04	FTG/SGP/FTG	8/3/8	1500 × 1500	Bolted connection	3
S05	HSG/SGP/FTG	8/3/8	1000 × 1000	Bolted connection	3
S06	FTG/SGP/FTG	8/5/8	1000 × 1000	Bolted connection	3

ID	Material	Make-up (o/m/i)	Dimensional of glass/mm	Support condition	Quantity
P01	FTG/PVB/FTG	8/1.52/8	1000 × 1000	Edge clamped	12
P02	FTG/PVB/FTG	8/1.52/8	1000 × 1000	Bolted connection	6
P03	FTG/PVB/FTG	8/1.52/8	1500 × 1500	Edge clamped	3
P04	FTG/PVB/FTG	8/1.52/8	1500 × 1500	Bolted connection	3
P05	HSG/PVB/HSG	8/1.52/8	1000 × 1000	Bolted connection	3
P06	FTG/PVB/FTG	8/0.76/8	1000 × 1000	Bolted connection	3
P07	FTG/PVB/FTG	8/3.04/8	1000 × 1000	Bolted connection	3
P08	FTG/PVB/FTG	8/1.52/10	1000 × 1000	Bolted connection	3
P09	FTG/PVB/FTG	6/1.52/10	1000 × 1000	Bolted connection	3
P10	FTG/PVB/HSG	8/1.52/8	1000 × 1000	Bolted connection	3
P11	HSG/PVB/FTG	8/1.52/8	1000 × 1000	Bolted connection	3
S01	ANG/SGP/FTG	8/3/8	1000 × 1000	Edge clamped	3
S02	FTG/SGP/ANG	8/3/8	1000 × 1000	Edge clamped	3
S03	FTG/SGP/FTG	8/3/8	1000 × 1000	Bolted connection	6
S04	FTG/SGP/FTG	8/3/8	1500 × 1500	Bolted connection	3
S05	HSG/SGP/FTG	8/3/8	1000 × 1000	Bolted connection	3
S06	FTG/SGP/FTG	8/5/8	1000 × 1000	Bolted connection	3

Material parameter	FTG	HSG	ANG	PVB	SGP
Density/(kg·m^-3)	-	2500.0	-	1100.00	950.00
Elasticity modulus/GPa	-	70.0	-	0.15	0.30
Poission ratio	-	0.2	-	0.45	0.45
Mean failure strength/MPa	157.4	104.0	42.0	-	-

Material parameter	FTG	HSG	ANG	PVB	SGP
Density/(kg·m^-3)	-	2500.0	-	1100.00	950.00
Elasticity modulus/GPa	-	70.0	-	0.15	0.30
Poission ratio	-	0.2	-	0.45	0.45
Mean failure strength/MPa	157.4	104.0	42.0	-	-

n	Input parameter	Outer layer state AUC (Ao n)	Inner layer state AUC (Ai n)
1	Thickness of interlayer	0.605	0.571
2	Thickness of outer layer	0.516	0.537
3	Thickness of inner layer	0.511	0.546
4	Type of interlayer	0.573	0.576
5	Type of outer layer	0.573	0.515
6	Type of inner layer	0.563	0.559
7	Side length	0.516	0.507
8	Boundary condition	0.541	0.573
9	Peak kinetic energy	0.654	0.675
10	State of outer layer	0.873	0.513
11	State of inner layer	0.472	0.714
12	Multiple input	0.916	0.842