基于卷积神经网络的夹层玻璃开裂后拉伸性能快速评估方法  

Fast Evaluation Method for Post-Fracture Tensile Properties of Laminated Glass Based on Convolutional Neural Network

作　　者：尹俊熙 彭沈楠 王星尔 杨健[1,2] YIN Junxi;PENG Shennan;WANG Xinger;YANG Jian(State Key Laboratory of Ocean Engineering,Shanghai Jiao Tong University,Shanghai 200240,China;Shanghai Key Laboratory for Digital Maintenance of Buildings and Infrastructure,School of Ocean and Civil Engineering,Shanghai Jiao Tong University,Shanghai 200240,China)

机构地区：[1]上海交通大学海洋工程全国重点实验室,上海200240 [2]上海交通大学船舶海洋与建筑工程学院,上海市公共建筑和基础设施数字化运维重点实验室,上海200240

出　　处：《硅酸盐通报》2024年第12期4588-4596,共9页Bulletin of the Chinese Ceramic Society

基　　金：国家自然科学基金(52078293)。

摘　　要：夹层玻璃开裂后的承载性能对持续风灾或爆炸时建筑玻璃结构安全性非常重要。玻璃裂纹形态作为开裂后性能的关键影响因素,对夹层玻璃的拉伸硬化效应和局部失效触发有着重要作用。对采用离子性中间层(SG)的钢化夹层玻璃进行碎裂试验及碎裂后单轴拉伸试验,获取图形和力学性能的试验数据集,并和精细有限元模型校核,确定裂纹形态对应的碎片密度、最小最近邻距离、有效粘结系数等关键参数。随后,建立由表面应力调控裂纹形态的夹层玻璃开裂后有限元模型批量化生成方法,基于维诺(Voronoi)形态近似来扩充模拟数据集。采用卷积神经网络方法,基于裂纹形态图像识别训练,得到具有良好精度的夹层玻璃开裂后等效拉伸性能快速评估方法。The post-fracture load-bearing properties of laminated glass is crucial for the safety of architectural glass structures under long-duration wind disasters or explosions.Crack morphology significantly impacts the tension-stiffening effect and the triggering of local failure in laminated glass.Fragmentation tests and post-fracture uniaxial tensile tests were conducted on laminated tempered glass with SG interlayer to obtain the testing datasets of crack morphological images and associated mechanical properties.The refined finite element(FE)models were then calibrated with experimental outcomes to determine key parameters such as fragment density,minimum nearest neighbour distance,and effective bonding coefficients corresponding to the crack morphology.Subsequently,the batch generation of FE models for fractured laminated glass by adjusting surface compressive stress was established,which utilized Voronoi tessellation to approximate the actual morphology to expand the simulation dataset.The fast evaluation method based on convolutional neural network(CNN)with high accuracy for the equivalent tensile properties of fractured laminated glass was developed via image recognition training on crack morphology.

关 键 词：夹层玻璃 SG中间层 开裂后性能 裂纹形态 快速评估方法 卷积神经网络 

分 类 号：TQ171[化学工程—玻璃工业]