基于神经网络的裂纹扩展过程预测
2021-02-18郑国君杜超群申国哲夏阳
郑国君 杜超群 申国哲 夏阳








摘要: 提出基于神经网络的裂纹扩展过程实时预测方法,其计算效率比近场动力学(peridynamic, PD)模型提高。使用PD算法获取裂纹扩展过程中的损伤云图,构建裂纹扩展数据集。基于数据集构建生成对抗网络(generative adversarial networks, GAN)模型,根据不同加载条件实时生成损伤云图,从而快速预测裂纹的扩展过程。将PD模型计算得到的损伤云图中的RGB值与相应位置处的损伤值结合,构建多层前馈神经网络模型的数据库,并使用多层前馈神经网络模型分析GAN模型产生的损伤云图,得到相应的损伤值。通过数值算例证明该方法的效率和准确性。
关键词: 近场动力学; 计算时间; 生成对抗网络; 裂纹扩展; 裂纹预测; 损伤云图; 前馈神经网络; 损伤值预测
中图分类号: TP391.99; TB115.1文献标志码: B
Prediction of crack propagation process
based on neural network
ZHENG Guojun, DU Chaoqun, SHEN Guozhe, XIA Yang
(State Key Laboratory of Structural Analysis for Industrial Equipment, School of Automotive Engineering,
Dalian University of Technology, Dalian 116024, Liaoning, China)
Abstract: A method of real-time prediction of the crack propagation process based on the neural network is proposed, and the calculation efficiency can be improved compared with peridynamic(PD) model. The damage contours of the crack propagation process are obtained using the PD model, and then the crack propagation dataset is built. Based on the dataset, the generative adversarial network(GAN) model is constructed. The damage contours are generated in real time according to differential loading conditions, thus the crack propagation process can be predicted quickly. The damage value database is built based on the RGB color values and the damage values at the corresponding position from the damage contours generated by the PD model. The damage contours generated by GAN model is analyzed using the multi-layer feedforward neural network model, and then the corresponding damage values are obtained. The efficiency and accuracy of the method are proved by numerical examples.
Key words: peridynamics; calculation time; generative adversarial network; crack propagation; crack prediction; damage contour; feedforward neural network; damage value prediction
收稿日期: 2021-08-16修回日期: 2021-10-21
基金項目: 国家自然科学基金(12072065);中央高校基本科研业务费专项资金(DUT20JC34)
作者简介: 郑国君(1981—),男,湖北恩施人,副教授,研究方向为车身结构轻量化,(E-mail)gj_zheng@dlut.edu.cn
通信作者: 夏阳(1987—),男,河南驻马店人,副教授,研究方向为近场动力学等数值模拟方法,(E-mail)yangxia@dlut.edu.cn0引言经典连续介质力学是一种局部理论,其运动方程基于微分形式进行求解,当位移不连续时无法确定当前的运动方程,因此到目前为止依然难以通过连续介质力学对材料中裂纹的形成与扩展进行准确预测。SILLING等[1-4]提出非局部近场动力学(peridynamics,PD)理论,使用积分形式构建运动方程,实现材料在多部位萌生并沿任意路径进行扩展的计算。PD理论可以应用于线弹性、非线弹性、塑性、黏塑性以及黏弹性等[5-8]材料的行为,同时也可以应用于多晶材料[9]和复合材料[10]。……
