摘要
为准确预测多相材料在任意变形模式下的力学响应,以双相钢细观模型为研究对象,通过高斯过程生成5000组任意加载路径,按19∶1的比例划分训练集与测试集,基于深度学习法预测双相钢细观模型在未知加载路径下的应力-应变响应。结果表明:基于时序的LSTM深度学习模型可捕捉到细观模型应力-应变非线性映射关系的转变,预测的正应力均方根误差为0.062,切应力均方根误差为0.108,局部最大累积塑性应变均方根误差为0.115,证明了基于深度学习模型与细观模型结合的数据驱动型材料本构关系预测的可行性。
In order to accurately predict the mechanical response of multiphase materials under any deformation mode,5000 sets of arbitrary loading paths were generated by Gaussian process for the dual-phase steel meso-model,and the training set and test set were divided according to the proportion of 19∶1.The stress and strain responses of the dual-phase steel meso-model under unknown loading paths were predicted based on deep learning method.The results show that the LSTM deep learning model based on time series data can capture the transformation of the nonlinear mapping relationship between stress and strain in the meso model.The root mean square error of normal stress is 0.062,the root mean square error of shear stress is 0.108,and the root mean square error of local maximum cumulative plastic strain is 0.115.The feasibility of data-driven material constitutive relationship prediction based on the combination of deep learning model and mesoscopic model is proved.
作者
李智琪
方宇轩
陆阿飞
沈涛
LI Zhiqi;FANG Yuxuan;LU Afei;SHEN Tao(Faculty of Mechanical Engineering and Mechanics,Ningbo University,Ningbo 315211,China)
出处
《兵器材料科学与工程》
CAS
CSCD
北大核心
2024年第5期117-123,共7页
Ordnance Material Science and Engineering
关键词
双相钢
深度学习
本构模型
duplex steel
deep learning
constitutive model
作者简介
第一作者:李智琪,男,硕士。E-mail:chen_990128@163.com;通信作者:沈涛,男,博士,讲师。E-mail:shentao@nbu.edu.cn。
