A genetic evolved machine learning approach for 3D DEM modelling of anisotropic materials with full consideration of particulate interactions
A genetic evolved machine learning approach for 3D DEM modelling of anisotropic materials with full consideration of particulate interactions
复制标题
DOI：
10.1016/j.compositesb.2022.110432
复制
发表时间：
2023-02
期刊：
影响因子：
--
通讯作者：
Zewen Gu;Xiaoxuan Ding;X. Hou;Jianqiao Ye
中科院分区：
文献类型：
--
作者：
Zewen Gu;Xiaoxuan Ding;X. Hou;Jianqiao Ye
多尺度建模技术的快速发展使人们对材料破坏的理解有了很大的提高。然而，对一般各向异性材料的精确模拟仍然是一个巨大的挑战。这是由于不同比例尺的模型所需的材料参数数量不平衡，从已知的宏观属性中提取微观材料属性是困难的，有时甚至是不可能的。针对一般各向异性复合材料，提出了一种新的三维离散单元模型(DEM)，以充分考虑材料颗粒之间的相互作用。将机器学习(ML)技术与遗传算法(GA)相结合，解决了三维DEM模型微观键性质确定的难题。通过将DEM预测的宏观材料性质与实验结果进行比较，验证了所学到的键性质。在此基础上，进一步利用微观粘结性能对螺栓连接复合材料搭接接头的强度进行了预测和裂纹形态模拟。ML模型的预测结果与节点试验结果吻合较好。
Rapid development of multiscale modelling techniques has enabled significant improvement in understanding material failure. However, accurate simulation of general anisotropic materials still remains a great challenge. This is due to the unbalanced number of material parameters required by models of different scales, and it is difficult and sometime impossible to extract micro material properties from known macro properties. This paper proposes a new 3D discrete element model (DEM) to take full interactions between material particles for general anisotropic composite materials. The challenging issue in determining the micro bond properties of the 3 DEM model is resolved by coupling machine learning (ML) technique with the genetic algorithm (GA). The learned bond properties are validated by comparing DEM predicted macro material properties with experimental results. The micro bond properties are further used to predict strength and simulate crack patterns of bolted composite lap joints. The predictions of the ML model agree well with experimental results of the joints.