Neural Network-Based Constitutive Model for Solid Materials
Neural Network-Based Constitutive Model for Solid Materials
复制标题
DOI：
10.23967/wccm-apcom.2022.079
复制
发表时间：
2022
期刊：
影响因子：
--
通讯作者：
R. Alhayki;E. Muttio-Zavala;W. Dettmer;D. Peric
中科院分区：
文献类型：
--
作者：
R. Alhayki;E. Muttio-Zavala;W. Dettmer;D. Peric
提出了一种新的基于神经网络的方法来再现固体材料的复杂非线性本构关系，包括弹性行为，塑性变形和损伤机制。提出了一种基于历史的策略，使用阿尔蒂神经网络来训练路径依赖的非弹性行为。网络的发展是基于一个一般的内部形式主义。结果表明，本文方法能准确地描述单轴应力状态下的von Mises弹塑性材料模型。该策略被施加到序列的训练和验证数据，这些数据是数值生成的弹塑性硬化和不硬化，以及弹塑性损伤。结果已与建立的数学模型进行了比较，并显示出在一维精确描述复杂的非线性固体材料行为的潜力
This proposes novel neural network-based approaches to reproduce the complex nonlin-ear constitutive relations of solid materials including elastic behavior, plastic deformation and damage mechanism. A history-based strategy has been suggested using an artificial neural network for training path-dependent inelastic behavior. The network development is based on a general internal formalism. of selected It is shown that the proposed methodology can represent exactly the von Mises elastoplastic material model in uni-axial stress state. The strategy was applied to sequences of training and validation data which were generated numerically for elastoplasticity with and without hardening as well as for elastoplastic damage. The results have been compared against established mathematical models and shown a potential of describing complex non-linear solid material behavior accurately in one-dimensional