摘要:
为准确表征金属材料在高应变速率下应力应变本构关系,提出了基于图神经网络(GNN)和KAN的本构关系的高精度预测模型。为解决传统Johnson-Cook(JC)模型不考虑温度、应变速率与应变之间的耦合效应问题,在GNN模型中构建图结构数据描述多维参数的非线性关联,KAN模型中基于Kolmogorov-Arnold定理实现高维输入空间的非线性映射。基于ODS铜合金的高应变率压缩实验数据,评估了GNN、KAN和JC的本构关系描述和预测精度。结果表明:GNN与KAN模型在测试集中的平均相对误差分别为9.2%与9.1%,决定系数(R²)均高于0.95,显著优于JC模型(MRE=0.38,R²=0.75);将上述本构关系模型应用在有限元仿真中,GNN和KAN模型预测的等效塑性应变与应力分布更符合理论特征,而JC模型对材料软化阶段描述不足且仿真结果偏差较大。提出的模型能有效捕捉高应变速率下材料的多场耦合特性,为极端载荷条件下的应力应变本构关系预测提供了新方法。
Abstract:
To accurately characterize the stress-strain constitutive relationship of metal materials under high strain rates, a high-precision constitutive relationship prediction model based on Graph Neural Networks (GNN) and Kolmogorov-Arnold Networks (KAN) was developed. Traditional Johnson-Cook (JC) models often fail to account for the coupling effects among temperature, strain rate, and strain, which are crucial for describing the dynamic behavior of materials under extreme conditions. This limitation was addressed by constructing graph-structured data in the GNN model to capture the nonlinear correlations of multidimensional parameters and leveraging the Kolmogorov-Arnold theorem in the KAN model to achieve precise mapping of high-dimensional input spaces. The research methodology involved several key steps. Experimental data from ODS copper from ODS copper under high strain rate compression were collected using a Split Hopkinson Pressure Bar (SHPB) system and subsequently preprocessed. The dataset included temperature, strain rate, strain, and stress. In the GNN model, when temperature and strain rate were constant, nodes were connected in sequence based on strain values to form edges. When temperature was constant, a reasonable threshold was set between nodes with adjacent strain rates, and nodes within this threshold were connected to form edges. The GNN employed a Message Passing Neural Network (MPNN) architecture to learn and predict material properties. Model parameters were optimized using the Adam optimizer, with the Root Mean Squared Error (RMSE) serving as the loss function. The KAN model was constructed based on the Kolmogorov-Arnold representation theorem and consisted of multiple KAN-Linear layers. Each KAN-Linear unit included base weights and spline weights. Base weights handled linear relationships through traditional linear transformations, while spline weights managed nonlinear mappings via B-spline interpolation. Both models were trained on the preprocessed dataset, and their performance was evaluated using metrics such as the Mean Relative Error (MRE), Root Mean Squared Error (RMSE), and the coefficient of determination (R²). The GNN model achieved an average MRE of 9.2% with an R² value exceeding 0.95, while the KAN model recorded an MRE of 9.1% with a similar R² value. Both models significantly outperformed the JC model, which had an MRE of 0.38 and an R² value of 0.75. Furthermore, the predictive capabilities of the GNN and KAN models were validated through finite element simulations. The simulation results demonstrated that the stress-strain distributions predicted by the GNN and KAN models were more consistent with theoretical expectations compared to those predicted by the JC model, particularly in capturing the material's softening phase. The findings highlight the potential of integrating advanced machine learning techniques, such as GNN and KAN, into the field of materials science to enhance the accuracy and efficiency of constitutive modeling. The models offer a promising alternative to traditional empirical models and hold significant implications for engineering applications in aerospace, automotive, and other industries where materials are subjected to high strain rates.