Existing numerical methods for complex composites, such as multiscale simulation and neural network algorithms, face significant limitations. Multiscale techniques are often prohibitively expensive for large models, w...Existing numerical methods for complex composites, such as multiscale simulation and neural network algorithms, face significant limitations. Multiscale techniques are often prohibitively expensive for large models, while neural networks struggle to represent underlying microscopic material properties. To overcome these challenges, a meso-micro scale numerical method using a virtual node approach is developed in this study. A Wbraid/Al/Epoxy functional structural material is fabricated, and a representative periodic unit cell is identified based on its architecture. The complex structure is then discretized into nodes, and mechanical interactions are governed by pre-defined computation rules. This virtual node method is systematically compared against both multiscale simulation and a neural network algorithm, with validation provided through mechanical experiments. The results demonstrate that the nodal operation strategy significantly reduces computational resource requirements. By quantifying microscopic bonding with coefficients, explicit interface treatment is avoided, granting the method strong adaptability to lattice materials. The method can simulate extremely complex structures using parameters from simple tests and is suited for large systems. Compared to three-point bending experiments, errors for multiscale, virtual node, and neural network methods were 12.4%, 6.9%, and 34.5%, respectively. Under dynamic compression, the errors were 2.7%, 9.3%, and 15.43%. The virtual node method demonstrated superior accuracy under static conditions, enabling efficient prediction and auxiliary development of complex structural materials.展开更多
文摘Existing numerical methods for complex composites, such as multiscale simulation and neural network algorithms, face significant limitations. Multiscale techniques are often prohibitively expensive for large models, while neural networks struggle to represent underlying microscopic material properties. To overcome these challenges, a meso-micro scale numerical method using a virtual node approach is developed in this study. A Wbraid/Al/Epoxy functional structural material is fabricated, and a representative periodic unit cell is identified based on its architecture. The complex structure is then discretized into nodes, and mechanical interactions are governed by pre-defined computation rules. This virtual node method is systematically compared against both multiscale simulation and a neural network algorithm, with validation provided through mechanical experiments. The results demonstrate that the nodal operation strategy significantly reduces computational resource requirements. By quantifying microscopic bonding with coefficients, explicit interface treatment is avoided, granting the method strong adaptability to lattice materials. The method can simulate extremely complex structures using parameters from simple tests and is suited for large systems. Compared to three-point bending experiments, errors for multiscale, virtual node, and neural network methods were 12.4%, 6.9%, and 34.5%, respectively. Under dynamic compression, the errors were 2.7%, 9.3%, and 15.43%. The virtual node method demonstrated superior accuracy under static conditions, enabling efficient prediction and auxiliary development of complex structural materials.
