The mapping of variables between old and new elements is a crucial aspect of large deformation finite element simulation. However, traditional mapping methods suffer from limited applicability and complex operations. In this study, a novel Gray Wolf Optimization-Machine Learning (GWO-ML) based variable mapping method for large deformation finite element analysis is proposed. This method achieves accurate mapping results without requiring explicit determination of element-to-element positional relationships. Furthermore, it effectively integrates a wider range of integration point information during the mapping process and maintains high prediction accuracy near boundaries with sparse mesh data. The remeshing and interpolation technique by small strain is employed as an illustrative example, wherein a set of automated operations is proposed to control the commercial software ABAQUS through Python scripting for conducting small strain analysis and GWO-ML mapping process. For machine learning models, Multilayer Perceptrons, Random Forest, Extreme Learning Machine, and Support Vector Regression, are compared, where the RF model has excellent computational accuracy and efficiency in terms of performance. The application of this method is investigated in scenarios with various foundation types, motion modes, loading modes, and multi-physics coupling mappings. The results confirm that the method exhibits high accuracy and robustness in simulation processes.
Read full abstract