Reduced-order concurrent multiscale modelling of composite structures

Concurrent multiscale simulation of the inelastic response of composite structures remains a challenging task due to the substantial computational cost. This complexity stems from the large number of degrees of freedom (DOFs) introduced by the fine mesh discretizations required for both the macroscopic structure and the underlying microstructural models. To address this challenge, we propose a finite-volume-driven self-consistent clustering analysis (FVSCA) framework for reduced-order, three-dimensional concurrent multiscale modeling of composites. In the offline stage, the repeating unit cell (RUC) is discretized into finely resolved subvolumes, and the elastic strain concentration matrices of each subvolume are computed using the finite-volume direct averaging micromechanics (FVDAM) method. These matrices are then used to identify material clusters via the k-means clustering, resulting in a reduced RUC representation. In the online stage, the self-consistent clustering approach is applied to evaluate the inelastic homogenized response of the reduced RUC and the local response of each material cluster. The FVSCA micromechanics model is implemented in ABAQUS as a meta user material subroutine (Meta-UMAT), providing an efficient multiscale bridge between the structural response and local constitutive behavior. Validation against conventional numerical simulations at both material-point and structural scales demonstrates the high accuracy of the proposed method. Moreover, the FE×FVSCA approach achieves at least a 13-fold reduction in computational time and decreases memory consumption by >83 times compared with FE×FVDAM. Comparison with experimental results for a graphite/aluminum composite further verifies the predictive capability of the proposed method.