Data-driven multifidelity multiscale topology optimization via phasor-based evolutionary de-homogenization

Multiscale topology optimization is crucial for designing porous infill structures with high stiffness-to-weight ratios and excellent energy absorption. Although gradient-based methods provide a rigorous framework, they are computationally expensive and struggle to capture cross-scale sensitivities in nonlinear settings. Moreover, the resulting hierarchical geometries are often overly complex and lack macroscopically meaningful features. To overcome these issues, we propose an evolutionary de-homogenization framework that couples MultiFidelity Topology Design with a phasor-based de-homogenization technique. The framework translates low-dimensional geometric descriptors into explicit, CAD-compatible high-resolution structures through a hybrid evolutionary algorithm integrating NSGA-II selection, PCA-VAE-enabled latent-space crossover, and a novel image-deformation-based mutation operator. This gradient-free approach supports multi-objective refinement through non-differentiable geometry processing and body-fitted high-fidelity analysis. The numerical studies show design trade-offs for stiffness, stress, and buckling objectives; the intended scope is design-set exploration rather than universal computational superiority over full-scale gradient-based optimization.