A multiscale topology optimization design framework with data driven surrogate model

Conventional topology optimization methods face intrinsic limitations in designing functionally graded hierarchical structures, primarily due to scale-decoupling assumptions and computational intractability. This research introduces a transformative offline-online multiscale framework that synergistically integrates parameterized lattice architectures with concurrent optimization. During offline preprocessing, Moving Least Squares surrogate models establish real-time mappings between lattice parameters and equivalent properties through multiscale finite element analysis, circumventing scale separation. The online phase pioneers a unified formulation where macroscopic topology variables and microscopic lattice parameters are co-optimized within a Discrete Material Optimization scheme, enabling simultaneous control over spatial configuration distribution and anisotropic property gradation. Validated across geometrically complex benchmarks, the framework demonstrates superior mechanical rationality through load-path-aligned configuration invariance and adaptive density modulation. By bridging high-dimensional design freedom with computational tractability while ensuring manufacturing-ready solutions, this work establishes a new paradigm for performance-driven metamaterials in aerospace and biomedical applications.