Evolutionary de-homogenization using a generative model for optimizing solid-porous infill structures considering the stress concentration issue

TL;DR

This paper introduces an evolutionary de-homogenization framework that combines data-driven generative models with multi-fidelity optimization to design porous infill structures with reduced stress concentration and improved geometric accuracy.

Contribution

It presents a novel hybrid design method that bridges low- and high-fidelity models using a generative approach for optimized porous structures considering stress issues.

Findings

Effective reduction of stress concentration in porous structures.

Enhanced geometric accuracy in the final design.

Demonstrated superiority over traditional topology optimization methods.

Abstract

The design of porous infill structures presents significant challenges due to their complex geometric configurations, such as the accurate representation of geometric boundaries and the control of localized maximum stress. In current mainstream design methods, such as topology optimization, the analysis is often performed using pixel or voxel-based element approximations. These approximations, constrained by the optimization framework, result in substantial geometric discrepancies between the analysis model and the final physical model. Such discrepancies can severely impact structural performance, particularly for localized properties like stress response, where accurate geometry is critical to mitigating stress concentration. To address these challenges, we propose evolutionary de-homogenization, which is a design framework based on the integration of de-homogenization and data-driven multifidelity optimization. This framework facilitates the hybrid solid-porous infill design by bridging the gap between low-fidelity analysis and high-fidelity physical realizations, ensuring both geometric accuracy and enhanced structural performance. The low-fidelity level utilizes commonly used density control variables, while the high-fidelity level involves stress analysis based on structures with precise geometric representations. By employing a de-homogenization-based mapping method, a side-by-side correspondence between low-fidelity and high-fidelity results is established. The low-fidelity control variables are iteratively adjusted to optimize the high-fidelity results by integrating deep generative model with multi-objective evolutionary algorithm. Finally, numerical experiments demonstrate the effectiveness of the proposed method.