A Diffusion Model Based Iterative Convolution Thresholding Method for Structural Topological Optimization

Abstract

In this study, we introduce the Diffusion Model with Iterative Convolution Thresholding Method (DICTM), a novel hybrid approach designed to address the minimum compliance problem in topology optimization. DICTM synergistically combines the robustness of diffusion models with the precision of threshold dynamics to tackle the complexities inherent in linear elasticity problems, while substantially enhancing computational efficiency. Our approach facilitates the generation of initial configurations via the diffusion model, which dramatically improves the efficiency of the subsequent threshold dynamics process, reducing the iteration count to about one-tenth of that required by traditional methods. This significant reduction in computational effort also enables more effective hyperparameter tuning without added cost. The integration of deep-generative models with a rigorous threshold dynamics framework positions DICTM as a powerful tool in topology optimization, producing designs not only with low compliance, but also in a computationally efficient way.