A Computationally Efficient Technique for Dynamic Sensitivity Filtering in Topology Optimization

Abstract

In this paper, we present a novel approach: a dynamic sensitivity filtering technique designed for density-based topology optimization. Sensitivity filtering methods have found extensive utility in mitigating numerical instabilities. Nevertheless, conventional sensitivity filtering can introduce numerous gray elements along the topology's boundaries, thereby impeding practical manufacturing within real-world engineering applications. To address this challenge, we propose a dynamic sensitivity filter that adjusts the sensitivity at each iterative step based on the optimization outcomes of the preceding iteration. Through diverse test examples, our method demonstrates the capacity to effectively solve the numerical instability problem while concurrently achieving a nearly pure black and white design, characterized by significantly reduced computational expense and improved structural stiffness.