Interface-filtering structural optimization (Q7020263)

This article presents an Interface-Filtering Structural Optimization (IFSO) method for shape and topology optimization that combines the sharp-interface characteristics of the level-set method with the simplicity of density-based optimization. The key idea is to evolve the designs only through subtle movements of the interface, rather than through widespread material redistribution. Throughout this process, the sharpness of the density field stays close to binary.\N\NThe interface movements are achieved via a combination of a nonuniform-radius Helmholtz-type PDE filter and the Heaviside projection. In contrast to conventional density-based methods, not the density itself is optimized but the filter radius. The authors derive a linear relationship between the maximal filter radius and the maximum interface movement, resulting in numerical stability. One key advantage of the IFSO method is the ability to start from a custom, sharp initial design that already has sharp interfaces. This enables shape refinement and also removes dozens of iterations to obtain a clear design from a gray field.\N\NThe proposed algorithm consists of an inner loop where the filter radius for an optimal interface movement is determined and an outer loop where the density field is updated via this filter radius to the new design. Furthermore, the authors introduce a radius re-initialization, which resets the filter radius to a uniform large value after a fixed number of iterations. This prevents the algorithm from being stuck in local minima caused by excessive sharpness.\N\NThe IFSO method is demonstrated on structural and thermal compliance and negative Poisson ratio problems. The results show that the algorithm achieves performance comparable to classical topology optimization while also maintaining high sharpness and providing control over the shape evolution.