Abstract
The paper presents a new method for shape and topology optimization based on an efficient and scalable boundary integral formulation for elasticity. To optimize topology, our approach uses iterative extraction of isosurfaces of a topological derivative. The numerical solution of the elasticity boundary value problem at every iteration is performed with the boundary element formulation and the kernel-independent fast multipole method. Providing excellent single node performance, scalable parallelization and the best available asymptotic complexity, our method is among the fastest optimization tools available today. The performance of our approach is studied on few illustrative examples, including the optimization of engineered constructions for the minimum compliance and the optimization of the microstructure of a metamaterial for the desired macroscopic tensor of elasticity.
Introduction
Topology optimization—the process of finding the most efficient material distribution within a domain to support specific loads—is a cornerstone of modern aerospace, automotive, and structural engineering. Traditionally, this is solved using the Finite Element Method (FEM). However, FEM faces significant challenges when scaling to massive 3D problems, including high computational costs due to volume discretization and numerical artifacts like "checkerboard" patterns that require artificial filtering.
The BEM-KIFMM Framework
This paper introduces a high-performance approach that shifts the computational focus from the entire volume to just the surface by combining the Boundary Element Method (BEM) with the Kernel-Independent Fast Multipole Method (KIFMM).
By using a direct boundary integral equation for elasticity, the dimensionality of the problem is effectively reduced. To solve the resulting dense BEM systems, the authors employ KIFMM, which uses an octree structure to approximate far-field interactions. This reduces the computational complexity from $O(N^2)$ to $O(N)$, making large-scale 3D optimization feasible.
Benchmarks and Performance
The method was rigorously tested using several engineering benchmarks, including a unit cube under tension/compression to verify single-node performance and parallel scalability.
A standout feature of this framework is its exceptional scalability. The implementation shows excellent "strong scaling," meaning that as more CPU cores are added, the problem solves significantly faster. In tests, the system supported up to 128 cores efficiently.
Key Results
- Speed: A single FMM pass achieved a performance of 30,000 Degrees of Freedom (DOF) per second on a single core.
- Scale: Models with millions of DOFs were successfully optimized; for instance, a $200^3$ voxel problem was solved in just two minutes on 128 cores.
- Accuracy: In metamaterial benchmarks, the resulting structures reached a bulk modulus within 95% of the Hashin-Shtrickmann upper bound, the theoretical physical limit.
- Numerical Stability: Unlike FEM, this BEM approach inherently avoids checkerboard patterns without needing explicit regularization filters.
Insights and Future Directions
The dimensionality advantage of BEM is naturally suited for shape optimization where the boundary defines the object. Furthermore, the "kernel-independent" nature of the FMM means this framework can be easily extended beyond elasticity to problems involving fluid flow (Stokes) or heat transfer.
While the current "hard-kill" greedy algorithm is highly effective, it may occasionally get stuck in local optima. Future work will focus on improving solver convergence for complex boundary conditions and implementing advanced surface smoothing techniques, such as Marching Cubes, to further refine the quality of the final optimized shapes.
Summary updated