Abstract
A wide variety of engineering design tasks can be formulated as constrained optimization problems where the shape and topology of the domain are optimized to reduce costs while satisfying certain constraints. Recent works have demonstrated the feasibility of the Boundary Element Method (BEM) as a tool for topological-shape optimization. However, high computational costs of the iterative process remain a significant drawback. This note suggests ways to address these limitations by supplementing complex variables BEM with quadtree-based sampling and incremental matrix updates, validated through numerical examples.
Introduction
In engineering, topology optimization seeks the most efficient material distribution for a structure under specific loads—for example, designing a bracket that uses minimal metal while maintaining structural integrity. Traditionally, the Finite Element Method (FEM) dominates this field. However, FEM requires a volumetric mesh, which often necessitates expensive re-meshing as the shape evolves.
The Boundary Element Method (BEM) offers an attractive alternative by only discretizing the surfaces (boundaries), simplifying geometry management. Historically, BEM's dense matrices led to high computational costs ($O(n^3)$). This research presents strategies to achieve $O(n^2)$ performance, making BEM a viable, mesh-flexible competitor for structural optimization.
Methodology
The proposed approach utilizes the Complex Variables Boundary Element Method (CVBEM) enhanced by three key techniques:
1. Quadtree-Based Sampling
The optimization process relies on the Topological Derivative (TD), which quantifies the sensitivity of total energy to the introduction of a small hole. Calculating TD across the entire domain is computationally intensive. By employing a quadtree algorithm, the authors concentrate high-resolution sampling near boundaries where changes are most likely to occur, significantly reducing the number of evaluation points.
2. Iterative Inverse Matrix Updates
Topology optimization is an incremental process. Instead of re-solving the entire system after minor shape changes, the authors apply blockwise matrix inversion formulas (such as the Sherman-Morrison-Woodbury formula). This allows for updating the previous solution by accounting only for added or removed elements, drastically cutting linear algebra overhead.
3. Hard-Kill Algorithm
A simple "hard-kill" strategy is used to remove material where the TD falls below a specific threshold. This process continues until the target volume fraction is achieved.
Experimental Validation
The method was tested on a standard 2D benchmark: a square domain fixed on one side with a point load applied to a corner. The goal is to minimize strain energy for a given volume of material.
Key Results
- Sampling Efficiency: Quadtree sampling reduced required points from 1600 to 676, cutting evaluation time from 11.3s to 4.8s.
- Computational Speedup: Blockwise matrix updates reduced linear algebra time from 41s to 15.5s.
- Convergence: The system reached an optimal, mesh-independent configuration in just 29 iterations, matching results from more expensive FEM studies.
Conclusion and Future Work
This work demonstrates that BEM can be a high-performance tool for topology optimization. By achieving $O(n^2)$ scaling, it becomes competitive with FEM while avoiding volumetric meshing issues and "checkerboard" instabilities. Future improvements could include integrating the Fast Multipole Method (FMM) to reach $O(n \log n)$ complexity and extending the framework to 3D problems where geometric complexity is significantly higher.
Summary updated