Source-linked AI summary
Topology optimization based on moving deformable components: A new computational framework
Xu Guo, Weisheng Zhang, Wenliang Zhong
TL;DR
Topology optimization faces limitations in geometry representation, feature-size control, and computational scalability. The paper proposes moving deformable components to incorporate geometry and mechanical information directly, with the potential to make optimization more flexible and substantially reduce computational burden. The introduction also notes that the framework remains subject to further study of efficiency, initial-design dependency, robustness, and convergence rate.
Problem
Pixel- and node-based topology optimization face geometry-representation and dimensionality limitations, including difficulty connecting directly with CAD systems and escaping the curse of dimensionality.
Method
The paper proposes a structural topology-optimization framework based on moving deformable components and explicit component parameters.
Results
The proposed paradigm directly incorporates more geometry and mechanical information into topology optimization, rendering the solution process more flexible and potentially reducing computational burden substantially.
Takeaways & Limitations
Moving deformable components provide a geometry-explicit solution framework intended to link structural topology optimization with CAD modeling systems.
Takeaways & Limitations
The framework is still at a primary-development stage, with efficiency, initial-design dependency, robustness, and convergence rate identified for further exploration.
Abstract
from arXiv · showhide
In the present work, a new computational framework for structural topology optimization based on the concept of moving deformable components is proposed. Compared with the traditional pixel or node point-based solution framework, the proposed solution paradigm can incorporate more geometry and mechanical information into topology optimization directly and therefore render the solution process more flexible. It also has the great potential to reduce the computational burden associated with topology optimization substantially. Some representative examples are presented to illustrate the effectiveness of the proposed approach.
1. Introduction
The introduction identifies geometry representation, feature-size control, and computational cost as challenges in pixel- and node-based topology optimization, then proposes moving deformable components as a more geometry-explicit framework.
- Pixel-based methods discretize the design domain into finite elements and optimize element-wise 0-or-1 material densities to represent structural topology.
- Pixel-based representations differ from CAD descriptions based on geometric primitives and Boolean operations, preventing direct topology optimization on CAD platforms.
- Explicit geometry information is absent from pixel-based approaches, making precise control of minimum or maximum length scales and curvature difficult despite manufacturing importance.
- Three-dimensional pixel-based optimization can require very large computational effort; a 100-by-100-by-100 discretization produces one million design variables.
- Node point-based level-set methods retain implicit geometry representations, cannot directly connect with CAD modeling systems, and remain subject to the curse of dimensionality.
- The proposed moving-deformable-component framework directly incorporates geometry and mechanical information, making topology optimization more flexible and potentially reducing computational burden substantially.
2. Moving deformable components-based topology optimization framework
The framework uses moving deformable structural components as topology-optimization building blocks, varying their geometry and layout to represent structural topologies. Components may overlap, enabling topology changes while incorporating explicit geometric information.
- Deformable components serve as primary building blocks in a topology-optimization framework distinct from traditional approaches.
- The framework represents topology through explicit component parameters and can incorporate more geometry information into the problem formulation.
- Structural topology is controlled by component shape, length, thickness, orientation, and layout or connectivity.
- Overlapping components changes structural layout and can make redundant components effectively disappear from structural responses.
- The initial formulation uses rectangular components in two dimensions, while extensions to curved geometries and three-dimensional cases are left for future work.
- A finite number of components is treated as practically useful because manufacturability and robustness favor structures with finite components.
3. Numerical solution aspects
The numerical framework uses a fixed background mesh with implicit level-set component boundaries and XFEM-based analysis. Overlap is handled directly, with localized remeshing used to improve boundary-near displacement and stress computations.
- A fixed background finite-element mesh establishes interactions among components before the optimized structure is obtained.
- Explicit level-set functions describe component boundaries and flexibly support overlapping components, the key mechanism for topology changes.
- XFEM analysis is adopted for level-set-described structural geometries.
- Remeshing is needed only near structural boundaries to improve displacement and stress computation accuracy.
- Weak material is introduced to mimic voids in the design domain.
- Sensitivity analysis uses the primary and adjoint displacement fields for variations of individual component descriptions.
4. Merits of the proposed topology optimization framework
The proposed framework emphasizes explicit, geometry-based component design, model independence, and black-and-white structural layouts. It is presented as flexible for integrating multiple optimization modes while potentially reducing design-space size and computational effort.
- Explicit component geometry supports seamless CAD integration and local control of structural features.
- The optimization model is independent of the analysis model, helping address numerical inconsistencies and supporting non-FEM methods for multiphysics applications.
- The framework can integrate shape, size, topology, and structural-type optimization through component geometry, connectivity, and element choices.
- Crisp boundaries and overlapping components combine boundary clarity with flexibility for topology changes and complex boundary conditions.
- Pure black-and-white layouts eliminate gray elements and the need for rational interpolation schemes associated with variable-density approaches.
- The reduced design space is independent of finite-element resolution and may improve efficiency, enable global optimization or surrogate models, and expose component-level parallelism.
5. Numerical examples
Numerical examples evaluate the moving deformable components framework on several dimensionless 2D plane-stress problems. The method achieves solutions comparable to classical approaches while using far fewer design variables and supporting substantial topology changes.
- Numerical setup: The examples use dimensionless material, load, and geometry data for 2D plane-stress problems with unit thickness and uniform bilinear square finite elements.MMA is used to solve the optimization problems numerically.
- Numerical setup: The initial designs contain movable components that can translate, rotate, dilate, and shrink while minimizing mean compliance under a solid-material constraint.One initial design contains 16 components.
- Results: The proposed method obtains results almost the same as SIMP and level set methods using only 80 design variables, versus more than 3000 for the same finite-element mesh.The reported reduction is associated with a substantial reduction in computational cost.
- Results: An optimal solution is achieved within 100 iterations even with a relatively small step length, demonstrating flexibility in handling drastic topological changes.Small steps are adopted because slight component changes can produce large topology changes.
- Interpretation of designs: A coarse-mesh contour extraction can make solution boundaries appear zigzag, whereas plotting the result in a CAD system reveals smooth boundaries.The CAD plot includes components narrower than one mesh width.
- Results: The method also produces an optimal design close to classical solutions for a different load location and shows components shrinking in low-strain-energy regions and merging centrally.The examples report comparable topologies for multiple loading cases.
- Results: For another benchmark, the result is very similar to traditional methods with only 120 design variables, compared with 4800 for the same finite-element mesh.Components can approach, overlap, and merge during optimization, providing a mechanism for topology change.
6. Concluding remarks
The paper proposes a topology-optimization framework that obtains structural topology by optimizing deformable component layouts rather than removing material or evolving boundaries. The examples indicate potential for flexible, CAD-integrated optimization, but broader efficiency and robustness remain to be studied.
- Contribution: The framework obtains optimal structural topology by optimizing the layout of deformable structural components.This differs from traditional approaches based on eliminating unnecessary material or evolving structural boundaries.
- Contribution: The authors identify the approach as a novel idea that had not been explored in the literature to their knowledge.The claim is explicitly qualified by the authors’ knowledge.
- Potential: The approach may integrate size, shape, and topology optimization seamlessly in CAD modeling systems.The paper presents this as having potential in engineering applications.
- Limitations and future work: The framework remains at an early stage, requiring further study of efficiency, initial-design dependency, robustness, and convergence rate.The authors specifically identify non-self-adjoint, large-scale, and multiphysics topology-optimization problems as areas for further investigation.