Real-time generative design of diverse, "truly" optimized structures with controllable structural complexities

TL;DR

The paper tackles the challenge of real-time, diverse generation of truly optimized structures by coupling a modified Moving Morphable Components (MMC) topology optimization method with a conditional Wasserstein GAN (WGAN). It introduces genus-based complexity labeling and three strategies to curate high-quality datasets with crisp boundaries and clear load paths, then trains an improved WGAN with gradient penalty and loading-position conditioning to map random noise and labels to optimized designs in real time. Key contributions include a large, labeled dataset (≈19,015 cantilever-like designs and 8,672 L-shaped designs), a topology-informed complexity measure, and a WGAN framework that generates novel, label-consistent designs that generalize beyond the training set. The work advances real-time generative design for engineering applications, enabling rapid exploration of diverse, near-optimal structures and supporting CAD-like workflows with potential 3D extensions.

Abstract

Compared with traditional design methods, generative design significantly attracts engineers in various disciplines. In thiswork, howto achieve the real-time generative design of optimized structures with various diversities and controllable structural complexities is investigated. To this end, a modified Moving Morphable Component (MMC) method together with novel strategies are adopted to generate high-quality dataset. The complexity level of optimized structures is categorized by the topological invariant. By improving the cost function, the WGAN is trained to produce optimized designs with the input of loading position and complexity level in real time. It is found that, diverse designs with a clear load transmission path and crisp boundary, even not requiring further optimization and different from any reference in the dataset, can be generated by the proposed model. This method holds great potential for future applications of machine learning enhanced intelligent design.

Figures (15)

• Figure 1: (a) Classic design process of equipment with the help of structural topology optimization; (b) a game-changing design process of equipment with the help of generative design.
• Figure 2: (a) Initial design of the node-driven based MMC method (red points - driven nodes); (b) the corresponding optimized design (the components with 0 thickness is eliminated); (c) a component is totally determined by five geometry parameters $\left( x_a^i, y_a^i, x_b^i, y_b^i, t^i \right)^\top$. The dimensionless size of design domain is $2 \times 1$, the amplitude of the external load is $1$, and the elstic constants of the base material is $E=1, \nu=0.3$, respectively.
• Figure 3: The cases account for $|{\bm A}_1|$, $|{\bm A}_2|$ and $|{\bm A}_3|$ (number of elements with nodal density $1/4$, $2/4$ and $3/4$, respectively).
• Figure 4: Some typical optimized designs classified by the complexity levels.
• Figure 5: Distribution of the objective function value, complexity level and genus of samples corresponding to loading position 0.