Integrating Geometric Data into Topology Optimization via Neural Style Transfer.

Area of Science:

• Engineering
• Computer Science
• Materials Science

Background:

• Traditional topology optimization focuses solely on structural performance.
• Geometric and aesthetic considerations are increasingly important in design.
• Integrating visual appeal with mechanical function presents a significant challenge.

Purpose of the Study:

• To develop a novel topology optimization method incorporating neural style transfer.
• To enable simultaneous optimization of structural performance and geometric similarity.
• To provide designers with a tool for balancing mechanical efficiency and aesthetic criteria.

Main Methods:

• Utilizing convolutional layers of pre-trained neural networks to extract design features.
• Implementing a weighted objective function to balance structural performance and neural style transfer.
• Applying the method to various loading conditions and reference designs through case studies.

Main Results:

• Optimized designs achieved structural performance within 10% of baseline without geometric constraints.
• Incorporated features from reference designs, such as member size and meshed elements.
• Demonstrated superior performance compared to optimizers lacking geometric similarity constraints.

Conclusions:

• The proposed method effectively integrates structural performance and geometric similarity in topology optimization.
• Neural style transfer offers a viable approach for incorporating aesthetic considerations into engineering design.
• This technique empowers designers to find optimal trade-offs between form and function.