Accelerating multi-objective optimization of concrete thin shell structures using graph-constrained GANs and NSGA-II.

Area of Science:

• Structural Engineering
• Computational Mechanics
• Artificial Intelligence in Design

Background:

• Optimizing concrete thin shell structures involves complex, non-linear multi-objective challenges.
• Traditional optimization methods struggle with design space exploration and computational demands.
• Existing techniques often yield suboptimal solutions for structural design.

Purpose of the Study:

• To present a novel hybrid approach for topology and thickness optimization of concrete thin shell structures.
• To combine deep learning generative capabilities with evolutionary algorithm refinement.
• To enhance the efficiency and effectiveness of structural optimization processes.

Main Methods:

• Utilized Graph-constrained conditional Generative Adversarial Networks (GANs) for design generation.
• Employed the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) for evolutionary refinement and constraint adherence.
• Integrated deep learning with evolutionary algorithms for a hybrid optimization strategy.

Main Results:

• Achieved a 33.3% reduction in total weight, 50% decrease in maximum deflection, and 20% reduction in strain energy.
• Demonstrated a 50% enhancement in convergence speed compared to traditional NSGA-II.
• Confirmed structural integrity via finite element analysis and prototyping with <3.5% performance variation.

Conclusions:

• The hybrid GANs and NSGA-II approach offers significant improvements in structural optimization.
• This methodology provides a new pathway for innovative, high-performance architectural solutions.
• Sophisticated machine learning and evolutionary algorithms hold great potential for advanced structural design.