Physics-Guided Generative Inverse Design of Low-Carbon Concrete via Multi-Objective Latent Space Optimization

Area of Science:

• Materials Science
• Civil Engineering
• Computational Science

Background:

• Designing concrete with reduced carbon footprint, high strength, and acceptable cost presents a significant multiobjective optimization challenge.
• Existing methods often struggle to balance competing objectives like environmental impact and mechanical performance.

Purpose of the Study:

• To develop an integrated generative inverse-design framework for low-carbon concrete.
• To optimize concrete mixtures for compressive strength, global warming potential (GWP), and cost simultaneously.

Main Methods:

• Utilized a Variational Autoencoder (VAE) for latent-space representation of concrete mixtures.
• Employed a physics-guided strength surrogate model incorporating regularization terms.
• Applied Non-dominated Sorting Genetic Algorithm II (NSGA-II) for Pareto-optimal solution generation in the latent space.

Main Results:

• Generated Pareto-optimal concrete mixtures balancing strength, GWP, and cost.
• Experimental validation showed good agreement between predicted and measured 28-day compressive strengths (relative errors < 5%).
• The framework successfully identified ecological optimum, cost optimum, and balanced solutions.

Conclusions:

• The proposed generative inverse-design framework can effectively generate feasible low-carbon concrete mixtures.
• Preliminary results support the framework's ability to meet specified strength, GWP, cost, and constraint targets.
• Further validation with broader performance indicators is needed for full practical applicability.