AI-Guided Auto-Discovery of Low-Carbon Cost-Effective Ultra-High Performance Concrete: Data Synthesis, Semi-supervised Learning, and Many-Objective Optimization

Abstract

This study presents an AI-guided approach to automatically discover low-carbon cost-effective ultra-high performance concrete (UHPC). The presented approach automates data augmentation, machine learning model generation, and mixture selection by integrating advanced techniques of generative modeling, automated machine learning, and many-objective optimization. New data are synthesized by generative modeling and semi-supervised learning to enlarge datasets for training machine learning models that are automatically generated to predict the compressive strength, flexural strength, mini-slump spread, and porosity of UHPC. The proposed approach was used to explore new UHPC mixtures in two design scenarios with different objectives. The first scenario maximizes the compressive and flexural strengths and minimizes porosity while retaining self-consolidation. The second scenario minimizes the life-cycle carbon footprint, embodied energy, and material cost, besides the objectives of the first scenario. The life-cycle carbon footprint, embodied energy, and material cost of the UHPC in the second scenario are respectively reduced by 73%, 71%, and 80%, compared with the UHPC in the first scenario. This research advances the capability of developing cementitious composites using AI-guided approaches.