Automated construction of knowledge graphs for accelerated design and understanding of ultra-high-performance concrete

Knowledge graphs have enabled incorporation of concrete domain knowledge into machine learning-accelerated material design via representing knowledge using digital, operable graphs, toward a knowledge-guided data-driven paradigm with essential interpretability, transparency, and trustworthiness. However, manual construction of knowledge graphs is time-consuming and labor-intensive. This paper presents a framework for the automated construction of material knowledge graphs that leverage the one-shot capability of large language models to address this challenge. The approach reduces the time required to build a comprehensive knowledge graph for ultra-high-performance concrete from 52 h to 1 h. The applications are demonstrated through three use cases, and its generalizability is validated via an application to the accelerated design of ultra-high-performance geopolymer. The framework reveals the complex physicochemical pathways between material components and performance properties, accelerating both the speed of material design and the depth of understanding in the process, pushing the boundaries of knowledge-guided data-driven design.