The Reincarnate project is happy to announce the publication of the conference paper Leveraging Large Language Models for Automated Knowledge Graphs Generation in Non-Destructive Testing by Ghezal Ahmad Jan, Z., Valdestilhas, A., Moreno Torres, B., & Kruschwitz, S. (2024, September 24).

Presented at SeMatS 2024: The 1st International Workshop on Semantic Materials Science, co-located with the 20th International Conference on Semantic Systems (SEMANTiCS) in Amsterdam, from September 17-19, the paper introduces an innovative method for utilising large language models (LLMs) to automate the generation of Knowledge Graphs (KGs) from scientific articles in the field of Non-Destructive Testing (NDT) – a crucial method used to evaluate the integrity of materials like concrete, wood, steel, and bricks without causing damage.

The research outlines how material-specific agents, equipped with carefully curated glossaries, extract relevant information on material degradation, physical changes, and applicable NDT methods. The extracted data is then organised into a Neo4j graph database, forming a comprehensive KG that visually maps out the relationships between different materials, their deterioration mechanisms, and corresponding NDT techniques. This allows for the automatic discovery of how various materials degrade and which NDT methods are best suited to detect these changes.

The study showcases the effectiveness of this approach in capturing the complex interactions between materials and NDT techniques. For instance, the KG can show how ultrasonic testing or infrared thermography is applied to detect cracks or other physical changes in building materials under specific conditions. The automated nature of the process simplifies knowledge discovery, providing a scalable framework that can be extended beyond NDT to other scientific fields.

Using LLMs to generate KGs in the NDT domain significantly improves the organisation and accessibility of complex information. The generated KGs offer a valuable tool for researchers and engineers, facilitating more informed decisions and enabling further innovation in material testing and scientific research.

Reincarnate partners involved: Bundesanstalt für Materialforschung und -prüfung and Technical University of Berlin.

Read the paper: https://zenodo.org/records/13834165