Accelerating Materials Discovery in Tire Manufacturing: A Knowledge Graph-Driven Optimization Framework

[Context] The automotive industry consumes over 75% of global rubber, with tire manufacturing demanding high-quality engineering and precision. Continental AG has accumulated extensive tire compound recipe knowledge since 1904, yet accessing this historical data remains challenging. Traditional recipe development relies on material scientists’ experience and iterative laboratory testing that can take months per formulation. With about 60 raw materials per tire, exploring ingredient combinations and predicting their interactions creates long feedback loops that slow innovation. A central problem is that accessing recipe knowledge requires technical database expertise, creating barriers for material scientists who could benefit from exploring historical formulations but lack query language proficiency. [Objective] To support material scientists, a knowledge graph-driven framework was developed specifically tailored to their needs. The framework enables exploration of tire compound formulations through natural language queries without requiring database expertise. Three key objectives guide this work: establish a comprehensive knowledge representation capturing recipe compositions and ingredient-property relationships; develop intelligent natural language interfaces for accessible data exploration; and build a practical framework integrating into Continental AG’s industrial workflows. [Method] The approach is based on close collaboration with Continental AG, following Gorschek and Wohlin’s technology transfer model, allowing material scientists to continuously influence the design. Five development phases were implemented, in which the solution was iteratively refined according to user requirements. These phases included knowledge graph construction with a flexible multi-hop schema, text-to-Cypher translation using few-shot learning, natural language response generation with limitation awareness, and correlation analysis for quantitative predictions. Evaluation was conducted through demonstration sessions involving seven material scientists from Continental AG’s tire compound development team. Both objective and subjective metrics were collected to assess the framework’s accuracy, usability, workflow integration, and adoption potential. [Results] The results indicate that the framework provides conditional value for material scientists. Time savings emerged as the strongest finding, with tasks requiring 15 minutes to 4 hours completed in seconds to minutes. Response time and query quality received positive ratings. However, ease of use showed variation, as some participants struggled with natural language interpretation. Adoption intent reflected mixed findings. Material scientists identified clear requirements before committing to adoption, including transparency about system reasoning and integration into existing workflows. [Conclusion] This resulted in a domain-specific framework that demonstrates how natural language interfaces can make technical knowledge graphs accessible in industrial contexts. The solution represents a trade-off between eliminating technical barriers and introducing natural language interpretation challenges. However, a critical limitation was revealed regarding trust and transparency. In safety-critical domains like tire compound development, material scientists need to understand how the system interprets queries and arrives at results before trusting it for important decisions, making transparency mechanisms essential for future deployment.