The Future of Refractory Material Selection—A Data-Driven Approach

Refractory materials are a crucial part of every aggregate used to produce many materials where temperatures exceed 1,200 °C, such as copper, nickel, steel, cement, lime, and glass. As most refractory materials are manufactured from various raw materials via different process routes, there is a huge variety available. Selecting the most suitable material for each application and furnace area is challenging and requires (1) a deep understanding of the dominant wear mechanisms—highly affected by local conditions (e.g. slag chemistry)—and (2) profound knowledge about refractory materials, including the relevant properties necessary to achieve the expected performance under process conditions. The large number of refractory products available, and the increased generation of public and proprietary know-how on refractory products and their properties (e.g. literature or internal R&D reports) turns the material selection process a task that requires a transformation from an expert-centred towards a data-centered approach. The article proposes an approach to transform the material selection process to make better informed decisions by combining existing databases with machine learning models and knowledge graphs. In addition to addressing the expected challenges, the article outlines the proposed architecture of a self-learning, interactive refractory recommendation system.