Turning AI based rubber compound predictions into practical formulations
Artificial intelligence (AI) is increasingly used in rubber compounding to accelerate formulation development and predict performance. While AI systems can efficiently analyze large historical databases, they often generate compound recipes that are mathematically correct, but impractical for production.
One of the main reasons lies in database structure. Historic compound databases typically contain a wide variety of polymers: multiple natural rubbers, numerous styrene butadiene rubber (SBR) grades, butadiene rubber (BR) types, recycled rubbers and different oils. In real manufacturing, however, compounds usually contain one polymer, occasionally two, and only rarely, three. Without practical constraints, AI algorithms may combine several similar polymers to achieve property targets.
This article describes a four-step workflow that must be followed by the user when working with the GrafCompounder software to transform AI generated predictions into manufacturable rubber compounds, while maintaining target performance.