AI-based models for intelligent PP compound development
The “KIOptiPack – Holistic AI-based optimisation of plastic packaging with recyclate content” project aims to streamline the formulation development process and fundamentally accelerate it through data-driven methods. At the core of this approach lies the targeted control of the mechanical and rheological properties of compounds, even with fluctuating raw material qualities.
Within a subproject on compounding at IKV, investigations examined how the complex interactions between polymers, fillers, and additives can be modelled to precisely predict relevant material properties. Various polypropylene homopolymers and copolymers, chalk, peroxide masterbatches, and impact modifiers were compounded at laboratory scale and subsequently characterised rheologically and mechanically. Target quantities included shear viscosity and MFR, as well as E-modulus and Charpy impact strength as mechanical properties. On this data basis, analytical models and AI-based methods – including artificial neural networks, symbolic regression, and genetic algorithms – were developed and compared. The combination of analytical prediction models with genetic algorithms proved particularly promising for identifying new formulations, enabling targeted generation of recipes that optimally fulfil specified target values (Fig. 1).
© IKVFor practical validation of this methodology, an industrially established PP compound was ultimately successfully replicated with respect to its mechanical and rheological properties using alternative formulation components. Thus, KIOptiPack demonstrated that data-driven methods enable substantial acceleration of formulation development while yielding reproducible and transparently traceable results.
Project data and funding
We would like to thank the BMFTR for funding the project (funding code 033KI101) and the project partners for their cooperation.
Project duration: 01.08.2022 – 31.12.2025
