Geert Postma

Francisco Souza, Tim Offermans, Ruud Barendse et al.

This work proposes a new data-driven model devised to integrate process knowledge into its structure to increase the human-machine synergy in the process industry. The proposed Contextual Mixture of Experts (cMoE) explicitly uses process knowledge along the model learning stage to mold the historical data to represent operators' context related to the process through possibility distributions. This model was evaluated in two real case studies for quality prediction, including a sulfur recovery unit and a polymerization process. The contextual mixture of experts was employed to represent different contexts in both experiments. The results indicate that integrating process knowledge has increased predictive performance while improving interpretability by providing insights into the variables affecting the process's different regimes.

Mahdiyeh Ghaffari, Gerjen H. Tinnevelt, Marcel C. P. van Eijk et al.

Plastic sorting is a very essential step in waste management, especially due to the presence of multilayer plastics. These monomaterial and multimaterial plastics are widely employed to enhance the functional properties of packaging, combining beneficial properties in thickness, mechanical strength, and heat tolerance. However, materials containing multiple polymer species need to be pretreated before they can be recycled as monomaterials and therefore should not end up in monomaterial streams. Industry 4.0 has significantly improved materials sorting of plastic packaging in speed and accuracy compared to manual sorting, specifically through Near Infrared Hyperspectral Imaging (NIRHSI) that provides an automated, fast, and accurate material characterization, without sample preparation. Identification of multimaterials with HSI however requires novel dedicated approaches for chemical pattern recognition. Non negative Matrix Factorization, NMF, is widely used for the chemical resolution of hyperspectral images. Chemically relevant model constraints may make it specifically valuable to identify multilayer plastics through HSI. Specifically, Multi Block Non Negative Matrix Factorization (MBNMF) with correspondence among different chemical species constraint may be used to evaluate the presence or absence of particular polymer species. To translate the MBNMF model into an evidence based sorting decision, we extended the model with an F test to distinguish between monomaterial and multimaterial objects. The benefits of our new approach, MBNMF, were illustrated by the identification of several plastic waste objects.