Development of a digital twin of the catalyst extrusion process from combining physics-based and data-driven models

Erik Tomas Holmen Olofsson

Research output: Book/ReportPh.D. thesis

24 Downloads (Pure)


This Ph.D. dissertation mainly focuses on optimizing starvefed singlescrew extruders using a combination of mechanistic and datadriven models, explicitly targeting industrial systems designed to produce hydroprocessing treatment catalysts. The importance of this research is highlighted by the industrial potential to significantly reduce both downtime and the generation of substantial waste while manufacturing catalyst products.

This dissertation is centered around three leading publications, each exploring different aspects of, and further developing on, the concept of digital twins for a starvefed singlescrew
extrusion system. Initial studies successfully extracted process data from laboratoryscale equipment, facilitating metrics for aligning mechanistic models with physical results. The research mainly aims to establish methods for realtime extraction of process signatures relating to respective extruder degree of fill and material residence time. The degree of fill in an extruder indicates how much of the screw length within the extruder barrel is fully occupied by the material, and the residence time is the duration that the material spends inside the extruder from the feed to the discharge point.

The research conducted during this PhD included the development of multiple numerical models to promote the process signatures mentioned above and enhance the understanding of the process itself. The initial model focused on the laboratoryscale extruder, establishing the process of extracting process signatures and assessing results concerning measurement data and different analytical expressions. The second numerical model scaled the numerical modeling to an industrial size system, incorporating a porous representation of the extruder die plate that could produce the corresponding permeability. A vital aspect of this work is the transition from highfidelity numerical simulations to simpler surrogates that approximate the behavior of the more complex and computationally expensive models. The last numerical model corresponded to a reduced dimensional representation of the second model, thereby generating residence time distribution results much faster. With the model referring to a velocity field of higher order, necessary corrections were established to transfer the velocity to the model of lower order.

Lastly, statistical response surface methodology and Machine Larning (ML) methods were evaluated to study the relative influence on predicting product quality and performance compared to a simpler baseline model currently in operation. Evaluation of the response surface model indicated that a slightly larger reference data set is needed to interpolate results with sufficient accuracy. Results from applying ML methods indicated that Random Forest is the most suitable model. Incorporating process signature data next to a reference data set, did not result in a significant enhancement of the machine learning model’s predictive capabilities, thus preventing the demonstration of improved correlation through the inclusion of this information. However, the established framework enables realtime corrections to the extruder process parameters based on promoting optimum extrudate porosity or a specific fill level.

This thesis focused mainly on a specific material and a single extrusion system, yet the developed numerical models, methodologies, and insights have potential broader relevance for analyzing different materials and extrusion methods. This extensive applicability is especially pertinent for the company Topsoe, considering its diverse collection of mixing and extrusion systems and the various manufactured catalyst products.
Original languageEnglish
Place of PublicationKgs. Lyngby
PublisherTechnical University of Denmark
Number of pages119
ISBN (Print)978-87-7475-784-9
Publication statusPublished - 2023


Dive into the research topics of 'Development of a digital twin of the catalyst extrusion process from combining physics-based and data-driven models'. Together they form a unique fingerprint.

Cite this