2023

Bayesian statistical techniques can be leveraged to combine previous knowledge (previous data, values from literature, modeller’s experience) with new data to estimate model parameters. This report shows Bayesian objective functions to allow incorporation of prior knowledge, demonstrated on a steady-state gas-phase n-hexane hydroisomerization case study. Knowledge about an old catalyst obtained from a large dataset is transferred in different ways to estimate kinetic parameters for a new, but similar catalyst. Three scenarios using small datasets for the new catalyst are considered. The results show that intelligent priors from reliable knowledge is useful if the dataset for the new catalyst is limited, but not with larger datasets and particularly not with priors based on mistaken beliefs.

2023

Model-based design of experiments (DoE) aims at identifying the most informative experimental points in order to estimate model parameters, based on certain model assumptions. However, the parameters entering these model assumptions are generally known only within a certain accuracy. This leads to the question in how far this affects the optimal experimental designs. This report shows how the effect of uncertain model parameters can be quantified by introducing the concept of design clusters. It turns out that the sharp experimental points are smeared out to such clusters, which size depends on the uncertainty level and the design quantity. We also suggest a modification of parallel coordinate plots to visualize the clusters.

2023

Plant data quality can be improved by using tools such as data reconciliation and optimal experimental design. This report provides an overview of the main features of these methodologies and their application to industrial case studies. Following some information on the project context (Section 1), this report is divided into four main sections: (Section 2) principles and available tools for data reconciliation, (Section 3) practical applications, challenges and perspectives, (Section 4) optimal experimental design and sensor placement and (Section 5) application to specific case studies. General conclusions and take-home messages are provided in Section 6. More information and links to the webpages on the software packages and of the review studies can be found in the Excel file available as well.

2023

Overview of results inquiry among Eurokin members on the future aspects of currently used software packages for kinetic parameter estimation from experimental data. Software packages included are: Aspen Custom Modeler, Athena Visual Studio, gPROMS, KASTER + microKinetic Engine, Matlab, Octave, Presto Kinetics, and Python. The results were collected in a table to give an easy overview. The inquiry showed that all packages included are being further improved in the course of time, on functionality as well as in digital trends, although not at similar speeds.

2023

Results of a new inventory and assessment of software packages for chemical reactor simulation, kinetic modelling and parameter estimation focusing on the available functionalities, options, and user-friendliness options. The inventory yielded the following commercial packages: Aspen PLUS, Aspen Custom Modeler, Athena Visual Studio, gPROMS, Matlab, and Presto Kinetics. Three non-commercial packages were included: Copasi, Octave, and microKinetic Engine. The information obtained from the software distributers and the users of the non-commercial packages were collected and compared, resulting in some conclusions. It was not attempted to rank the packages since the purpose of the packages varies too widely. In 2023 the packages KASTER (the follow-up of microKinetic Engine) and Python were added. Additionally, the outlook was updated with an explanation why a more detailed assessment was not carried out. Addition of an appendix containing an overview of the reports and codes on the implementation of the four Eurokin parameter estimation problems from 2000 in the various packages and other evaluations of software packages.

2023

Model-based design of experiments (DoE) aims at identifying the most informative experimental points in order to estimate model parameters, based on certain model assumptions. However, the parameters entering these model assumptions are generally known only within a certain accuracy. This leads to the question in how far this affects the optimal experimental designs. This report shows how the effect of uncertain model parameters can be quantified by introducing the concept of design clusters. It turns out that the sharp experimental points are smeared out to such clusters, which size depends on the uncertainty level and the design quantity. We also suggest a modification of parallel coordinate plots to visualize the clusters. The corresponding Python codes are available as well.

2021

Python codes and instructions and explanation belonging to the report on model-based optimal Design of Experiments, illustrated with a case study on cinnamaldehyde hydrogenation. This short manual provides a setup instruction for a virtual Anaconda environment, such that the provided data creation and design of experiments code can be run in the Spyder Editor or the Anaconda Prompt. For this setup we require a python version >= 3.6 and < 3.9. If a higher/smaller version is used, we cannot guarantee that all dependencies are set correctly.

2022

In this work a theoretical and experimental study was carried out to observe the instability in a reactor. The first instability concerns the multiplication of stationary states. This instability was observed by simulation and experiment. The second instability concerning the observation of a sustained periodic instability called limit cycle, which appears as a periodical variation maintained by the state of the system. This second instability was observed by simulation, but could not be obtained in the experiments (exothermic reaction between Na2S2O3 and H2O2), also not in a new reactor with improved cooling system. This can be explained by the complex conditions to obtain the limit cycle as well as the sensitivity of this zone to the operating conditions and probably due to the high inertia of the coil.

2022

This review focuses on kinetic models of depolymerization, which is defined here as thermal, catalytic or solvolytic degradation of polymers to form monomers, oligomers, or otherwise “small” molecules to be used as chemical feedstock or reactant. We split the kinetic models into 4 general categories: machine learning models, continuum models which do not track polymer detail, models which explicitly track the polymer chain length distribution (CLD) or its moments, and kinetic Monte Carlo models. Indicative statistics of the model types and analytical techniques employed in literature are reported. Experimental methods to gather data for building kinetic models are reviewed, including analytical techniques, reactor design considerations, and methods for identifying contaminants. Finally, some challenges facing the field are discussed.

2022

An overview is given of specific challenges to track macromolecules as opposed to small molecules for polymerization processes. It is stressed that the macromolecular variety complicates the kinetic study and thus much more kinetic variables exist than in small molecule systems. The 1st part (Polymer classification and characterization) addresses in detail the macromolecular and macroscopic properties and various reaction mechanisms. The 2nd part (polymerization kinetic models: principles) addresses the related kinetic approaches, ending with a summary table with advantages and disadvantages. The 3rd part focuses on polymerization kinetic models and kinetic parameters.