| dc.description.abstract | Many measures have been taken to combat climate change. Amine-based carbon capture technology plays an essential role in the reduction of CO2 emissions. However, one issue of this technology is solvent degradation. There are three main kinds of degradation: oxidative degradation, thermal degradation, and CO2 induced degradation. Among them, around 90% of the degradation is due to oxidative degradation. This study focuses on gaining a better understanding of oxidative degradation.
A significant amount of research efforts have been dedicated to studying the principles of oxidative degradation, and some kinetic models have been proposed. However, the mechanism of oxidative degradation is not fully understood. In this study, a machine learning technique, the eXtreme Gradient Boosting (XGBoost), has been investigated as an alternative approach to model the degradation rate to understand oxidative degradation better.
The data-driven model is built based on data from laboratory experiments. The assessment of the data-driven model is completed by R-squared(R2). R2 is a statistical measure representing how much independent variables can explain the variance for a dependent variable in a regression model. The R2 of the data-driven model is 77.2% for predicting test data set where the conditions are 40°C, CO2 loading 0.272 and 0.42.
This data-driven model is further integrated into a process model to explore oxidative degradation in different parts of the carbon capture process. The carbon capture process model is divided into five main parts: water wash, absorber, absorber sump and piping, heat exchanger, and regenerator. Due to degassing of oxygen before the regenerator, it is assumed that the degradation in the regenerator is negligible. Based on the integration of the data-driven model in the process model, it can be estimated that oxidative degradation mainly happens in absorber sump and piping, and accounts for 74% of the total degradation. Subsequently, the proportions for absorber, heat exchanger, and water wash are 17%, 9%, and 0.25%, respectively. It can also be estimated that when the capture plant starts with virgin MEA, the MEA loss is 11.13g when capturing 1 ton CO2.
This data-driven model has also been used to extrapolate in other cases. For example, the case of pilot plant RWE Niederaussem is mimicked. The MEA loss when capturing 1 ton CO2 in absorber and absorber sump in this study is 10.14g, less than 320g listed in public data for RWE. When the data-driven model is integrated into the carbon capture process model, the result is different from the experimental result. Several possible reasons for this are discussed in this thesis.
Finally, possible countermeasures for oxidative degradation are shortly discussed. Based on the data-driven model's prediction, removing oxygen before flue gas enters the absorber is recommended. ``Bleed and feed" is a possibility but not practical. The use of plastic plants could also be a choice but is still under investigation. | |