Multivariate Data Analysis and Data Fusion Techniques for Modeling Spectroscopic Data During CO2 Capture with Amines
Not peer reviewed
MetadataShow full item record
This master thesis is a collaboration between CO2 Technology Center Mongstad and the University of Bergen. The goal of the project was to use multivariate data analysis methods to generate predictive models for the total alkalinity (TOT_ALK), total inorganic carbon (TIC) and density of anime solvent samples. This was to be done using three different instrument types, namely ATR-FTIR, NIR and Raman spectrometers. An individual predictive model was to be created for each instrument type, and then data fusion techniques were to be employed in order to create fused models which combine traits from all of the different instruments. The goal of this was to determine which individual instrument provided measurement data which was most useful in creating predictive models, as well as to determine if the use of multiple instruments in tandem is a worthwhile endeavor. In order to accomplish this, a partial least squares regression (PLS-R) algorithm was programmed in MATLAB, as well as multiple subroutines for data preprocessing and analysis, including Monte-Carlo cross-validation (MCCV), variable selection, outlier detection, extended multiplicative signal correction (EMSC) and Savitzky-Golay filtering. The models that were generated indicate that the use of Raman spectroscopy is not advised for the stated intent, and that a combination of ATR-FTIR and NIR instruments provide the best and most consistent overall results.
PublisherThe University of Bergen
Copyright the Author. All rights reserved