| dc.description.abstract | Furfural (FUR) and hydroxymethylfurfural (HMF) are furanic derivatives from dehydration of carbohydrates found in many types of biomasses. These molecules are often referred to as platform chemicals because of their wide application area as petroleum-based resource substitution. The carbohydrate composition of biomass is a complex system to analyze, due to the innate molecular similarities causing analysis result separation and interpretation to be challenging. In this thesis two predictive models for carbohydrate composition in a mixture were developed. This was done by the application of the experimental design, mixture design and multivariate analysis principal competent analysis (PCA) and partial-least square (PLS) regression analysis. Pure standards of ten sugar molecules (arabinose (Ara), fructose (Fru), galactose (Gal), glucose (Glu), lactose (Lac), maltose (Mal), mannose (Man), ribose (Rib), sucrose (Suc) and xylose (Xyl)) and mixture design samples, consisting of six components (Ara, Fru, Glu, Mal, Man and Xyl) at three levels, were analyzed using proton nuclear magnetic resonance (1H-NMR) spectroscopy and the results were analyzed with PCA and two different approaches of PLS regression analysis, univariable (PLS1) and multivariable (PLS2) calibration. Both models were validated using a training/test split and exhibited promisingly high statistical values. Each dependent variable (DV) for the PLS1 predictive model had a R2 value >0.982, RMSEC < 0.001 M and RMSEP < 0.001 M. Each DV for the PLS2 predictive model had a R2 value > 0.998, RMSEC < 0.001 M and RMSEP < 0.004. Both prediction models were applied to the aqueous phase of thermochemically converted waste biomass (plum rejects). This was performed with the use of a biphasic system developed according to the principles of green chemistry. Most of the predictions were as expected compared to literature, however both models need further development for this purpose. | |
