NEAR-INFRARED SPECTROSCOPY FOR REFUSE DERIVED FUEL : Classification of waste material components using hyperspectral imaging and feasibility study of inorganic chlorine content quantification

Abstract: This degree project focused on examining new possible application of near-infrared (NIR) spectroscopy for quantitative and qualitative characterization of refuse derived fuel (RDF). Particularly, two possible applications were examined as part of the project. Firstly, use of NIR hyperspectral imaging for classification of common materials present in RDF. The classification was studied on artificial mixtures of materials commonly present in municipal solid waste and RDF. Data from hyperspectral camera was used as an input for machine learning models to train them, validate them, and test them. Three classification machine learning models were used in the project; partial least-square discriminant analysis (PLS-DA), support vector machine (SVM), and radial basis neural network (RBNN). Best results for classifying the materials into 11 distinct classes were reached for SVM (accuracy 94%), even though its high computational cost makes it not very suitable for real-time deployment. Second best result was reached for RBNN (91%) and the lowest accuracy was recorded for PLS-DA model (88%). On the other hand, the PLS-DA model was the fastest, being 10 times faster than the RBNN and 100 times faster than the SVM. NIR spectroscopy was concluded as a suitable method for identification of most common materials in RDF mix, except for incombustible materials like glass, metals, or ceramics. The second part of the project uncovered a potential in using NIR spectroscopy for identification of inorganic chlorine content in RDF. Experiments were performed on samples of textile impregnated with a water solution of kitchen salt representing NaCl as inorganic chlorine source. Results showed that contents of 0.2-1 wt.% of salt can be identified in absorbance spectra of the samples. Limitation appeared to be water content of the examined samples, as with too large amount of water in the sample, the influence of salt on NIR absorbance spectrum of water was too small to be recognized.