Application of chemometric tools for coal classification and multivariate calibration by transmission and drift...

The aim of this paper focuses on the determination of nine coal properties related to combustion power plants (moisture (%), ash (%), volatile matter (%), fixed carbon (%), heating value (kcal kg(-1)), carbon (%), hydrogen (%), nitrogen (%) and sulphur (%)) by mid-infrared spectroscopy. For that, a wide and diverse coal sample set has been clustered into new homogeneous coal subgroups by the use of hierarchical clustering analysis. This process was performed including property values and spectral data (scores of principal component analysis, PCA) as independent variables. Once the clusters were defined, the corresponding property calibration models were performed by partial least squares regression. Several mathematical pre-treatments were applied to the original spectral data in order to cope with some non-linearities. The accuracy and precision levels for each property were studied. The results revealed that coal properties related to organic components presented relative error values around 2% for some clusters, comparable to those provided by commercial online analysers. Finally, the discrimination level between those groups of samples was evaluated by linear discriminant analysis (LDA). The sensitivity of the system was studied accomplishing percentages close to 100% when the samples were classified attending only to their mid-infrared spectra.