MULTIVARIATE CALIBRATIONS IN FOURIER-TRANSFORM INFRARED SPECTROMETRY FOR PREDICTION OF KEROSENE PROPERTIES

Seven aircraft fuel quality properties: density, freezing point, flash point, aromatic content, initial boiling point, final boiling point a nd viscosity, have been predicted from the Fourier transform infrared (FT-IR) spectra in the range of 4000 to 600 cm-1, using three multivar iate techniques. Multiple linear regression (using the all-variables a nd stepwise methods), principal components regression (using the all-v ariables and stepwise methods) and partial least squares (PLS) models, have been employed and their predictive capabilities evaluated. Altho ugh the standard error of prediction (SEP) has been the main parameter considered to select the ''best model'', repeatability and reproducib ility have been also considered. FT-IR-PLS repeatability and reproduci bility values fall well within the ASTM ranges and SEP values are real ly good. Sample manipulation was improved by using a stopped-now syste m.