Influence of Temperature on the Predictive Ability of near Infrared Spectroscopy Models

Abstract

Temperature changes alter the position and intensity of near infrared (NIR) spectral absorption bands and thus affect the predictive ability of the associated calibration models. Achieving accurate control of this variable in industrial processes is difficult and variations can have a strong impact on their analytical monitoring. In this work, the effect of temperature changes over the range 25–90°C on the predictions for the ingredients of the esterification reaction between acetic acid and butanol was examined. Spectra for mixtures of the different reactants and products were used to construct calibration models by partial least-squares (PLS) regression and stepwise principal component regression (stepwise PCR). The models were constructed from the temperature ranges, wavelengths, numbers of factors and spectral treatments leading to the highest predictive ability. Based on the results, the variable temperature can also be modelled and the predictive ability of calibration models improved by including partially or completely the effect of temperatures.