Near infrared (NIR) spectroscopy is a fast and reliable technique for assessing properties of amino resins. One important property that defines the cost and performance of these resins is the solids content (SC). This work studied the prediction of SC of amino resins by combining NIR spectroscopy with partial least squares (PLS) regression. A total of 990 industrial NIR spectra of amino resins were obtained and split randomly by a ratio of 2/3 for calibration and 1/3 for validation. The best model achieved a root mean-square error of prediction (RMSEP) of 0.32% (m/m) and a coefficient of determination of prediction (Rp2) of 81%. standard normal variate (SNV) was found to be the NIR pre-processing that provided the best results for model construction. Addition of water to two amino resins showed that the NIR model does not respond to the water addition, despite water making great contribution to the SC value. An inference that can be obtained from this is that the NIR model of amino resins uses NIR properties of amino resins that relate to the SC and from there predict the most probable SC, instead of looking at all the components that affect the SC of amino resins.
© 2020 The Author(s)PDF Article