The potential of near-infrared diffuse reflectance spectrometry for quality control analyses in the textile industry was explored with a view to the quantification of finishing oils in acrylic fibres by partial least-squares regression, using a rotary cuvette system for recording spectra. Calibration was performed with a set of samples that encompassed every source of variability (linear density of the fibres, colour, concentration of the finishing oil), and the wavelength region where absorption was mostly due to the oil was used to construct several models from which that leading to the minimum relative standard error for a sample test set was selected. The results provided by various mathematical treatments [second- derivative, standard normal variate (SNV)] used to minimize scattering resulting from the differential linear density of the samples revealed no significant differences between prediction errors (only in the number of partial least-squares components). The model was used to quantify levels of finishing oil in routinely manufactured samples for a period of 6 months, during which time two batches showed poor predictions due to a new component appearing in the product. Modification of the calibration model to account for this component substantially increased robustness and allowed the accurate quantification of all batches manufactured after the model has been developed.
A fast and accurate procedure has been developed for the simultaneous determination of resorcinol, m-aminophenol, o-cresol, phenol and m-cresol in waters. The method is based on the reaction between the aforementioned phenolic compounds, at concentration levels of ppm, and 300 ppm of p-aminophenol in 0.05 mol l–1 NaOH and in the presence of 0.004 mol l–1 KIO4. The treatment of absorbance data, between 400 and 700 nm, obtained at interval times of 20 s between 40 and 120 s and 40 s between 120 and 600 s, by means of the use of Partial Least Squares 2 (PLS-2), using the UNSCRAMBLER program, suggested that the use of 7 factors can explain more than 99.5% of the residual variance. The method has been applied to the analysis of synthetic samples, not employed for calibration, and average errors lower than 4% were found.