Quantification of butanol and ethanol in aqueous phases by reflectometric interference spectroscopy - different approaches to multivariate calibration

This paper presents several methods for analysis of data from reflectometri c interference spectroscopic measurements (RUS) of water samples. The set-u p consists of three sensors with different polymer layers. Mixtures of buta nol and ethanol in water were measured from 0 to 12,000 ppm each. The data space was characterized by principal component analysis (PCA). Calibration and prediction were achieved by multivariate methods, e.g. multiple linear regression (MLR), partial least squares (PLS) with additional predictors, a nd quadratic partial least squares (Q-PLS), and by use of artificial neural networks. Artificial neural networks gave the best results of all the cali bration methods used. Calibration and prediction of the concentration of th e two analytes by artificial neural nets were robust and the set-up could b e reduced to only two sensors without deterioration of the prediction.