Computational Estimation of the Composition of Fat/Oil Mixtures Containing Interesterifications from Gas and Liquid Chromatography Data Martin H. van Vlieta, Geert M.P. van Kempenb,*, Marcel J.T. Reindersa, and Dick de Riddera, aInformation and Communication Theory Group, Faculty
...
Computational Estimation of the Composition of Fat/Oil Mixtures Containing Interesterifications from Gas and Liquid Chromatography Data Martin H. van Vlieta, Geert M.P. van Kempenb,*, Marcel J.T. Reindersa, and Dick de Riddera, aInformation and Communication Theory Group, Faculty of Electrical Engineering, Mathematics and Computer Science, Delft University of Technology, 2628 CD Delft, The Netherlands, and bOil-based Product Technology, Unilever R&D, Vlaardingen, 3130 AC Vlaardingen, The Netherlands.
A mathematical framework is introduced that relates analytical data to the composition of fat and oil mixtures. Within this framework, the noise characteristics of four common analytical techniques [FAME, FAME2-pos, CN (carbon number), and AgLC] were investigated and modeled by both additive and multiplicative noise terms. The fat blend recognition (FBR) performance was investigated under these two types of noise, both qualitatively and quantitatively. Furthermore, an extension is proposed that makes it possible to detect interesterifications of unknown mixtures, which was impossible before. The proposed procedure is divided into a qualitative estimation stage, which is focused on identifying the raw materials (RM), followed by a quantitative estimation stage, which is focused on quantifying the levels of the RM identified. We compared two qualitative strategies and four quantitative methods for their ability to correctly estimate simulated mixtures under the noise characteristics determined. The comparison of methods was extended to actual mixtures, revealing promising results. Our analysis presents multiple directions for further adulteration and FBR studies.
Paper no. J11060 in JAOCS 82, 707¿716 (October 2005).@en