Aspectos e aplicações da análise de fatores paralelos: clean-up virtual, reconstrução de sinal e colinearidade unimodal

Abstract

Parallel factor analysis (PARAFAC) is a trilinear algorithm for higher-order data analysis that relies on the second-order advantage. In this dissertation, four di erent studies were made to explore di erent aspects of PARAFAC application. The rst evaluates the use of sinchronic uorescence matrices converted into excitation-emission matrices, which are trilinear. The proposed method was applied to a mixture of HPAs and PARAFAC was able to retrieve the pro les of pure components, and due to the trilinearity of the system, rebuild the signal in the regions containing empty spaces. The second study explores the second-order advantage by performing a virtual clean-up of a complex matrix by using a tensor containing one sample at a time and a calibration set consisting of analyte standards. This technique was successfully applied for the determination of melamine in adulterated milk in a method consistent with green chemistry precepts. The third study evaluated the use of reaction kinetics alongside UV-Vis absorption spectroscopy as source of second-order data. The degradation of food dyes by fenton reaction and dopamine oxidation in the presence of levodopa as an interferent were the model systems studied. Simulated systems were also evaluated. It was veri ed that kinetic processes that result in correlated components can hinder deconvolution by PARAFAC, generating problems of convergence and loss of the uniqueness of the solution. The fourth study used simulations to explore cases in which one of the modes is colinear for two factors. The simulations showed that the colinear mode of the a ected components can be recovered consistently, but the other may present considerable distortions. Some strategies were developed to break the colinearity in the concentration mode, which common in spectrokinetic matrices, and one of them, the use of additional matrices displaced in time, was tested with oxidation of the dopamine and showed good recovery of the trilinear consistency of the model, which signals the correction of distortions in spectral and kinetic modes. These applications demonstrate that a deeper understanding of the PARAFAC modeling process can be used to develop innovative analytical strategies.