| Name: | Description: | Size: | Format: | |
|---|---|---|---|---|
| 8.18 MB | Adobe PDF |
Authors
Advisor(s)
Abstract(s)
O objetivo deste trabalho foi aplicar uma língua eletrónica potenciométrica na
análise de águas minerais com e sem sabor adquiridas em várias superfícies comerciais,
para averiguar se a informação obtida permitia fazer análise qualitativa e quantitativa.
As águas minerais naturais e de nascente estão disponíveis comercialmente com uma
grande variedade de paladares, associados à diversidade da composição físico-‐química.
As águas minerais de sabor são conhecidas como refrigerantes por terem adição de
ingredientes autorizados (ex., corantes, conservantes, aromatizantes, edulcorantes,
acidulantes, antioxidantes, etc).
A língua eletrónica potenciométrica utilizada neste trabalho incluía um elétrodo
de referência Ag/AgCl de dupla junção e dois sistemas de multi-‐sensores, conectados a
um datalogger para aquisição dos sinais. Cada sistema de multi-‐sensores era constituído
por 20 membranas lipídicas diferentes, de sensibilidade cruzada, preparadas com vários
compostos lipídicos (3,0%), plastificantes (65,0%) e com o polímero PVC (32,0%). O
segundo sistema era uma réplica do primeiro sistema de multi-‐sensores.
A caracterização físico-‐química das 34 amostras minerais recolhidas envolveu
análises de pH e condutividade em todas as amostras e a análise por cromatografia
líquida nas amostras de sabor (16 amostras), para determinar conteúdos em ácidos
orgânicos (ácidos cítrico, málico e ascórbico) e açúcares (glucose, frutose e sacarose).
Num primeiro estudo sobre os dados analíticos obtidos verificou-‐se que as amostras
poderiam ser agrupadas em 4 grupos de acordo com a informação dos rótulos: águas
minerais com e sem gás e águas de sabor com e sem gás. A análise de variâncias
mostrou que havia diferenças significativas entre os quatros grupos considerando as
variáveis da caracterização físico-‐química das amostras. A análise de componentes
principais usando os parâmetros químicos obtidos por HPLC mostrou que os dados
analíticos separavam naturalmente as amostras de águas de sabor com e sem gás.
Numa segunda fase, usaram-‐se os perfis de sinais obtidos com a língua
eletrónica e, no âmbito da análise qualitativa, aplicou-‐se a análise dos componentes
principais para verificar a distribuição espacial das amostras e a interdependência com
os parâmetros físico-‐químicos determinados; posteriormente, aplicou-‐se análise
discriminante linear acoplada ao algoritmo de arrefecimento simulado para seleção de variáveis e estabeleceu-‐se o melhor modelo discriminante para os quatros grupos
definidos. O desempenho do modelo obtido ao nível da precisão foi avaliado usando a
validação cruzada “leave-‐one-‐out”. Verificou-‐se que com uma só função discriminante
(explica 99,94% da variância total dos dados), contendo 24 sensores (12 sensores do
primeiro sistema e 12 do segundo sistema), permitia obter 100% de classificações
corretas, por validação cruzada. Neste estudo obteve-‐se também informação de que os
sinais obtidos da língua eletrónica tinham informação sobre os valores de pH e
condutividade e, por isso, estabeleceram-‐se modelos de regressão linear múltipla para
prever resultados de pH e condutividade nas amostras. Como a regressão linear
múltipla é sensível à multicolinearidade dos sinais dos sensores também, foi necessário
selecionar o melhor modelo através da seleção das variáveis independentes usando o
algoritmo arrefecimento simulado. Os modelos foram testados na robustez de previsão
através da validação cruzada com a técnica “leave-‐one-‐out”.
Os melhores modelos de previsão de pH e condutividade incluíam 27 sensores
(13 sensores do primeiro sistema e 14 do segundo sistema) e 28 sensores (14 sensores
do primeiro sistema e 14 do segundo sistema), respetivamente, mostrando que a
informação estava presente nos perfis dos sinais. Verificou-‐se que na relação linear
entre os valores previstos pelo modelo selecionado e os valores experimentais, quer
para a variável pH quer para a condutividade, obtiveram-‐se valores de declive e
ordenada na origem que, estatisticamente, podem ser considerados os teóricos (declive
igual a 1 e ordenada na origem igual a zero).
Os resultados mostraram que a língua eletrónica pode ser uma ferramenta
analítica prática na análise de águas minerais com e sem sabores, ao nível do controlo
analítico de classificação ou descriminação das amostras, bem como, da análise de pH e
condutividade.
The objective of this study was to apply a potentiometric electronic tongue in the analysis of mineral waters with and without flavors acquired in various commercial areas, to ascertain whether the information obtained allowed qualitative and quantitative analysis. Mineral waters without flavors, natural or spring water (still and sparkling), are commercially available with a large variety of flavors associated with the diversity of physical and chemical composition. Mineral waters with flavors are allowed to have added ingredients (ex., dyes, preservatives, flavorings, sweeteners, acidulants, antioxidants, etc.). The potentiometric electronic tongue used in this work included a reference electrode Ag/AgCl with a double-‐junction and two multi-‐sensor systems, connected to a datalogger for signal acquisition. Each multi-‐sensor system consisted of 20 different lipid membranes with cross-‐sensitivity prepared with various lipid compounds (3.0%), plasticizers (65.0%) and PVC polymer (32.0%). The second system was a replica of the first multi-‐sensor system. The physico-‐chemical characterization of the 34 samples acquired involved analysis of pH and conductivity in all samples and HPLC analysis in the samples of flavors (16 samples) to determine contents in organic acids (citric, malic and ascorbic acids) and sugars (glucose, fructose and sucrose). As a first study on the data obtained it was found that the samples could be grouped into 4 groups according to information on labels: mineral waters and sparkling water and water with flavors with or without gas. The analysis of variance showed there were significant differences between the four groups for the variables of physico-‐chemical characterization of the samples. The principal component analysis using the chemical parameters obtained by HPLC showed that the analytical data naturally separated the samples of flavored water with and without gas. In a second step, the signal profiles obtained from the electronic tongue were used, within the qualitative analysis, and principal component analysis was applied to verify the spatial distribution of the samples and the interdependence with certain physico-‐chemical parameters; subsequently, linear discriminant analysis coupled to simulated annealing algorithm for selection of variables was used to obtain the best model to discriminate the 4 groups defined. The performance of the model at the precision level was evaluated using cross-‐validation "leave-‐one-‐out". It was found that with only one discriminant function (explain 99.46% of data total variance), having 21 sensors (12 sensors of the first system and 12 of the second system), allowed to get 100% of correct classification for cross-‐validation. This study also allowed to verify that signals obtained from the electronic tongue had information about pH and conductivity values and, therefore, models were established using multiple linear regression to predict pH and conductivity values in the samples. As multiple linear regression is sensitive to multicollinearity of sensor signals, it was also was necessary to select the best model through the selection of independent variables using the simulated annealing algorithm. The models were tested in the robustness of prediction by cross-‐validation with the "leave-‐one-‐out" technique. The best models for predicting pH and conductivity sensors included 27 (12 sensors of the first system) and 28 (14 sensors of the first system), respectively, showing that the information was present in the profiles of the signals sensors. It was found that the linear relation between the values provided by the selected model and the experimental values, both for the variables pH and conductivity, provided slope and intercept values that may statistically be regarded as the theoretical (slope equal to 1 and intercept equal to zero). The results showed that the electronic tongue can be a practical analytical tool in the analysis of mineral water with or without flavors in the analytical control of samples by classification or discrimination sample, as well as the analysis of pH and conductivity of the samples.
The objective of this study was to apply a potentiometric electronic tongue in the analysis of mineral waters with and without flavors acquired in various commercial areas, to ascertain whether the information obtained allowed qualitative and quantitative analysis. Mineral waters without flavors, natural or spring water (still and sparkling), are commercially available with a large variety of flavors associated with the diversity of physical and chemical composition. Mineral waters with flavors are allowed to have added ingredients (ex., dyes, preservatives, flavorings, sweeteners, acidulants, antioxidants, etc.). The potentiometric electronic tongue used in this work included a reference electrode Ag/AgCl with a double-‐junction and two multi-‐sensor systems, connected to a datalogger for signal acquisition. Each multi-‐sensor system consisted of 20 different lipid membranes with cross-‐sensitivity prepared with various lipid compounds (3.0%), plasticizers (65.0%) and PVC polymer (32.0%). The second system was a replica of the first multi-‐sensor system. The physico-‐chemical characterization of the 34 samples acquired involved analysis of pH and conductivity in all samples and HPLC analysis in the samples of flavors (16 samples) to determine contents in organic acids (citric, malic and ascorbic acids) and sugars (glucose, fructose and sucrose). As a first study on the data obtained it was found that the samples could be grouped into 4 groups according to information on labels: mineral waters and sparkling water and water with flavors with or without gas. The analysis of variance showed there were significant differences between the four groups for the variables of physico-‐chemical characterization of the samples. The principal component analysis using the chemical parameters obtained by HPLC showed that the analytical data naturally separated the samples of flavored water with and without gas. In a second step, the signal profiles obtained from the electronic tongue were used, within the qualitative analysis, and principal component analysis was applied to verify the spatial distribution of the samples and the interdependence with certain physico-‐chemical parameters; subsequently, linear discriminant analysis coupled to simulated annealing algorithm for selection of variables was used to obtain the best model to discriminate the 4 groups defined. The performance of the model at the precision level was evaluated using cross-‐validation "leave-‐one-‐out". It was found that with only one discriminant function (explain 99.46% of data total variance), having 21 sensors (12 sensors of the first system and 12 of the second system), allowed to get 100% of correct classification for cross-‐validation. This study also allowed to verify that signals obtained from the electronic tongue had information about pH and conductivity values and, therefore, models were established using multiple linear regression to predict pH and conductivity values in the samples. As multiple linear regression is sensitive to multicollinearity of sensor signals, it was also was necessary to select the best model through the selection of independent variables using the simulated annealing algorithm. The models were tested in the robustness of prediction by cross-‐validation with the "leave-‐one-‐out" technique. The best models for predicting pH and conductivity sensors included 27 (12 sensors of the first system) and 28 (14 sensors of the first system), respectively, showing that the information was present in the profiles of the signals sensors. It was found that the linear relation between the values provided by the selected model and the experimental values, both for the variables pH and conductivity, provided slope and intercept values that may statistically be regarded as the theoretical (slope equal to 1 and intercept equal to zero). The results showed that the electronic tongue can be a practical analytical tool in the analysis of mineral water with or without flavors in the analytical control of samples by classification or discrimination sample, as well as the analysis of pH and conductivity of the samples.
Description
Keywords
Língua eletrónica Águas minerais com e sem sabor Análise de componentes principais Análise discriminante linear Regressão linear múltipla Algoritmo de arrefecimento simulado