Classificação de carvão comercial para uso doméstico por espectroscopia no infravermelho próximo

• Autor(a) principal: Costa, Anna Carolina Pereira Rezende
• Data de Publicação: 2017
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFLA
• Texto Completo: http://repositorio.ufla.br/jspui/handle/1/28160
• topic: Carvão vegetal – Qualidade; Análise espectral – Métodos estatísticos; Análise multivariada; Radiação infravermelha; Charcoal – Quality; Spectrum analysis – Statistical methods; Multivariate analysis; Infrared radiation; Carvão

Charcoal for domestic use presents great variation in quality because in most furnaces the carbonization process is difficult to control and often inadequate raw material is used, generating a product with heterogeneous and dubious quality. The state of São Paulo created the Premium Seal in 2003 to certify the charcoal marketed in the state and assigned minimum reference values for such certification. Given the need for fast and reliable methods to classify charcoal for domestic use an alternative is near infrared (NIR) spectroscopy. Thus, the aim of this study was to apply NIR spectroscopy coupled with multivariate statistics in order to classify into categories and estimate the quality of commercial vegetable charcoal for domestic use. 76 charcoal samples from nine suppliers were selected for NIR spectra acquisition on the transverse face (TV) and rolling surface (SR) using the integration sphere and optical fiber from raw (untreated) and sanded charcoal and for chemical characterization by method of the Immediate Chemical Analysis (AQI). Analysis of variance was performed in the results of the AQI to assign quality classes to the coal as a function of the fixed carbon content (TCF). Principal Component Analysis (PCA) was done from the spectra of the charcoals to verify if it is possible to group samples with similar chemical characteristics. Subsequently, Discriminant Analysis based on Partial Least Squares (PLS-DA) was carried out to classify the charcoals according to their suppliers and to predict their quality through the TCF. The charcoal specimens had an average moisture content of 5.53%, volatile material content (TMV) of 23.84% and TCF of 76.14%. The PCA made from the spectra of the raw and sanded charcoals was not able to distinguish groups in this data set indicating the heterogeneity of the samples. PLS-DA resulted in a high percentage of correct classifications for both suppliers and quality classes. From the spectra obtained by the integrating sphere, the model generated to classify by supplier yielded 95.45% for SR signals and 100% on TV surface. From the spectra via fiber optics, the model hit 96.05% for SR and 97.37% for TV of sanded charcoal. For the quality of the charcoal using the optical fiber probe the models correctly classified 98.68% and 100% for SR and TV, respectively. For the independent validation the SR model was the best one, which correctly classified 78.95% of the samples of the external set. It was possible to conclude that the NIR associated to the multivariate statistics presented potential to be an efficient and rapid technique to classify charcoal. None of the suppliers analyzed is within the minimum reference values of the Premium Seal. The PCA was not efficient to exploit the data of this specimens set and the PLS-DA resulted in reliable models and that can be applied in unknown samples of charcoal.