USO DE ESPECTROSCOPIA ÓPTICA E APRENDIZADO DE MÁQUINA PARA DIFERENCIAÇÃO ENTRE CLASSES DE RESISTÊNCIA À ABRASÃO SUPERFICIAL EM PLACAS CERÂMICAS ESMALTADAS

Abstract

The identification of wood species is usually done based on human sensory analysis, observing its color, smell, texture, quantity and pore distribution as some of the factors that form the identity of each wood species. In this aspect, researches aimed at the association of Machine Learning (ML) techniques, together with the Fourier Transform Infrared Spectroscopy (FTIR) technique to discriminate different types of materials, which solve the difficulties of traditional methods and present advantages for implementation, such as speed, cost, has been growing in recent years. In this work we associate the FTIR and ML technique to solve the problems found in the incorrect identification of commercial wood, optimizing the speed of analysis and sample preparation in relation to the traditional method. Infrared absorption spectra of five wood species: Hymenolobium petraeum Ducke, Angelim-stone (ANG); Gochnatia polymorpha, Cambara (CAM); Erisma uncinatum, Cedrinho (CED)Dipteryx odorata, Champagne (CHA); Goupia glabra Aubl, Northern Peroba (PER) were used as in multivariate analysis using ML. The results showed that the FTIR technique together with multivariate analysis were able to differentiate the five wood species with 100% sensitivity and specificity. The developed method makes it possible to verify the potential of the technique associated with ML so that industries, laboratories, companies and/or control bodies can identify the nature of the product after extracted and semimanufactured.