Validação de modelos baseados em espectroscopia no infravermelho próximo para estimativa da densidade básica da madeira

• Autor(a) principal: Costa, Emylle Veloso Santos
• Data de Publicação: 2016
• Tipo de documento: Dissertação
• Idioma: por
• Título da fonte: Repositório Institucional da UFLA
• Texto Completo: http://repositorio.ufla.br/jspui/handle/1/11825
• topic: Madeira – Densidade; Análise espectral; Radiação infravermelha; Análise de regressão; Wood – Density; Spectrum analysis; Infrared radiation; Regression analysis; Propriedades Físico-Mecânicas da Madeira

The basic density of the wood is important for characterizing the material as it is directly related to its quality. In addition to the classical method, one of the quickest ways to determine it is by Near Infrared Spectroscopy (NIR). By having several advantages, NIRS technique has been used to determine various properties of wood. However, even with promising results, it needs to be further investigated to assess the robustness of its estimates. Thus, the aim of this study was to develop regression models from NIR spectra to estimate the basic density of wood and verify its robustness through cross and independent validations. Initially, the NIR spectra was obtained by an integrating sphere and optical fiber in the tangential and radial faces, besides the transversal faces produced by circular saw and band saw. Then determined the basic wood density by the conventional method in 278 wood samples of Eucalyptus sp. clones, aged 6 to 6,5 years old. The basic density values were correlated with the NIR spectra by means of partial least squares regression. Finally, the model calibrated on the basis of different types of spectra was validated and applied in different databases. The best model to estimate the wood density was generated from spectra obtained in the transversal face produced by band saw both the integrating sphere via (R²cv = 0.88 and RPD = 2.97) and via optical fiber (R²cv RPD = 0.74 and = 2.01). The application of the models obtained by optical fiber on the basis of spectral data recorded by integrating sphere presented the best estimates in the three faces of the samples, and the transversal face produced by the band saw that got better fit to the data (R² = 0.80). Thus, NIR spectroscopy is effective for estimating the wood density spectra measured using the three faces of the timber using the two acquisition pathways. In general, precise estimates are generated when spectra measured via acquisition optical fiber are applied on models calibrated from the spectra recorded by integrating sphere, but otherwise did not occur. The integrating sphere and the transversal or radial face are the most suitable way to measure NIR spectra and generate models to estimate the basic density of Eucalyptus.