Estimativa do teor de extrativos da madeira por espectroscopia no infravermelho próximo

Abstract

Wood is an abundant and renewable material and has been used as raw material for many industries due to its properties. However, chemical characterization of this material is usually carried out using time-consuming and costly processes. For these reason, the search for faster, technically and economically viable solutions are required. Near infrared spectroscopy (NIR) is a technique that has been applied in different areas for in-line monitoring of several characteristics from organic materials. This is a fast method, that requires minimal sample preparation and combined with multivariate statistics, can obtain accurate and fast results (30 seconds). Thus, this study aimed to associate the NIR spectroscopy and multivariate analysis in order to develop mathematical models capable of monitoring the total extractive content in wood, based on spectral signatures of solid and powder wood. Then, wood samples from native forest (Cedrela sp., Jacaranda sp., Apuleia sp. and Aspidosperma sp.) and clones of Eucalyptus hybrids from commercial plantations were analyzed. The spectra were obtained from radial surface of solid wood and, in the powder wood, after processing the material, using two methods of spectral acquisition: integration sphere and optical fiber. The chemical analyzes were performed by standardized method to quantify the total extractive content of wood, using an extraction sequence: toluene/ethanol (2:1), pure ethanol and hot water. The principal component analysis (PCA) was performed in the spectra in order to cluster the different species or woods group. The partial least squares regression (PLS-R) was performed to fit models, correlating the spectral signatures from the NIR with the total extractives values determined in laboratory. By using the PCA, the best samples separation was obtained from the solid wood spectra measured by integration sphere. The global PLS -R models were also developed based on the spectra obtained by integration sphere in solid wood (R 2 cv = 0.87 and RMSECV = 1.08%) and powder wood (R 2 cv = 0.85 and RMSECV = 1.19%) presented satisfactory statistics for extractive content prediction. The independent validation of models was performed with 25% of the samples and presented R 2 p = 0.93 and RMSEP = 0.95% (for solid wood) and R 2 p = 0.87 and RMSEP = 1.40% (for wood powder). The results showed the integration sphere is the most indicated spectral acquisition method in NIR to estimate the wood extractive content. A small difference in the statistics of the models generated with spectra of the solid and powder wood was observed, but both results were satisfactory, and from the operational view, the models based in solid wood spectra are more interesting.