MULTIVARIATE MODELS BASED ON SPECTRAL DATA FOR THE CLASSIFICATION OF CHARCOALS PRODUCED AT DIFFERENT FINAL CARBONIZATION TEMPERATURES

Abstract

Background: The prediction of apparent relative density (ARD) based on spectral data obtained in the near-infrared (NIR) region has not achieved high accuracy in the literature, emphasizing the need for alternatives to enable the use of NIR technology for classifying charcoal by ARD. In this context, the present study aimed to investigate the precision of NIR in estimating ARD classes for charcoal produced at different carbonization temperatures. Wood wastes from six tropical species (Dinizia excelsa, Licania sp., Brosimum gaudichaudii, Caryocar sp., Simaba guianensis, and Parkia sp.) were carbonized at four final temperatures (400, 500, 600, and 700°C) under laboratory conditions. The spectral data collected from the radial and transverse surfaces of the charcoals were analyzed using Principal Component Analysis (PCA) and Partial Least Squares Discriminant Analysis (PLS-DA).

Results: While PCA was ineffective in distinguishing the ARD classes, PLS-DA demonstrated relevant accuracy for classification. The quality of the PLS-DA models varied depending on the final carbonization temperature and the surface from which the spectra were collected. Spectral data from the radial and transverse surfaces showed high accuracy in classifying charcoals (>70%) up to 500°C for ARD. A global model with data from temperatures of 400–500°C achieved accuracy rates of 74% (transverse) and 80% (radial).

Conclusion: Using NIR technology for ARD classification represents significant progress for the energy sector, especially in the pursuit of bio-reducers with suitable quality for industrial applications.