Avaliação de parâmetros da qualidade do carvão vegetal, utilizando de espectroscopia no infravermelho próximo e de aprendizado de máquina

Abstract

Sectors that incorporate the use of charcoal in their industrial processes must invest in the characterization of this material, as the search for agile and precise methods, especially in the factory floor environment, intensifies itself. Among the technologies applied, it stands out near-infrared spectroscopy (NIR) and machine learning. In this context, the objective of this study was to develop multivariate and machine learning models to estimate the immediate chemical properties of charcoal, which include fixed carbon content, volatile matter, ash and moisture. To build the NIR model, they were used charcoal samples from three clones belonging to the Eucalyptus genus, produced in the UFMG oven -furnace system and from private companies. They were collected spectra on the transverse and tangential surfaces of the charcoal pieces and the we determined values of the characteristics using conventional laboratory techniques. From this data, models were developed, using both the Partial Least Squares (PLS) technique and Support Vector regression (SVR). During the validation stage, the models demonstrated satisfactory adjustments, with emphasis on the tangential surface. For the moisture, volatile matter and fixed carbon parameters, the PLS models achieved R² values of 89.52%, 88.8% and 94.09%, respectively, in addition to presenting low RMSE values, high RPD and an appropriate number of latent variables. The SVR models also obtained impressive results on the tangential surface, with R² of 95.57% for moisture, 90.48% for volatile matter and 95.10% for fixed carbon. However, it is important to highlight that the ash models did not reach significance, with a R² of less than 1%. The application of the near-infrared spectroscopy technique, especially in conjunction with machine learning, has proven to be highly promising. Furthermore, the tangential surface of charcoal stood out as the most suitable for determining the contents of volatile material, fixed carbon and moisture in energy production environments on the factory floor.