Análise de dados de rochas de cromita visando a melhoria de um algoritmo de classificação baseada em sensores

Abstract

The sensor-based sorting technology has been used to separate chromite lump (chromite ore) from low grade rocks for about seven years in a mining company at the northeast of Brazil. This technology has improved the quality and the productivity, replacing a hand-picking operation for an automatic sorting equipment. In a tentative to better understand the correlation of the Cr2O3 grade and the sensors readings, one thousand and eight hundred rocks have been individually scanned and assayed, on a “Rock by Rock” methodology. This study approaches a data analysis of the x-ray and laser readings, together with the chemical assays, to better understand the behavior and influence of the chromite content on the sensor’s responses. The actual algorithm is compared also with a neural network, indicating a possible improvement on the classification using supervised learning.