Estudo da identificação automática do cristal de C3S em imagens microscópicas de clínquer e sua relevância para o monitoramento do funcionamento de fornos rotativos em plantas de cimento

Abstract

According to the Brazilian Portland Cement Association, cement can be defined as a fine powder with binding properties. In turn, clinker, the main component of Portland cement, can be defined as a granulated product, obtained by burning a suitable mixture of limestone and clay, called flour, in a rotary kiln, whose internal temperature reaches 1450°C. This work presents the use of reflected light microscopy for identification and grouping according to idiomorphism of alite crystals (C3S) in the microscopic images of clinker. Through this, propose manual groupings of crystals, to be used in the training of Mask Region based Convolutional Neural Network (MRCNN), a machine learning algorithm, which has, by intuition, to identify crystals automatically. As expected, in the analyzed images, the MRCNN managed to identify only of subdiomorphic crystals. This is due to the large amount of this type of crystal present in the analyzed images. With regard to idiomorphic and xenomorphic, due to their characteristics specials, mainly in terms of form, combined with a low number of images, it was not possible to carry out a proper training of the algorithm so that it would be able to identify. Another point was the non-detection of crystals in images with large amounts of crystals, due to low number of images. Therefore, the automation of the detection process of crystals, even with the low number of images available, it proved to be a viable alternative, since it tends to reduce and greatly facilitate human work, since after the identification of the crystals, it is possible to do the counting, the measurement of the crystals using other specific algorithms, if trained with a good number of images.