Per: Giulio Antunes de Medeiros (COMpanhia siderúrgica nacional), leonardo martins da silva (universidade federal fluminense), José Adilson de Castro (universidade federal fluminense)
Abstract:
Most of the world's production of metallic iron is used for steel production, however, among the so-called conventional processes, units such as coke ovens, sintering plants and, above all, blast furnaces are considered highly polluting, with high levels of emission of substances such as carbon dioxide (CO2), plychlorinated dibenzodioxins and polychlorinated dibenzofurans (PCDD/F), NOx and SOx, moving in the opposite direction of resolutions such as the Kyoto Protocol (1997), a complement to the United Nations Framework Convention on Climate Change (UNFCCC) that define the reduction in the emission of greenhouse gases. Appearing as an alternative for sustainable production in several processes, the use of self-reducing pellets is strongly based on eliminating the need for an external reducing agent, including it directly in the agglomerate and allowing a reduction in coke consumption when replacing it with cleaner more efficient alternatives such as biomasses. In the present work, the global kinetics of reduction for self-reducing pellets containing biomass as a carbon source was analyzed through the construction of multilayer perceptron neural network models and experimental kinetic tests used as a reference. For this, neural networks built from 100 or 1000 fixed neurons were developed, as well as networks with variable numbers of neurons according to an algorithm, which had between 3 and 100 neurons, or between 10 and 1000 neurons. The prediction of the reduced fraction through neural networks showed considerable performance for predicting the kinetic tests for time inputs of up to 12 minutes at all analyzed temperature levels. The fixed network of 1000 neurons showed the best accuracy in predicting the extent of reduction.
