PROCESS USING ARTIFICIAL NEURAL NETWORK FOR PREDICTIVE CONTROL IN SINTER MACHINE

摘要

The sinter production in compliance with the standards is of a fundamentally economical importance for the steel industry because the blast furnace productivity depends on it and, consequently, for the whole plant productivity. Although the improvements of the sinterings have significant savings of great economical and ecological importance, such as the use of the mine rejects and the viabilization of the mines whose ores tend to produce great numbers of strips in their processes of milling and stonebreaking, the thermodynamics of the sinter process require the pellet layers to be sintered have the level kept within strict limits, which, if not obeyed, causes stops of slow recuperation and material non-compliance, implying in reprocessing and a series of productivity losses. The big problem of the State of the Art that this patent comes to advance is that the traditional controls of the hopper (5) level, the sinter machine (6) feeder, has a response time of about 250 seconds, whfch is too long for a continuous and safe operation. The 'PROCESS USING ARTIFICIAL NEURAL NETWORK FOR PREDICTIVE CONTROL IN SINTER MACHINE', object of this patent, has a specific software as its neuro-fuzzy artificial intelligence core supported by preferably the tools MATLAB and ADALINE, being able, however, to use countless other tools and platforms of the ANN, as the ANN is trained to predict the filling level of the hopper (5) 250 seconds or more ahead, for the case of its specific application. The Artificial Neural Network was trained with the pieces of information of the process such as the weight of the materials (10) fed by the feeder silos of the pellets (2), the material density (11 ), the volume of the production by time unit (12), which, as they are sent to the specific software, allow the control of the system with an advance of 250 seconds, or more, and this specific software (9) provides the interfaces (13) to the control panels and it relates with the database to allow a continuous learning, since the ANN can operate with values of variables that have not been provided to it during its training process.