Mathematical Modelling to Control the Chemical Composition of Blast Furnace Slag Using Artificial Neural Networks and Empirical Correlation
Abstract Portland cement additions have been used for many years with the main objective of reducing the amount of clinker. Among the additions, blast furnace slag, resulting from the production of pig iron, that is, reusing this by-product, reduces the emission of carbon dioxide as well as decreases the exploitation of natural limestone and clay reserves, which are raw materials for Portland clinker. In order to reduce these emissions and increase the availability of raw materials, research has been directed to study clinker-free binders, as is the case with activated alkali cements and supersulfated cements. In this way, alkali-activated cements can only involve the reuse of industry by-products and do not require the calcination of the raw material, thus reducing the emission of polluting gases into the atmosphere. Supersulfated cement are composed of up to 90% blast furnace slag, in addition to 10 to 20% calcium sulfate. One of the most important characteristics of blast furnace slag is the ratio of the content of CaO and SiO2, also known as the simplified basicity index (B2). This paper proposes the mathematical modeling of an artificial neural network to predict the final chemical composition of the blast furnace slag to be produced based on the operational parameters of the blast furnace aiming its use in the production of special cements such as alkali-activated cements and supersulfated cements. The high values of (R) associated with low values (RMSE) show the good statistical performance of ANN demonstrating that the mathematical model is efficient to carry out the forecast of the production of blast furnace slag.