Investigação das Varíaveis que Impactam a Previsão do Diferencial de Pressão em um Moinho de Cimento

Abstract

Cement, widely used in construction, has its origins in ancient civilizations and has evolved over the centuries, culminating in the development of Portland cement during the Industrial Revolution. In Brazil, Portland cement is the most produced and consumed type. According to data from the Statistical Yearbook of Brazil, the production of Portland cement between 2020 and 2022 reached approximately 114 million tons. The cement mill is a fundamental piece of equipment in the cement industry, responsible for processing raw materials to obtain the final product. The mill’s pressure differential is a critical monitoring parameter, since its increase may indicate instability in the equipment, resulting in an atypical increase in the vibration of its structure. In the industrial context, this instability compromises the quality of the cement, reduces the equipment’s feed and leads to higher energy costs, in addition to negatively impacting productivity. In view of this scenario, this study analyzes the operational variables associated with the pressure differential in a vertical roller mill, using a historical database from a manufacturing unit. Statistical techniques were applied for preprocessing and smoothing the data, including time lag analysis using lag and moving average, in addition to Pearson, Spearman and Mutual Information correlation studies. Subsequently, machine learning was used, with the Random Forest Regressor model, to predict the behavior of the pressure differential. The results indicate that the Pearson and Spearman correlations corroborated the relevance of the analyzed variables, while the Mutual Information showed more complex interactions and informational redundancies, suggesting the possibility of simplifying the predictive model. The modeling showed predictive potential, but the model’s performance metrics indicate the need for additional adjustments to improve the accuracy of the predictions.