Classificação de padrões de escoamento bifásico líquido-gás baseada em séries de Fourier e técnicas de aprendizagem de máquina

Abstract

The goal of this academic work is to explore the application of the Fourier transform on a signal obtained by the Wire-Mesh sensor and to use the Random Forest method for the classification of liquid-gas two-phase flow patterns in the context of the petroleum industry. The study aims to investigate the effectiveness of this approach as a tool for classifying flow patterns in two-phase petroleum systems. The choice of this technique is based on the Fourier transform’s ability to describe characteristics in the frequency domain of the signal, facilitating the categorization of obtained frequencies. The Random Forest algorithm is known for its robust classification capability and adaptability, particularly in differentiating similar data due to the branches formed by the method. The methodology includes signal processing, graph analysis, and evaluation of final results. The study reveals that the feature importance method, specifically Mean Decrease in Accuracy (MDA), achieved the best result, obtaining an accuracy of 81% with 380 relevant features of the model. These results indicate a promising combination of techniques for classification in the petroleum industry, providing an effective alternative for the analysis and categorization of data. Finally, this work aims to highlight the potential of this classification, contributing to the advancement of knowledge and technology application in the production, supply, and manipulation of petroleum fluid, promoting the use of these and related tools for control and monitoring of associated industrial processes.