Uso de redes neurais artificiais no controle de qualidade da gasolina – uma revisão sistemática da literatura

Abstract

This monograph is based on the main gaps and trends on the subject - Application of Artificial Neural Networks (ANNs) in gasoline quality control and aims to find ANN models used as an alternative for analyzing the quality of gasoline in the last 10 years. Thus, the main purpose of this research is to analyze the ANN methods that are being used for gasoline quality inspection through a systematic review of the literature. The research was conducted through a systematic review of the literature, using the IEEE, Scopus and ScienceDirect as search sources. Regarding the operationalization and extraction of articles, 3 search strings formed by the keywords were used: neural network, gasoline, octane number prediction, antiknock performance, density. The results found in the databases were restricted to publications in the time period of the last 10 years (2012-2022) in the English language, which were submitted to a prior analysis, and posteriori, selected for reading in full and introduced the criteria of exclusion and inclusion. Through the analysis of the articles that composed this systematic review of the literature, it was possible to subdivide them into two empirical categories, that is, research that used an existing ANN in the literature, and research that sought to introduce modifications in the ANN models present in the literature. Then, the articles were synthesized in order to explain the main characteristics of each model, in addition to highlighting the data used in the processing of the model and gasoline compounds predicted in the analysis through the use of ANNs. As conclusions, the research allowed identifying which ANN models and structure adaptations are most suitable for the analysis of gasoline quality, in addition to demonstrating that to obtain accurate results, ANNs need adequate training algorithms and structure modifications.