bachelorThesis
Uso de algoritmos de aprendizado de máquina para predição e classificação de variáveis de interesse em processo industrial de abate de frango
Fecha
2021-05-14Registro en:
BORTOLETI, Gabriel Balieiro. Uso de algoritmos de aprendizado de máquina para predição e classificação de variáveis de interesse em processo industrial de abate de frango. 2021. Trabalho de Conclusão de Curso (Bacharel em Engenharia Química) – Universidade Tecnológica Federal do Paraná, Francisco Beltrão, 2021.
Autor
Bortoleti, Gabriel Balieiro
Resumen
In the industrial chicken slaughtering process, the most important and most required step is the cooling step, where the final water absorption content, the final temperature, as the physical characteristics of the meat and controls, the microbiological index of the product is defined. In this step, mass and energy transfer processes take place, in a transient regime with control volume without a fixed format and with different process variables that influence the final result, so the calculations to estimate the temperature, final water absorption in the carcass and classify which of the process variables are the most influential, end up being complicated and involve several approximations that end up making the final result not to be consistent with the real industrially processed. The focus of the study is to use artificial intelligence methods, more precisely, machine learning system to perform the prediction and classification of the variables involved in the chicken meat cooling process. The objective was to predict the water absorption values in the carcass at the end of the process, using data obtained from Sant'anna's thesis in 2008, applying three different algorithms, among them: KStar, Minimum Sequential Optimization for Regression (SMOReg) and Artificial Neural Networks. In total, 206 tests were carried out involving the three algorithms. For the validation of the results, the correlation coefficient (R) and the mean square error (MSE) were used. For the KStar algorithm, a prediction of the results and a classification of the results variables of 50.11% correlation and an ability to classify two of the three most important variables. For the SMOReg algorithm, a function that stands out with the best result for a Radial Base function (RBF) with 99.87% correlation and an ability to classify as three most important variables. For the Artificial Neural Networks algorithm, the best configuration of its structure obtained a result of 99.09% correlation and an ability to classify as three most important variables. Finally, analyzing the individual results of the correlation coefficient (R) and the ability to classify the variables of the three methods, the one that presents the best result was the SMOReg algorithm for both tasks.