Tese: Artificial Intelligence Methods Applied to Mechanical Engineering Problems
Aluno(a) : Pedro Henrique Leite da Silva Pires DominguesOrientador(a): Helon Ayala
Área de Concentração: Mecânica Aplicada
Data: 13/04/2020
Link para tese/dissertação: http://doi.org/10.17771/PUCRio.acad.48474
Resumo: The majority real-world engineering design, management and decision making problems, may be seen as a multi-objective optimization problem, while others may comprehend classification, prediction or regression tasks. The use of artificial intelligence (AI) techniques to address these problems is attractive since they i) are most recommended for MO problems when compared with classical mathematical programming methods, which requires significant computational cost and informations of the problem domain; and ii) present better results with a simpler structure, adaptability and the possibility to extract information from the problems, when compared with other methods. Therefore, this work considers the application of AI techniques in anti-lock braking system (ABS) improvement, heat exchanger design optimization and theft-sensitive leak detection system (LDS) problems to i) develop and improve AI techniques; ii) constitute an AI application guide for control and design MO problems, as well as for the feature extraction process and machine learning classifier training through supervised learning approach; and mainly iii) demonstrate AI methods potential in solving realworld mechanical engineering problems. The results showed that the novel AI optimization algorithm versions proposed performed well considering the Pareto front dominance, spacing, Euclidian distance and hypervolume for the ABS problem and two IGD-based metrics compared through wilcoxon rank sum test for the HE design problem. Also, the decision tree-based machine learning classifier, trained through supervised learning and principal component analysis feature extraction technique, detcted all fuel theft situations with no false alarms.