Dissertation defense (September 19, 2024): Danielle Rodrigues Pinna
Student: Danielle Rodrigues Pinna
Title: Identificação de Falhas em Turbinas Eólicas: Uma Abordagem de Aprendizado de Máquina Centrada em Dados
Advisors: Diego Nunes Brandão and Rodrigo Franco Toso
Committee: Diego Nunes Brandão (PPCIC), Rodrigo Franco Toso (Microsoft), Rafaelli de Carvalho Coutinho (CEFET/RJ), Gustavo Silva Semaan (UFF), Ângela Ferreira (IPB/Portugal)
Day/Time: September 19, 2024 / 11 a.m.
Abstract: The last few years have been marked by the transition of the world energy matrix, predominantly with wind and solar sources, which are considered clean energies. Wind turbines, responsible for the energy conversion process, are complex and expensive equipment, susceptible to various failures due to multiple operational and environmental factors. Continuous monitoring of turbine components is essential for early fault detection, which can significantly reduce maintenance costs and increase operational efficiency. This work uses a data-centric approach to apply and compare machine learning techniques for fault detection in wind turbines. The research emphasizes the importance of data preprocessing, highlighting techniques such as class balancing, data partitioning, and attribute selection. Additionally, different machine learning algorithms are compared, focusing on hyperparameter optimization. The results demonstrate that adequate data preprocessing is crucial for the performance of machine learning models. The importance of computational time in optimizing hyperparameters and selecting the most appropriate algorithm for the specific problem context is also highlighted.
